Defining the optimum strategy for identifying adults and children with coeliac disease: systematic review and economic modelling

Notes

Permissions

Chapter 1 Objectives

The overall aim of this project was to define at-risk groups and determine the cost-effectiveness of active case-finding strategies in primary care.

We defined the following six objectives to address this overall aim:

1. systematic review of the accuracy of potential diagnostic indicators for coeliac disease (CD) (see Chapter 3)

2. routine data analysis to develop a prediction model to identify people who should be tested for CD (see Chapter 4)

3. systematic review of the accuracy of diagnostic tests for CD (see Chapter 5)

4. systematic review of the accuracy of genetic tests for CD (see Chapter 6)

5. online survey to identify diagnostic thresholds for testing, starting treatment and referral for biopsy (see Chapter 7)

6. economic modelling to identify the cost-effectiveness of different active case-finding strategies, informed by the findings of previous objectives (see Chapter 8).

Chapter 2 Background

Chapter 3 Accuracy of diagnostic indicators for coeliac disease

Parts of this chapter have been reproduced from Elwenspoek et al. 20 This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) license, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: https://creativecommons.org/licenses/by/4.0/. The text below includes minor additions and formatting changes to the original text.

We conducted a systematic review to assess the accuracy of various symptoms and risk factors in ‘diagnosing’ CD, considering the potential of these as initial screening tools prior to serological testing. In this chapter, we will refer to these symptoms and risk factors as ‘diagnostic indicators’. We defined diagnostic indicators as signs, symptoms and risk factors that may help clinicians identify patients for whom further testing for CD is warranted. We did not consider factors that are difficult to determine at an initial consultation, such as perinatal risk factors or age at gluten introduction, or experimental factors that are not measured in clinical practice [i.e. tests for susceptibility genes; these are currently not widely available to clinicians and therefore are not (yet) useful in aiding diagnosis].

The review was registered with the international prospective register of systematic reviews (PROSPERO) under the registration number CRD42020170766. We published the protocol for the review, which predefined the objectives and methods for this review. 21 This review followed the recommendations from the Centre for Reviews and Dissemination22 and the Cochrane Handbook for Systematic Reviews of Diagnostic Test Accuracy Version 1.0.0,23 and is reported in accordance with the Preferred Reporting Items for a Systematic Review and Meta-analysis of Diagnostic Test Accuracy Studies (PRISMA-DTA) statement. 24

Results of assessment of diagnostic accuracy of indicators

We identified 12,027 records, after deduplication, through searching scientific databases and screening the reference lists of four recent guidelines on CD9–11,41,42 and 22 systematic reviews. Of these, 709 records were selected for full-text assessment, of which 241 studies fulfilled the inclusion criteria. These studies contained 369 reports of 90 distinct diagnostic indicators. Table S1 (see Report Supplementary Material 1) provides a list of diagnostic indicators (including study references) for which we did not extract data because they reported on a rare indicator (reported by fewer than five studies). We included 183 studies reporting on 25 distinct indicators in our meta-analysis (see Appendix 2, Figure 26).

Study characteristics

The included diagnostic indicators consisted of seven symptoms, 17 risk conditions and family history (see Appendix 3, Table 25, and Appendix 4, Tables 26–51). The symptoms that were most often investigated as diagnostic indicators were abdominal pain (n = 12) and diarrhoea (n = 12), while type 1 diabetes (n = 31) and thyroid disease (n = 23) were the most commonly studied risk conditions. Studies investigating symptoms associated with CD predominantly used a cohort or cross-sectional design, using a serological test to detect CD. Studies looking at risk conditions mainly used case–control designs, that is, people with the diagnostic indicator were compared with a healthy control group without the diagnostic indicator. Most studies included adult participants, although many diagnostic indicators were also studied in a population of children or a mixed population.

Although sample sizes for each meta-analysis ranged between 1004 and 55,500 participants, some meta-analyses were based on a small number of CD participants, as prevalence was often low. For instance, in the case of multiple sclerosis and systemic lupus erythematosus, estimates of sensitivity are based on only 12 and nine people with CD, respectively.

Risk of bias

Several studies reported more than one diagnostic indicator, resulting in 281 risk-of-bias judgements. Most studies had methodological issues, and only one study was judged as having a low overall risk of bias (see Appendix 5, Figure 27). In total, only 19 study reports were judged to be at low risk of bias regarding patient selection. The main source of potential bias in patient selection was the use of a case–control study design. The index test domain was judged to be at low risk of bias if it was the study’s main aim to investigate the diagnostic indicator of interest, which was the case for most studies. In total, 167 study reports on diagnostic indicators were judged to have a high risk of bias for the reference standard. This was mainly driven by studies using serology tests without biopsy confirmation to determine whether or not participants had CD; this risks misclassifying participants as having or not having CD. Flow and timing were judged to be at high risk of bias in studies that did not use the same combination of diagnostic tests for CD for all participants (reference standard), for example in studies in which a biopsy was performed only in participants who had a positive serology test result.

Accuracy of diagnostic indicators to detect coeliac disease

We found large variation in sensitivity, specificity and PPV estimates between studies for most diagnostic indicators (Figure 1) (see Appendix 6, Figures 28–51; see also table S2 and figure S1 in Report Supplementary Material 1). Estimates of sensitivity were particularly variable, often ranging from 0% to almost 100%, owing to very small numbers of CD participants for some indicators.

FIGURE 1.

Sensitivity, specificity and PPVs. Meta-analysis results are shown per diagnostic indicator. PPVs were calculated for a population with a CD prevalence of 1% (red dotted line) using the estimated sensitivities and specificities from the meta-analyses. Diagnostic indicators are ordered from high to low PPV per diagnostic indicator group. The area of the box size is proportional to the total number of participants. Negative predictive values are not shown because these diagnostic indicators should not be used to rule out CD. FN, false negative; FP, false positive; TN, true negative; TP, true positive.

The PPVs for the symptoms included in this review are similar to the baseline CD prevalence, with 95% CIs providing no evidence that the presence of any of these symptoms increases the chance that an individual has CD beyond the levels found in the general population (see Figure 1).

Figure S1 (see Report Supplementary Material 1) shows meta-analysis results in receiver operating characteristic (ROC) space. A diagnostic indicator with a SROC curve closely following the diagonal line is no better at predicting CD than a coin toss, which is approximately the case for all symptoms.

Of the risk conditions, dermatitis herpetiformis had the highest estimated sensitivity, specificity and PPV (the estimated PPV at 1% prevalence of CD was 29%, 95% CI 3% to 72%). However, the uncertainty around these estimates was substantial, that is the 95% CIs were very wide. In addition, dermatitis herpetiformis is a rare condition, so it will be a clinically useful diagnostic indicator in only a minority of cases. We estimated PPVs of > 2% for migraine, anaemia, type 1 diabetes, osteoporosis and chronic liver disease. These estimates were relatively precise for anaemia, type 1 diabetes, osteoporosis and chronic liver disease, but there was considerable uncertainty for migraine. People with thyroid disease, subfertility or recurrent pregnancy loss, or IBS were 1.5–2 times more likely to have CD than the general population, with 95% CIs lying entirely above the population prevalence of 1%. Although the estimated PPVs of psoriasis, epilepsy, inflammatory bowel disease, systemic lupus erythematosus, fracture, arthritis and type 2 diabetes suggest an increased likelihood of CD among people with these conditions, there was considerable uncertainty in these estimates. The 95% CIs crossed or touched the line of population prevalence, indicating that the likelihood of CD may be similar to that of the general population. We found no evidence of an increased likelihood of CD among people with multiple sclerosis (see Figure 1).

Similarly, arthritis, fracture and type 2 diabetes appear to have no diagnostic ability when judging sensitivity and specificity in ROC space (see figure S1 in Report Supplementary Material 1). For multiple sclerosis, systemic lupus erythematosus, psoriasis and inflammatory bowel disease, there was not enough evidence to estimate a reliable SROC curve. For chronic liver disease, epilepsy, migraine, IBS, and dermatitis herpetiformis, there was substantial uncertainty in summary estimates because of a high level of variation between the study estimates. The SROC plots for type 1 diabetes, anaemia, subfertility or recurrent pregnancy loss, thyroid disease and osteoporosis suggest a greater accuracy in predicting CD than a coin toss.

People with a family history of CD were 2.7% (95% CI 1.2 to 3.0%) more likely to have CD than the general population.

Subgroup and sensitivity analyses

There were sufficient data on five diagnostic indicators to stratify the meta-analyses by age group (see Appendix 7, Figure 52; see also table S3 in Report Supplementary Material 1). Estimated PPVs were similarly low, at around 1%, for abdominal pain, arthritis, constipation and diarrhoea for adults and children. The results suggest that arthritis may be more predictive of CD in children than in adults, and abdominal pain and constipation may be more predictive of CD in adults than in children. The PPV for type 1 diabetes appeared to be higher for adults [3.4% (95% CI 1.9% to 5.6%)] than for children or mixed populations [1.8% (95% CI 1.4% to 2.3%) and 2.1% (95% CI 1.6% to 2.9%), respectively]. However, each of these differences should be interpreted with caution because the CIs overlap.

There were sufficient data on seven diagnostic indicators to stratify the analysis on CD diagnosis, comparing studies that used a serology-only approach with studies that included a confirmation duodenal biopsy (see Appendix 7, Figure 52; see also table S3 in Report Supplementary Material 1). Estimated PPVs were similar between the subgroups.

A sensitivity analysis was performed restricted to studies using a cohort or cross-sectional design for abdominal pain, anaemia, bloating or abdominal distension, constipation and diarrhoea (see Appendix 8, Figure 53; see also table S4 in Report Supplementary Material 1). Although case–control studies are more prone to bias than cohort studies, removing case–control studies did not affect the sensitivity, specificity or PPV estimates among these diagnostic indicators. It was not possible to perform a sensitivity analysis restricted to studies with a low risk of bias, because all included studies were judged to be at an overall high risk of bias.

Finally, a post hoc sensitivity analysis was performed on the diagnostic indicator ‘family history of CD’, restricted to studies that included only first-degree relatives. This increased the estimated PPV from 2.7% (95% CI 1.2% to 3.9%) to 3.0% (95% CI 1.6% to 3.7%), although the CIs overlap.

Chapter 4 Prediction rule for coeliac disease diagnosis

This chapter describes the development and internal validation of diagnostic prediction models for men, women and children in a routinely collected primary care data set to estimate the probability of having CD. We also describe the development of a model for children in a birth cohort, with external validation in the primary care data set. The aim of each prediction model is to help clinicians in primary care decide whether or not a patient should be offered a serological test for CD based on their pre-existing conditions and current/recent symptoms. To demonstrate the potential clinical usefulness of each model, we present the PPVs and percentage of CD patients missed at different thresholds.

Results

Clinical Practice Research Datalink participants

Appendix 10, Figure 54, shows the patient flow diagrams for the CPRD development data set and the external validation data set. Cases and controls had an average follow-up time of 7 years prior to CD diagnosis [median 7 years, interquartile range (IQR) 3–11 years, range 1–31 years].

Table 2 describes the characteristics of the participants and the prevalence of selected indicators in the development data set and the external validation data set for children, women and men. In both data sets, the prevalence of CD was significantly higher among girls than boys in the cohort of children, with almost two-thirds of CD patients being girls. The median age and age ranges were similar between CD patients and controls in the cohort of children in the development data set; however, in the validation data set, children with CD were, on average, 1 year younger than controls. In the development data set for women, the median age was 49 years for both cases and controls. In the external validation data set, the median age was similar, at 47 years, although cases were, on average, 1 year younger than controls. In contrast to the data sets for children and women, in both data sets for men, cases were significantly older than controls, with a median age of 55 years for cases and 47 years for controls. On average, CD male patients were 8 and 7 years older than controls in the development and validation data sets, respectively.

TABLE 2 - Cohort of children, women and men: characteristics

NR, not reported.

Note

p-values show the result of a Welch two-sample t-test for continuous variables and of a Pearson’s chi-squared test with Yates’ continuity correction for categorical variables.

Cohort and characteristic	Development data set (CPRD GOLD)				External validation data set (CPRD Aurum)
	Control	CD	p-value	Overall	Control	CD	p-value	Overall
Children	(N = 12,948)	(N = 3237)		(N = 16,185)	(N = 28,131)	(N = 7033)		(N = 35,164)
Age (years)
Mean (SD)	9.04 (4.62)	8.89 (4.77)	0.127	9.01 (4.65)	8.96 (4.61)	8.66 (4.74)	< 0.001	8.90 (4.64)
Median (minimum, maximum)	9.00 (1.00, 17.0)	9.00 (1.00, 17.0)		9.00 (1.00, 17.0)	9.00 (1.00, 17.0)	8.00 (1.00, 17.0)		9.00 (1.00, 17.0)
Sex, n (%)
Male	6862 (53.0)	1249 (38.6)	< 0.001	8111 (50.1)	15024 (53.4)	2702 (38.4)	< 0.001	17,726 (50.4)
Female	6086 (47.0)	1988 (61.4)		8074 (49.9)	13107 (46.6)	4331 (61.6)		17,438 (49.6)
Ethnicity, n (%)
Non-white	438 (3.4)	98 (3.0)	0.004	536 (3.3)	686 (2.4)	156 (2.2)	< 0.001	842 (2.4)
White	3816 (29.5)	1205 (37.2)		5021 (31.0)	3478 (12.4)	1103 (15.7)		4581 (13.0)
Missing	8694 (67.1)	1934 (59.7)		10,628 (65.7)	23,967 (85.2)	5774 (82.1)		29,741 (84.6)
Deprivation (IMD 201567 quintiles), n (%)
1	1104 (8.5)	385 (11.9)	0.059	1489 (9.2)	1532 (5.4)	391 (5.6)	0.057	1923 (5.5)
2	896 (6.9)	270 (8.3)		1166 (7.2)	1154 (4.1)	310 (4.4)		1464 (4.2)
3	871 (6.7)	264 (8.2)		1135 (7.0)	1000 (3.6)	280 (4.0)		1280 (3.6)
4	742 (5.7)	219 (6.8)		961 (5.9)	836 (3.0)	188 (2.7)		1024 (2.9)
5	641 (5.0)	165 (5.1)		806 (5.0)	858 (3.1)	185 (2.6)		1043 (3.0)
Missing	8694 (67.1)	1934 (59.7)		10,628 (65.7)	22,751 (80.9)	5679 (80.7)		28,430 (80.8)
Anaemia, n (%)
Present	46 (0.4)	188 (5.8)	< 0.001	234 (1.4)	25 (0.1)	84 (1.2)	< 0.001	109 (0.3)
Arthritis, n (%)
Present	5 (0.0)	7 (0.2)	0.003	12 (0.1)	5 (0.0)	3 (0.0)	0.426	8 (0.0)
Delayed puberty, n (%)
Present	1 (0.0)	5 (0.2)	< 0.001	6 (0.0)	2 (0.0)	2 (0.0)	0.382	4 (0.0)
Down syndrome, n (%)
Present	6 (0.0)	18 (0.6)	< 0.001	24 (0.1)	2 (0.0)	15 (0.2)	< 0.001	17 (0.0)
Failure to thrive, n (%)
Present	52 (0.4)	84 (2.6)	< 0.001	136 (0.8)	21 (0.1)	31 (0.4)	< 0.001	52 (0.1)
Fatigue, n (%)
Present	210 (1.6)	253 (7.8)	< 0.001	463 (2.9)	165 (0.6)	150 (2.1)	< 0.001	315 (0.9)
Fatigue (count, 1 year), n (%)
Once	53 (0.4)	130 (4.0)	< 0.001	183 (1.1)	34 (0.1)	48 (0.7)	< 0.001	82 (0.2)
Twice	5 (0.0)	17 (0.5)		22 (0.1)	8 (0.0)	14 (0.2)		22 (0.1)
Three times	0 (0)	11 (0.3)		11 (0.1)	2 (0.0)	8 (0.1)		10 (0.0)
First-degree relative with CD, n (%)
Present	98 (0.8)	496 (15.3)	< 0.001	594 (3.7)	NR	NR		NR
GI symptoms, n (%)
Present	3684 (28.5)	1924 (59.4)	< 0.001	5608 (34.6)	1371 (4.9)	794 (11.3)	< 0.001	2165 (6.2)
GI symptoms (count, 1 year), n (%)
Once	785 (6.1)	588 (18.2)	< 0.001	1373 (8.5)	293 (1.0)	218 (3.1)	< 0.001	511 (1.5)
Twice	192 (1.5)	294 (9.1)		486 (3.0)	82 (0.3)	126 (1.8)		208 (0.6)
Three times	55 (0.4)	154 (4.8)		209 (1.3)	24 (0.1)	60 (0.9)		84 (0.2)
Four times	52 (0.4)	203 (6.3)		255 (1.6)	39 (0.1)	98 (1.4)		137 (0.4)
IBS, n (%)
Present	20 (0.2)	29 (0.9)	< 0.001	49 (0.3)	6 (0.0)	12 (0.2)	< 0.001	18 (0.1)
IgA deficiency, n (%)
Present	0 (0)	4 (0.1)	< 0.001	4 (0.0)	1 (0.0)	4 (0.1)	0.005	5 (0.0)
Iron, vitamin B12 or folate deficiency, n (%)
Present	8 (0.1)	35 (1.1)	< 0.001	43 (0.3)	8 (0.0)	24 (0.3)	< 0.001	32 (0.1)
Mood disorders, n (%)
Present	256 (2.0)	143 (4.4)	< 0.001	399 (2.5)	91 (0.3)	59 (0.8)	< 0.001	150 (0.4)
Type 1 diabetes, n (%)
Present	16 (0.1)	275 (8.5)	< 0.001	291 (1.8)	11 (0.0)	87 (1.2)	< 0.001	98 (0.3)
Thyroid disorders, n (%)
Present	16 (0.1)	61 (1.9)	< 0.001	77 (0.5)	12 (0.0)	21 (0.3)	< 0.001	33 (0.1)
Turner syndrome, n (%)
Present	0 (0)	8 (0.2)	< 0.001	8 (0.0)	1 (0.0)	5 (0.1)	< 0.001	6 (0.0)
Weight loss, n (%)
Present	23 (0.2)	93 (2.9)	< 0.001	116 (0.7)	8 (0.0)	22 (0.3)	< 0.001	30 (0.1)
Women	(N = 37,079)	(N = 12,051)		(N = 49,130)	(N = 77,422)	(N = 26,164)		(N = 103,586)
Age (years)
Mean (SD)	49.5 (17.0)	49.7 (17.2)	0.313	49.6 (17.0)	48.5 (17.0)	47.4 (17.6)	< 0.001	48.2 (17.2)
Median (minimum, maximum)	49.0 (18.0, 111)	49.0 (18.0, 104)		49.0 (18.0, 111)	47.0 (18.0, 108)	46.0 (18.0, 99.0)		47.0 (18.0, 108)
Ethnicity, n (%)
Non-white	807 (2.2)	212 (1.8)	< 0.001	1019 (2.1)	1262 (1.6)	399 (1.5)	< 0.001	1661 (1.6)
White	12,286 (33.1)	4612 (38.3)		16,898 (34.4)	11,210 (14.5)	4496 (17.2)		15,706 (15.2)
Missing	23,986 (64.7)	7227 (60.0)		31,213 (63.5)	64,950 (83.9)	21,269 (81.3)		86,219 (83.2)
Deprivation (IMD 2015 quintiles), n (%)
1	3199 (8.6)	1262 (10.5)	0.174	4461 (9.1)	4255 (5.5)	1424 (5.4)	0.276	5679 (5.5)
2	2939 (7.9)	1077 (8.9)		4016 (8.2)	3696 (4.8)	1254 (4.8)		4950 (4.8)
3	2761 (7.4)	1001 (8.3)		3762 (7.7)	2975 (3.8)	991 (3.8)		3966 (3.8)
4	2292 (6.2)	819 (6.8)		3111 (6.3)	2642 (3.4)	966 (3.7)		3608 (3.5)
5	1902 (5.1)	665 (5.5)		2567 (5.2)	2146 (2.8)	702 (2.7)		2848 (2.7)
Missing	23,986 (64.7)	7227 (60.0)		31,213 (63.5)	61,708 (79.7)	20,827 (79.6)		82,535 (79.7)
Anaemia, n (%)
Present	1038 (2.8)	1969 (16.3)	< 0.001	3007 (6.1)	484 (0.6)	968 (3.7)	< 0.001	1452 (1.4)
Cardiovascular disease, n (%)
Present	1601 (4.3)	883 (7.3)	< 0.001	2484 (5.1)	602 (0.8)	317 (1.2)	< 0.001	919 (0.9)
Chronic liver disease, n (%)
Present	376 (1.0)	253 (2.1)	< 0.001	629 (1.3)	233 (0.3)	180 (0.7)	< 0.001	413 (0.4)
Down syndrome, n (%)
Present	5 (0.0)	5 (0.0)	0.132	10 (0.0)	4 (0.0)	4 (0.0)	0.229	8 (0.0)
Epilepsy, n (%)
Present	194 (0.5)	116 (1.0)	< 0.001	310 (0.6)	109 (0.1)	46 (0.2)	0.24	155 (0.1)
Fatigue, n (%)
Present	4638 (12.5)	3027 (25.1)	< 0.001	7665 (15.6)	1755 (2.3)	1165 (4.5)	< 0.001	2920 (2.8)
Fatigue (count, 1 year), n (%)
Once	945 (2.5)	953 (7.9)	< 0.001	1898 (3.9)	333 (0.4)	341 (1.3)	< 0.001	674 (0.7)
Twice	145 (0.4)	174 (1.4)		319 (0.6)	86 (0.1)	118 (0.5)		204 (0.2)
Three times	38 (0.1)	73 (0.6)		111 (0.2)	56 (0.1)	79 (0.3)		135 (0.1)
First-degree relative with CD, n (%)
Present	108 (0.3)	416 (3.5)	< 0.001	524 (1.1)	NR	NR		NR
Fractures (count, 1 year), n (%)
Once	320 (0.9)	202 (1.7)	< 0.001	522 (1.1)	139 (0.2)	65 (0.2)	< 0.001	204 (0.2)
Twice	169 (0.5)	84 (0.7)		253 (0.5)	92 (0.1)	60 (0.2)		152 (0.1)
GI symptoms, n (%)
Present	12,364 (33.3)	6994 (58.0)	< 0.001	19,358 (39.4)	4520 (5.8)	2603 (9.9)	< 0.001	7123 (6.9)
GI symptoms (count, 1 year), n (%)
Once	2605 (7.0)	2206 (18.3)	< 0.001	4811 (9.8)	949 (1.2)	774 (3.0)	< 0.001	1723 (1.7)
Twice	773 (2.1)	996 (8.3)		1769 (3.6)	310 (0.4)	349 (1.3)		659 (0.6)
Three times	308 (0.8)	507 (4.2)		815 (1.7)	126 (0.2)	179 (0.7)		305 (0.3)
Four times	265 (0.7)	679 (5.6)		944 (1.9)	162 (0.2)	302 (1.2)		464 (0.4)
Inflammatory bowel disease, n (%)
Present	160 (0.4)	104 (0.9)	< 0.001	264 (0.5)	82 (0.1)	54 (0.2)	< 0.001	136 (0.1)
IBS, n (%)
Present	1716 (4.6)	1346 (11.2)	< 0.001	3062 (6.2)	757 (1.0)	545 (2.1)	< 0.001	1302 (1.3)
IgA deficiency, n (%)
Present	2 (0.0)	6 (0.0)	0.004	8 (0.0)	1 (0.0)	7 (0.0)	< 0.001	8 (0.0)
Iron, vitamin B12 or folate deficiency, n (%)
Present	505 (1.4)	938 (7.8)	< 0.001	1443 (2.9)	235 (0.3)	394 (1.5)	< 0.001	629 (0.6)
Mouth ulcers (count, 1 year), n (%)
Once	110 (0.3)	121 (1.0)	< 0.001	231 (0.5)	43 (0.1)	32 (0.1)	< 0.001	75 (0.1)
Twice	12 (0.0)	38 (0.3)		50 (0.1)	11 (0.0)	16 (0.1)		27 (0.0)
Neuropathy or ataxia, n (%)
Present	84 (0.2)	55 (0.5)	< 0.001	139 (0.3)	56 (0.1)	36 (0.1)	0.003	92 (0.1)
Osteoporosis, n (%)
Present	915 (2.5)	898 (7.5)	< 0.001	1813 (3.7)	367 (0.5)	305 (1.2)	< 0.001	672 (0.6)
Systemic lupus erythematosus, n (%)
Present	42 (0.1)	40 (0.3)	< 0.001	82 (0.2)	23 (0.0)	18 (0.1)	< 0.001	41 (0.0)
Type 1 diabetes, n (%)
Present	99 (0.3)	141 (1.2)	< 0.001	240 (0.5)	223 (0.3)	147 (0.6)	< 0.001	370 (0.4)
Thyroid disorders, n (%)
Present	2042 (5.5)	1442 (12.0)	< 0.001	3484 (7.1)	815 (1.1)	623 (2.4)	< 0.001	1438 (1.4)
Turner syndrome, n (%)
Present	3 (0.0)	5 (0.0)	0.037	8 (0.0)	0 (0)	5 (0.0)	< 0.001	5 (0.0)
Weight loss, n (%)
Present	500 (1.3)	672 (5.6)	< 0.001	1172 (2.4)	105 (0.1)	145 (0.6)	< 0.001	250 (0.2)
Men	(N = 35,264)	(N = 6035)		(N = 41,299)	(N = 76,775)	(N = 12,385)		(N = 89,160)
Age (years)
Mean (SD)	47.5 (16.4)	53.9 (16.3)	< 0.001	48.4 (16.6)	46.6 (16.5)	52.4 (17.1)	< 0.001	47.4 (16.7)
Median (minimum, maximum)	47.0 (18.0, 103)	55.0 (18.0, 94.0)		48.0 (18.0, 103)	46.0 (18.0, 107)	53.0 (18.0, 98.0)		47.0 (18.0, 107)
Ethnicity, n (%)
Non-white	535 (1.5)	94 (1.6)	0.022	629 (1.5)	976 (1.3)	155 (1.3)	< 0.001	1131 (1.3)
White	10,083 (28.6)	2315 (38.4)		12,398 (30.0)	8967 (11.7)	2100 (17.0)		11,067 (12.4)
Missing	24,646 (69.9)	3626 (60.1)		28,272 (68.5)	66,832 (87.0)	10,130 (81.8)		76,962 (86.3)
Deprivation (IMD 2015 quintiles), n (%)
1	2563 (7.3)	624 (10.3)	0.005	3187 (7.7)	4205 (5.5)	682 (5.5)	0.55	4887 (5.5)
2	2389 (6.8)	597 (9.9)		2986 (7.2)	3745 (4.9)	625 (5.0)		4370 (4.9)
3	2264 (6.4)	450 (7.5)		2714 (6.6)	3038 (4.0)	478 (3.9)		3516 (3.9)
4	1852 (5.3)	399 (6.6)		2251 (5.5)	2745 (3.6)	431 (3.5)		3176 (3.6)
5	1550 (4.4)	339 (5.6)		1889 (4.6)	2139 (2.8)	316 (2.6)		2455 (2.8)
Missing	24,646 (69.9)	3626 (60.1)		28,272 (68.5)	60,903 (79.3)	9853 (79.6)		70,756 (79.4)
Anaemia, n (%)
Present	208 (0.6)	733 (12.1)	< 0.001	941 (2.3)	104 (0.1)	305 (2.5)	< 0.001	409 (0.5)
Cardiovascular disease, n (%)
Present	2081 (5.9)	860 (14.3)	< 0.001	2941 (7.1)	822 (1.1)	321 (2.6)	< 0.001	1143 (1.3)
Chronic liver disease, n (%)
Present	421 (1.2)	184 (3.0)	< 0.001	605 (1.5)	311 (0.4)	140 (1.1)	< 0.001	451 (0.5)
Down syndrome, n (%)
Present	6 (0.0)	6 (0.1)	0.002	12 (0.0)	1 (0.0)	4 (0.0)	< 0.001	5 (0.0)
Epilepsy, n (%)
Present	207 (0.6)	69 (1.1)	< 0.001	276 (0.7)	101 (0.1)	45 (0.4)	< 0.001	146 (0.2)
Fatigue, n (%)
Present	1941 (5.5)	881 (14.6)	< 0.001	2822 (6.8)	710 (0.9)	346 (2.8)	< 0.001	1056 (1.2)
Fatigue (count, 1 year), n (%)
Once	396 (1.1)	288 (4.8)	< 0.001	684 (1.7)	113 (0.1)	87 (0.7)	< 0.001	200 (0.2)
Twice	50 (0.1)	63 (1.0)		113 (0.3)	28 (0.0)	35 (0.3)		63 (0.1)
Three times	17 (0.0)	22 (0.4)		39 (0.1)	15 (0.0)	34 (0.3)		49 (0.1)
First-degree relative with CD, n (%)
Present	103 (0.3)	157 (2.6)	< 0.001	260 (0.6)	NR	NR		NR
GI symptoms, n (%)
Present	7850 (22.3)	3164 (52.4)	< 0.001	11,014 (26.7)	2974 (3.9)	1154 (9.3)	< 0.001	4128 (4.6)
GI symptoms (count, 1 year), n (%)
Once	1482 (4.2)	1030 (17.1)	< 0.001	2512 (6.1)	535 (0.7)	306 (2.5)	< 0.001	841 (0.9)
Twice	421 (1.2)	483 (8.0)		904 (2.2)	170 (0.2)	172 (1.4)		342 (0.4)
Three times	134 (0.4)	247 (4.1)		381 (0.9)	60 (0.1)	76 (0.6)		136 (0.2)
Four times	115 (0.3)	272 (4.5)		387 (0.9)	100 (0.1)	144 (1.2)		244 (0.3)
IBS, n (%)
Present	597 (1.7)	339 (5.6)	< 0.001	936 (2.3)	288 (0.4)	142 (1.1)	< 0.001	430 (0.5)
Iron, vitamin B12 or folate deficiency, n (%)
Present	193 (0.5)	429 (7.1)	< 0.001	622 (1.5)	104 (0.1)	163 (1.3)	< 0.001	267 (0.3)
Mouth ulcers, n (%)
Present	331 (0.9)	170 (2.8)	< 0.001	501 (1.2)	142 (0.2)	76 (0.6)	< 0.001	218 (0.2)
Mouth ulcers (count, 1 year), n (%)
Once	50 (0.1)	42 (0.7)	< 0.001	92 (0.2)	32 (0.0)	19 (0.2)	< 0.001	51 (0.1)
Twice	4 (0.0)	9 (0.1)		13 (0.0)	3 (0.0)	8 (0.1)		11 (0.0)
Osteoporosis, n (%)
Present	118 (0.3)	145 (2.4)	< 0.001	263 (0.6)	42 (0.1)	62 (0.5)	< 0.001	104 (0.1)
Psoriasis, n (%)
Present	722 (2.0)	237 (3.9)	< 0.001	959 (2.3)	290 (0.4)	85 (0.7)	< 0.001	375 (0.4)
Type 1 diabetes, n (%)
Present	119 (0.3)	126 (2.1)	< 0.001	245 (0.6)	324 (0.4)	150 (1.2)	< 0.001	474 (0.5)
Thyroid disorders, n (%)
Present	389 (1.1)	287 (4.8)	< 0.001	676 (1.6)	166 (0.2)	130 (1.0)	< 0.001	296 (0.3)
Weight loss, n (%)
Present	340 (1.0)	467 (7.7)	< 0.001	807 (2.0)	85 (0.1)	100 (0.8)	< 0.001	185 (0.2)

Data on ethnicity and deprivation, through linkages with hospital records, were available for approximately one-third of the development data set and one-fifth of the validation data set. In all cohorts, of the patients with known ethnicity, 90–95% were white, and CD patients were more likely to be white than controls. People with CD in the development data set lived in more deprived areas (IMD quintiles 1 and 2) than the controls. This was not the case for CD patients in the validation data set.

All selected indicators were significantly different between cases and controls in all three cohorts of the development data set, except for Down syndrome among women. Although the prevalence of Down syndrome among women with CD was 4 in 10,000, compared with 1 in 10,000 among controls in both data sets, this difference did not reach statistical significance because of the small number of individuals with Down syndrome. In the validation data set for women, epilepsy was also more prevalent in the CD group, but this did not reach statistical significance because of small numbers. In the validation data set with children, arthritis and delayed puberty were not related to CD. In both data sets with men, all indicators were significantly more prevalent among people with CD than among controls.

The most important difference between the development and the validation data sets was that first-degree relatives with CD was not recorded in the validation data set. There are also small differences in prevalences of indicators, with most indicators being more prevalent in the development data set than in the validation data set. For instance, the most prevalent indicator was GI symptoms in all three cohorts. In the development data set, these prevalences ranged from 27% to 40%, whereas the prevalences in the validation data set ranged from 5% to 7%.

Diagnostic indicator selection

The following candidate diagnostic indicators (see Appendix 9, Table 52) could not be considered in the model because there were no observations with the relevant codes: hyposplenism or functional asplenia, raised liver enzymes, multiple sclerosis, pancreatitis, pulmonary haemosiderosis, subfertility and recurrent pregnancy loss among children; delayed puberty and pulmonary haemosiderosis in women; amenorrhoea and Turner syndrome among men. There were no observations of Williams–Beuren syndrome or dental enamel defects in any of the samples.

Table S8 (see Report Supplementary Material 1) presents the proportion of bootstrap samples that included each indicator and the median beta coefficient across all bootstrap samples.

The following indicators were dropped from the model because of an apparent inverse relationship with CD: amenorrhoea, arthritis, irritability, mood disorders, multiple sclerosis, subfertility and type 2 diabetes for women; and type 2 diabetes for men. Attention deficit hyperactivity disorder, headaches, migraines, hyposplenism or functional asplenia, IgA nephropathy, irritability, pancreatitis, type 2 diabetes and multiple sclerosis were not selected as important indicators in any of the models.

Model specification

The model intercepts and coefficients of the final prediction models are presented in Table 3. For children, having type 1 diabetes, Turner syndrome, IgA deficiency or a first-degree relative with CD were estimated to be the strongest diagnostic indicators (i.e. had the highest estimated coefficients). For women and men, the strongest predictors were having a first-degree relative with CD, and anaemia. All three models included first-degree relatives (with CD); anaemia; type 1 diabetes; iron, vitamin B₁₂ or folate deficiency; thyroid disorders; weight loss; Down syndrome; GI symptoms; fatigue; IBS; and age. Epilepsy, cardiovascular disease, chronic liver disease, mouth ulcers and osteoporosis were estimated to be important indicators for adults, but not for children; arthritis, failure to thrive, mood disorders and delayed puberty were estimated to be predictive of CD in children, but not in adults. Fractures, inflammatory bowel disorder, systemic lupus erythematosus, and neuropathy or ataxia were selected indicators for women only. Appendix 11, Tables 53–55, report the effect of the applied shrinkage, showing coefficients with and coefficients without shrinkage, and report the adjusted and unadjusted odds ratios for each indicator for children, women and men, respectively.

TABLE 3 - Model specification

NS, not selected.

Highest to lowest coefficient.

Selected diagnostic indicator	Children			Women			Men
	Coefficient	Ranka	200 bootstrapped samples, median (IQR)	Coefficient	Ranka	200 bootstrapped samples, median (IQR)	Coefficient	Ranka	200 bootstrapped samples, median (IQR)
(Intercept)	–5.119		–5.127 (–5.146 to –5.108)	–5.063		–5.062 (–5.080 to –5.042)	–5.478		–5.488 (–5.526 to –5.460)
Age	0.011	19	0.011 (0.007–0.014)	–0.006	25	–0.006 (–0.006 to –0.005)	0.01	20	0.011 (0.010–0.011)
Anaemia	2.645	5	2.618 (2.522–2.751)	1.63	2	1.635 (1.605–1.661)	2.685	1	2.689 (2.632–2.753)
Arthritis	1.318	12	1.371 (0.949–1.738)	NS	NS	NS	NS	NS	NS
Cardiovascular disease	NS	NS	NS	0.196	20	0.190 (0.139–0.222)	0.253	18	0.243 (0.214–0.282)
Chronic liver disease	NS	NS	NS	0.326	16	0.324 (0.245–0.383)	0.321	16	0.321 (0.236–0.396)
Delayed puberty	1.995	10	1.997 (1.537–2.577)	NS	NS	NS	NS	NS	NS
Down syndrome	2.429	6	2.428 (2.096–2.763)	1.163	5	1.269 (1.161–1.358)	1.293	7	1.344 (0.856–1.813)
Epilepsy	NS	NS	NS	0.258	17	0.252 (0.232–0.268)	0.259	17	0.290 (0.147–0.384)
Failure to thrive	1.382	11	1.398 (1.215–1.540)	NS	NS	NS	NS	NS	NS
Fatigue	0.613	16	0.605 (0.500–0.698)	0.153	22	0.151 (0.127–0.178)	0.185	19	0.186 (0.136–0.233)
Fatigue (count 1 year)	1.111	14	1.090 (0.967–1.233)	0.545	14	0.544 (0.518–0.571)	0.663	12	0.652 (0.592–0.714)
First-degree relative with CD	3.1	4	3.109 (3.037–3.172)	2.459	1	2.449 (2.378–2.517)	2.347	2	2.362 (2.282–2.461)
Fractures (count 1 year)	NS	NS	NS	0.196	19	0.203 (0.167–0.241)	NS	NS	NS
GI symptoms	0.582	17	0.584 (0.550–0.613)	0.249	18	0.251 (0.173–0.360)	0.448	13	0.442 (0.414–0.472)
GI symptoms (count 1 year)	0.794	15	0.792 (0.775–0.817)	0.604	12	0.604 (0.594–0.615)	0.787	10	0.789 (0.772–0.807)
IgA deficiency	3.21	3	3.185 (2.287–3.563)	1.127	6	1.170 (0.765–1.596)	NS	NS	NS
Inflammatory bowel disease	NS	NS	NS	0.138	23	0.112 (0.000–0.227)	NS	NS	NS
Iron, vitamin B12 or folate deficiency	2.016	9	2.013 (1.704–2.363)	1.323	3	1.383 (0.554–2.113)	1.81	3	1.828 (1.754–1.917)
IBS	1.127	13	1.135 (0.934–1.377)	0.478	15	0.474 (0.450–0.505)	0.709	11	0.714 (0.651–0.776)
Mood disorders	0.363	18	0.343 (0.250–0.448)	NS	NS	NS	NS	NS	NS
Mouth ulcers	NS	NS	NS	NS	NS	NS	0.412	14	0.401 (0.305–0.514)
Mouth ulcers (count 1 year)	NS	NS	NS	0.857	10	0.841 (0.767–0.907)	0.934	8	0.919 (0.849–0.994)
Neuropathy or ataxia	NS	NS	NS	0.179	21	0.178 (0.074–0.311)	–	–	–
Osteoporosis	NS	NS	NS	1.028	8	1.040 (1.000–1.077)	1.554	5	1.549 (1.433–1.673)
Psoriasis	NS	NS	NS	0.048	24	0.047 (0.000–0.097)	0.335	15	0.339 (0.265–0.401)
Sex (male)	–0.477	20	–0.472 (–0.502 to –0.447)	NS	NS	NS	NS	NS	NS
Systemic lupus erythematosus	NS	NS	NS	0.699	11	0.698 (0.532–0.856)	NS	NS	NS
Thyroid disorders	2.144	8	2.185 (2.000–2.395)	0.599	13	0.598 (0.563–0.629)	0.91	9	0.913 (0.753–1.074)
Turner syndrome	3.949	2	3.908 (3.715–4.084)	1.08	7	1.057 (0.422–1.681)	NS	NS	NS
Type 1 diabetes	4.153	1	4.182 (4.062–4.278)	1.277	4	1.337 (1.293–1.375)	1.746	4	1.749 (1.650–1.868)
Weight loss	2.316	7	2.302 (2.142–2.485)	0.91	9	0.895 (0.848–0.950)	1.49	6	1.489 (1.431–1.552)

Model performance

The amount of variability explained (R²), the overall model fit (Brier score), discrimination (c-statistic), and calibration measures (intercept and slope of calibration curve) are shown in Table 4.

TABLE 4 - Model performance

FDR, first-degree relative.

Calibration statistics were estimated using an inflated control group to adjust for sampling frequency.

Data	Apparent model performance: original data set (CPRD GOLD)	Internally validated model performance: 200 bootstrapped samples of original data, median (IQR)	Externally validated model performance: independent data set (CPRD Aurum)
Children
R 2	0.407	0.408 (0.401–0.413)	0.065
Brier score	0.167	0.167 (0.165–0.169)	Without FDR: 0.190
			With FDR: 0.156
c-statistic	0.821	0.821 (0.818–0.824)	0.600
Calibration intercepta	0.147	0.161 (0.134–0.181)	Without FDR: 0.433
			With FDR: –2.676
Calibration slopea	0.964	0.986 (0.959–1.014)	0.655
Women
R 2	0.237	0.248 (0.242–0.254)	0.032
Brier score	0.227	0.225 (0.223–0.227)	Without FDR: 0.245
			With FDR: 0.217
c-statistic	0.756	0.764 (0.761–0.767)	0.551
Calibration intercepta	–0.161	–0.153 (–0.169 to –0.143)	Without FDR: 0.433
			–2.676
Calibration slopea	0.822	0.836 (0.817–0.855)	0.655
Men
R 2	0.286	0.284 (0.278–0.291)	0.056
Brier score	0.122	0.124 (0.122–0.126)	Without FDR: 0.134
			With FDR: 0.118
c-statistic	0.798	0.796 (0.793;0.801)	0.619
Calibration intercepta	–0.505	–0.515 (–0.534 to–0.497)	Without FDR: 0.112
			With FDR: –2.250
Calibration slopea	0.934	0.840 (0.817–0.867)	0.668

The development model for children shows the best overall model fit and ability to discriminate between those with and those without CD, compared with the models for men and women. The calibration slope is < 1 for all three models, meaning that all three models overestimate the risk of CD, on average (see Appendix 12, Figure 55). For instance, the model for children estimates a 40% chance of CD, compared with a prevalence in the data set of 20%. At higher risks, the model performs better. The estimated model performance appears to be stable, as the internal model performance in 200 bootstrapped samples was similar, with narrow CIs (see Table 4).

External validation

The prevalence of each predictor was generally lower in the external validation data set. The models performed less well in the validation data set. The amount of variability explained by the model dropped to < 7% in all models. The c-statistics were > 0.5, suggesting that the models discriminated better than chance. Calibration intercepts were further away from 0 and calibration slopes further away from 1 than for the apparent model performance. The R², c-statistic and calibration slope were not different for people with or people without a first-degree relative with CD.

Clinical usefulness

Table 5 shows the sensitivity and specificity of the prediction model at several thresholds to achieve a 1.5%, 2%, 5%, 10% or 20% PPV of the prediction model. This can also be considered as the pre-test probability for serological testing, the next stop on the diagnostic pathway. The lowest pre-test probability (1%) is the estimated prevalence of CD in the general population, so this strategy is the same as testing everyone. By definition, this strategy is 100% sensitive and 0% specific because a test is offered to anyone regardless of their diagnostic indicators. Currently in the UK, only one in three people with CD is believed to be diagnosed, so a prediction rule that picks up more than one in three (i.e. sensitivity of > 33%) might already improve case-finding. As can be see in Table 5, this can be achieved at a pre-test probability of > 20% for children, > 5% for women and > 10% for men. Table 6 shows examples of the combination of predictors among patients at these model thresholds.

TABLE 5 - Clinical usefulness in development data

FN, false negative; FP, false positive; NA, not applicable; NPV, negative predictive value; TN, true negative; TP, true positive.

Note

In a population of 10,000 people.

Population	PPV (%)	Threshold	TP (n)	FP (n)	FN (n)	TN (n)	Sensitivity (%)	Specificity (%)	NPV (%)	CD patients missed (%)
Children	1	0	100	9900	0	0	100.0	0.0	NA	0
	1.5	0.0038	88	5776	12	4124	88.2	41.7	99.7	11.8
	2	0.0042	81	3865	19	6035	80.7	61.0	99.7	19.3
	5	0.0077	67	1271	33	8629	66.7	87.2	99.6	33.3
	10	0.0170	53	478	47	9422	53.3	95.2	99.5	46.7
	20	0.0800	33	129	67	9771	33.1	98.7	99.3	66.9
Women	1	0	100	9900	0	0	100.0	0.0	NA	0
	1.5	0.0053	84	5468	16	4432	84.1	44.8	99.6	15.9
	2	0.0062	76	3687	24	6213	75.8	62.8	99.6	24.2
	5	0.0233	39	731	61	9169	38.7	92.6	99.3	61.3
	10	0.1070	11	96	89	9804	10.7	99.0	99.1	89.3
	20	0.7550	0	1	100	9899	0.2	100.0	99.0	99.8
Men	1	0	100	9900	0	0	100.0	0.0	NA	0
	1.5	0.007	87	5634	13	4266	87.0	43.1	99.7	13
	2	0.008	79	3858	21	6042	79.0	61.0	99.7	21
	5	0.0185	58	1095	42	8805	57.9	88.9	99.5	42.1
	10	0.0610	32	290	68	9610	32.2	97.1	99.3	67.8
	20	0.2820	11	43	89	9857	10.7	99.6	99.1	89.3

TABLE 6 - Examples of the combination of predictors among patients at several model thresholds

CVD, cardiovascular disease; FDR, first-degree relative.

Symptoms that occurred within the previous 10 years.

Risk (%)	Children	Women	Men
> 1.5	All female children	CVD Neuropathy or ataxia Fatiguea GI symptomsa	Fatiguea
> 2	Mood disorders GI symptomsa Fatiguea	GI symptomsa and psoriasis CVD, GI symptomsa Chronic liver disease IBS Thyroid disease	CVD IBS GI symptomsa Mouth ulcersa Epilepsy
> 5	Fatigue within previous year IBS Arthritis Failure to thrive	Fatiguea, GI symptomsa and once last year, and IBS Anaemia Fatiguea and thyroid disorder First-degree relative with CD	GI symptoms,a and chronic liver disease or epilepsy Down syndrome Weight loss
> 10	GI symptomsa (and once in previous year) Failure to thrive and GI symptomsa Iron/folate/vitamin B12 deficiency Thyroid disorders Down syndrome Anaemia	Anaemia, GI symptoms,a iron/folate/B12 deficiency GI symptomsa (and four times in previous year), IBS Chronic liver disease, fatiguea (and once in previous year), GI symptomsa (and three times in previous year) GI symptoms,a IBS, osteoporosis	GI symptomsa (and twice in previous year) Type 1 diabetes, fatigue,a GI symptomsa Fatigue, FDR with CD GI symptoms,a osteoporosis Anaemia
> 20	FDR with CD IgA deficiency Turner syndrome Type 1 diabetes	Anaemia, fatigue,a GI symptomsa (and four times in previous year), iron/vitamin B12/folate deficiency, thyroid disorder Anaemia, fatiguea (and three times in previous year), GI symptomsa (and twice in previous year), inflammatory bowel disease, osteoporosis, thyroid disorder Anaemia, CVD, GI symptomsa (and four times in previous year), iron/vitamin B12/folate deficiency, weight loss	Fatigue,a GI symptoms,a iron/vitamin B12/folate deficiency Fatiguea (and once in previous year), GI symptoms,a thyroid disorders GI symptomsa (and four times in previous year), IBS CVD, GI symptomsa (and once in previous year), mouth ulcersa (and twice in previous year)

The ROC curves in Figure 2 show that the model performs less well with the external data and the ROC curves are closer to the chance line, corresponding to a lower c-statistic. When applying the prediction rule in the external validation data sets, at the 20% threshold for children, 95% of people with CD are missed; at the 5% threshold for women, 86% of people with CD are missed; and at the 10% threshold for men, 94% of people with CD are missed. However, lower thresholds still appear to be able to pick up more than the one in three people with CD, which is how many people with CD are thought to currently have a CD diagnosis (see Appendix 13, Table 56).

FIGURE 2.

The ROC curves model development. (a) Children: development sample; (b) children: external validation; (c) women: development sample; (d) women: external validation; (e) men: development sample; and (f) men: external validation. Thresholds are shown that result in a 1%, 1.5%, 2%, 5%, 10% and 20% PPV. The same thresholds are applied on the external data. FPR, false-positive rate; TPR, true-positive rate.

Sensitivity analysis on coeliac disease patients diagnosed after 1997

The vast majority of patients in the CPRD GOLD data set were diagnosed after 1997, so limiting the analysis to these patients did not make a big impact on sample size. For this sensitivity analysis, 495 (3%) children, 2039 (4%) women and 1591 (4%) men were removed from the development data sets. Although there were minor changes in variable selection and model performance measures, the new models did not perform substantially better or worse than the original models.

Children

The same predictors and two additional predictors, amenorrhoea and mouth ulcers (as count, 2 years prior to CD diagnosis), were included in the model. However, this did not seem to improve model performance: all performance measures were similar between the new and the original models (R² = 0.415, Brier score = 0.166, c-statistic = 0.827, calibration intercept = 0.113 and calibration slope = 1.011).

Women

In the new model, inflammatory bowel disease, psoriasis, and neuropathy or ataxia were not included, following the same model selection criteria as for the original model. The new model performed similarly to the original model (R² = 0.246, Brier score = 0.222, c-statistic = 0.763, calibration intercept = –0.162 and calibration slope = 0.814).

Men

The results were similar for men. With the same model selection procedure, epilepsy was not included in the new model, whereas abnormal liver function was included (which was not included in the original model). However, model performance was similar to that of the original model (R² = 0.282, Brier score = 0.124, c-statistic = 0.797, calibration intercept = –0.543 and calibration slope = 0.829).

Sensitivity analysis including ethnicity and deprivation as predictions

Information on ethnicity and 2015 IMD67 was available for only one-third of the CPRD GOLD cohort. The linked data set consisted of 4254 controls and 1303 CD patients for children; 13,093 controls and 4824 CD patients for women; and 10,618 controls and 2409 CD patients for men. CD prevalence was higher in the linked data sets (at 23.4%, 26.9% and 18.5% for children, women and men, respectively) than in the original data sets (20%, 24.5%, and 14.6% for children, women and men, respectively). Although ethnicity and 2015 IMD67 quintiles were significantly associated with CD in all three samples, the updated model did not perform substantially better (see Appendix 14, Table 57).

The Avon Longitudinal Study of Parents and Children analysis

Table 7 describes the prevalence of candidate predictors among children with (n = 46) and children without CD (n = 5071). A significantly higher proportion of children with CD were female (67% female vs. 33% male); the mean age of children with and children without CD was similar (7.5 years). There were no significant differences in the prevalence of any of the other candidate predictors between children with and children without CD. The most prevalent predictors were male sex (52%), fatigue (23%), mood disorder (12%) and mouth ulcers (10%). With the exception of age, candidate predictors were missing for between 0.2% and 60% of children.

TABLE 7 - Cohort of children: prevalence of predictors

p-values show the result of a Welch two-sample t-test for continuous variables and Fisher’s exact test for categorical variables.

This may include zero.

Predictor	Children without CD (N = 5071)	Children with CD (N = 46)	p-valuea
Type 1 diabetes, n (%)
Absent	2665 (52.6)	30 (65.2)	0.119
Present	< 5 (< 0.01)	< 5 (< 10.9)b
Missing	2404 (47.4)	16 (34.8)
Anaemia, n (%)
Absent	4881 (96.3)	43 (93.5)	0.326
Present	150 (3.0)	< 5 (< 10.9)b
Missing	40 (0.8)	< 5 (< 10.9)b
Thyroid disorder, n (%)
Absent	3297 (65.0)	30 (65.2)	1.000
Present	37 (0.7)	< 5 (< 10.9)b
Missing	1737 (34.3)	16 (34.8)
GI symptom count, n (%)
0	3840 (75.7)	40 (87.0)	0.225
1	207 (4.1)	< 5 (< 10.9)b
2–4	79 (1.6)	< 5 (< 10.9)b
Missing	945 (18.6)	< 5 (< 10.9)b
Sex, n (%)
Female	2444 (48.2)	31 (67.4)	0.027
Male	2618 (51.6)	15 (32.6)
Missing	9 (0.2)	< 5 (< 10.9)b
Fatigue, n (%)
Absent	2767 (54.6)	25 (54.3)	0.960
Present	1165 (23.0)	10 (21.7)
Missing	1139 (22.5)	11 (23.9)
Mouth ulcers, n (%)
Absent	1554 (30.6)	16 (34.8)	0.857
Present	497 (9.8)	< 5 (< 10.9)b
Missing	3020 (59.6)	26 (56.5)
GI symptoms, n (%)
Absent	3840 (75.7)	40 (87.0)	0.235
Present	306 (6.0)	< 5 (< 10.9)b
Missing	925 (18.2)	< 5 (< 10.9)b
Mood disorders, n (%)
Absent	3573 (70.5)	38 (82.6)	0.229
Present	608 (12.0)	< 5 (< 10.9)b
Missing	890 (17.6)	5 (10.9)
Age (years), mean (SD)	7.49 (0.19)	7.45 (0.12)	0.030

Among children for whom complete data for the outcome and each candidate predictor were available, univariable associations indicated that males were less likely to have CD than females (coefficient –0.78, 95% CI –1.39 to –0.17; see Appendix 15, Table 58) (Figure 3). Children with type 1 diabetes, anaemia, a thyroid disorder or two to four GI symptoms were more likely to have CD; however, the 95% CIs contain zero. None of the remaining candidate predictors showed evidence of association with an increased likelihood of CD.

FIGURE 3.

Coefficients and 95% CIs from complete-case univariable analysis of candidate predictors and CD.

Male sex was the only predictor that showed evidence of an association with CD in the ALSPAC data set, with male children less likely than female children to be diagnosed with CD. This finding is consistent with the association between male sex and CD in the CPRD data. For the predictors age, mood disorder, fatigue, mouth ulcers, GI symptoms and GI symptom count of 1, the direction of effects estimated in the ALSPAC cohort contradicts the associations identified in the CPRD data; therefore, inclusion in the model is likely to result in inaccurate risk predictions. Low counts of children with type 1 diabetes and thyroid disorder precluded multiple imputation of these predictors and, because of the large number of missing data (34% and 47%, respectively), inclusion in the model would almost halve the sample size. A prediction model consisting of the remaining predictors (anaemia, two to four GI symptoms and male sex) would not be clinically useful; therefore, the decision was taken not to fit a prediction model in this data set.

Chapter 5 Accuracy of diagnostic tests for coeliac disease

We conducted a systematic review of the accuracy of serological tests for CD. The review was registered with the international prospective register of systematic reviews (PROSPERO): registration number CRD42019115506. The review followed the recommendations from the Centre for Reviews and Dissemination22 and the Cochrane Handbook for Systematic Reviews of Diagnostic Test Accuracy Version 1.0.0,23 and is reported according to the PRISMA-DTA statement. 24

Systematic review methods

Eligibility criteria

Inclusion criteria were defined during protocol development and piloted on a subset of 500 articles during title and abstract screening to ensure that they could be applied objectively. We included studies that met the following criteria.

Study design

Diagnostic cohort studies (also known as ‘one-gate design’) were included. Diagnostic case–control studies (also known as ‘two-gate’ designs) that enrolled a group of people with known CD and a group without CD were excluded, as this type of design has been associated with inflated estimates of accuracy. We anticipated a substantial evidence base from cohort studies and so restricted to this more methodologically robust design. 68

Participants

People presenting with symptoms of CD (e.g. diarrhoea, abdominal pain, fatigue) were included. After piloting the inclusion criteria, we decided to exclude studies conducted with healthy individuals (i.e. screening) or those restricted to single risk groups (e.g. people with diabetes) to ensure that the review was conducted in a clinically relevant population that would be eligible to be tested for CD.

Index test

Any of the following serological tests for CD were eligible: IgA tTG, IgG tTG, IgA EMA, IgG EMA, IgA DGP, IgG DGP and IgA actin antibodies. Combined serological tests, such as IgA/IgG tTG (which detect the presence of IgA tTG or IgG tTG in a serum sample) or IgA tTG followed by IgA EMA, were also eligible for inclusion. GAs were not considered in this review, as they are not recommended for use in the diagnosis of CD by NICE as their accuracy has been shown to be poor. 9 Point-of-care or rapid serological tests were excluded as a systematic review of their accuracy was published in 2019. 69

Reference standard

Studies had to use duodenal biopsy to confirm the diagnosis of CD; at least some seronegative patients had to have received biopsy for the study to be included. Studies in which serology formed part of the reference standard were excluded as this could lead to overestimation of test accuracy.

Search strategy

The following databases were searched from January 1990 to August 2020 for published studies and relevant reviews:

• MEDLINE

• Embase

• Cochrane Library

• Web of Science

• Kleijnen Systematic Reviews (KSR) Evidence.

To identify completed and ongoing trials, we searched the following trial registries from January 1990 to August 2020:

• WHO ICTRP

• NIH Clinical Trials database.

Internet searches were carried out using terms such as ‘celiac’, ‘coeliac’ and ‘serological tests’. The reference lists of relevant systematic reviews identified during the literature search were used as sources of potentially relevant studies.

No language or publication restrictions were applied. Date restrictions were applied to limit searches to 1990 onwards, as serological tests for CD first became available in 1990. 70

We combined terms for ‘antibodies’ with terms for ‘coeliac disease’. Search strategies were adapted for each database searched. Full details of the search strategies are available in Appendix 16. Results of the searches were downloaded and saved to an EndNote X9 library.

Study selection

Titles and abstracts were uploaded to the Rayyan QCRI (Doha, Qatar; https://rayyan.qcri.org/) systematic review software platform71 and screened independently by two reviewers. Any discrepancies were discussed; where disagreement remained, articles were considered to be potentially relevant. Articles considered potentially relevant were obtained as full-text articles and assessed by one reviewer and checked by a second reviewer against the review inclusion criteria using a Microsoft Access form designed specifically for this review. Any discrepancies between reviewers were resolved through discussion or referral to a third reviewer.

Data collection process

Standardised data extraction forms were developed using Microsoft Access. These were piloted on a small sample of papers and adapted as necessary before use. Data extraction was performed by one reviewer and checked by a second. Disagreements were resolved through discussion or referral to a third reviewer when necessary.

We extracted data on the following from each included study:

• participant characteristics – adults/children, reason for biopsy, number of participants

• serological tests – test(s) evaluated, threshold(s) for positivity

• biopsy procedures – definition of CD (biopsy threshold), proportion biopsied.

Results data were extracted as 2 × 2 tables of test results (numbers of true positives, false negatives, false positives and true negatives) for each index test against the biopsy reference standard. When reported, 2 × 2 tables for testing strategies involving combinations of serological tests were also extracted. When 2 × 2 data were reported at multiple thresholds within a study, data relating to the cut-off point prespecified by the manufacturer or author were extracted. Two-by-two data were extracted at a biopsy cut-off point of Marsh grade 3a, if available; otherwise, any reported biopsy threshold was accepted.

Risk of bias

Studies were assessed for methodological quality using the QUADAS-2 tool. This assesses the risk of bias and concerns regarding applicability across the following four domains: patient selection, index test, reference standard, and flow and timing. For this review, we omitted the assessment of applicability as the research question was defined such that any study meeting the inclusion criteria was applicable to the research question. We tailored the signalling questions to our review; the exact signalling questions used are reported in table S9 (see Report Supplementary Material 1). For comparative accuracy studies, we added additional signalling questions based on a draft version of the quality assessment of diagnostic accuracy studies–comparative (QUADAS-C) tool,72 which was under development at the time of our review, to assess the comparative nature of the studies. These are also summarised in table S9 (see Report Supplementary Material 1).

Each study was judged as having a ‘high’, ‘low’ or ‘unclear’ risk of bias in each risk-of-bias domain. If at least one of the domains was rated as having a ‘high’ risk of bias, the study was considered as being at high risk of bias; if all domains were judged as having a ‘low’ risk of bias, the study was considered as being at low risk of bias; otherwise, the study was considered as having an ‘unclear’ risk of bias.

When a study reported accuracy data for two or more tests, the ‘index test’ and ‘flow and timing’ domains were assessed separately for each. When a study reported accuracy data for adults and children separately, all domains were assessed separately for each patient group.

Synthesis of results

We grouped studies according to age group [adults aged > 16 years, children aged ≤ 16 years, mixed (adults and children) and age unspecified] and stratified all analyses by age group and serological test, or test combination. Meta-analyses were performed in Stata^® version 16.0 (StataCorp LP, College Station, TX, USA) using the metandi command.

Primary analyses

For data sets that included at least four studies, a bivariate random-effects meta-analysis of sensitivity and specificity was performed,29 assuming binomial likelihoods for the numbers of true-positive and true-negative test results. 30 When there were fewer than four studies in a data set, univariate fixed-effect meta-analyses of sensitivity and specificity were performed. When only a single study was available, the sensitivity and specificity reported in that study are presented.

When the extracted data on a test relate to a range of thresholds, we report results from two separate meta-analyses. First, we fitted the bivariate model to studies reporting at the most commonly reported threshold only. From these models, we report summary estimates of sensitivity and specificity at that threshold. Second, we fitted the bivariate model to the full data set. From these models, we present the SROC curve, which represents the trade-off between sensitivity and specificity across thresholds.

The sensitivity and specificity reported in each study were plotted in ROC space, with colour-coding allowing comparisons to be made between different thresholds. Summary estimates of sensitivity and specificity with 95% CIs at the most commonly reported threshold and SROC curves across all reported thresholds are presented.

Summary PPVs were calculated from estimates of sensitivity and specificity, for a hypothetical prevalence of 1%, the estimated prevalence of CD in the UK population. We then used a hypothetical population of 10,000 people tested for CD, for a pre-test probability of 1%, to produce a test consequence graphic based on natural frequencies. 73 Values were estimated based on summary sensitivity and specificity from the meta-analyses restricted to the most commonly reported threshold for the most commonly evaluated tests.

Direct comparisons and test combinations

To investigate direct comparisons between tests, we performed additional analyses restricted to studies that reported accuracy estimates for multiple tests based on the same participants (comparative accuracy studies). These studies used the design outlined in Figure 4.

FIGURE 4.

Comparative accuracy study design.

As all patients undergo both index tests and the reference standard, these studies provide more reliable estimates of the comparative accuracy of different tests. 74 For the two most commonly assessed tests, IgA tTG and IgA EMA, we estimated the relative sensitivity and specificity within each study, which are measures of the comparative accuracy of the two tests. Relative sensitivity is the ratio of two sensitivities; for example, if relative sensitivity is 1, then the sensitivity of the two tests is the same (similarly for specificity). We report the observed range of these measures across the set of studies that evaluated both of these tests in the same group of patients.

A subset of these studies also reported accuracy data for the combined accuracy of IgA tTG and IgA EMA. These provided estimates of the accuracy of considering an individual to be positive for CD if ‘both tests were positive’ (and negative otherwise) or of considering an individual to be positive for CD if ‘either test was positive’ (and negative only if both tests were negative). We tabulated estimates of sensitivity and specificity of these testing strategies, but did not pool across these studies using meta-analysis because of variation in the diagnostic thresholds used for each test.

To allow comparisons across the testing strategies that were under consideration for inclusion in the economic model (see Chapter 8), we performed additional analyses of the accuracy of IgA tTG in isolation and of IgA EMA in isolation, based on this same subset of comparative accuracy studies. This allowed us to make direct comparisons between the accuracy of singular tests and that of combined testing strategies. We then restricted to the most commonly reported threshold. These estimates of accuracy were selected for inclusion in the economic model (see Chapter 8).

Sensitivity analyses

We used sensitivity analyses to investigate whether or not estimates of accuracy varied according to study quality, patient spectrum and potential for partial verification bias. We restricted analyses to the following subsets of studies to investigate the impact of these variables:

• studies judged to be at low risk of bias overall based on the QUADAS-2

• studies that included only symptomatic patients

• studies in which all patients received a biopsy to confirm whether or not they had CD.

Deviations from the protocol

• In the protocol, we described our target population as ‘adults or children at risk of CD’. After piloting the inclusion criteria at title and abstract screening, we chose to exclude studies in healthy populations (i.e. screening) or single-risk groups only, as described in Eligibility criteria.

• We described the intervention as ‘any serological test for CD’, including HLA-DQ typing. We decided not to include GAs as they are not recommended for use in the diagnosis of CD by NICE. 9 We did not include HLA-DQ typing in this chapter, as this is covered in a separate review (see Chapter 6). We did not include point-of-care or rapid serological tests as a systematic review of their accuracy was published in 2019. 69

• We described our comparator as ‘any reported reference standard’. After piloting the exclusion criteria at title and abstract screening, we decided to exclude studies in which serology formed part or all of the reference standard, as this would lead to overinflation of test accuracy estimates.

In the strategy for data synthesis, we stated, ‘If a test is reported at a single threshold for test positivity across studies, summary operating points will be used to measure the test’s accuracy. If a test is reported at differing thresholds across studies, SROC curves showing the trade-off between sensitivity and specificity at the various thresholds will be produced’. 75 In the review, we produced both summaries of the evidence for completeness: a SROC curve across all reported thresholds and summary sensitivity/specificity at the most commonly reported threshold.

Results of the assessment of diagnostic accuracy of serological tests

The electronic search identified 15,170 articles. After removing duplicates, 7956 unique reports remained. Following title and abstract screening of these 7956 reports, 398 reports were considered potentially relevant and full-text reports were requested (see Appendix 17, Figure 56). We were unable to obtain full texts for four studies, and translation was not possible for a further five studies; this left 389 reports, which were assessed for inclusion. We included 113 studies (3351 participants) reported in 131 unique publications, with a total of 203 sets of 2 × 2 data reported in these studies.

Study characteristics

The included studies were published between 1990 and 2019. Studies were conducted in the UK (n = 19), Italy (n = 13), Argentina (n = 10), Spain (n = 9), Sweden (n = 8), the USA (n = 8), Canada (n = 6), India (n = 4), Pakistan (n = 4), Islamic Republic of Iran (n = 4), Ireland (n = 3), the Netherlands (n = 3), New Zealand (n = 2), Germany (n = 2), Brazil (n = 2) and Israel (n = 2), with single studies from Slovakia, Türkiye, Denmark, France, Finland, Australia, Czechia, Oman, Switzerland, Austria, Egypt, Poland, Lebanon and Serbia. Twenty-nine studies were conducted with adults, 48 with children and 33 with mixed populations of adults and children. Those that did not specify the age of included participants were assumed to have included a mixed population. A further three studies reported accuracy data separately for adults and children; these were treated as if they were two separate studies and data were extracted separately for adults and children. When reported, the mean age was 43.6 years (SD 13.9 years, range 13–94 years) for adults and 6.3 years (SD 4.4 years, range 2 months to 19 years) for children. On average, 66% of adults and 52% of children were female. Fifty-six studies were prospective and 57 were retrospective in design (see table S10 in Report Supplementary Material 1).

Risk of bias

A total of 137 sets of 2 × 2 data were judged to be at high risk of bias, 22 were judged to be at low risk of bias and 44 were judged to be at unclear risk of bias (see Appendix 18, Figure 57, and table S11 in Report Supplementary Material 1).

The main reason for sets of 2 × 2 data being judged as having a high risk of bias was because biopsy results were interpreted with knowledge of (or not explicitly blinded to) serology results; this was the case for 118 sets of 2 × 2 data. In 28 sets of 2 × 2 data, there was potential for partial verification bias because not all patients, particularly those with a negative test result, underwent biopsy to verify their true disease status. A further 23 sets of 2 × 2 data were judged as having a high risk of bias because of concerns about patient selection (e.g. through inappropriate study exclusions or patients not adhering to a gluten-free diet prior to testing) and 12 sets were judged as having a high risk of bias because of concerns about the index test (e.g. threshold not prespecified).

Twenty-four sets of 2 × 2 data were judged to be at an unclear risk of bias because of missing information on patient selection (e.g. study exclusion criteria), 23 sets were judged to be at an unclear risk of bias because details of serological testing (e.g. threshold for test positivity) were not reported and 40 sets were judged to be at an unclear risk of bias because information on flow and timing (e.g. interval between serology and biopsy, or whether or not patients maintained a gluten-free diet between tests) was not reported. The prevalence of CD varied greatly between studies, ranging from 1.8% to 92.6%.

Accuracy of serological tests for diagnosing coeliac disease

All thresholds

The raw 2 × 2 data extracted from each study, together with details on test and test positivity threshold, are summarised in table S10 (see Report Supplementary Material 1). The ranges in the estimates of sensitivity, specificity, and positive and negative predictive values are summarised in Table 8. The most commonly evaluated test was IgA tTG, which was evaluated in 27 studies with adults76–114 and in 37 studies with children. 115–151 The next most frequently evaluated test was IgA EMA, which was evaluated in 19 studies with adults77,81,83,87,88,90,91,93,97,100,107,108,110–112,152–162 and in 28 studies with children. 82,97,108,115,117,118,121–123,131–133,140,144,145,147,151,152,154–157,163–174

TABLE 8 - Study estimates of test accuracy, stratified by age group and test

NR, not reported.

Serological test	Studies (n)	Participants (N) (with CD, n)	Threshold	Sensitivity (range) (%)	Specificity (range) (%)
Adults
IgA tTG	27	11,355 (2566)	5–25 U/ml	35–100	0–100
IgG tTG	1	65 (14)	10 U/ml	71	96
IgA EMA	19	7122 (1028)	1 : 5–1 : 20	61–100	88–100
IgG EMA	1	96 (28)	1 : 20	39	99
IgA DGP	3	885 (154)	10–20 U/ml	86–98	92–96
IgG DGP	4	1046 (217)	10–20 U/ml	90–97	99–100.0
IgA/IgG DGP	4	1161 (280)	20 U/ml	86–98	96–98.8
IgA/IgG tTG/DGP	3	1849 (173)	20 U/ml	72–96	80–97.4
IgA actin antibodies	2	820 (140)	25 U/ml	80–86	92–95.1
Children
IgA tTG	37	7944 (4164)	3–100 U/ml	29–100	8–100
IgG tTG	5	599 (278)	3–7 U/ml	31–97	71–100
IgA/IgG tTG	2	742 (282)	6–45.1 U/ml	94–96	86–99
IgA EMA	28	4974 (2472)	1 : 5–1 : 40	40–100	29–100
IgA/IgG EMA	2	173 (131)	1 : 2.5–1 : 5	95–96	74–91
IgA DGP	1	212 (109)	20 U/ml	85	88
IgG DGP	3	1135 (669)	10–25 U/ml	77–92	84–94
IgA/IgG DGP	6	941 (464)	16–20 U/ml	88–100	22–96
IgA/IgG tTG/DGP	4	986 (415)	3–32.7 U/ml	88–98	61–99
Mixed or unspecified
IgA tTG	25	4564 (1414)	2–89.5 U/ml	38–100	9.5–100
IgG tTG	2	432 (122)	10–18.9 U/ml	41–85	78.0–89
IgA/IgG tTG	1	254 (26)	7.8 U/ml	92	82.9
IgA EMA IgA DGP	15	2884 (843)	1 : 2.5–1 : 10	68–100	38.9–100
IgG DGP	2	561 (58)	19.9 U/ml	77–90	93.4–97
IgA/IgG DGP	2	562 (56)	19.9 U/ml	76–80	92.0–99
IgA actin antibodies	3	480 (48)	NR	71–86	92.9–99

Study estimates of sensitivity and specificity at the most commonly reported thresholds are shown in ROC space in Figures 5 and 6. These plots also show SROC curves, which are based on all data (i.e. including studies reporting at other thresholds). There was substantial variation in estimates of accuracy across included studies, particularly for IgA tTG and IgA EMA, the most commonly evaluated tests. This made it difficult to draw conclusions regarding the accuracy of these tests. Owing to the range in thresholds used to define a positive test result, it was not appropriate to produce pooled estimates of sensitivity and specificity based on all studies.

FIGURE 5.

Study estimates of test sensitivity and specificity among adults plotted in ROC space. (a) IgA tTG; (b) IgA EMA; (c) IgA DGP; (d) IgG DGP; (e) IgA/IgG DGP; and (f) IgA/IgG tTG/DGP. SROC curves are estimated from a meta-analysis of all data, across thresholds. Summary point estimates are estimated from a meta-analysis of data reporting at the most commonly reported threshold only.

FIGURE 6.

Study estimates of test sensitivity and specificity among children plotted in ROC space. (a) IgA tTG; (b) IgG tTG; (c) IgA EMA; (d) IgG DGP; (e) IgA/IgG DGP; and (f) IgA/IgG tTG/DGP. SROC curves are estimated from a meta-analysis of all data, across thresholds. Summary point estimates are estimated from a meta-analysis of data reporting at the most commonly reported threshold only.

Accuracy of tests at the most commonly reported thresholds

Table 9 shows pooled estimates of the accuracy of each serological test evaluated at the most commonly reported threshold. Summary estimates restricted to the most common threshold are also shown in Figures 5 and 6.

TABLE 9 - Summary estimates of sensitivity and specificity, restricted to the most commonly reported threshold only

Univariate fixed-effects meta-analysis.

Serological test	Studies (n)	Participants (N) (with CD, n)	Threshold	Sensitivity (95% CI) (%)	Specificity (95% CI) (%)
Adults
IgA tTG	5	4310 (454)	15 U/ml	90.7 (87.3 to 93.2)	87.4 (84.4 to 90.0)
IgA EMA	5	927 (446)	1 : 5	88.0 (75.2 to 94.7)	99.6 (92.3 to 100.0)
IgA DGPa	2	820 (140)	20 U/ml	96.4 (91.7 to 98.5)	95.4 (93.6 to 96.8)
IgG DGPa	3	981 (203)	20 U/ml	93.6 (88.6 to 96.5)	99.4 (98.5 to 99.7)
IgA/IgG DGP	4	1161 (280)	20 U/ml	91.5 (84.7 to 95.4)	96.7 (95.3 to 97.7)
IgA/IgG tTG/DGPa	2	851 (155)	20 U/ml	93.5 (88.4 to 96.5)	86.3 (79.7 to 91.0)
IgA actin antibodiesa	2	820 (140)	25 U/ml	82.9 (75.7 to 88.2)	92.5 (90.3 to 94.3)
Children
IgA tTG	6	2232 (1051)	20 U/ml	97.7 (91.0 to 99.4)	70.2 (39.3 to 89.6)
IgA EMA	5	1257 (685)	1 : 10	94.5 (88.9 to 97.3)	93.8 (85.2 to 97.5)
IgA/IgG DGP	5	533 (276)	20 U/ml	96.4 (91.7 to 98.5)	77.4 (44.0 to 93.7)
IgA/IgG tTG/DGPa	2	244 (133)	20 U/ml	95.6 (83.9 to 98.9)	62.2 (52.8 to 70.7)

Although tTG was evaluated in 27 studies with adults, only five studies reported at a common threshold of 15 U/ml. 90,100,110,175,176 At this threshold, IgA tTG had a summary sensitivity of 90.7% (95% CI 87.3% to 93.2%) and summary specificity of 87.4% (95% CI 84.4% to 90.0%) for CD among adults (see Table 9). Only 5 of the 19 studies of IgA EMA among adults reported accuracy data at a common threshold of 1 : 5. 83,93,107,111,112 These studies reported that the EMA was highly specific, with a summary specificity of 99.6% (95% CI 92.3% to 100%) in adults, but summary sensitivity was lower, at 88.0% (95% CI 75.2% to 94.7%). IgA and IgG DGP also showed very good sensitivity, with summary estimates for IgA, IgG and IgA/IgG DGP all > 90%, but these were evaluated in four or fewer studies. 91,101,109 Specificity for these tests was also high, with all estimates > 95%.

Six of the 37 studies with children reported data at a common threshold of 20 U/ml. 121,134,136,140,141,146 These studies found that IgA tTG was very sensitive, with a summary sensitivity of 97.7% (95% CI 91.0% to 99.4%), but summary specificity was lower, at 70.2% (95% CI 39.3% to 89.6%). Six of the 28 studies of EMAs with children reported data at a threshold of 1 : 10. 121,135,140,144,163,174 Summary sensitivity was 94.5% (95% CI 88.9% to 97.3%) and summary specificity was 93.8% (95% CI 85.2% to 97.5%). IgA/IgG DGP was also evaluated at a common threshold of 20 U/ml in five studies and showed good sensitivity (summary estimate 96.4%, 95% CI 91.7% to 98.5%), but poorer specificity (summary estimate 77.4%, 95% CI 44% to 93.7%). 124,128,136,138,145,150

Direct comparisons and test combinations

Comparative accuracy studies provided little evidence of differences in accuracy between tests (Table 10). Fourteen studies with adults and 16 studies with children provided direct comparison of IgA tTG and IgA EMA. There was a suggestion that IgA EMA was more specific than IgA tTG in adults, with similar estimates of sensitivity; estimates in children were similar for both tests. However, studies reported results at different thresholds and, therefore, formal statistical comparison was not appropriate.

TABLE 10 - Study estimates of sensitivity and specificity, restricted to comparative studies only

Serological test	Studies (n)	Participants (N) (with CD, n)	Threshold (range)	Sensitivity (range) (%)	Relative sensitivity (range) (%)	Specificity (range) (%)	Relative specificity (range) (%)
Adults
IgA tTG vs. IgA EMA	14	6575 (881)
IgA tTG			5–25 U/ml	64–100		81–99
IgA EMA			1 : 5	61–100	0.81–1.22	88–100	1.00–1.17
IgA tTG vs. IgA DGP	3	885 (154)
IgA tTG			10–20 U/ml	64–95		88–98
IgA DGP			10–20 U/ml	86–98	1.04–1.33	92–96	0.96–1.04
IgA tTG vs. IgG DGP	4	1046 (217)
IgA tTG			10–20 U/ml	64–95		88–98
IgG DGP			10–20 U/ml	90–96	0.99–1.44	99–100	1.02–1.13
IgA tTG vs. IgA/IgG DGP	4	1161 (280)
IgA tTG			5–20 U/ml	76–95		95–99
IgA/IgG DGP			20 U/ml	86–98	1.01–1.14	96–99	0.99–1.01
Children
IgA tTG vs. IgA EMA	16	3021 (1746)
IgA tTG			5.5–21 U/ml	29–99		24–100
IgA EMA			1 : 5 to 1 : 10	50–100	0.91–2.25	29–100	0.76–1.25

Other test pairs were compared directly in only three or four studies. IgG DGP and IgA/IgG DGP appeared to be slightly more sensitive and specific than IgA tTG; however, this difference was much smaller than indirect comparisons suggested (see Table 10). This suggests that studies providing a direct comparison between DGP and other serological tests may be subject to bias, resulting in overestimated accuracy for all tests evaluated in these studies.

A subset of the studies that provided a direct comparison of IgA tTG with IgA EMA also reported data that could be used to estimate the accuracy of these tests when used in combination. We identified six such studies with adults81,82,88,90,107,108 and six with children. 82,108,121,123,144,145 However, none of these reported accuracy estimates for the same thresholds. We therefore selected the studies that were judged to be at lowest risk of bias and that had the largest sample sizes. These were Hopper et al. 90 (n = 2000 adults) and Wolf et al. 121 (n = 949 children). Wolf et al. 121 was judged to be at low risk of bias for all domains; Hopper et al. 90 was judged to be at low risk of bias for all domains except the reference standard domain, as it was unclear whether or not the reference standard results were interpreted without knowledge of the index test results. We consider these studies to provide the most reliable comparative estimates of the accuracy of IgA tTG and IgA EMA, alone and in combination, among adults and among children. Table 11 summarises the accuracy data from these studies; these estimates of accuracy were selected for inclusion in the economic model (see Chapter 8). Among both adults and children, the IgA tTG test had the highest sensitivity, although estimates among children were very similar, and the IgA EMA test had the highest specificity. There was little improvement in either sensitivity or specificity when the tests were used in combination. Despite thresholds being the same as those used in the meta-analyses for the individual tTG and EMA tests, there were minor differences between estimates from these studies and summary estimates when all studies that reported at the same thresholds were pooled (Table 12).

TABLE 11 - Estimates of accuracy for the two selected studies that provided information on the accuracy of the tests when used in combination

Study	Population	Participants with complete data (n)	Test	Threshold	Sensitivity (95% CI) (%)	Specificity (95% CI) (%)
Hopper et al.90 2008	Adults	2000	Both positive	15 U/ml (tTG) 1 : 4 (EMA)	86 (76 to 92)	99 (98 to 99)
			Either positive	15 U/ml (tTG) 1 : 4 (EMA)	92 (84 to 96)	90 (89 to 92)
			tTG	15 U/ml	91 (82 to 95)	91 (90 to 92)
			EMA	1 : 4	87 (78 to 93)	98 (97 to 99)
Wolf et al.121 2017	Children	873	Both positive	20 U/ml (tTG) 1 : 10 (EMA)	95 (93 to 97)	95 (92 to 97)
			Either positive	20 U/ml (tTG) 1 : 10 (EMA)	97 (95 to 98)	89 (86 to 92)
			tTG	20 U/ml	97 (95 to 98)	89 (86 to 92)
			EMA	1 : 10	96 (94 to 97)	94 (91 to 96)

TABLE 12 - Study estimates of sensitivity and specificity limited to specific subgroups, stratified by age group and test

Serological test	Studies (n)	Participants (N) (with CD, n)	Threshold (range)	Sensitivity (range) (%)	Specificity (range) (%)
Symptomatic patients only
Adults
IgA tTG	7	4244 (325)	5–20 U/ml	64–100	88–98
IgA EMA	8	3786 (327)	1 : 5–1 : 20	61–100	98–100
Children
IgA tTG	8	1126 (615)	4–20 U/ml	40–100	8–100
IgA EMA	8	1327 (753)	1 : 5–1 : 20	40–98	43–100
All patients underwent biopsy
Adults
IgA tTG	26	11,183 (2444)	5–25 U/ml	64–100	0–100
IgA EMA	18	7010 (1021)	1 : 5–1 : 20	61–100	88–100
Children
IgA tTG	31	5824 (3330)	3–100 U/ml	29–100	24–100
IgA EMA	22	3685 (2015)	1 : 5–1 : 40	46–100	29–100
Low risk of bias
Adults
IgA tTG	5	1577 (426)	5–20 U/ml	76–100	88–98
IgA EMA	2	268 (103)	1 : 5–1 : 20	61–89	100–100
Children
IgA tTG	4	1319 (823)	20–21 U/ml	95–100	24–99
IgA EMA	1	873 (528)	1 : 10	95	94

Sensitivity analyses

Table 12 shows results from the sensitivity analyses; further details are reported in Appendix 19 (see Figures 58–61). Estimates of sensitivity and specificity in analyses were restricted to studies of symptomatic patients, studies judged to be at low risk of bias overall on the QUADAS-2 tool and studies in which all patients received a biopsy to confirm whether or not they had CD were similar to the estimates of analyses that included all studies. There were insufficient studies to allow comparisons when analyses were restricted to those that reported a common threshold.

Chapter 6 Accuracy of genetic tests for coeliac disease

We conducted a systematic review of the accuracy of genetic tests for CD. We had originally planned to include this as part of the review of the accuracy of serological tests but decided that these tests were sufficiently different from serological tests that a separate review would be appropriate. The review followed recommendations from the Centre for Reviews and Dissemination22 and the Cochrane Handbook for Systematic Reviews of Diagnostic Test Accuracy Version 1.0.023 and is reported in accordance with the PRISMA-DTA statement. 24

Results of assessment of diagnostic accuracy of indicators

Full details of the search results are reported in Chapter 3. We identified four studies reported in six publications (n = 12,087);177–182 one study, based on the Dutch Generation R birth cohort study, was reported in three publications. 180–182 Although none of these specifically aimed to evaluate the accuracy of HLA testing for CD, all reported data that could be used to estimate accuracy of HLA testing. We selected the most recent report, which also provided data on the greatest number of participants, for inclusion in the analysis. 180

Study characteristics

Table 14 provides an overview of the included studies. All studies were conducted with children. Two studies used a nested case–control design and two used a cohort design. The median age was 12 years in two studies,178,179 6 years in one study180 and 3 years in the remaining study. 177 Two studies were conducted in Sweden,177,179 one in Finland178 and one in the Netherlands. Three studies177–179 confirmed the diagnosis of CD by a combination of serology with biopsy of those who were positive on serology, the other used serology alone. 180

TABLE 14 - Details of studies that evaluated the accuracy of HLA genotyping

NA, not applicable.

Study	Population	Study design	Genetic test details	Total sample (N) (analysed, n)	Age (years)	Sex (% female)	Control group	Reference standard	Setting	Location	Risk of bias
Björck et al.177 (2010)	Children	Nested case–control	HLA-DQ2, HLA-DQ8 or both	3435 (3435)	3	NR	Cases: HLA positive	Serology and only positive patients biopsied	Community	Sweden	Low
							Controls: HLA negative
Mäki et al.178 (2003)	Children	Cohort	HLA-DQ2, HLA-DQ8 or both	3654 (3617)	Median 12 (range 7–16)	NR	NA	Serology and only positive patients biopsied	Community	Finland	Low
Sandström et al.179 (2013)	Children	Nested case–control	HLA-DQ2, HLA-DQ8 or both	1320	12	52	Cases: serology positive, referred for biopsy	Serology and only positive patients biopsied	Community	Sweden	Low
							Controls: serology negative
Wahab et al.180 (2019)	Children	Cohort	HLA-DQ2, HLA-DQ8 or both	4442 (3715)	Median 6 (range 5–9)	48	NA	tTG positive	Community	The Netherlands	High

Risk of bias

Three studies were judged to be at low risk of bias across all domains (see Appendix 20, Figure 62). 177–179 The other was judged to be at high risk of bias as CD diagnosis was based on serology alone. 180 This study was judged to be at low risk of bias for all other QUADAS-2 domains.

Accuracy of diagnostic indicators to detect coeliac disease

Results were broadly consistent across studies, with visual examination of the forest plot (Figure 7) and the SROC plot (Figure 8) suggesting little evidence of heterogeneity. The presence of either HLA-DQ2 or HLA-DQ8 had a very high summary sensitivity of 99.2% (95% CI 83.4% to 100%) and summary specificity of 55.6% (95% CI 50.2% to 60.9%). These figures were included in the economic model in Chapter 8.

FIGURE 7.

Forest plots of individual study and summary estimates of sensitivity and specificity. FN, false negative; FP, false positive; TN, true negative; TP, true positive.

FIGURE 8.

The SROC curve for HLA-DQ2/-DQ8 for diagnosing CD. FPR, false-positive rate; TPR, true-positive rate.

Chapter 7 Establishing diagnostic thresholds

This chapter describes the collaborative work with patients and the public through focus groups and an online survey to investigate how confident people want to be in their diagnosis before starting a gluten-free diet or undergoing a biopsy.

The aims of the survey were as follows:

• to estimate at what level of diagnostic certainty people are willing to start a gluten-free diet without a confirmation biopsy if they are symptomatic

• to estimate at what level of diagnostic certainty people are willing to start a gluten-free diet without a confirmation biopsy if they are asymptomatic

• to understand what factors influenced their answers on the above

• to estimate the quality of life among people with CD

• to understand how strictly people with CD followed the gluten-free diet

• to understand how difficult people with CD found it to follow a gluten-free diet

• to understand the current assumptions and understandings of CD diagnosis and how that affects confidence in diagnosis before starting a gluten-free diet.

Results

Respondent characteristics

The survey was completed by 472 people. Five respondents did not live in the UK and were removed from the analysis. A total of 244 (52%) respondents had CD, with the condition confirmed by a blood test and/or biopsy. An additional 36 respondents believed that they had CD, although this had not been confirmed by a test. These respondents were included in the group without confirmed CD (n = 223, 48%). Thirty-two (6.9%) respondents were children (aged < 18 years); either they filled out the questionnaire by themselves or it was done by a parent on their behalf. The majority of respondents were aged between 26 and 64 years (n = 310, 66.2%). Questions on sex, ethnicity, qualifications and postcode were optional; these were answered by most CD patients, whereas most respondents without CD did not answer these questions. Among those who answered the optional questions, the vast majority were white (n = 264, 95%) and female (n = 239, 86%). Most respondents went to university or college (n = 159, 58%). The highest proportion lived in south-west England (n = 98, 36%). Respondents tended to live in less deprived areas (median deprivation index of 7) than the national average; this was similar across respondents with and those without CD. A total of 304 (65%) respondents were following a gluten-free diet at the time of filling out the survey and 28 (6%) had done so in the past (see Appendix 22, Table 59).

Scenarios

The scenario questions were answered by 88.5–90.0% of respondents (Table 15). In scenario 1, in which respondents were asked to imagine having typical CD symptoms, 387 (92.8%) respondents opted for a blood test. If the blood test showed a negative result for CD, 223 (53.5%) respondents chose to follow a gluten-free diet anyway to find out if it reduced their symptoms, 93 (22.3%) respondents wanted further testing for CD and 101 (24.2%) accepted that it was unlikely that they had CD. The free-text answers revealed that some respondents misunderstood this question and believed that the negative test result followed two positive test results, which may have affected their answers.

TABLE 15 - Responses to scenarios

GFD, gluten-free diet.

Scenario	Risk of CD (%)	Test result	Preferred option per scenario in response to risk of having CD, n (%)
			Start GFD	Further testing	No further testing or diet
GI symptoms and fatigue	1	Negative	223 (53.5)	93 (22.3)	101 (24.2)
GI symptoms and fatigue	5	Pre test	21 (5)	387 (92.8)	12 (2.9)
GI symptoms and fatigue	33	Positive	112 (26.9)	295 (70.7)	9 (2.2)
GI symptoms and fatigue	75	Strong positive	184 (44.1)	224 (53.7)	6 (1.4)
First-degree relative	10	Pre test	21 (5)	369 (88.5)	30 (7.2)
First-degree relative	50	Positive	94 (22.5)	297 (71.2)	28 (6.7)
First-degree relative	90	Strong positive	195 (46.8)	207 (49.6)	17 (4.1)
Osteoporosis	2	Pre test	15 (3.6)	376 (90.2)	30 (7.2)
Osteoporosis	15	Positive	51 (12.2)	336 (80.6)	33 (7.9)
Osteoporosis	55	Strong positive	99 (23.7)	311 (74.6)	7 (1.7)

If the test result showed a positive result for CD, corresponding to a 33% risk of having CD, the majority of respondents opted for a confirmation biopsy (n = 295, 70.7%), rather than starting a gluten-free diet immediately. If the blood test showed a strong positive result, increasing their risk of CD to 75%, 184 (44.1%) respondents chose to start a gluten-free diet without confirmation biopsy, whereas 224 (53.7%) respondents chose to wait for a biopsy appointment.

In scenario 2, in which respondents were asked to imagine that a first-degree relative had CD, the majority opted for a blood test (n = 369, 88.5%). If the test was positive and the risk of CD was 50%, 94 (22.5%) respondents would start a gluten-free diet without a confirmation biopsy, but most (n = 297, 71.2%) would opt for a confirmation biopsy. If the test was strongly positive and the risk of CD was 90%, 195 (46.8%) respondents would start the gluten-free diet, whereas 207 (49.6%) respondents still preferred to wait for a confirmation biopsy.

In scenario 3, in which the respondents were asked to imagine being diagnosed with risk conditions that increased their risk of CD to 2%, 376 (90.2%) respondents opted to have the blood test. At 15% and 55% risks of CD, after a positive or strongly positive test result, the majority opted for a confirmation biopsy [n = 336 (80.6%) and n = 311 (74.6%) for 15% and 55% risks, respectively], rather than starting a gluten-free diet immediately.

Certainty of coeliac disease diagnosis

When respondents were asked to imagine that they had symptoms, they wanted to be a median of 66% (IQR 33–90%) certain of the diagnosis before starting a gluten-free diet. Without symptoms, respondents wanted to be more certain (median 90%, IQR 66–99%) before committing to a gluten-free diet.

We found no evidence that age group or having had a biopsy affected the level of certainty of CD diagnosis that respondents would like before starting a gluten-free diet. However, respondents without CD and those familiar with a gluten-free diet accepted a lower certainty. In the scenario in which respondents were asked to imagine that they had symptoms, this was 20% lower for people without CD [standard error (SE) 3.8, t = –5.1; p < 0.001] and 15% lower for those familiar with a gluten-free diet (SE 4.0, t = –3.7; p < 0.001). In the scenario in which respondents were asked to imagine that they did not have symptoms, this was 13% lower for people without CD (SE 3.5, t = –3.7603.8; p < 0.001) and 12% lower for those familiar with a gluten-free diet (SE 3.6; t = –3.2; p = 0.001) (Figure 9).

FIGURE 9.

Preferred certainty of CD diagnosis before starting a GFD. GFD, gluten-free diet.

Gluten-free diet adherence and difficulty

Figure 10 shows the results from CD patients regarding the difficulty of following a gluten-free diet and how strictly they adhered to the diet. A total of 131 (53.5%) respondents found it easy or very easy to follow the gluten-free diet, whereas 69 (28.2%) respondents found it difficult or very difficult. Most respondents with confirmed CD reported that they were strict or very strict in their adherence to the gluten-free diet (n = 222, 90.6%).

FIGURE 10.

Gluten-free diet adherence and difficulty. (a) Difficulty following a GFD; and (b) adherence to a GFD. GFD, gluten-free diet.

Quality of life among coeliac disease patients

The median overall quality-of-life score among CD patients was estimated at 59.38 (IQR 43.75–76.25), based on the CD-QOL. Quality of life is considered good when the score is > 59, medium when the score is 37–59 and poor when the score is < 37. Although there was large variation between CD patients’ quality-of-life scores, most scores indicated a good or medium quality of life (Figure 11). CD patients had a high score in the dysphoria domain (median 87.50, IQR 68.75–93.75), suggesting that dysphoria did not affect quality of life for the majority of respondents. CD patients also had a good score in the health concerns domain (median 60.0, IQR 35.0–80.0), although there was more variation, and some respondents felt that their quality of life was affected by health concerns. CD patients had a medium score for the domains ‘CD-related limitations’ (median 55.56, IQR 36.11–72.22) and ‘inadequate treatment’ (median 50.00, IQR 37.50–50.00). About one-quarter of the respondents with CD had a low score (associated with a poor quality of life) for the CD-related limitations and health concerns domains.

FIGURE 11.

Quality-of-life scores (CD-QOL).

Qualitative analysis

Of the 468 respondents, 353 (75.4%) used at least one of the free-text boxes to explain their answer to a scenario question. Of these, 185 (52.4%) had a CD diagnosis confirmed by a blood test and/or biopsy and 167 (47.3%) did not. Of those who answered, the highest proportion fell into the 41–64 years age group (42.2%), followed by the 26–40 years age group (28.0%). Twenty-six children (aged < 18 years), or parents of children (7.4%), explained their answers to the scenario questions. The response rate was slightly higher for those with CD than for without [CD, n = 17/185 (9.2%); no CD, n = 9/167 (5.4%)] (see Appendix 22, Table 60).

Free-text analysis

Four key themes and 29 subthemes were identified in the explanations respondents gave. The four key themes were factors that prompt CD diagnosis, the diagnostic process, how to respond to a negative test result and opinions on the gluten-free diet (see Appendix 23, Table 61). Themes raised by adults and children were similar.

Factors that prompt coeliac disease diagnosis

The most frequently mentioned prompt for seeking CD diagnosis was being diagnosed with osteoporosis, with 44 (9.4%) respondents remarking that this would be a worrying development:

If a young man gets an old woman’s disease they’d be clutching at lots of straws to try and figure it out.

Thirty-seven (7.9%) respondents stated throughout the scenarios that the (more general) future risks of CD would prompt seeking or continuing testing for diagnosis. A total of 59 (12.6%) respondents said they would look into CD diagnosis immediately and rule it out early if it was suggested to them by their doctor. For 29 (6.2%) respondents, symptoms would decide whether or not they looked into getting or continuing a diagnosis. Twenty-four (5.1%) respondents said that family has affected them considering a CD diagnosis for themselves.

A total of 49 (10.5%) respondents said that having an official diagnosis is important to them, stating that there are benefits to having an official record:

If I have it, I want the diagnosis on my medical records so that I can be allowed DEXA [dual-energy X-ray absorptiometry] scans and flu jabs recommended for people with coeliac disease.

In contrast, only four (0.9%) respondents said that they felt that there is no point to an official diagnosis:

As the ‘benefit of having a diagnosis’ is irrelevant as gluten-free prescriptions aren’t allowed anymore and little no support or understanding from professionals.

The diagnostic process

A total of 184 (39.3%) respondents mentioned that they wanted or were happy to have a blood test for CD, as it is not invasive and is seen as easy to have:

It is easy and quick to get a blood test done which would give a much clearer idea of the situation.

As for opinion on biopsies, 116 (24.8%) respondents said they feel that a biopsy is necessary for confidence in their diagnosis, whereas 27 (5.8%) feel that biopsy is invasive or unpleasant and would want to avoid it. The majority of those who believe that a biopsy is necessary for certainty of their diagnosis are those with CD [CD, n = 77 (31.63%); no CD, n = 39 (17.5%)]. Forty-one (8.8%) respondents said they would continue a gluten diet for biopsy even if they did not have symptoms. A total of 76 (16.2%) respondents mentioned that, when having no symptoms, a biopsy was more important for confidence in diagnosis and they were more willing to continue a gluten-containing diet during the waiting time for a biopsy appointment.

Twenty-eight (6.0%) respondents stated that 6–8 weeks is an acceptable wait time for a biopsy, compared with 13 (2.8%) who stated that the wait time was unacceptable, and 76 (16.2%) respondents expressed that, if the blood test showed a > 50% chance of having CD, they would start a gluten-free diet without waiting for a biopsy:

This is a very high answer from a blood test. I feel I have got the answer I need without needing an invasive biopsy.

How to respond to a negative test result

When presented with a negative or low chance of having CD, 37 (7.9%) respondents said that they would try a gluten-free diet regardless of diagnosis:

If the gluten-free diet could potentially stop the pain or improve quality of life I would start the diet regardless of diagnosis.

Thirty-two (6.8%) respondents would ask for further testing or retesting if they had a negative result. A total of 26 (5.6%) respondents commented that their doctor’s guidance and opinion would be important to them, and 26 (5.6%) respondents would accept not having CD.

Opinions on a gluten-free diet

A total of 128 (27.4%) respondents stated that they would want a definitive diagnosis (through a biopsy) before starting a gluten-free diet, whereas 36 (7.7%) would start a gluten-free diet without a definitive diagnosis.

Twenty-seven (5.8%) respondents said that the lowest likelihood of having CD would not be enough for them to start a gluten-free diet, and 25 (5.3%) stated that a 50/50 likelihood would not be enough to start a gluten-free diet. However, 20 (4.3%) would start a gluten-free diet at the lowest likelihood (≤ 10% chance) of having CD. The majority of those who stated this were those without a CD diagnosis [CD, n = 1 (0.2%); no CD, n = 19 (8.5%)]:

I would want my symptoms to get better as soon as possible, so if the [gluten-free diet] helped to eliminate the uncomfortable symptoms, that would be good enough for me. I wouldn’t need the tests.

A total of 77 (16.5%) respondents said they would start a gluten-free diet and see if CD symptoms improved (even if not diagnosed with CD); of these, 22 (4.7%) also said that they may consider restarting a gluten diet and having a biopsy if a gluten-free diet did not work:

I could try the gluten-free diet and if my symptoms improve I would continue, if it makes no difference I may go back to eating gluten and then opt for a biopsy or other tests.

Thirty-five (7.5%) respondents have said that whether or not they live with another person who has CD would affect their gluten-free diet adherence, and 48 (10.3%) expressed a negative opinion on following a gluten-free diet or see a gluten-free diet as a big commitment. More of those with a CD diagnosis expressed negativity towards a gluten-free diet than those without CD [CD, n = 33 (13.5%); no CD, n = 15 (6.7%)]:

I have 30 years on a [gluten-free diet], I know what it is like. You would want to be sure before changing your diet for good. It isn’t much fun, long term.

Chapter 8 Assessment of cost-effectiveness of diagnostic strategies

This chapter describes the decision tree and Markov modelling to estimate the cost-effectiveness of screening strategies for CD among men, women and children. This builds substantially on previous chapters. Individual diagnostic indicators from Chapter 3 and combinations of diagnostic indicators from the prediction model in Chapter 4 are investigated in combination with serological and genetic tests. Serological test accuracies come from Chapter 5 and genetic test accuracy comes from Chapter 6. Results of the patient survey in Chapter 7 inform the key assumptions of the model. The objectives of this analysis were to determine the following:

• the most cost-effective combination of sensitivity and specificity for a pre-test probability above which all patients should be screened

• the most cost-effective screening strategy.

Results were based on a lifetime time horizon and take a UK NHS perspective.

Methods for cost-effectiveness analysis

Our model development was based on a review of previously published models; the results of the diagnostic indicator review (see Chapter 3) and CPRD analysis of diagnostic indicators for CD (see Chapter 4), which identified important patient characteristics for modelling; the results of the meta-analysis on serological and genetic test accuracies (see Chapters 5 and 6), which identified tests to model; the survey of patient opinion (see Chapter 7); and discussion with our clinical and patient advisers.

Review of previous cost-effectiveness models

To identify important elements of CD to model, and potentially identify existing models we could adapt, we reviewed previously published models as a first step in our method of model development.

Search strategy

We searched Embase and combined terms for CD with an economic model filter. This identified 390 papers, which were screened independently by two reviewers. We identified 13 papers as relevant. Any studies that contained any type of decision-analysis model for CD were eligible for inclusion. We excluded economic evaluations not based on cost-effectiveness models. The full search strategy is provided in Appendix 24.

Summary of identified models

Of the 13 economic models identified, seven were decision trees, four were Markov models and two were decision trees followed by Markov models. Only the NICE 2015 model was in a UK setting. 193 Of the others, one was set in the Netherlands194 and the rest were based in the USA. 195–203,206,207 The studies are discussed in chronological order and details are shown in Appendix 25, Table 62.

Harewood and Murray195 developed a decision tree to compare the costs of different screening strategies for the detection of CD. This study found screening with EMA to be the less costly than both screening with GAs and small bowel biopsy in a low- to medium-risk population. We note, however, that GAs are not recommended by NICE because of their low accuracy. 193 Mein and Ladabaum196 used a decision tree to explore the cost-effectiveness of different screening strategies for CD in patients with IBS symptoms. They concluded that testing for tTG to diagnose CD in patients with a diagnosis of IBS is cost-effective at thresholds of US$50,000 (assuming a prevalence of CD in IBS patients of 2%) and US$100,000 (assuming a prevalence of CD in IBS patients of 1.1%) per QALY gained. Spiegel et al. 197 built a decision tree followed by a Markov model to evaluate different screening strategies for CD in IBS patients with predominant diarrhoea symptoms. They found that serological testing for CD resulted in 51.6% of the patients achieving symptomatic improvement at 10 years (at an average cost of US$4100 per patient treated). Starting IBS therapy without testing for CD resulted in 50.9% of the population achieving symptomatic improvement (at an average cost of US$4023 per patient treated). The incremental cost-effectiveness ratio (ICER) of CD testing, compared with IBS treatment alone, was US$11,000 to achieve one additional symptomatic improvement. Testing for CD was, therefore, considered cost-effective.

Shamir et al. 198 developed a lifetime Markov model to conduct a cost-effectiveness analysis comparing seven screening strategies for CD in the adult population. The ICER was US$44,941 per life-year gained for screening, compared with no screening, using an IgA EMA testing strategy. All remaining six strategies were dominated and thus excluded. Swigonski et al. 199 used a decision tree to evaluate the cost-effectiveness of screening for CD among asymptomatic children with Down syndrome to prevent lymphoma. They found that screening for CD among children with Down syndrome is more costly and less effective than not screening, and decreases quality of life.

Dorn and Matchar200 evaluated the cost-effectiveness of five strategies for diagnosing CD. The IgA tTG-alone strategy was the least costly (US$22 per person), but also the least accurate. IgA tTG then HLA then oesophagogastroduodenoscopy (OGD) cost US$2233 per additional correctly diagnosed case. IgA tTG followed by further IgA testing and then OGD biopsy was US$32,605 per additional correctly diagnosed case. OGD with biopsy alone had an ICER of > US$1M. Chang and Green201 developed a decision tree to evaluate the cost of genetic testing before serological screening among relatives of people with CD. The cost of initial screening with IgA tTG of relatives was US$434 per person. Genetic screening before serological screening of relatives was more costly, at US$750 per person. In terms of cost per correct diagnosis of CD, it would cost US$2668 per case in the initial IgA tTG branch, and US$4422 per case in the HLA branch. The incremental cost per additional case of CD diagnosed would be approximately US$449,000 for genetic testing when compared with IgA tTG alone.

Hershcovici et al. 202 developed a Markov model to estimate whether or not mass screening using serological tests followed by biopsy is cost-effective compared with no screening. The ICER of screening compared with no screening was US$48,960 per QALY gained. However, at a US$50,000 threshold, the probability of being cost-effective was 60%. Mohseninejad et al. 194 conducted a cost-effectiveness analysis of targeted screening for CD among IBS patients and compared serological screening with no screening. They concluded that screening patients with confirmed IBS symptoms for CD is cost-effective (ICER of €6200).

Park et al. 203 developed a Markov model to determine the cost-effectiveness of universal serological screening to prevent non-traumatic hip and vertebral fractures among patients with CD. They found that universal screening, compared with standard care (i.e. screening only symptomatic or at-risk patients), is not cost-effective. Standard care was associated with lower costs (US$8472) and more QALYs (25.515). Yang et al. 206 conducted a cost-effectiveness analysis of routine duodenal biopsy for CD during endoscopy for gastro-oesophageal reflux, compared with no biopsy, using a lifetime decision tree. They found that performing a biopsy to detect CD among patients with gastro-oesophageal reflux disease is not cost-effective (ICER of US$121,875 per QALY gained). Broide et al. 207 built a Markov model to determine the cost-effectiveness of routine duodenal biopsy to detect CD among patients with IDA. The intervention (biopsy during an OGD in all patients with IDA, irrespective of serological test results) resulted in 19.888 QALYs gained and an average cost of US$218.10, dominating the comparator (i.e. performing biopsy only in patients with a positive serological test), which had a gain of 19.887 QALYs at an average cost of US$234.17.

In 2015, NICE conducted a cost-effectiveness analysis using a decision tree followed by a Markov model to estimate which serological test is the most appropriate to diagnose CD among both children and adults. 193 For adults, the most effective strategy was the most sensitive: considering people serologically positive if they are positive on either IgA tTG or IgA EMA (17.6004 total QALYs). However, the ICER was £173,484 per QALY gained. The incremental analysis suggested that most benefits could be achieved at a lower cost by a strategy that tests IgA tTG in all people and reserves IgA EMA to classify cases in which IgA tTG results are weakly positive. All other strategies were strongly dominated. For children, the most effective strategy was IgA tTG plus IgA EMA plus HLA (a combination of serological tests for IgA tTG and IgA EMA and HLA-DQ2/-DQ8 genotyping), with a total QALY gain of 21.3823. The ICER of this strategy was £33,800 per QALY gained, compared with the next cheapest non-dominated option. NICE also conducted a health economic evaluation of active case-finding strategies. It found that screening first-degree relatives of people with CD was cost-effective among adults and children, that screening people with type 1 diabetes was cost-effective among adults and potentially cost-effective among children, and that screening those with autoimmune thyroid disease was not cost-effective among adults or children.

Implications for cost-effectiveness modelling in coeliac disease

A Markov model, rather than a decision tree, is most appropriate for modelling a long-term health-care condition such as CD, as comorbidities, including non-Hodgkin lymphoma (NHL), may not develop for a long time after the initial screening. 208,209 A decision tree is an appropriate structure to model the initial set of testing strategies, but must be combined with a Markov model to capture long-term impacts. 210,211 Only three of the Markov models considered mass screening, rather than screening focused on patients with fractures, IDA, IBS or Down syndrome. Shamir et al. 198 used states for CD undiagnosed, CD diagnosed on gluten-free diet and CD diagnosed but off gluten-free diet, while Hershcovici et al. 202 considered states related to whether patients were diagnosed, undiagnosed, adhered to a gluten-free diet, had IDA or had IBS. The NICE 2015 Markov model was the only model in the UK setting and its Markov model included states only for health conditions with a strong impact on costs and effects. 193 They used CD on gluten-free diet and CD not on gluten-free diet as primary states, with different levels of IDA, IBS and other symptoms in these states. They additionally included subfertility, osteoporosis, NHL and other cancers as health states. As the NICE model considered more events and had a richer evidence base than Shamir et al. 198 or Hershcovici et al. ,202 it was taken as the starting point for our model.

Model structure

As explained previously, we adopted a decision tree followed by a Markov model, as this can fully describe testing strategies and model long-term consequences of CD, and we largely follow the evidence-based model of NICE 2015. 193,211 Initial prevalence of CD is determined by a selected pre-test probability of CD, representing the target population for screening, and this drives the maximum number of true positives (i.e. diagnosed CD) that our testing strategy can identify. The decision tree captures the sensitivity and specificity of serological and genetic tests from Chapters 5 and 6, respectively, along with costs and impacts on health of these tests and the confirmatory biopsy. The Markov model captures the costs and health impacts of having diagnosed or undiagnosed CD, and thus measures the benefit of initially detecting CD with the decision tree. Modifications were made to the Markov model used by NICE following discussion with our clinical and patient advisers, and the results of our CPRD analysis of risk factors for CD (see Chapter 4); these are described in more in Structure of the Markov model.

Structure of the decision tree

The decision tree in which various serological and genetic test combinations are included is illustrated in Figure 12. Patients enter the tree either with CD (CD+) or without CD (CD–) and then follow one of three branches representing the three strategies: screening by serological testing, screening by serological and genetic testing, or no screening (representing standard of care). If a strategy combines serological and genetic testing, the second test is used only if post-test probability following the first was < 90% (see Chapter 7). In each case in which a person is tested, a confirmatory biopsy is used only if the post-test probability of having CD is < 90%. This does not apply to the combination of IgA tTG and IgA EMA, which were based on the results of joint testing, reported in Chapter 5. The model assumes that biopsy is perfectly accurate.

FIGURE 12.

Decision tree representing serological and genetic testing among patients with suspected CD. A test is judged to be positive if a patient’s post-test probability of having CD is > 90%; confirmatory biopsy is applied to all strategies if the post-test probability is below this threshold.

At the end of the decision tree, patients have either a positive or a negative result for CD or remain undiagnosed if they followed the no-screening strategy or had a false-negative test. If they are diagnosed with CD through the testing process, they would be advised to follow a gluten-free diet. It is assumed that, beyond the initial costs and disutility associated with testing, patients without CD who are tested will not have different costs or health outcomes from non-CD patients who were not screened. As a result, we model only the long-term consequences of patients who have diagnosed or undiagnosed CD. Costs and effects associated with false-positive CD patients are scaled by the ratio of false positives to true positives and false negatives (i.e. the number of CD patients who enter the cohort Markov model), so they are on the same scale as the diagnosed/undiagnosed CD cohort.

Structure of the Markov model

The decision tree is followed by a lifetime time-inhomogeneous discrete-time Markov model. 208,212 In this model, health states are assumed to be mutually exclusive. In other words, a patient can be in only one health state at a time. Costs and utilities are assigned to each health state, and are accrued for each year spent in that state. As in the NICE 2015 model,193 our model has annual cycles and a lifetime horizon.

Following the NICE model, complications from biopsy are not modelled, thus implicitly assuming that there is no elevated risk of major complications or death from biopsy, although a small disutility is included to represent the procedure and waiting time for diagnosis, during which an undiagnosed CD patient is less likely to be on a gluten-free diet. This assumption is supported by a 2018 study213 that looked at 13,233 patients undergoing outpatient OGD with biopsy over 5 years in the USA. The authors concluded that no patient was admitted because of complications that could be ascribed to conscious sedation, upper GI endoscopic access or mucosal biopsy and that these data confirm that OGD biopsy is safe. 213

Unlike the NICE 2015 model,193 our model does not distinguish between people with diagnosed CD following a gluten-free diet and people diagnosed with CD not following a gluten-free diet. We instead modelled CD patients as being diagnosed or undiagnosed, with different rates of complications in these states due to different levels of adherence to a gluten-free diet. This modelling choice is because of methodological challenges associated with assessing dietary adherence and a lack of reliable data on the health impacts of following a gluten-free diet or not. Reported gluten-free diet adherence rates vary substantially depending on how adherence is measured and the population studied. 214 Adherence to a gluten-free diet is affected by age at diagnosis, type and severity of symptoms, quality of counselling, mental health, local or societal support levels, and the cost and availability of gluten-free foods. 215 Biopsy is the gold standard when assessing adherence, but the procedure is invasive and costly. For this reason, various other methods are used including validated and non-validated questionnaires, serology, faecal or urine tests, dietitian’s assessment, interviews and patient-reported adherence. 215 Identifying and assigning an average adherence rate to a cohort of individuals is therefore very difficult. Even if this were possible, studies informing the risk of developing complications among those on a gluten-free diet, compared with those who are not, are based on cohorts of people with unknown, and presumably mixed, adherence.

A further distinction from the NICE model193 is that we do not include subfertility as a separate state. Our CPRD analysis found that the prevalence of subfertility among women is the same regardless of CD status; this is supported by a growing literature that there is no association between CD and subfertility. 216–219

Iron-deficiency anaemia was identified as important in our CPRD analysis, but, in line with NICE, was not modelled as a separate state as our patient and clinical advice was that costs were mostly for over-the-counter medications, and therefore not included from an NHS perspective. The impact of IDA was also captured in utility values in diagnosed and undiagnosed states, aligned with our modelling of the efficacy of a gluten-free diet, as major CD utility studies included large proportions of people with IDA.

Further risk factors of CD identified by the diagnostic indicator review (see Chapter 3) and the CPRD analysis (see Chapter 4) (1) are captured as general symptoms (e.g. fatigue, GI symptoms, IBS, mood disorders), (2) cannot be affected by a gluten-free diet and are thus unaffected by diagnosis (e.g. Down syndrome or having a first-degree relative with CD) or (3) are either rare or not plausibly caused by CD (e.g. thyroid disorders, Turner syndrome, type 1 diabetes).

Our final Markov model is illustrated in Figure 13.

FIGURE 13.

Markov model structure.

The initial distribution of patients between diagnosed and undiagnosed CD states is determined by the decision tree. People can transition to death from any state in the model. Patients have an annual probability of transitioning from no complications to health states representing osteoporosis, NHL and death. Patients can also be in any of these health states at diagnosis. As in the NICE model,193 and to avoid modelling multiple comorbidities on which CPRD evidence is limited, patients who develop osteoporosis can transition to NHL or death only, and no distinction is made between NHL patients with and NHL patients without prior osteoporosis. Patients who develop NHL are assumed to stay in this health state until death (i.e. they do not return to the less severe osteoporosis state). Patients with undiagnosed CD transition through the health states and can be diagnosed at any point, with a proportion receiving biopsy.

Input parameters

In this section we describe the parameters of the decision tree and Markov model. These are informed by the DTA meta-analysis (see Chapters 5 and 6), the literature review of coeliac screening cost-effectiveness models, UK CPRD analyses, estimates used in the NICE 2015 model193 or targeted literature searches. A summary of all parameters with evidence sources is shown in Table 18.

Prevalence of osteoporosis, non-Hodgkin lymphoma and iron-deficiency anaemia in a population with coeliac disease

The prevalence rates of NHL, osteoporosis and IDA were estimated using the CPRD Aurum data of patients with a CD diagnosis, linked to HES, and these rates were assumed to be age and gender dependent. Records of complications were identified in HES and the CPRD using the International Statistical Classification of Diseases and Related Health Problems, Tenth Revision (ICD-10), and Aurum medical codes (full details of the analysis and results are in tables S12–S16 in Report Supplementary Material 1). The initial proportions of people with each complication among newly diagnosed and undiagnosed CD patients in the Markov model are assumed to follow these prevalences. It is a limitation that CPRD data cannot generate prevalence rates of NHL, osteoporosis and IDA in those with undiagnosed CD; however, rates of developing each of these complications are modelled to be different from those in patients with a CD diagnosis (see Risk of developing osteoporosis, iron-deficiency anaemia and non-Hodgkin lymphoma). Patients without these complications begin in the no-complications state. IDA is modelled independently of NHL and osteoporosis, so an independent beta distribution is used to present the initial probability of having IDA. Although NHL and osteoporosis are separate states in the Markov model, the prevalence estimates are not mutually exclusive. The prevalence of NHL is low (< 2%), so overlap with osteoporosis is negligible and each prevalence is modelled by an independent beta distribution. Appendix 26, Table 63, includes estimates of the prevalence of each of these complications in a mixed-gender cohort of > 48,046 people with a CD diagnosis recorded in CPRD Aurum. This is stratified into men and women in Appendix 26, Tables 64 and 65, respectively. The adult men and adult women analyses use the stratified estimates whereas the children analyses use the mixed-gender estimates.

Percentage receiving biopsy

The percentage of patients with a delayed diagnosis assumed to require a confirmatory biopsy is 70%, based on clinical advice and in accordance with recent research indicating high diagnostic accuracy of tests alone. 70,89,220,221 This percentage is varied within a 95% credible interval (CrI) of 60% to 80%. The 60% lower bound is based on the findings of a retrospective study of 270 adult patients. 89 This study found that, at an IgA tTG antibody cut-off point of > 45 U/ml, the PPV for CD was 100%, and that 40% of cases were above this cut-off point.

Diagnostic accuracy of tests

The diagnostic accuracy parameters for IgA tTG, IgA EMA and IgA tTG plus IgA EMA were informed by the DTA meta-analysis (see Chapter 5). The diagnostic accuracy of HLA tests was informed by the genetic test accuracy meta-analysis (see Chapter 6). In the serological followed by HLA testing and HLA followed by serological testing combinations, the second test was used only with patients whose post-(first)-test probability was < 90%, aligning with the patient survey result that this was a threshold above which they would be willing to go on a gluten-free diet (see Chapter 7). Estimates of the test accuracy of serological tests were assumed to be independent of HLA test accuracy. All tests are followed by a confirmatory biopsy among those with a post-test probability that is still < 90%.

Costs

The model includes costs associated with diagnostic tests, biopsy, annual health-care resource use costs among diagnosed CD patients, and annual costs of osteoporosis and NHL. A price year of 2018/19 was assumed, aligning with NHS reference costs. 222 The cost of IgA tTG, IgA EMA and HLA tests were sourced from correspondence with hospitals and testing centres affiliated with the research team. The cost of biopsy was informed by NHS reference costs.

Along with test and biopsy costs, to get to a diagnosis, patients would also have nurse, GP and/or specialist consultations. Our clinical and patient team advised on the likely pathway to diagnosis. We assumed that a person’s first appointment is with either a nurse or a GP, at a cost of £10.85 or £39.23, respectively, with average £25.04, reported by the Personal Social Services Research Unit. 223 This is assumed to be followed by phlebotomy at a cost of £4, following NHS reference costs 2018/19, and then by a further GP appointment. 222 Following that, we assumed patients have, on average, one gastroenterology outpatient appointment costing £141 and two dietitian appointments at a cost of £85 each before diagnosis. 223 This gives a total cost of £379.27 (£25.04 + £4 + £39.23 + £141 + £170) for a patient to reach diagnosis. This cost is applied to both true positives and false positives in the model.

Annual health-care resource use costs for patients with CD were based on analysis of CPRD data, which updated the 2012 analysis of Violato et al. ;50 full details can be found in Report Supplementary Material 1, chapter 8. Following this earlier analysis, we quantified the volume of health-care resources in terms of primary care consultations, tests, referrals to outpatient hospital care and prescriptions used by individuals diagnosed with CD up to 10 years before and after diagnosis in a UK primary care setting. The volumes of resources, along with unit costs, have been used to estimate overall medical costs associated with a true-positive or false-negative diagnosis. A limitation of this analysis is that referrals to gastroenterology are likely to include only adult referrals, as children referred to gastroenterology are normally recorded under paediatrics. The unit costs used and a breakdown of costs are provided in tables S17 and S18 (see Report Supplementary Material 1).

The model assumed that patients falsely diagnosed with CD would follow a gluten-free diet for a limited time before stopping. Although such patients are not included in the Markov model, we count this short-term impact in the total costs (and total effects). Following input from clinicians and patient advisors on the likely time taken to decide that the diet is not of benefit, we assumed that this time would be 1 year with a log-normal distribution to reflect skew towards longer durations. The only additional cost incurred by people with a false-positive diagnosis of CD was the cost of non-NHS-reimbursed gluten-free products. We assumed that no other costs are incurred by people with a true-negative diagnosis.

The cost associated with NHL was based on a study by Wang et al. ,224 which followed all patients newly diagnosed with diffuse large B-cell lymphoma in the UK’s population-based Haematological Malignancy Research Network from 2007 to 2013 (n = 271). Mapped treatment pathways, alongside cost information derived from the National Tariff 2013/14,225 were incorporated into a patient-level simulation model to reflect heterogeneities in patient characteristics and treatment options. As treatment for NHL is usually complete within 5 years, most costs were estimated to fall within 5 years. The mean cost per patient was £18,096 (95% CI £18,078 to £18,114) over a 5-year time horizon and £18,396 (95% CI £18,377 to £18,415) over a lifetime time horizon. Using lifetime costs is problematic for a cohort Markov model as some patients may develop NHL at an advanced age and not have many additional years to incur costs. However, the NHL model factors in different survival times when estimating the treatment costs, so patients would incur less cost if they died within a shorter time. We therefore assumed a one-off cost of £18,396 (95% CI £18,377 to £18,415) with a log-normal distribution, relating to the lifetime cost estimated by Wang et al. 224

For the cost associated with osteoporosis, a similar approach to that of the NICE model193 was taken, with the cost being equal to the annual probability of hip, vertebral and wrist fracture multiplied by the cost of each type of fracture. Probabilities were estimated from Curtis et al. ,226 who used CPRD data to look at trends in fracture rates in the UK over a 24-year period from 1988 to 2012. 226 The rates of hip, clinical vertebral and carpus (wrist) fractures found among those aged 18–49 and ≥ 50 years are shown in Appendix 27, Table 66.

For the cost associated with hip fractures, Leal et al. 227 estimated hospital costs up to 2 years post fracture using a cohort of 33,152 patients aged > 60 years admitted with a hip fracture in a UK region between 2003 and 2013 who were identified from hospital records and followed until death or administrative censoring. The mean censor-adjusted 2-year hospital cost after index hip fracture was £16,302 (95% CI £16,097 to £16,515). 227 For costs associated with vertebral and wrist fractures, we used estimates from Dolan and Torgerson,228 who estimated the cost per annum to be £468 per wrist fracture and £479 per vertebral fracture. These were based on health and social care costs of fractures among women aged ≥ 50 years in the UK using a variety of data sources including published estimates, a survey of resource use among fracture patients before and after hip fracture and a case–control study using the General Practice Research Database. All costs are inflated to 2021 prices using the Composite Price Index published by the Office for National Statistics. 229 The overall assumed annual cost of osteoporosis is shown in Appendix 27, Table 67. Uncertainty was included by fitting gamma distributions to the cost of hip, vertebral and wrist fractures, with the SE set to one-tenth of the mean. 212

The cost of IDA is only that related to over-the-counter iron tablets, which incur no NHS cost.

Utilities

The model includes health-state utility values associated with undiagnosed CD and diagnosed CD, and disutilities associated with osteoporosis and NHL. As explained, we assume that some proportion of diagnosed CD patients will adhere to a gluten-free diet, so utilities for diagnosed CD should represent this mixed population. We assumed that the health-state utilities for osteoporosis and NHL are the same as for the general (i.e. not CD-specific) population. The model also includes disutilities associated with the biopsy procedure, with the waiting time associated with a biopsy and with a lifetime false-positive diagnosis.

Quality-adjusted life-year norms for the UK are used to reflect background age-specific quality of life, which decreases naturally with age. All state QALYs are multiplied by their norms to reflect this age reduction. The values used, sourced from Janssen and Szende,230 are shown in Appendix 28, Table 68.

The utility values for undiagnosed CD and diagnosed CD were sourced from Violato and Gray. 231 The values are based on 1584 EuroQol-5 Dimensions, three-level version (EQ-5D-3L), questionnaires retrospectively completed by people with CD in the UK. Participants rated their health using the validated questionnaire before and after their diagnosis. Not all of the diagnosed cohort adhered perfectly to a gluten-free diet, with 90.8% reporting adherence all of the time, 8.3% most of the time and 0.9% some/little/none of the time. The estimated utility values for diagnosed CD and undiagnosed CD were 0.65 (95% CI 0.63 to 0.67) and 0.85 (95% CI 0.84 to 0.86), respectively, in the total sample. In the subgroup aged < 18 years, the values were 0.57 (95% CI 0.51 to 0.64) for undiagnosed CD and 0.88 (95% CI 0.85 to 0.92) for diagnosed CD on gluten-free diet. An alternative source for utility in diagnosed and undiagnosed CD was identified in Casellas et al. ,232 who carried out a prospective study of 163 CD patients on a gluten-free diet and 177 newly diagnosed CD patients on a normal diet from seven hospitals in different areas of Spain, with all completing the EuroQol-5 Dimensions (EQ-5D). A similar pattern was found to our selected source, with the median EQ-5D value being 0.93 (95% CI 0.85 to 1.0) among gluten-free diet patients and 0.72 (95% CI 0.58 to 1.0) in the normal diet group.

For osteoporosis, a similar approach is taken as with costs, with total disutility being equal to the annual probability of hip, vertebral and wrist fractures multiplied by the disutility associated with each type of fracture. The results are presented in Appendix 28, Table 69. These disutilities are estimated from the Si et al. 233 study, which was a systematic review and meta-analysis of 62 studies reporting utility-based quality of life for osteoporosis-related conditions. Most of the studies used EQ-5D health-state utility values, followed by a visual analogue scale. Si et al. 233 calculated utilities for the first year after hip, vertebral and wrist fractures as 0.59 (95% CI 0.54 to 0.65), 0.55 (95% CI 0.50 to 0.60) and 0.78 (95% CI 0.72 to 0.84), respectively. These utilities are subtracted from the UK population norm of 0.817 for the group aged 55–64 years. Uncertainty was modelled by moment-matching beta distributions to the hip, vertebral and wrist fracture disutilities, with the SE for each set to one-tenth of their means. 212

The utility associated with NHL is based on an observational cross-sectional comparative study carried out by Fargier et al. 234 in three French teaching hospitals. In this study, 73 patients with follicular lymphoma were receiving either subcutaneous or intravenous rituximab maintenance monotherapy (73% as first-line treatment, 21% as second-line treatment and 6% as third-line treatment). Health-related quality of life was evaluated using the French version of the self-administered EQ-5D-3L questionnaire. The aim was to compare the impact of using subcutaneous rituximab with that of using intravenous rituximab as maintenance therapy. The mean EQ-5D score was 0.8 for intravenous rituximab and 0.7 for subcutaneous rituximab. 234 The average of these two values is subtracted from the French EQ-5D population norm for people in the 55–64 years age group (0.836230), giving a disutility of 0.086. The upper and lower bounds, of 0.8 (disutility 0.036) and 0.7 (disutility 0.136) respectively, are used to represent uncertainty in this parameter using a uniform distribution.

The disutility associated with a biopsy in adults is assumed to be equal to 1 quality-adjusted life-day. This is the same assumption made in the NICE model193 and aims to account for any anxiety associated with the biopsy as well as potential side effects. For children, the loss of 2 quality-adjusted life-days is assumed because of the need for general anaesthetic. The disutility associated with waiting for a biopsy is assumed to be equal to the difference between the utility of being diagnosed with CD and following a gluten-free diet and the utility of having undiagnosed CD. This disutility is assumed to last for 6 weeks in accordance with average waiting times for a biopsy.

In line with the NICE model,193 we assume an annual disutility for patients falsely diagnosed with CD of 0.009. This was based on an estimated change in social function score on the Short Form questionnaire-36 items of –8.3 (SE 3.83) and a mapping for this score to utility scale of 0.0011 (SE 0.0002). We modelled these uncertain factors with normal distributions and the mean of their product was –0.009 (SE 0.004). Note that we assume that any disutility from a gluten-free diet in the true CD population is captured by the Violato and Gray138 2019 survey.

For IDA, we assume that disutility is already captured in the utility values for undiagnosed CD and diagnosed CD sourced from Violato and Gray,138 as they reported that 64.8% of their population had anaemia prior to diagnosis and 14.5% after.

Probability of late detection of coeliac disease after false-negative diagnosis

All patients in the undiagnosed CD state have an annual probability of transitioning to the diagnosed CD state. Violato and Gray138 asked patients about the duration of their symptoms prior to diagnosis and found that the average duration was 12.8 years (SD 15.3 years), based on 1584 completed questionnaires. The average duration across the sample aged ‘< 18’ was shorter, at 3.3 years (SD 3.7 years). As our adult cohort begins at age 18 years, we need the time to diagnosis only as an adult. We use these figures to calculate this average duration in patients aged > 18 years of 10.9 [(12.8 – 3.3) × proportion aged < 18 years) × (proportion aged > 18 years ÷ 1584)]. Although our children cohort begins age 10 years, it was not reported what proportion of patients were under the age of 10 years, so the ‘< 18’ years figure had to be used. We model the average duration as a log-normal distribution moment matched to the reported mean and SE calculated using SD and sample size. Assuming a constant rate, λ, of diagnosis (i.e. an exponential survival model), the rate is the inverse of the sampled duration. This is used to calculate an annual probability of diagnosis by year of 1 – e^–λt.

Risk of developing osteoporosis, iron-deficiency anaemia and non-Hodgkin lymphoma

Patients who are diagnosed have a lower risk of developing long-term complications than patients who are undiagnosed, owing to the greater likelihood of following a gluten-free diet and receiving improved care and medical advice. To inform the risk of developing long-term complications, we use rates in non-CD patients estimated using CPRD data (full details are in Report Supplementary Material 1, chapter 8), differences in rates between this general population and patients who are diagnosed with CD, and differences in rates between the general population and those with undiagnosed CD. Evidence on these differences primarily comes from the NICE 2015 model. 193 In each case, the literature was searched for more recent or higher-quality studies, but none was identified.

The NICE model193 based the risk of developing osteoporosis when following a gluten-free diet on Swedish data from a study by Ludvigsson et al. ,235 which comprised > 13,000 people diagnosed with CD and compared this group with a matched control group. This study did not distinguish between patients diagnosed with CD and on a gluten-free diet and those diagnosed but not on a gluten-free diet, so it represents our merged diagnosed CD state. The study reported an odds ratio of fractures of any type in the diagnosed CD population, compared with the controls, of 1.40 (95% CI 1.30 to 1.50). The NICE model193 based the risk of developing osteoporosis when not following a gluten-free diet on a study by Godfrey et al. ,236 who tested blood samples for CD and compared the medical records of 129 US people with not diagnosed with CD, but with positive coeliac serology, with those of seronegative individuals. They found an odds ratio for developing osteoporosis of 2.59 (95% CI 1.32 to 5.09) in the undiagnosed CD group, compared with those without CD, and this is the ratio we used in our model. This finding contradicts that of Ludvigsson et al. ,235 who found no association between date of diagnosis, used as a proxy for initiating gluten-free diet, and risk of osteoporosis. However, Ludvigsson et al. 235 did not have data on adherence to gluten-free diet in either diagnosed or undiagnosed states, and so is confounded in comparison with Godfrey et al. 236

The NICE model193 based the probability of patients developing NHL from CD when not on a gluten-free diet on a study of 1968 previously undiagnosed patients who were subsequently diagnosed with CD at 20 Italian gastroenterology referral centres. 237 The incidence ratio of NHL in the population with undiagnosed CD compared with the general population was found to be 4.7 (95% CI 2.9 to 7.3). The incidence ratio of NHL in the population with diagnosed CD (3.28, 95% CI 1.49 to 6.28), relative to the general population, was based on an analysis of a cohort of patients with diagnosed CD from a linked statistical database of hospital and mortality data in an area in southern England. 238

Normal distributions were used to model the log-odds ratios and incidence ratios of osteoporosis and NHL, respectively (i.e. log-normal distributions). We added these log ratios to the log of rates of developing long-term conditions among control (i.e. non-CD) patients from our CPRD analysis (Table 16). Rates were calculated from the number of outcome events divided by the person-time at risk stratified by decile of age at index date (date of CD diagnosis in matched case) for osteoporosis and IDA. For NHL we grouped patients into those aged < 18 years and those aged ≥ 18 years because of limited sample size. Time at risk was calculated from a patient’s index date to date of outcome event, date of death or the end date of HES coverage. The rates were used to calculate the annual probabilities of developing the conditions, with uncertainty based on a log-normal distribution of the rates mapped to a log scale, consistent with the ratios for osteoporosis and NHL.

TABLE 16 - Age-stratified rates per 1000 (95% CI) among non-CD controls from CPRD analysis

Age group (years)	Rate (95% CI)
	Mixed gender	Men	Women
NHL
< 18	0.03 (0.01 to 0.07)	0.02 (0.01 to 0.09)	0.04 (0.01 to 0.12)
≥ 18	0.28 (0.25 to 0.32)	0.30 (0.26 to 0.35)	0.27 (0.23 to 0.32)
Osteoporosis
0–9	0.02 (0.01 to 0.09)	0.02 (0.00 to 0.15)	0.02 (0.00 to 0.17)
10–19	0.03 (0.01 to 0.10)	0.04 (0.01 to 0.15)	0.03 (0.00 to 0.18)
20–29	0.16 (0.11 to 0.25)	0.10 (0.05 to 0.21)	0.24 (0.14 to 0.41)
30–39	0.42 (0.34 to 0.52)	0.17 (0.10 to 0.27)	0.66 (0.52 to 0.83)
40–49	1.40 (1.20 to 1.50)	0.55 (0.43 to 0.70)	2.20 (1.90 to 2.50)
50–59	3.90 (3.60 to 4.20)	1.20 (1.00 to 1.50)	6.60 (6.10 to 7.10)
60–69	7.50 (7.00 to 7.90)	2.70 (2.30 to 3.10)	12.00 (11.00 to 13.00)
70–79	13.00 (12.00 to 14.00)	4.80 (4.10 to 5.60)	20.00 (19.00 to 22.00)
80–89	21.00 (19.00 to 23.00)	9.60 (7.50 to 12.00)	29.00 (26.00 to 33.00)
90–99	27.00 (19.00 to 38.00)	5.90 (1.50 to 24.00)	36.00 (25.00 to 51.00)

Assuming that events occur at a constant rate r per unit of time t, the probability that an event will occur during time t is given by p = 1 –e^rt, equivalent to an exponential survival distribution. 239

Mortality

Mortality in the model is assumed to be equal to general population mortality, with excess mortality assigned to those patients who develop NHL or osteoporosis. Mortality in the general population came from the 2019 Office for National Statistics life tables. 240

Mortality from NHL is based on data from the Office for National Statistics on adults diagnosed with NHL between 2013 and 2017, which show that 1-year survival is 79.4% (95% CI 79.0% to 79.7%), 5-year survival is 65.6% (95% CI 65.0% to 66.3%) and 10-year survival is 54.7% (95% CI 53.2% to 56.3%). 241 The Markov model cannot allow the probability of NHL mortality to change over time, as patients may enter the NHL state at different times. These estimates are therefore used to calculate a single annual probability capturing average survival.

This synthesis was performed in the Bayesian software OpenBUGS. 242 We first converted observed probabilities to hazards under the assumption of constant hazards, using the following equation:

⁽¹⁾ P ( t ) = e − h t t .

These hazards, h_t, were then converted to a natural log scale (log-hazard y_t) and the upper and lower bounds of the 95% reference range were used to estimate the SE (SE_t) by assuming normality via the central limit theorem. We then treated them as independent observations of a common log-hazard λ:

⁽²⁾ y t ∼ N ( λ , S E t ) ,

with a vague prior distribution:

⁽³⁾ λ ∼ N ( 0 , σ 2 = 1000 ) .

We fit this model in OpenBUGS using two chains, 50,000 burn-in iterations and 10,000 sampling iterations. This gave a posterior mean for λ of –2.092 (SE 0.006378). These are used in the economic model by sampling from a normal distribution with mean –2.092 and SD 0.006378.

This has the limitation that the hazards are changing over time (Table 17). This overestimates the hazard in later years and underestimates it in earlier years, but is the closest approximation possible. This is a necessary assumption for a cohort Markov model and, owing to the rarity of NHL, is unlikely to affect the conclusions.

TABLE 17 - Office for National Statistics data on the probability of survival following a diagnosis of NHL

Year	Probability survival (95% reference range)	Hazard (95% reference range)	Log-hazard (SE)
1	0.794 (0.79 to 0.797)	0.231 (0.227 to 0.236)	–1.467 (0.00973)
5	0.656 (0.650 to 0.663)	0.084 (0.632 to 0.657)	–2.473 (0.009661)
10	0.547 (0.532 to 0.653)	0.06 (1.08 to 1.154)	–2.808 (0.016893)

For mortality from hip fracture, we used the study by Klop et al. ,243 which used CPRD data to examine all-cause and cause-specific mortality post hip fracture among 31,495 patients with a first hip fracture, compared with general population mortality, from 2000 to 2010. The mean age was 74.1 ± 14.8 years for male hip fracture patients and 80.5 ± 10.5 years for female hip fracture patients. During the total study period, the hazard ratio for 1-year all-cause mortality was 3.5 times (95% CI 3.28 to 3.74 times) greater for male hip fracture patients than for control subjects after adjustment for age, comorbidities, medication use and lifestyle factors. For female patients, this risk was 2.4-fold (95% CI 2.31- to 2.50-fold) greater than that of controls.

Table of model inputs

A summary of all input parameters, assumed distributions and evidence sources is provided in Table 18.

TABLE 18 - Model parameters and data sources

BNF, British National Formulary; FN, false negative; OR, odds ratio; TP, true positive.

SE assumed to be one-tenth of the mean.

Parameter	Estimate	Distribution	Source
Decision tree and Markov initial probabilities
Probability of late detection of CD after FN diagnosis	Mean time (years) to diagnosis: Adults – 10.93 (SD 13.10) Children – 3.34 (SD 3.71)	Normal	Violato et al.231
Probability that confirmatory biopsy needed, mean (minimum, maximum)	0.7 (0.6, 0.8)	Uniform	Studies on percentage needing biopsy.70,89,220,221 The 60% lower bound is based on the findings of a retrospective analysis of 270 adult patients89
Proportion with osteoporosis, NHL or IDA at diagnosis	Age dependent (see table S13 in Report Supplementary Material 1)	Independent beta for each condition	CPRD age- and gender-stratified prevalence
Diagnostic accuracy of tests from Chapters 5 and 6
Costs
Unit cost of IgA EMA test	£14.92 (SE £1.87)	Gamma	Personal communication with various laboratories offering CD testing in the UK
Unit cost of IgA tTG test	£10.77 (SE £2.15)	Gamma	Personal communication with various laboratories offering CD testing in the UK
Unit cost of HLA test	£122.34 (SE £24.47)	Gamma	Personal communication with various laboratories offering CD testing in the UK
Endoscopic biopsy, adults: diagnostic endoscopic upper GI tract procedures with biopsy, aged ≥ 19 years	£530	Fixed	NHS reference costs (2018/19)222 FE21Z
Endoscopic biopsy, children: endoscopic or intermediate, upper GI tract procedures, aged between 5 and 18 years	£823	Fixed	NHS reference costs (2018/19)222 FE23C
Other diagnosis costs (e.g. nurse, GP, gastroenterologist, dietitian appointments)	£379.27 (SE £37.93)	Gamma	NHS reference costs (2018/19),222 Unit Costs of Health and Social Care 2020223
Annual health-care resource use costs by diagnosis – TP	Adults: £757 (SE £5.3)a Children: £452 (SE £20.6)a	Gamma	CPRD analysis based on update of Violato et al.50
Annual health-care resource use costs by diagnosis – FN	Adults: £421 (SE £3.34)a Children: £248 (SE £4.97)a	Gamma	CPRD analysis based on update of Violato et al.50
Treatment for NHL	£18,396 (95% CI £18,377 to £18,415)	Log-normal	Wang et al.224 One-off lifetime cost of treatment
Osteoporosis	£39.04 (SE £0.27)	Weighted average of gamma distributions (see Appendix 27, Table 67)	Curtis et al.,226 Leal et al.227 and Dolan and Torgerson228
Anaemia	£0 under NHS (£17.89 if counting over-the-counter expenses)	Fixed	Cost of over-the-counter iron tablets. Costed as 200 mg of ferrous sulfate tablet twice per day; 1000-tablet pack = £17.89, BNF 2020244
Utilities
Diagnosed CD	Whole population: 0.85 (95% CI 0.84 to 0.86) Aged < 18 years: 0.88	Beta	Violato et al 2019231
Undiagnosed CD	Whole population: 0.65 (95% CI 0.63 to 0.67) Aged < 18 years: 0.57	Beta	Violato et al 2019231
Osteoporosis (annual disutility)	0.0008 (SE 0.000067)	Weighted average of beta distributions (see Appendix 27, Table 69)	Si et al.233 and Curtis et al.226
NHL (annual disutility)	0.086 (95% CI 0.036 to 0.136)	Uniform	Fargier et al.234
Disutility associated with biopsy in adults	–0.003 (95% CI –0.005 to 0)	Triangular	NICE guideline193
Disutility associated with biopsy in children	–0.006 (95% CI –0.010 to 0)	Triangular	NICE guideline193
Disutility associated with waiting for a biopsy	–0.023 (95% CrI –0.0232 to –0.0227)	Combination	Assumed equal to difference between utility diagnosed CD and utility undiagnosed CD for 6 weeks
Disutility associated with false-positive diagnosis	–0.009 (SE 0.004)	Product of a normal distribution with a mean of –8.3 and a SE of 3.83, and a normal distribution with a mean of 0.0011 and a SE of 0.0002	NICE guideline combination of social function and mapping to utility score193
Disutility associated with anaemia	0	Fixed	Assumed already captured in overall health-state utility values
Transition probabilities
Probability of developing osteoporosis with undiagnosed CD	OR vs. general population, 2.59 (95% CI 1.32 to 5.09)	Log-normal on both OR and general population rate	Godfrey et al. 236 CPRD rate for general population
	General population rates in Table 16
Probability of developing osteoporosis with diagnosed CD	OR vs. general population, 1.40 (95% CI 1.30 to 1.50)	Log-normal on both OR and general population rate	Ludvigsson et al. 235 CPRD rate for general population
	General population rates in Table 16
Probability of developing NHL with undiagnosed CD	Incidence ratio vs. general population, 4.7 (95% CI 2.9 to 7.3)	Log-normal on both OR and general population rate	Silano et al. 237 CPRD rate for general population
	General population rates in Table 16, and in tables s14–16 (see Report Supplementary Material 1)
Probability of developing NHL with diagnosed CD	Incidence ratio vs. general population, 3.28 (95% CI 1.49 to 6.28)	Log-normal on both OR and general population rate	Goldacre et al. 238 CPRD rate for general population
	General population rates in Table 16
Mortality probability from osteoporosis	Hazard ratio 3.5 (95% CI 3.28 to 3.74) times greater for male hip fracture patients than for controls and 2.4 (95% CI 2.31 to 2.50) times greater for female patients than for controls	Normal distributions on log-hazard ratios	Klop et al.243
Mortality probability from NHL	Log-hazard 2.092 (SD 0.006378)	Normal	Office for National Statistics 2018241

Analyses

Cost–benefit analysis to identify the optimal screening strategy

The objective of the model was to calculate the expected costs and health outcomes over a patient’s lifetime for all screening strategies. Risk prediction strategies with sensitivity and specificity combinations of 0.5, 0.6, 0.7, 0.8, 0.9 and 1.0 were explored, followed by serological testing alone (IgA tTG alone, IgA tTG plus IgA EMA, and IgA EMA alone), combination of serological testing with genetic testing (HLA before or after the three serological testing strategies, with double-testing only for those with a 90% post-test probability after the first test) and no screening (routine diagnosis only) in adult men, adult women and children. Half-cycle corrections are applied to the Markov model; costs and QALYs were discounted at 3.5%, as recommended by NICE. 245 The analysis was fully probabilistic, with 1000 samples being generated for each population; no deterministic results are presented as the Markov model is non-linear. 246

The total costs and QALYs were calculated for each of these strategies, along with the net benefit at £20,000 per QALY (i.e. total QALYs multiplied by £20,000 per QALY minus the total costs). Incremental costs, QALYs and net benefit were calculated relative to no screening. The probabilistic samples were summarised by their means and Bayesian 95% CrIs. The probability of each decision being cost-effective, compared with no screening, at £20,000 per QALY was calculated by counting the proportion of samples for which the incremental net benefit was positive.

Further investigation was conducted in the testing combinations (e.g. IgA EMA plus HLA or IgA tTG) with the greatest net benefit and probability of being cost-effective for each population. If cost-effectiveness was similar to that of other tests, IgA tTG alone was selected because of its greater availability in UK laboratories. 247 These testing combinations were applied to the risk prediction strategies identified in Table 5. These were strategies with pre-test probabilities for blood test of 1%, 1.5%, 2%, 5%, 10% and 20%, with implied sensitivities (i.e. proportion of those with CD who have the diagnostic indicator combination) and specificities (i.e. proportion of those without CD who would have the diagnostic indicator combination) that can be run through the economic model. The cost-effectiveness of these strategies of interest were compared by total costs, QALYs and net benefits. The probability that each strategy has the greatest net benefit at various willingness-to-pay thresholds was also plotted, which is the cost-effectiveness acceptability curve (CEAC).

Value-of-information analysis to identify further research priorities

Decision uncertainty was quantified using a value-of-information analysis following the best-practice recommendations of the International Society for Pharmacoeconomics and Outcomes Research. 248–250 The total decision uncertainty was quantified using the expected value of perfect information (EVPI), whereas decision uncertainty related to a single parameter, or subset of parameters, was quantified using the expected value of partial perfect information (EVPPI). The ratio of the EVPPI for each parameter to the total EVPI was calculated to represent the relative importance of each parameter. 251 Generalised additive models were used to approximate the EVPPI for each parameter when calculating ratios with the EVPI. 252,253 As the exact values of the EVPPI may not be reliable, the EVPPI for parameters and parameter sets of interest were further estimated using multilevel Monte Carlo (MLMC) methods. 254

The EVPI and the EVPPI are per person and must be scaled to the size of the population of interest. For CD, this is the total population of the UK, or 67,081,000 as of 2020. 255 It was assumed that 79% are aged ≥ 18 years and that 50% are female. 256 The prevalence of CD was taken as 1%. 2 This gives a total population of 263,465 for adult men or adult women, and 140,070 for children.

The EVPI and the EVPPI must also be summed, with discounting at 3.5%, over a ‘technology horizon’, which is the length of time for which the screening strategy recommendations are likely to remain relevant before newer testing techniques emerge. The technology horizon was conservatively set at 10 years. This gives a 10-year discounted total population of 2,267,824 for adult men or adult women, and of 1,205,679 for children.

The total population EVPI over 10 years was calculated. The total population EVPPI was also calculated for all utilities and disutilities, all rates of osteoporosis and NHL, all parameters related to a gluten-free diet (i.e. utility and effects of a gluten-free diet on NHL and osteoporosis rates), sensitivity and specificity of tests used in selected strategies of interest, and the probability of late diagnosis.

Software

The model was implemented in the R statistical programming language; the lower-level C/C± programming language was used for optimisation. This has well-documented advantages of efficiency, transparency and flexibility over the more commonly employed Microsoft Excel^® (Microsoft Corporation). 257 The R package ‘Bayesian cost-effectiveness analysis’ (BCEA) was used to generate CEACs and estimate EVPPI via generalised additive models for individual parameters. The BCEA package was also used to plot the ratios of EVPPI to EVPI for each parameter using the info.rank() function.

Summary of modelling assumptions

A summary of modelling assumptions is provided in Table 19.

TABLE 19 - Key assumptions of the cost-effectiveness model

Assumption	Justification
Biopsy is not associated with any major complications or death	No recent studies give evidence of major complications or death. One recent study of 13,233 patients undergoing outpatient OGD with biopsy over 5 years in the USA concluded that no patient was admitted because of complications.213 A small disutility associated with the procedure and waiting time for diagnosis is assumed
Biopsy is perfectly accurate	In reality, a biopsy will not always provide perfect results; however, it is the current gold standard and people are treated on the basis of their biopsy results. The 2015 NICE health economic analysis193 of screening strategies also made this assumption
It is assumed that the percentage of patients being diagnosed with CD after screening (i.e. initially missed) requiring a confirmatory biopsy before diagnosis is 70%	Based on clinical advice that 30%, with a plausible range, of patients diagnosed with CD do not require a confirmatory biopsy. This is in accordance with recent research indicating high diagnostic accuracy of tests alone70,89,220,221
The only additional cost incurred by people with a false-positive diagnosis of CD is the cost of a gluten-free diet for a short period of time	Patients without CD who are screened for CD are assumed not to have costs or health outcomes that are different from those of other non-CD patients who were not screened for CD. As such, we model only the long-term consequences of patients who have CD (whether diagnosed or undiagnosed)
No further costs are incurred by people with a true-negative diagnosis
NHL is associated with a one-off lifetime cost of £18,396 (95% CI £18,377 to £18,415) with a log-normal distribution	Treatment for NHL is usually complete within 5 years, with most costs estimated to fall within 5 years. The model was used to estimate the cost factors of different survival times when estimating the treatment costs, so patients would incur fewer costs if they died within a shorter time
Annual cost/disutility of osteoporosis is equal to annual probability of hip, vertebral and wrist fractures multiplied by cost/disutility of each type of fracture	This is the approach taken in the NICE model.193 The cost and disutility of osteoporosis mainly relates to the risk of fracture, of which hip, vertebral and wrist are common types
Treatment for IDA among CD patients is by over-the-counter iron tablets only.	This was based on clinical opinion
IDA disutility is already captured in utility values for undiagnosed CD/diagnosed CD sourced from Violato and Gray231	Violato and Gray231 reported that 64.8% of their population had anaemia prior to diagnosis and 14.5% had anaemia after
Mortality in model is equal to general population mortality, with excess mortality assigned to patients who develop NHL or osteoporosis
To get to a diagnosis of CD, a person’s first appointment is with either a nurse or a GP, with a 50% chance of this being with one or the other. This is followed by an appointment with a phlebotomist/nurse and then by another GP appointment. Following that, a patient has, on average, one gastroenterologist and two dietitian appointments before diagnosis	This was the most common route to diagnosis, as agreed by clinicians and patients on the research team

Results

In this section we present all the results of the cost-effectiveness analysis of case-finding strategies for adult men, adult women and children.

Adult men

We first compared a range of combinations of sensitivities and specificities for a hypothetical diagnostic indicator (or indicator combinations) in population of the adult men. Figure 14 plots the incremental net benefit at £20,000 per QALY of each sensitivity/specificity combination of an indicator combined with each serological testing strategy, compared with no screening, among adult men. The central estimates are the mean incremental net benefit, with the upper and lower Crls represented by the upper and lower lines.

FIGURE 14.

Incremental net benefit at £20,000 per QALY, compared with no screening, at each combination of sensitivity (x-axis) and specificity (line colour) among adult men. (a) IgA EMA; (b) IgA tTG plus EMA; (c) IgA tTG; (d) IgA EMA plus HLA; (e) IgA tTG plus EMA plus HLA; (f) IgA tTG plus HLA; (g) HLA plus IgA EMA; (h) HLA plus IgA tTG plus EMA; and (i) HLA plus IgA tTG. Note that thick lines indicate the mean and narrow lines indicate the 95% CrI.

Figure 14 shows that, for strategies using only serological tests (i.e. without HLA), incremental net benefit is positive only when diagnostic indicator sensitivity is > 0.9 or both the specificity and sensitivity are > 0.8. For the serological tests including HLA, the incremental net benefit is positive regardless of the accuracy of the diagnostic indicator, although the lower limit of the 95% CrI shows that there is some uncertainty. The combinations using only the IgA EMA serological test plus HLA (i.e. IgA EMA plus HLA or HLA plus IgA EMA) are more cost-effective than strategies using IgA tTG plus HLA and comparable in cost-effectiveness to tests using all three of IgA EMA, IgA tTG and HLA. HLA being administered before or (as confirmation if post-test probability < 90%) after IgA EMA or IgA tTG plus EMA does not affect cost-effectiveness, but it is most cost-effective if used before IgA tTG.

Figure 15 plots the probability that each combination of diagnostic indicator sensitivity and specificity is cost-effective for each test combination at £20,000 per QALY (i.e. proportion of simulations having a positive incremental net benefit at £20,000 per QALY, compared with no screening). It again shows that the only strategies using serological tests alone that have a probability of > 50% of being cost-effective are those diagnostic indicators with a sensitivity of > 0.9 or both specificity and sensitivity of > 0.8. As expected, the combinations of tests with the highest probability of being cost-effective are those with a specificity of 0.9 and a sensitivity of 0.9 or 1. The serological tests including HLA have a high probability of being cost-effective regardless of the combination of sensitivity and specificity of the diagnostic indicator. Strategies using IgA EMA or IgA tTG alone with HLA have a probability of cost-effectiveness that is similar to that of combinations using all three of IgA EMA, HLA and IgA tTG.

FIGURE 15.

Probability of being cost-effective at £20,000 per QALY, compared with no screening, at each combination of sensitivity (x-axis) and specificity (line colour) among adult men. (a) IgA EMA; (b) IgA tTG plus EMA; (c) IgA tTG; (d) IgA EMA plus HLA; (e) IgA tTG plus EMA plus HLA; (f) IgA tTG plus HLA; (g) HLA plus IgA EMA; (h) HLA plus IgA tTG plus EMA; and (i) HLA plus IgA tTG.

In general, strategies employing IgA tTG and IgA EMA have similar incremental net benefits and probabilities of being cost-effective; given the greater availability of IgA tTG in the UK, we chose this as one of our strategies of interest. 247 Although all strategies employing HLA have higher incremental net benefit and probability of cost-effectiveness, IgA tTG plus HLA performs worst. There is also limited benefit to adding IgA tTG to strategies using IgA EMA and HLA, and limited difference between using IgA EMA or HLA first, so we used IgA EMA followed by HLA for our second testing combination of interest. Again, owing to its wide availability and its having high incremental net benefit and probability of cost-effectiveness, we chose HLA plus IgA tTG as a final test of interest. These selections were combined with the risk prediction strategies of Table 5.

Table 20 shows the estimated costs, QALYs and incremental net benefit, compared with no screening, for each of these strategies of interest. All combinations have positive incremental net benefit compared with no screening. If using only the serological test IgA tTG, with a pre-test probability of 1%, it is the most cost-effective strategy (greatest incremental net benefit at £20,000 per QALY), with the highest costs but also the largest number of QALYs. The net benefits of the IgA EMA plus HLA and the HLA plus IgA tTG strategies are very similar to each other and to those of IgA tTG with a pre-test probability of 1%, and the 95% CrIs are completely overlapping. This indicates that there is little or no difference between these strategies in cost-effectiveness.

TABLE 20 - Costs, QALYs and incremental net benefits, at £20,000 per QALY, associated with strategies of interest for adult men

Pre-test probability for blood test (%)	Sensitivity of indicator	Specificity of indicator	Strategy	Mean (95% Crl)
				Costs (£)	QALYs	Incremental net benefit at £20,000 per QALY vs. no screening (£)
No screening				17,389 (14,011 to 19,332)	18.31 (16.39 to 19.54)	0 (0 to 0)
1	1	0	IgA tTG	20,468 (20,152 to 20,778)	19.68 (19.14 to 20.04)	24,331 (5080 to 56,493)
1.5	0.87	0.431	IgA tTG	18,325 (15,069 to 20,186)	18.36 (16.52 to 19.56)	161 (–676 to 1530)
2	0.79	0.61	IgA tTG	18,236 (14,990 to 20,091)	18.37 (16.53 to 19.56)	355 (–549 to 1838)
5	0.579	0.889	IgA tTG	18,270 (15,286 to 19,963)	18.49 (16.83 to 19.6)	2838 (187 to 7218)
10	0.322	0.971	IgA tTG	18,567 (16,137 to 19,955)	18.72 (17.37 to 19.66)	7179 (1292 to 16,949)
20	0.107	0.996	IgA tTG	19,267 (18,020 to 20,053)	19.21 (18.43 to 19.82)	16,114 (3482 to 37,161)
1	1	0	IgA EMA plus HLA	20,841 (20,546 to 21,147)	19.73 (19.19 to 20.1)	25,060 (5050 to 58,568)
1.5	0.87	0.431	IgA EMA plus HLA	20,535 (20,229 to 20,844)	19.71 (19.17 to 20.08)	24,884 (5204 to 57,564)
2	0.79	0.61	IgA EMA plus HLA	20,430 (20,123 to 20,737)	19.71 (19.17 to 20.08)	24,981 (5300 to 57,672)
5	0.579	0.889	IgA EMA plus HLA	20,330 (20,034 to 20,641)	19.72 (19.18 to 20.09)	25,262 (5415 to 58,394)
10	0.322	0.971	IgA EMA plus HLA	20,350 (20,053 to 20,655)	19.72 (19.18 to 20.09)	25,237 (5331 to 58,488)
20	0.107	0.996	IgA EMA plus HLA	20,412 (20,123 to 20,716)	19.71 (19.18 to 20.09)	25,135 (5186 to 58,444)
1	1	0	HLA plus IgA tTG	20,837 (20,545 to 21,148)	19.73 (19.19 to 20.1)	25,059 (5055 to 58,561)
1.5	0.87	0.431	HLA plus IgA tTG	20,541 (20,235 to 20,850)	19.71 (19.18 to 20.08)	24,957 (5217 to 57,825)
2	0.79	0.61	HLA plus IgA tTG	20,437 (20,131 to 20,747)	19.71 (19.18 to 20.08)	25,048 (5311 to 57,918)
5	0.579	0.889	HLA plus IgA tTG	20,338 (20,035 to 20,645)	19.72 (19.18 to 20.09)	25,273 (5409 to 58,467)
10	0.322	0.971	HLA plus IgA tTG	20,358 (20,056 to 20,665)	19.72 (19.18 to 20.09)	25,231 (5316 to 58,506)
20	0.107	0.996	HLA plus IgA tTG	20,417 (20,124 to 20,726)	19.71 (19.18 to 20.09)	25,126 (5176 to 58,438)

Figure 16 plots the CEACs, which show the probability that each testing strategy is optimal (i.e. has the highest net benefit) at each willingness-to-pay threshold for an additional QALY. None of the probabilities is > 60%, suggesting limited certainty that any of the strategies is most cost-effective. At £10,000–30,000 per QALY, the strategy with greatest probability of being cost-effective is HLA plus IgA tTG at a pre-test probability of 5% (i.e. 5% HLA plus IgA tTG); above this range, 1% IgA EMA plus HLA has the greatest probability of being cost-effective. However, the CEAC does not account for the magnitude of differences between net benefits, which, as reported in Table 20, are negligible with overlapping 95% CrIs.

FIGURE 16.

Cost-effectiveness acceptability curves for strategies of interest for adult men.

Table S19 (see Report Supplementary Material 1) provides the proportion of time spent in each state for adult men on the strategies of interest. Compared with other pre-test probability strategies with the same serological tests, the 1% IgA EMA strategy has the greatest proportion of time in the CD, gluten-free diet, no-complications state and the least time in the undiagnosed states. Osteoporosis appears to be the complication most patients spend time in, with NHL being up to six times less common. The HLA strategies have a greater proportion of time in the CD, gluten-free diet, no-complications state and less time in the undiagnosed states than any IgA EMA strategies. These strategies all have very similar proportions of time across the different pre-test probabilities, with only minor differences between the CD, gluten-free diet state and the undiagnosed CD, no-complications state. Overall, cost-effectiveness appears to be mostly driven by time spent in the CD, gluten-free diet, no-complications state; the CD, gluten-free diet, osteoporosis state; and the undiagnosed CD, osteoporosis state.

The ratio of EVPPI to EVPI for each parameter is illustrated in Figure 17. The probability of late diagnosis and the sensitivity of HLA testing are the most influential parameters. The impact of gluten-free diet on the probability of developing osteoporosis and the accuracy of IgA EMA testing have low influence, and all other parameters appear to have no influence, on the results.

FIGURE 17.

Ratio of EVPPI for each parameter to total EVPI among adult men. Only the 10 most influential parameters are included; the remaining parameters have minimal influence on the results. GFD, gluten-free diet; log_or_osteoporosis_noGFD, log-odds ratio of developing osteoporosis, not on a GFD, compared with being on a GFD; log_rr_NHL_noGFD, log-risk ratio of developing NHL, not on a GFD, compared with being on a GFD, for coeliac patients; probability_IDA_50, prevalence of IDA at age 50 years; probability_IDA_90, prevalence of IDA at age 90 years; sens, sensitivity; spec, specificity.

The total EVPI and EVPPI for parameter sets of interest are summarised in Table 21. The total population EVPI for men is £25.7M, indicating potential value in conducting further research. The EVPPI for parameter sets indicates that the greatest potential value lies in a study of probabilities of late diagnosis or on test-accuracy parameters. This aligns with the ranking in Figure 17.

TABLE 21 - Total EVPI and EVPPI for parameter sets of interest

GFD, gluten-free diet; NR, not reported.

MLMC method estimates for children failed to converge. Gaussian process estimates are presented, but were run only on the utilities estimate and in this set were also unstable.

Notes

Values for adult men and women were estimated using the MLMC method; Gaussian process estimates are presented for children. All EVPI and EVPPI values are in Great British pounds. Higher values indicate greater impact on results and greater value in conducting further research.

Parameter set	Men		Women		Childrena
Total population size (n)	263,465		263,465		140,070
Discounted population size (n)	2,267,824		2,267,824		1,205,679
	Per person (£)	Population (£)	Per person (£)	Population (£)	Per person (£)	Population (£)
Total EVPI	11.32	25,679,685	34.85	79,040,182	15.28	18,420,293
Utilities and disutilities	3.72	8,425,612	1.62	3,663,585	0.93	1,117,066
Rates of osteoporosis and NHL	2.84	6,449,814	10.08	22,849,058	0.40	486,521
GFD effect	3.02	6,850,323	10.89	24,702,170	0.26	316,551
Test accuracies	6.81	15,447,753	18.98	43,035,778	6.48	7,812,603
Probability of late diagnosis	8.01	18,159,091	30.22	68,534,495	8.76	10,561,800

Adult women

Figure 18 plots the incremental net benefit of each sensitivity/specificity combination of a diagnostic indicator (or combination of indicators) combined with each serological test, compared with no screening, for adult women. The central estimates are the mean incremental net benefit, with the upper and lower Crls represented by the upper and lower lines.

FIGURE 18.

Incremental net benefit at £20,000 per QALY, compared with no screening, at each combination of sensitivity (x-axis) and specificity (line colour) for adult women. (a) IgA EMA; (b) IgA tTG plus EMA; (c) IgA tTG; (d) IgA EMA plus HLA; (e) IgA tTG plus EMA plus HLA; (f) IgA tTG plus HLA; (g) HLA plus IgA EMA; (h) HLA plus IgA tTG plus EMA; and (i) HLA plus IgA tTG. Note that thick lines indicate the mean and narrow lines represent the 95% CrI.

The results are extremely similar to those for adult men. Figure 18 shows that, for the serological tests without HLA, the incremental net benefit at £20,000 per QALY is positive only when the diagnostic indicator sensitivity is > 0.9 or both the specificity and sensitivity are > 0.8. For the serological tests including HLA, the incremental net benefit is positive regardless of the accuracy of the diagnostic indicator. The combinations using only the IgA EMA serological test plus HLA are more cost-effective than strategies using IgA tTG plus HLA and are comparable in cost-effectiveness to tests using all of EMA, tTG and HLA. Using HLA before or after EMA or tTG plus EMA does not affect cost-effectiveness, but it appears to be most cost-effective if used before IgA tTG alone.

Figure 19 plots the probability that each combination of diagnostic indicator sensitivity and specificity is cost-effective for each test combination. Results are again nearly identical to those for adult men. The serological tests including HLA have a high probability of being cost-effective regardless of the combination of sensitivity and specificity of the risk test. Strategies using IgA EMA or IgA tTG alone and HLA have a similar probability of cost-effectiveness to combinations using all three of IgA EMA, HLA and IgA tTG.

FIGURE 19.

Probability of being cost-effective at £20,000 per QALY, compared with no screening, at each combination of sensitivity (x-axis) and specificity (line colour) for adult women. (a) IgA EMA; (b) IgA tTG plus EMA; (c) IgA tTG; (d) IgA EMA plus HLA; (e) IgA tTG plus EMA plus HLA; (f) IgA tTG plus HLA; (g) HLA plus IgA EMA; (h) HLA plus IgA tTG plus EMA; and (i) HLA plus IgA tTG.

As for adult men, we select the testing strategies IgA tTG alone, IgA EMA plus HLA, and HLA plus IgA tTG to combine with the various risk prediction strategies of Table 5. This is again justified as strategies using IgA tTG and IgA EMA have similar incremental net benefits and probabilities of being cost-effective, IgA tTG plus HLA performs worst among strategies with HLA, there is limited difference between IgA tTG plus EMA plus HLA and IgA EMA plus HLA (i.e. not employing tTG), and HLA plus the widely available IgA tTG has similar cost-effectiveness to other strategies. There is also limited difference between strategies applying HLA before or after serological tests.

Table 22 shows the estimated costs, QALYs and incremental net benefit, compared with no screening, for each of these strategies of interest. All combinations, apart from pre-test probabilities of 1.5% with IgA tTG, have positive incremental net benefits, compared with no screening. If using only IgA tTG, a pre-test probability of 1% is the most cost-effective strategy (greatest incremental net benefit at £20,000 per QALY), with the highest costs but also the largest number of QALYs. As with men, the net benefits of all IgA EMA plus HLA and HLA plus IgA tTG strategies are very similar to each other and to those of IgA tTG with a pre-test probability of 1%, and the 95% CrIs are completely overlapping. This again indicates little difference between the cost-effectiveness of these strategies.

TABLE 22 - Costs, QALYs and incremental net benefits, at £20,000 per QALY, associated with strategies of interest for adult women

Pre-test probability for blood test (%)	Sensitivity	Specificity	Strategy	Mean (95% Crl)
				Costs (£)	QALYs	Incremental net benefit at £20,000 per QALY vs. no screening (£)
No screening				17,673 (14,159 to 19,645)	18.68 (16.72 to 19.9)	0 (0 to 0)
1	1	0	IgA tTG	20,771 (20,469 to 21,082)	20.05 (19.59 to 20.42)	24,382 (4829 to 59,154)
1.50	0.841	0.448	IgA tTG	18,585 (15,184 to 20,501)	18.72 (16.84 to 19.91)	–27 (–716 to 1197)
2	0.758	0.628	IgA tTG	18,500 (15,118 to 20,407)	18.73 (16.86 to 19.91)	228 (–575 to 1660)
5	0.387	0.926	IgA tTG	18,458 (15,268 to 20,262)	18.81 (17.05 to 19.93)	1880 (–63 to 5360)
10	0.107	0.99	IgA tTG	18,706 (15,970 to 20,247)	19 (17.48 to 19.98)	5428 (807 to 13,728)
20	0.002	1	IgA tTG	19,214 (17,329 to 20,319)	19.35 (18.28 to 20.1)	11,835 (2322 to 28,874)
1	1	0	IgA EMA plus HLA	21,143 (20,863 to 21,447)	20.11 (19.62 to 20.46)	25,115 (4766 to 61,286)
1.50	0.841	0.448	IgA EMA plus HLA	20,817 (20,522 to 21,131)	20.08 (19.6 to 20.44)	24,844 (4934 to 60,463)
2	0.758	0.628	IgA EMA plus HLA	20,714 (20,419 to 21,030)	20.08 (19.6 to 20.44)	24,981 (5033 to 60,663)
5	0.387	0.926	IgA EMA plus HLA	20,598 (20,319 to 20,904)	20.09 (19.61 to 20.44)	25,244 (5118 to 61,069)
10	0.107	0.99	IgA EMA plus HLA	20,629 (20,358 to 20,927)	20.09 (19.61 to 20.45)	25,259 (5033 to 61,175)
20	0.002	1	IgA EMA plus HLA	20,683 (20,407 to 20,993)	20.09 (19.61 to 20.44)	25,210 (4927 to 61,189)
1	1	0	HLA plus IgA tTG	21,139 (20,856 to 21,442)	20.11 (19.62 to 20.46)	25,114 (4770 to 61,277)
1.50	0.841	0.448	HLA plus IgA tTG	20,824 (20,533 to 21,138)	20.08 (19.61 to 20.44)	24,935 (4951 to 60,666)
2	0.758	0.628	HLA plus IgA tTG	20,722 (20,434 to 21,030)	20.08 (19.61 to 20.44)	25,058 (5043 to 60,805)
5	0.387	0.926	HLA plus IgA tTG	20,607 (20,333 to 20,915)	20.09 (19.61 to 20.45)	25,268 (5110 to 61,127)
10	0.107	0.99	HLA plus IgA tTG	20,638 (20,366 to 20,938)	20.09 (19.61 to 20.45)	25,259 (5018 to 61,192)
20	0.002	1	HLA plus IgA tTG	20,690 (20,415 to 20,998)	20.09 (19.61 to 20.44)	25,203 (4914 to 61,189)

Figure 20 plots the CEACs, which show the probability that each testing strategy is optimal (i.e. has the highest net benefit) at each willingness-to-pay threshold for an additional QALY. None of the probabilities exceeds 50%, indicating little certainty that the 1% IgA tTG, the IgA EMA plus HLA or the HLA plus IgA tTG strategies differ in cost-effectiveness. At £10,000–30,000 per QALY, the strategy with the greatest probability of being cost-effective is 5% HLA plus IgA tTG; above this range, 1% IgA tTG has the greatest probability of being cost-effective. However, the CEAC does not account for the magnitude of differences between net benefits, which, as reported in Table 22, are negligible, with overlapping 95% CrIs.

FIGURE 20.

Cost-effectiveness acceptability curves for strategies of interest for adult women.

Table S20 (see Report Supplementary Material 1) provides the proportion of time spent in each state for adult women on the strategies of interest. Compared with other pre-test probability strategies with the same serological tests, the 1% IgA EMA strategy has the greatest proportion of time in the CD, gluten-free diet, no-complications state and the least time in the undiagnosed states. Osteoporosis appears to be the complication most patients spend time in; women appear to spend up to four times as much time with this complication as men. The HLA strategies have a greater proportion of time in the CD, gluten-free diet, no-complications state and less time in the undiagnosed states than any IgA EMA strategies. They have almost identical proportions of time in the states across the different pre-test probabilities, with only minor differences between the undiagnosed CD, no-complications state and the undiagnosed CD, osteoporosis state. Overall, cost-effectiveness appears to be mostly driven by time spent in the CD, gluten-free diet, no-complications state; the CD, gluten-free diet, osteoporosis state; and the undiagnosed CD, osteoporosis state.

The ratio of EVPPI to EVPI for each parameter is illustrated in Figure 21. As in the population of adult men, the probability of late diagnosis and the sensitivity of the HLA test are important parameters. However, the specificity of the IgA EMA test among adults is also important. The specificity of the HLA test, test cost of IgA tTG and cost of osteoporosis have low influence, and the remaining parameters have almost no influence on results.

FIGURE 21.

Ratio of EVPPI for each parameter to total EVPI for adult women. Only the 10 most influential parameters are included; the remaining parameters have minimal influence on the results. GFD, gluten-free diet; log_rr_NHL_noGFD, log-risk ratio of developing NHL, not on GFD, compared with being on GFD, for coeliac patients; death_log_hr_osteoporosis_female, log-hazard ratio of mortality with osteoporosis, compared with general population, for females; sens, sensitivity; spec, specificity.

The total EVPI and EVPPI for parameter sets of interest are summarised in Table 21. The total population EVPI for women is £79.0M, indicating potential value in conducting further research. This is greater than the potential value for men (£25.7M). As with men, the EVPPI for parameter sets indicates that the greatest potential value lies in a study of probabilities of late diagnosis (£68.5M) and on test accuracies (£43M). These values are two to three times greater than the corresponding values for men, suggesting that research should be focused on the adult women population.

Children

Figure 22 plots the incremental net benefit at £20,000 per QALY for risk prediction strategies. Results are similar to those in adults, but uncertainty is greater as the 95% CrIs are wider and, for strategies without HLA, now include 0 (i.e. the strategy is not cost-effective when compared with no screening). Figure 22 shows that, for the serological tests without HLA, the incremental net benefit at £20,000 per QALY is positive only when the diagnostic indicator sensitivity is > 0.9 or if the specificity is > 0.9 and sensitivity is > 0.8. For the tests including HLA, the incremental net benefit is positive regardless of the accuracy of the diagnostic indicator, except for IgA tTG plus HLA, which requires both sensitivity and specificity of the diagnostic indicator to be > 0.6. Cost-effectiveness is the same if HLA is applied before or after EMA or tTG plus EMA, except when combined with IgA tTG alone, in which case only HLA plus IgA tTG (i.e. IgA tTG as confirmatory test after HLA) is cost-effective, compared with no screening.

FIGURE 22.

Incremental net benefit at £20,000 per QALY, compared with no screening, at each combination of sensitivity (x-axis) and specificity (line colour) for children. (a) IgA EMA; (b) IgA tTG plus EMA; (c) IgA tTG; (d) IgA EMA plus HLA; (e) IgA tTG plus EMA plus HLA; (f) IgA tTG plus HLA; (g) HLA plus IgA EMA; (h) HLA plus IgA tTG plus EMA; and (i) HLA plus IgA tTG. Note that thick lines indicate the mean and narrow lines indicate the 95% CrI.

Figure 23 plots the probability that each combination of diagnostic indicator sensitivity and specificity is cost-effective for each serological test (i.e. the proportion of simulations that have a positive incremental net benefit, compared with no screening, at various willingness-to-pay thresholds). It shows that the only IgA EMA, IgA tTG plus EMA or IgA tTG strategies with probabilities of > 50% of being cost-effective are those with diagnostic indicator sensitivity of > 0.9 or both specificity and sensitivity of > 0.8. As expected, the combinations with the highest probabilities of being cost-effective are those with a specificity of 0.9 and a sensitivity of 0.9 or 1. The serological tests including HLA have a high probability of being cost-effective regardless of the combination of sensitivity and specificity of the risk test. The only exception is IgA tTG plus EMA, which requires a sensitivity of > 0.50 to have a high probability of being cost-effective.

FIGURE 23.

Probability of being cost-effective at £20,000 per QALY, compared with no screening, at each combination of sensitivity (x-axis) and specificity (line colour) for children. (a) IgA EMA; (b) IgA tTG plus EMA; (c) IgA tTG; (d) IgA EMA plus HLA; (e) IgA tTG plus EMA plus HLA; (f) IgA tTG plus HLA; (g) HLA plus IgA EMA; (h) HLA plus IgA tTG plus EMA; and (i) HLA plus IgA tTG.

As with the adult population, we selected the testing strategies IgA tTG alone, IgA EMA plus HLA and HLA plus IgA tTG to combine with the various pre-test probabilities listed in Table 5. This is justified, as strategies using IgA tTG and IgA EMA have similar incremental net benefits and probabilities of being cost-effective, IgA tTG plus HLA performs worst among strategies with HLA, there is limited difference between IgA tTG plus EMA plus HLA and IgA EMA plus HLA (i.e. not employing tTG), the order in which HLA and EMA tests are conducted has little impact, and HLA plus IgA tTG has high cost-effectiveness and employs a widely available test.

Table 23 shows the estimated costs, QALYs and incremental net benefits, compared with no screening, for each of these strategies of interest. All combinations, apart from pre-test probabilities of 1.5% and 2% with IgA tTG, have a positive incremental net benefit at £20,000 per QALY, compared with no screening. If using only IgA tTG, a pre-test probability of 1% is the most cost-effective strategy (i.e. has greatest incremental net benefit at £20,000 per QALY), with highest costs but also the largest number of QALYs. If using IgA EMA plus HLA or HLA plus IgA tTG, the pre-test probability of 10% is the most cost-effective. As with the adult population, however, the net benefits of all IgA EMA plus HLA and HLA plus IgA tTG strategies are very similar to each other and to 1% IgA tTG, and the 95% CrIs overlap. This indicates little difference between the cost-effectiveness of these strategies, but the differences are greater than they are for adults.

TABLE 23 - Costs, QALYs and incremental net benefits associated with strategies of interest in children. Mean estimates with 95% CIs

Pre-test probability for blood test (%)	Sensitivity	Specificity	Strategy	Mean (95% CI)
				Costs (£)	QALYs	Incremental net benefit at £20,000 per QALY vs. no screening (£)
No screening				11,746 (10,204 to 12,966)	20.6 (19.18 to 21.6)	0 (0 to 0)
1	1	0	IgA tTG	13,537 (12,558 to 14,637)	21.29 (20.29 to 22.14)	12,203 (3257 to 34,639)
1.5	0.882	0.417	IgA tTG	13,074 (11,577 to 14,280)	20.62 (19.25 to 21.61)	–841 (–1296 to 242)
2	0.807	0.61	IgA tTG	12,914 (11,431 to 14,119)	20.62 (19.27 to 21.62)	–574 (–1074 to 638)
5	0.667	0.872	IgA tTG	12,713 (11,343 to 13,890)	20.71 (19.48 to 21.67)	1290 (–226 to 5072)
10	0.533	0.952	IgA tTG	12,679 (11,477 to 13,793)	20.87 (19.8 to 21.78)	4517 (1002 to 13,401)
20	0.331	0.987	IgA tTG	12,699 (11,686 to 13,797)	21.09 (20.08 to 21.95)	8896 (2610 to 24,765)
1	1	0	IgA EMA plus HLA	13,808 (12,814 to 14,911)	21.32 (20.32 to 22.16)	12,502 (3202 to 35,895)
1.5	0.882	0.417	IgA EMA plus HLA	13,403 (12,418 to 14,497)	21.26 (20.25 to 22.1)	11,577 (3050 to 33,066)
2	0.807	0.61	IgA EMA plus HLA	13,226 (12,242 to 14,325)	21.26 (20.26 to 22.12)	11,878 (3293 to 33,524)
5	0.667	0.872	IgA EMA plus HLA	13,018 (12,030 to 14,117)	21.3 (20.29 to 22.14)	12,872 (3840 to 35,630)
10	0.533	0.952	IgA EMA plus HLA	12,994 (12,003 to 14,103)	21.31 (20.31 to 22.14)	13,085 (3919 to 36,248)
20	0.331	0.987	IgA EMA plus HLA	13,034 (12,033 to 14,133)	21.31 (20.31 to 22.15)	13,040 (3820 to 36,375)
1	1	0	HLA plus IgA tTG	13,800 (12,808 to 14,901)	21.32 (20.32 to 22.16)	12,509 (3210 to 35,884)
1.5	0.882	0.417	HLA plus IgA tTG	13,397 (12,416 to 14,492)	21.26 (20.26 to 22.1)	11,618 (3029 to 33,293)
2	0.807	0.61	HLA plus IgA tTG	13,221 (12,241 to 14,321)	21.26 (20.27 to 22.12)	11,914 (3273 to 33,675)
5	0.667	0.872	HLA plus IgA tTG	13,014 (12,027 to 14,114)	21.3 (20.29 to 22.14)	12,885 (3844 to 35,682)
10	0.533	0.952	HLA plus IgA tTG	12,991 (11,998 to 14,100)	21.31 (20.31 to 22.14)	13,090 (3929 to 36,260)
20	0.331	0.987	HLA plus IgA tTG	13,032 (12,029 to 14,132)	21.31 (20.31 to 22.15)	13,042 (3835 to 36,372)

Figure 24 plots the CEACs, which show the probability that each testing strategy is optimal (i.e. has the highest net benefit) at each willingness-to-pay threshold for an additional QALY. This shows that, at any threshold > £5000 per QALY, the strategy with the greatest probability of being cost-effective is 10% HLA plus IgA tTG. This probability is between 60% and 80%, which, although high, does not suggest certainty that any of the IgA EMA plus HLA or HLA plus IgA tTG strategies differ in cost-effectiveness.

FIGURE 24.

Cost-effectiveness acceptability curves for strategies of interest for children.

Table S21 (see Report Supplementary Material 1) provides the proportion of time spent in each state for children on the strategies of interest. Compared with other IgA tTG strategies, the 1% IgA tTG has the greatest proportion of time spent in the CD, gluten-free diet, no-complications state and the least time in the undiagnosed states, and similarly for the 5% HLA strategies compared with other HLA strategies. There is little difference in the proportions of time spent in the undiagnosed, complication states, as most child patients are eventually diagnosed. Time spent in the death state is the same across strategies for this reason. Overall, cost-effectiveness appears to be driven by time spent in the CD, gluten-free diet, no-complications state and the undiagnosed CD, no-complications state.

The total EVPI and EVPPI for parameter sets of interest are summarised in Table 21. The total population EVPI is £18.4M, indicating potential value in conducting further research. However, this potential value is less than that for adult men (£25.7M) and adult women (£79.0M). As with adults, the EVPPI is greatest for the probability of late diagnosis (£10.6M) and test accuracies (£7.8M). These are half of the corresponding values for adult men and about six times lower than for adult women, again suggesting that research should be focused on adult women.

The ratio of EVPPI to EVPI for each parameter is illustrated in Figure 25. As with the adult population, probability of late diagnosis, disutility of biopsy and specificity of IgA EMA are important parameters. Note that the effect of a gluten-free diet (i.e. being diagnosed with CD) on osteoporosis is picked out as highly influential, and actually its estimated EVPPI is greater than the total EVPI (note that proportion exceeds 1.00). This is due to an algorithm failure. The EVPPI estimated in Table 21 is based on MLMC methods, whereas the ratios in Figure 25 are based on generalised additive models, and the latter have been found to be unstable in comparison with the former in the literature. 254

FIGURE 25.

Ratio of EVPPI for each parameter to total EVPI for children. Only the 10 most influential parameters are included. Note that log_or_osteoporosis_GFD is high, but this is due to numerical failure of the algorithm. The remaining parameters have minimal influence on the results. GFD, gluten-free diet; log_or_osteoporosis_GFD, log-odds ratio of developing osteoporosis on GFD, compared with not on GFD, for coeliac patients; probability_IDA_20, prevalence of IDA at age 20 years; probability_IDA_10, prevalence of IDA at age 10 years; sens, sensitivity; spec, specificity.

Chapter 9 Discussion

Parts of this chapter are reproduced from Elwenspoek et al. 20 This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) license, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: https://creativecommons.org/licenses/by/4.0/. The text below includes minor additions and formatting changes to the original text.

Chapter 10 Conclusions

Acknowledgements

We would like to thank James McKernon (Research Associate and CPRD Database Manager at the University of Bristol) and Tim Jones (Senior Research Associate in Quantitative Research and Health Economics at the University of Bristol) for extracting the CPRD data sets that were used in Chapters 4 and 8. We are grateful for Zoe Trinder-Widdess’ (Head of Communications at NIHR ARC West) and Lucy Cordon’s (Project Support Administrator at NIHR ARC West) help with recruiting survey participants for Chapter 7, and to Lucy Cordon for developing the videos for the survey (see Chapter 7) and for retrieving full texts for the review in Chapter 3. We would also like to thank Lauren J Scott (Senior Research Associate in Quantitative Research and Evidence Synthesis), Victoria Corfield (GP trainee) and Anthony Sinobas (medical student) who contributed to eligibility screening and data extraction for the systematic reviews in Chapters 3, 5 and 6. Finally, we would like to thank Jonathan Banks (Research Fellow in Qualitative Research at NIHR ARC West) for advising on the qualitative analysis in Chapter 7.

References

1. BMJ Best Practice . Coeliac Disease n.d. https://bestpractice.bmj.com/topics/en-gb/636 (accessed 2 June 2022).
2. Bingley PJ, Williams AJ, Norcross AJ, Unsworth DJ, Lock RJ, Ness AR, et al. Avon Longitudinal Study of Parents and Children Study Team. Undiagnosed coeliac disease at age seven: population based prospective birth cohort study. BMJ 2004;328:322-3. https://doi.org/10.1136/bmj.328.7435.322.
3. West J, Otete H, Sultan AA, Crooks CJ. Changes in testing for and incidence of celiac disease in the United Kingdom: a population-based cohort study. Epidemiology 2019;30:e23-e24. https://doi.org/10.1097/EDE.0000000000001006.
4. West J, Fleming KM, Tata LJ, Card TR, Crooks CJ. Incidence and prevalence of celiac disease and dermatitis herpetiformis in the UK over two decades: population-based study. Am J Gastroenterol 2014;109:757-68. https://doi.org/10.1038/ajg.2014.55.
5. Green PH. Mortality in celiac disease, intestinal inflammation, and gluten sensitivity. JAMA 2009;302:1225-6. https://doi.org/10.1001/jama.281.24.2344.
6. Green PH, Jabri B. Coeliac disease. Lancet 2003;362:383-91. https://doi.org/10.1016/S0140-6736(03)14027-5.
7. Caio G, Ciccocioppo R, Zoli G, De Giorgio R, Volta U. Therapeutic options for coeliac disease: what else beyond gluten-free diet?. Dig Liver Dis 2020;52:130-7. https://doi.org/10.1016/j.dld.2019.11.010.
8. Norström F, Myléus A, Nordyke K, Carlsson A, Högberg L, Sandström O, et al. Is mass screening for coeliac disease a wise use of resources? A health economic evaluation. BMC Gastroenterol 2021;21. https://doi.org/10.1186/s12876-021-01737-1.
9. National Institute for Health and Care Excellence (NICE) . Coeliac Disease: Recognition, Assessment and Management 2015.
10. Husby S, Koletzko S, Korponay-Szabó I, Kurppa K, Mearin ML, Ribes-Koninckx C, et al. European Society Paediatric Gastroenterology, Hepatology and Nutrition Guidelines for Diagnosing Coeliac Disease 2020. J Pediatr Gastroenterol Nutr 2020;70:141-56. https://doi.org/10.1097/MPG.0000000000002497.
11. Al-Toma A, Volta U, Auricchio R, Castillejo G, Sanders DS, Cellier C, et al. European Society for the Study of Coeliac Disease (ESsCD) guideline for coeliac disease and other gluten-related disorders. United European Gastroenterol J 2019;7:583-61. https://doi.org/10.1177/2050640619844125.
12. Giersiepen K, Lelgemann M, Stuhldreher N, Ronfani L, Husby S, Koletzko S, et al. ESPGHAN Working Group on Coeliac Disease Diagnosis. Accuracy of diagnostic antibody tests for coeliac disease in children: summary of an evidence report. J Pediatr Gastroenterol Nutr 2012;54:229-41. https://doi.org/10.1097/MPG.0b013e318216f2e5.
13. Rostom A, Dubé C, Cranney A, Saloojee N, Sy R, Garritty C, et al. The diagnostic accuracy of serologic tests for celiac disease: a systematic review. Gastroenterology 2005;128:38-46. https://doi.org/10.1053/j.gastro.2005.02.028.
14. Schyum AC, Rumessen JJ. Serological testing for celiac disease in adults. United European Gastroenterol J 2013;1:319-25. https://doi.org/10.1177/2050640613502788.
15. van der Windt DAWM, Jellema P, Mulder CJ, Kneepkens CM, van der Horst HE. Diagnostic testing for celiac disease among patients with abdominal symptoms: a systematic review. JAMA 2010;303:1738-46. https://doi.org/10.1001/jama.2010.549.
16. Lebwohl B, Sanders DS, Green PHR. Coeliac disease. Lancet 2018;391:70-81. https://doi.org/10.1016/S0140-6736(17)31796-8.
17. Cecilio LA, Bonatto MW. The prevalence of HLA DQ2 and DQ8 in patients with celiac disease, in family and in general population. Arq Bras Cir Dig 2015;28:183-5. https://doi.org/10.1590/S0102-67202015000300009.
18. Husby S, Koletzko S, Korponay-Szabó IR, Mearin ML, Phillips A, Shamir R, et al. European Society for Pediatric Gastroenterology, Hepatology, and Nutrition guidelines for the diagnosis of coeliac disease. J Pediatr Gastroenterol Nutr 2012;54:136-60. https://doi.org/10.1097/MPG.0b013e31821a23d0.
19. British Society of Gastroenterology . BSG Interim Guidance: COVID-19 Specific Non-Biopsy Protocol for Those With Suspected Coeliac Disease 2020.
20. Elwenspoek MMC, Jackson J, O’Donnell R, Sinobas A, Dawson S, Everitt H, et al. The accuracy of diagnostic indicators for coeliac disease: a systematic review and meta-analysis. PLOS ONE 2021;16. https://doi.org/10.1371/journal.pone.0258501.
21. Elwenspoek MMC, Jackson J, Dawson S, Everitt H, Gillett P, Hay AD, et al. Accuracy of potential diagnostic indicators for coeliac disease: a systematic review protocol. BMJ Open 2020;10. https://doi.org/10.1136/bmjopen-2020-038994.
22. Centre for Reviews and Dissemination . Systematic Reviews – CRD’s Guidance for Undertaking Reviews in Healthcare 2009.
23. Deeks JJ, Bossuyt PM, Gatsonis C. Cochrane Handbook for Systematic Reviews of Diagnostic Test Accuracy Version 1.0.0. London: The Cochrane Collaboration; 2013.
24. McInnes MDF, Moher D, Thombs BD, McGrath TA, Bossuyt PM, Clifford T, et al. Preferred Reporting Items for a Systematic Review and Meta-analysis of Diagnostic Test Accuracy Studies: the PRISMA-DTA statement. JAMA 2018;319:388-96. https://doi.org/10.1001/jama.2017.19163.
25. Rutjes AW, Reitsma JB, Vandenbroucke JP, Glas AS, Bossuyt PM. Case–control and two-gate designs in diagnostic accuracy studies. Clin Chem 2005;51:1335-41. https://doi.org/10.1373/clinchem.2005.048595.
26. Geersing GJ, Bouwmeester W, Zuithoff P, Spijker R, Leeflang M, Moons KG, et al. Search filters for finding prognostic and diagnostic prediction studies in MEDLINE to enhance systematic reviews. PLOS ONE 2012;7. https://doi.org/10.1371/journal.pone.0032844.
27. Ingui BJ, Rogers MA. Searching for clinical prediction rules in MEDLINE. J Am Med Inform Assoc 2001;8:391-7. https://doi.org/10.1136/jamia.2001.0080391.
28. Whiting PF, Rutjes AW, Westwood ME, Mallett S, Deeks JJ, Reitsma JB, et al. QUADAS-2: a revised tool for the quality assessment of diagnostic accuracy studies. Ann Intern Med 2011;155:529-36. https://doi.org/10.7326/0003-4819-155-8-201110180-00009.
29. Reitsma JB, Glas AS, Rutjes AW, Scholten RJ, Bossuyt PM, Zwinderman AH. Bivariate analysis of sensitivity and specificity produces informative summary measures in diagnostic reviews. J Clin Epidemiol 2005;58:982-90. https://doi.org/10.1016/j.jclinepi.2005.02.022.
30. Chu H, Cole SR. Bivariate meta-analysis of sensitivity and specificity with sparse data: a generalized linear mixed model approach. J Clin Epidemiol 2006;59:1331-2. https://doi.org/10.1016/j.jclinepi.2006.06.011.
31. Ludvigsson JF, Leffler DA, Bai JC, Biagi F, Fasano A, Green PH, et al. The Oslo definitions for coeliac disease and related terms. Gut 2013;62:43-52. https://doi.org/10.1136/gutjnl-2011-301346.
32. Singh P, Arora A, Strand TA, Leffler DA, Catassi C, Green PH, et al. Global prevalence of celiac disease: systematic review and meta-analysis. Clin Gastroenterol Hepatol 2018;16:823-36.e2. https://doi.org/10.1016/j.cgh.2017.06.037.
33. R Core Team . R: A Language and Environment for Statistical Computing 2020.
34. Bates D, Maechler M, Bolker B, Walker S. Fitting linear mixed-effects models using lme4. J Stat Softw 2015;67:1-48. https://doi.org/10.18637/jss.v067.i01.
35. Scherer R. PropCIs: Various Confidence Interval Methods for Proportions. R Package Version 0.3-0 2018. https://CRAN.R-project.org/package=PropCIs (accessed 23 June 2022).
36. Gordon M, Lumley T. Advanced Forest Plot Using ‘grid’ Graphics [R Package Version 1.10.1] 2020.
37. Gohel D. Flextable: Functions for Tabular Reporting. R Package Version 0.5.11 2020.
38. Wickham H. ggplot2: Elegant Graphics for Data Analysis. New York, NY: Springer-Verlag; 2016.
39. Doebler P. Mada: Meta-Analysis of Diagnostic Accuracy. R Package Version 0.5.10 2020.
40. Murdoch D, Chow ED. Ellipse: Functions for Drawing Ellipses and Ellipse-Like Confidence Regions. R Package Version 0.4.2 2020.
41. Ludvigsson JF, Bai JC, Biagi F, Card TR, Ciacci C, Ciclitira PJ, et al. Diagnosis and management of adult coeliac disease: guidelines from the British Society of Gastroenterology. Gut 2014;63:1210-28. https://doi.org/10.1136/gutjnl-2013-306578.
42. National Institute for Health and Care Excellence (NICE) . 2019 Surveillance of Coeliac Disease: Recognition, Assessment and Management (NICE Guideline NG20) n.d. www.nice.org.uk/guidance/ng20/resources/2019-surveillance-of-coeliac-disease-recognition-assessment-and-management-nice-guideline-ng20-pdf-9097196600005 (accessed 19 January 2021).
43. Elwenspoek MMC, Jones HE, Thom H, Watson JC, Hay AD, Everitt H, et al. Prediction Model for Coeliac Disease Diagnosis: Statistical Analysis Plan (SAP) n.d. https://osf.io/egfr5/#! (accessed 25 May 2022).
44. Steyerberg EW. Clinical Prediction Models. A Practical Approach to Development, Validation, and Updating. Cham: Springer Nature Switzerland; 2019.
45. Collins GS, Reitsma JB, Altman DG, Moons KG. Transparent Reporting of a multivariable prediction model for Individual Prognosis Or Diagnosis (TRIPOD). Ann Intern Med 2015;162:735-6. https://doi.org/10.7326/L15-5093-2.
46. Herrett E, Gallagher AM, Bhaskaran K, Forbes H, Mathur R, van Staa T, et al. Data Resource Profile: Clinical Practice Research Datalink (CPRD). Int J Epidemiol 2015;44:827-36. https://doi.org/10.1093/ije/dyv098.
47. Clinical Practice Research Datalink . CPRD GOLD June 2021 (Version 2021.06.001) 2021. https://doi.org/10.48329/7308-xf35.
48. Ministry of Housing, Communities and Local Government . National Statistics: English Indices of Deprivation 2019 n.d. www.gov.uk/government/statistics/english-indices-of-deprivation-2019 (accessed 30 May 2022).
49. Abdul Sultan A, Crooks CJ, Card T, Tata LJ, Fleming KM, West J. Causes of death in people with coeliac disease in England compared with the general population: a competing risk analysis. Gut 2015;64:1220-6. https://doi.org/10.1136/gutjnl-2014-308285.
50. Violato M, Gray A, Papanicolas I, Ouellet M. Resource use and costs associated with coeliac disease before and after diagnosis in 3,646 cases: results of a UK primary care database analysis. PLOS ONE 2012;7. https://doi.org/10.1371/journal.pone.0041308.
51. Riley RD, Snell KIE, Ensor J, Burke DL, Harrell FE, Moons KGM, et al. Minimum sample size for developing a multivariable prediction model: PART II – binary and time-to-event outcomes. Stat Med 2019;38:1276-96. https://doi.org/10.1002/sim.7992.
52. World Organization of Family Doctors’ International Classification Committee . International Classification of Primary Care (ICPC-2) 2015.
53. Holmes GKT, Muirhead A. Epidemiology of coeliac disease in a single centre in Southern Derbyshire 1958–2014. BMJ Open Gastroenterol 2017;4. https://doi.org/10.1136/bmjgast-2017-000137.
54. Biesheuvel CJ, Vergouwe Y, Oudega R, Hoes AW, Grobbee DE, Moons KG. Advantages of the nested case–control design in diagnostic research. BMC Med Res Methodol 2008;8. https://doi.org/10.1186/1471-2288-8-48.
55. Moons KG. Criteria for scientific evaluation of novel markers: a perspective. Clin Chem 2010;56:537-41. https://doi.org/10.1373/clinchem.2009.134155.
56. Moons KGM, Wolff RF, Riley RD, Whiting PF, Westwood M, Collins GS, et al. PROBAST: a tool to assess risk of bias and applicability of prediction model studies: explanation and elaboration. Ann Intern Med 2019;170:W1-W33. https://doi.org/10.7326/M18-1377.
57. Wolf A, Dedman D, Campbell J, Booth H, Lunn D, Chapman J, et al. Data resource profile: Clinical Practice Research Datalink (CPRD) Aurum. Int J Epidemiol 2019;48:1740-1740g. https://doi.org/10.1093/ije/dyz034.
58. Clinical Practice Research Datalink . CPRD Aurum June 2021 (Version 2021.06.001) 2021. https://doi.org/10.48329/pyc2-we97.
59. Golding J, Pembrey M, Jones R, Team AS. ALSPAC – the Avon Longitudinal Study of Parents and Children. I. Study methodology. Paediatr Perinat Epidemiol 2001;15:74-87. https://doi.org/10.1046/j.1365-3016.2001.00325.x.
60. Boyd A, Golding J, Macleod J, Lawlor DA, Fraser A, Henderson J, et al. Cohort profile: the ‘Children of the 90s‘ – the index offspring of the Avon Longitudinal Study of Parents and Children. Int J Epidemiol 2013;42:111-27. https://doi.org/10.1093/ije/dys064.
61. Fraser A, Macdonald-Wallis C, Tilling K, Boyd A, Golding J, Davey Smith G, et al. Cohort profile: the Avon Longitudinal Study of Parents and Children: ALSPAC mothers cohort. Int J Epidemiol 2013;42:97-110. https://doi.org/10.1093/ije/dys066.
62. Great Britain . Human Tissue Act 2004 2004.
63. University of Bristol . Explore Data and Samples n.d. www.bristol.ac.uk/alspac/researchers/our-data/ (accessed November 2021).
64. Heinze G, Schemper M. A solution to the problem of separation in logistic regression. Stat Med 2002;21:2409-19. https://doi.org/10.1002/sim.1047.
65. Firth D. Bias reduction of maximum likelihood estimates. Biometrika 1993;80:27-38. https://doi.org/10.1093/biomet/80.1.27.
66. Harrell FE. Regression Modeling Strategies: With Applications to Linear Models, Logistic and Ordinal Regression, and Survival Analysis. Cham: Springer International Publishing; 2015.
67. Ministry of Housing, Communities and Local Government . National Statistics: English Indices of Deprivation 2015 n.d. www.gov.uk/government/statistics/english-indices-of-deprivation-2015 (accessed 27 May 2022).
68. Whiting PF, Rutjes AW, Westwood ME, Mallett S. QUADAS-2 Steering Group . A systematic review classifies sources of bias and variation in diagnostic test accuracy studies. J Clin Epidemiol 2013;66:1093-104. https://doi.org/10.1016/j.jclinepi.2013.05.014.
69. Singh P, Arora A, Strand TA, Leffler DA, Mäki M, Kelly CP, et al. Diagnostic accuracy of point of care tests for diagnosing celiac disease: a systematic review and meta-analysis. J Clin Gastroenterol 2019;53:535-42. https://doi.org/10.1097/MCG.0000000000001081.
70. Holmes G, Ciacci C. The serological diagnosis of coeliac disease – a step forward. Gastroenterol Hepatol Bed Bench 2018;11:209-15.
71. Ouzzani M, Hammady H, Fedorowicz Z, Elmagarmid A. Rayyan – a web and mobile app for systematic reviews. Syst Rev 2016;5. https://doi.org/10.1186/s13643-016-0384-4.
72. Yang B, Mallett S, Takwoingi Y, Davenport CF, Hyde CJ, Whiting PF, et al. QUADAS-C: a tool for assessing risk of bias in comparative diagnostic accuracy studies. Ann Intern Med 2021;174:1592-9. https://doi.org/10.7326/M21-2234.
73. Whiting P, Davenport C. Understanding test accuracy research: a test consequence graphic. Diagn Progn Res 2018;2. https://doi.org/10.1186/s41512-017-0023-0.
74. Yuan J, Zhou C, Gao J, Li J, Yu F, Lu J, et al. Prevalence of celiac disease autoimmunity among adolescents and young adults in China. Clin Gastroenterol Hepatol 2017;15:1572-9.e1. https://doi.org/10.1016/j.cgh.2017.04.025.
75. Sheppard A, Whiting P, Corfield V, Elwenspoek M, Richards A, Jones H, et al. What Is the Accuracy of Serological Testing for Diagnosing Coeliac Disease in Adults and Children? A Systematic Review n.d. www.crd.york.ac.uk/prospero/display_record.php?ID=CRD42019115506 (accessed 8 June 2022).
76. Abrams JA, Brar P, Diamond B, Rotterdam H, Green PH. Utility in clinical practice of immunoglobulin A anti-tissue transglutaminase antibody for the diagnosis of celiac disease. Clin Gastroenterol Hepatol 2006;4:726-30. https://doi.org/10.1016/j.cgh.2006.02.010.
77. Al Saidi SS, Al Harthi SO, Mula-Abed WA. Diagnostic utility of coeliac disease: a descriptive study in a tertiary care hospital, Oman. Oman Med J 2013;28:232-6. https://doi.org/10.5001/omj.2013.68.
78. Ashorn S, Raukola H, Välineva T, Ashorn M, Wei B, Braun J, et al. Elevated serum anti-Saccharomyces cerevisiae, anti-I2 and anti-OmpW antibody levels in patients with suspicion of celiac disease. J Clin Immunol 2008;28:486-94. https://doi.org/10.1007/s10875-008-9200-9.
79. Barada K, Habib RH, Malli A, Hashash JG, Halawi H, Maasri K, et al. Prediction of celiac disease at endoscopy. Endoscopy 2014;46:110-19. https://doi.org/10.1055/s-0033-1359200.
80. Bayram Y, Parlak M, Aypak C, Bayram I, Yilmaz D, Cikman A. Diagnostic accuracy of IgA anti-tissue transglutaminase in celiac disease in Van – Turkey. East J Med 2015;20:20-3.
81. Carroccio A, Vitale G, Di Prima L, Chifari N, Napoli S, La Russa C, et al. Comparison of anti-transglutaminase ELISAs and an anti-endomysial antibody assay in the diagnosis of celiac disease: a prospective study. Clin Chem 2002;48:1546-50. https://doi.org/10.1093/clinchem/48.9.1546.
82. Carroccio A, Di Prima L, Pirrone G, Scalici C, Florena AM, Gasparin M, et al. Anti-transglutaminase antibody assay of the culture medium of intestinal biopsy specimens can improve the accuracy of celiac disease diagnosis. Clin Chem 2006;52:1175-80. https://doi.org/10.1373/clinchem.2005.061366.
83. Dahle C, Hagman A, Ignatova S, Ström M. Antibodies against deamidated gliadin peptides identify adult coeliac disease patients negative for antibodies against endomysium and tissue transglutaminase. Aliment Pharmacol Ther 2010;32:254-60. https://doi.org/10.1111/j.1365-2036.2010.04337.x.
84. Dutta AK, Chacko A, Avinash B. Suboptimal performance of IgG anti-tissue transglutaminase in the diagnosis of celiac disease in a tropical country. Dig Dis Sci 2010;55:698-702. https://doi.org/10.1007/s10620-009-0789-1.
85. Emami MH, Karimi S, Kouhestani S, Hashemi M, Taheri H. Diagnostic accuracy of IgA anti-tissue transglutaminase in patients suspected of having coeliac disease in Iran. J Gastrointestin Liver Dis 2008;17:141-6.
86. Eremic N, Deic M, Hadnadev L. Diagnostic accuracy of IGA anti-tissue transglutaminase antibody testing in celiac disease. J Med Biochem 2012;31:100-6. https://doi.org/10.2478/v10011-011-0047-x.
87. Fernández ML, Vivas S, Ruiz de Morales JM, Marugán JM. Usefulness of anti-transglutaminase antibodies in the diagnosis of celiac disease. Gastroenterol Hepatol 2005;28:437-40. https://doi.org/10.1157/13078992.
88. Hadithi M, von Blomberg BM, Crusius JB, Bloemena E, Kostense PJ, Meijer JW, et al. Accuracy of serologic tests and HLA-DQ typing for diagnosing celiac disease. Ann Intern Med 2007;147:294-302. https://doi.org/10.7326/0003-4819-147-5-200709040-00003.
89. Holmes GKT, Forsyth JM, Knowles S, Seddon H, Hill PG, Austin AS. Coeliac disease: further evidence that biopsy is not always necessary for diagnosis. Eur J Gastroenterol Hepatol 2017;29:640-5. https://doi.org/10.1097/MEG.0000000000000841.
90. Hopper AD, Hadjivassiliou M, Hurlstone DP, Lobo AJ, McAlindon ME, Egner W, et al. What is the role of serologic testing in celiac disease? A prospective, biopsy-confirmed study with economic analysis. Clin Gastroenterol Hepatol 2008;6:314-20. https://doi.org/10.1016/j.cgh.2007.12.008.
91. Jarmi V, Cejas N, Kiener O, de Elías R, Balzola S, Cordoba C, et al. Diagnostic accuracy of antibodies to deamidated gliadin peptides. Acta Bioquim Clin Latinoam 2010;44:47-52.
92. Javaeed A, Shah W, Akhtar R, Ghauri SK, Ghauri SK, Alvi AH. Diagnostic accuracy of IgA anti-tissue transglutaminase antibodies in comparison with histopathological findings in celiac disease in Pakistan. Medical Forum Mon 2014;25:15-9.
93. Johnston SD, McMillan SA, Collins JS, Tham TC, McDougall NI, Murphy P. A comparison of antibodies to tissue transglutaminase with conventional serological tests in the diagnosis of coeliac disease. Eur J Gastroenterol Hepatol 2003;15:1001-4. https://doi.org/10.1097/00042737-200309000-00010.
94. Kashif T, Kashif A, Kashif Z. Prevalence of celiac disease and role of anti-tissue transglutaminase antibodies in different types of intestinal damage in celiac disease according to Marsh classification. Pakistan J Medical Health Sci 2017;11.
95. Kurien M, Johnston A, Avgerinos A, Sanders DS. Does IgA tissue transglutaminase antibody levels correlate with histological findings of coeliac disease (CD)?. United European Gastroenterol J 2013;1:A22-A3.
96. Lau MS, Mooney PD, White WL, Rees MA, Wong SH, Hadjivassiliou M, et al. Office-based point of care testing (IgA/IgG-deamidated gliadin peptide) for celiac disease. Am J Gastroenterol 2018;113:1238-46. https://doi.org/10.1038/s41395-018-0143-3.
97. Llorente MJ, Sebastián M, Fernández-Aceñero MJ, Serrano G, Villanueva S, Prieto G. IgA antibodies against tissue transglutaminase in the diagnosis of celiac disease: concordance with intestinal biopsy in children and adults. Clin Chem 2004;50:451-3. https://doi.org/10.1373/clinchem.2003.024976.
98. Longarini GI, Sugai E, Hwang H, Nachman F, Vazquez H, Costa A, et al. How does specific serology match with esphgan serologic guidelines for diagnosis of celiac disease in a prospective cohort of adults with high pretest probability?. United European Gastroenterol J 2014;2.
99. Mooney PD, Kurien M, Johnston AJ, Wong SH, Avgerinos A, Hadjivassiliou M, et al. Point of care testing for adult celiac disease: a novel role in endoscopy. Gastroenterology 2014;146:S-472. https://doi.org/10.1016/S0016-5085%2814%2961693-0.
100. Mooney PD, Wong SH, Johnston AJ, Kurien M, Avgerinos A, Sanders DS. Increased detection of celiac disease with measurement of deamidated gliadin peptide antibody before endoscopy. Clin Gastroenterol Hepatol 2015;13:1278-84.e1. https://doi.org/10.1016/j.cgh.2015.01.010.
101. Niveloni S, Sugai E, Cabanne A, Vazquez H, Argonz J, Smecuol E, et al. Antibodies against synthetic deamidated gliadin peptides as predictors of celiac disease: prospective assessment in an adult population with a high pretest probability of disease. Clin Chem 2007;53:2186-92. https://doi.org/10.1373/clinchem.2006.081364.
102. Pallav K, Kabbani T, Tariq S, Vanga R, Kelly CP, Leffler DA. Clinical utility of celiac disease-associated HLA testing. Dig Dis Sci 2014;59:2199-206. https://doi.org/10.1007/s10620-014-3143-1.
103. Previtali G, Licini L, D’Antiga L, Marseglia A, Ravasio R, Nembrini F, et al. Celiac disease diagnosis without biopsy: is a 10× ULN anti-transglutaminase result suitable for a chemiluminescence method?. J Pediatr Gastroenterol Nutr 2018;66:645-50. https://doi.org/10.1097/MPG.0000000000001773.
104. Rahmati A, Shakeri R, Sohrabi M, Alipour A, Boghratian A, Setareh M, et al. Correlation of tissue transglutaminase antibody with duodenal histologic marsh grading. Middle East J Dig Dis 2014;6:131-6.
105. Reeves GE, Squance ML, Duggan AE, Murugasu RR, Wilson RJ, Wong RC, et al. Diagnostic accuracy of coeliac serological tests: a prospective study. Eur J Gastroenterol Hepatol 2006;18:493-501. https://doi.org/10.1097/00042737-200605000-00006.
106. Santaolalla R, Fernández-Bañares F, Rodríguez R, Alsina M, Rosinach M, Mariné M, et al. Diagnostic value of duodenal antitissue transglutaminase antibodies in gluten-sensitive enteropathy. Aliment Pharmacol Ther 2008;27:820-9. https://doi.org/10.1111/j.1365-2036.2008.03652.x.
107. Sayed SK, Imam HM, Mahran AM, Refaiy AM. Diagnostic utility of deamidated gliadin peptide antibody in celiac disease compared to anti-tissue transglutaminase and IgA- endomysium antibodies. Egypt J Immunol 2012;19:41-52.
108. Scoglio R, Di Pasquale G, Pagano G, Lucanto MC, Magazzu G, Sferlazzas C. Is intestinal biopsy always needed for diagnosis of celiac disease?. Am J Gastroenterol 2003;98:1325-31. https://doi.org/10.1111/j.1572-0241.2003.07455.x.
109. Sugai E, Moreno ML, Hwang HJ, Cabanne A, Crivelli A, Nachman F, et al. Celiac disease serology in patients with different pretest probabilities: is biopsy avoidable?. World J Gastroenterol 2010;16:3144-52. https://doi.org/10.3748/wjg.v16.i25.3144.
110. Swallow K, Wild G, Sargur R, Sanders DS, Aziz I, Hopper AD, et al. Quality not quantity for transglutaminase antibody 2: the performance of an endomysial and tissue transglutaminase test in screening coeliac disease remains stable over time. Clin Exp Immunol 2013;171:100-6. https://doi.org/10.1111/cei.12000.
111. Tesei N, Sugai E, Vázquez H, Smecuol E, Niveloni S, Mazure R, et al. Antibodies to human recombinant tissue transglutaminase may detect coeliac disease patients undiagnosed by endomysial antibodies. Aliment Pharmacol Ther 2003;17:1415-23. https://doi.org/10.1046/j.1365-2036.2003.01595.x.
112. Volta U, Granito A, Parisi C, Fabbri A, Fiorini E, Piscaglia M, et al. Deamidated gliadin peptide antibodies as a routine test for celiac disease: a prospective analysis. J Clin Gastroenterol 2010;44:186-90. https://doi.org/10.1097/MCG.0b013e3181c378f6.
113. Whyte M, Smale S, Heil PM, Prasad P, Davis T, Sewell C, et al. IgA antibody to tissue transglutaminase is a highly sensitive predictor of adult onset coeliac disease in clinical practice. Gut 2001;48.
114. Zanini B, Magni A, Caselani F, Lanzarotto F, Carabellese N, Villanacci V, et al. High tissue-transglutaminase antibody level predicts small intestinal villous atrophy in adult patients at high risk of celiac disease. Dig Liver Dis 2012;44:280-5. https://doi.org/10.1016/j.dld.2011.10.013.
115. Pretto LE, Hernández MA, Eusebi SM, Tamagnini GA, Tibaldo AE, Pezzarini E, et al. Utility of Anti Deamidated Gliadin Peptides Antibodies in the Screening of Celiac Disease in Children n.d.
116. Rabbani MW, Aziz MT, Basit A, Iqbal I. Diagnostic accuracy of serum anti-tissue transglutaminase IgA, in the diagnosis of celiac disease, taking small bowel biopsy as a gold standard. Pak Paediatr J 2015;39:203-8.
117. Roca M, Donat E, Marco-Maestud N, Masip E, Hervás-Marín D, Ramos D, et al. Efficacy study of anti-endomysium antibodies for celiac disease diagnosis: a retrospective study in a Spanish pediatric population. J Clin Med 2019;8. https://doi.org/10.3390/jcm8122179.
118. Saneian H, Gorgani AM. Diagnostic value of serologic tests in celiac screening. Int J Prev Med 2012;3:S58-63.
119. Tronconi GM, Parma B, Bazzigaluppi E, Corsin P, Barera G. Antibodies against synthetic deamidated gliadin peptides for the diagnosis of childhood celiac disease: do they play a role?. Dig Liver Dis 2010;42. https://doi.org/10.1016/S1590-8658(10)60615-9.
120. Weile B, Heegaard NH, Høier-Madsen M, Wiik A, Krasilnikoff PA. Tissue transglutaminase and endomysial autoantibodies measured in an historical cohort of children and young adults in whom coeliac disease was suspected. Eur J Gastroenterol Hepatol 2002;14:71-6. https://doi.org/10.1097/00042737-200201000-00012.
121. Wolf J, Petroff D, Richter T, Auth MKH, Uhlig HH, Laass MW, et al. Validation of antibody-based strategies for diagnosis of pediatric celiac disease without biopsy. Gastroenterology 2017;153:410-19.e17. https://doi.org/10.1053/j.gastro.2017.04.023.
122. Mubarak A, Houwen R. Tissue transglutaminase antibody levels ≥ 100 U/mL and celiac disease: the first prospective study. Gastroenterology 2012;142. https://doi.org/10.1016/S0016-5085(12)61048-8.
123. Panetta F, Torre G, Colistro F, Ferretti F, Daniele A, Diamanti A. Clinical accuracy of anti-tissue transglutaminase as screening test for celiac disease under 2 years. Acta Paediatr 2011;100:728-31. https://doi.org/10.1111/j.1651-2227.2010.02124.x.
124. Parizade M, Bujanover Y, Weiss B, Nachmias V, Shainberg B. Performance of serology assays for diagnosing celiac disease in a clinical setting. Clin Vaccine Immunol 2009;16:1576-82. https://doi.org/10.1128/CVI.00205-09.
125. Pettie MJ, Levine JJ. The use of tissue transglutaminase IGG antibody (TTG-G) in the diagnosis of celiac disease. J Pediatr Gastroenterol Nutr 2009;49.
126. Poddar U, Thapa BR, Nain CK, Singh K. Is tissue transglutaminase autoantibody the best for diagnosing celiac disease in children of developing countries?. J Clin Gastroenterol 2008;42:147-51. https://doi.org/10.1097/MCG.0b013e31802fc1e3.
127. Polanco I, Esteban MM, Larrauri J. Relation of anti-tissue transglutaminase IgA antibodies with the morphology of the intestinal mucosa in children with celiac disease. Pediatrika 2001;21:9-20.
128. Olen O, Gudjónsdóttir AH, Browaldh L, Hessami M, Elvin K, Liedberg AS, et al. Antibodies against deamidated gliadin peptides and tissue transglutaminase for diagnosis of pediatric celiac disease. J Pediatr Gastroenterol Nutr 2012;55:695-700. https://doi.org/10.1097/MPG.0b013e3182645c54.
129. Nguyen CMT, Portolese O, Yazdani PA, . Validation of the ESPGHAN guidelines for the diagnosis of coeliac disease in a Canadian population. J Pediatr Gastroenterol Nutr 2019;130.
130. Kutty JS, McDermott M, O’Sullivan M, Quinn S, Broderick A, Bourke B, et al. Coeliac serology testing in Irish children – a substitute for biopsy?. Arch Dis Child 2014;99:A3-A4. https://doi.org/10.1136/archdischild-2014-306237.7.
131. Mubarak A, Gmelig-Meyling FH, Wolters VM, Ten Kate FJ, Houwen RH. Immunoglobulin G antibodies against deamidated-gliadin-peptides outperform anti-endomysium and tissue transglutaminase antibodies in children &lt; 2 years age. APMIS 2011;119:894-900. https://doi.org/10.1111/j.1600-0463.2011.02817.x.
132. Kabatova J, Hust’ak R, Blazickova S, Bosak V. Evaluation and applicability of nonbiopsy criteria in children and adolescents according to ESPGHAN for diagnosis of celiac disease. Gastroenterol Hepatol 2017;71:469-75. https://doi.org/10.14735/amgh2017csgh.info07.
133. Iwańczak F, Iwanczak B, Matusiewicz K, Mowszet K. Assessment of serum antibodies against gliadin, endomysium and tissue transglutaminase in diagnosis of celiac disease in children. Gastroenterol Pol 2003;10:323-8.
134. Jora R, Raghuvanshi V, Payal V, Sharma P, Vishnoi SK. Correlation of tissue transglutaminase with modified Marsh grading in celiac disease: a prospective cohort study. Indian J Pediatr 2017;84:515-20. https://doi.org/10.1007/s12098-017-2323-3.
135. Hollén E, Farnebäck M, Forslund T, Magnusson KE, Sundqvist T, Fälth-Magnusson K. Evaluation of multiple diagnostic indicators in comparison to the intestinal biopsy as the golden standard in diagnosing celiac disease in children. Med Sci 2016;4. https://doi.org/10.3390/medsci4040020.
136. Hojsak I, Mozer-Glassberg Y, Segal Gilboa N, Weinberger R, Hartman C, Shamir R. Celiac disease screening assays for children younger than 3 years of age: the performance of three serological tests. Dig Dis Sci 2012;57:127-32. https://doi.org/10.1007/s10620-011-1857-x.
137. Clouzeau H, Taupin JL, Le Bail B, Michaud L, Bridoux-Henno L, Olives J, et al. Diagnostic value of HLA DQ2/DQ8 and antitransglutaminase antibodies for the diagnosis of coeliac disease: is jejunal biopsy still mandatory?. J Pediatr Gastroenterol Nutr 2007;44.
138. Dahlbom I, Nyberg BI, Berntson L, Hansson T. Simultaneous detection of IgA and IgG antibodies against tissue transglutaminase: the preferred pre-biopsy test in childhood celiac disease. Scand J Clin Lab Invest 2016;76:208-16. https://doi.org/10.3109/00365513.2015.1137348.
139. Diamanti A, Colistro F, Calce A, Devito R, Ferretti F, Minozzi A, et al. Clinical value of immunoglobulin A antitransglutaminase assay in the diagnosis of celiac disease. Pediatrics 2006;118:e1696-700. https://doi.org/10.1542/peds.2006-0604.
140. Donaldson MR, Firth SD, Wimpee H, Leiferman KM, Zone JJ, Horsley W, et al. Correlation of duodenal histology with tissue transglutaminase and endomysial antibody levels in pediatric celiac disease. Clin Gastroenterol Hepatol 2007;5:567-73. https://doi.org/10.1016/j.cgh.2007.01.003.
141. Gidrewicz D, Potter K, Trevenen CL, Lyon M, Butzner JD. Evaluation of the ESPGHAN celiac guidelines in a North American pediatric population. Am J Gastroenterol 2015;110:760-7. https://doi.org/10.1038/ajg.2015.87.
142. Hashmi MA, Hussain T, Masood N, Younas M, Asghar RM, Shafi MS. Accuracy of anti-tissue transglutaminase IgA antibody in the diagnosis of paediatric celiac disease. J Coll Physicians Surg Pak 2016;26:263-6.
143. Bozzato D, Rossi E, Pelloso M, Guariso G, Moz S, Padoan A, et al. Aeskulisa and H-TTG/DGP screen for celiac disease diagnosis. Clin Chem Lab Med 2011;49.
144. Chan AW, Butzner JD, McKenna R, Fritzler MJ. Tissue transglutaminase enzyme-linked immunosorbent assay as a screening test for celiac disease in pediatric patients. Pediatrics 2001;107. https://doi.org/10.1542/peds.107.1.e8.
145. Bishop J, Reed P, Austin P, Hurst M, Ameratunga R, Craigie A, et al. Prospective evaluation of the ESPGHAN guidelines for diagnosis of celiac disease in New Zealand children. J Pediatr Gastroenterol Nutr 2018;67:749-54. https://doi.org/10.1097/MPG.0000000000002065.
146. Barker CC, Mitton C, Jevon G, Mock T. Can tissue transglutaminase antibody titers replace small-bowel biopsy to diagnose celiac disease in select pediatric populations?. Pediatrics 2005;115:1341-6. https://doi.org/10.1542/peds.2004-1392.
147. Basso D, Gallo N, Guariso G, Pittoni M, Piva MG, Plebani M. Role of anti-transglutaminase (anti-tTG), anti-gliadin, and anti-endomysium serum antibodies in diagnosing celiac disease: a comparison of four different commercial kits for anti-tTG determination. J Clin Lab Anal 2001;15:112-15. https://doi.org/10.1002/jcla.1012.
148. Alonso M, Saenz M, Echaniz P, Prada A, De Juan MD, Urreta I, et al. Evaluation of intraepithelial duodenal subsets, marsh pathological grades, and anti transglutaminase antibodies in diagnosis of celiac disease in a paediatric population. Eur J Immunol 2009;39.
149. Ali Z, Hagglund K, Lyons A, Sheikh U, Lyons H. Is small intestinal biopsy always necessary to diagnose celiac disease in children?. Am J Gastroenterol 2012;107. https://doi.org/10.14309/00000434-201210001-01930.
150. Aberg AK, Olcén P. Serologic screening for celiac disease in children: a comparison between established assays and tests with deamidated gliadin-derived peptides plus conjugates for both IgA and IgG antibodies. APMIS 2009;117:808-13. https://doi.org/10.1111/j.1600-0463.2009.02541.x.
151. Abdollahi A, Mehrazma M, Talachian E. The diagnostic value of serum IGG-antigliadin, IGA anti-endomysial and IGA anti-tissue transglutaminase antibodies for pediatric celiac disease. Pediatr Res 2011;70. https://doi.org/10.1038/pr.2011.1000.
152. Bowron A, Moorghen M, Morgan JE, Osborne JR, Stansbie D, Stone JE. Cost-effective strategy for the serological investigation of coeliac disease. Ann Clin Biochem 2000;37:467-70. https://doi.org/10.1177/000456320003700406.
153. Corrao G, Corazza GR, Andreani ML, Torchio P, Valentini RA, Galatola G, et al. Serological screening of coeliac disease: choosing the optimal procedure according to various prevalence values. Gut 1994;35:771-5. https://doi.org/10.1136/gut.35.6.771.
154. Atkinson K, Tokmakajian S, Watson W, Gregor J. Evaluation of the endomysial antibody for celiac disease: operating properties and associated cost implications in clinical practice. Can J Gastroenterol 1997;11:673-7. https://doi.org/10.1155/1997/813148.
155. Dickey W, McMillan SA, McCrum EE, Evans AE. Association between serum levels of total IgA and IgA class endomysial and antigliadin antibodies: implications for coeliac disease screening. Eur J Gastroenterol Hepatol 1997;9:559-62. https://doi.org/10.1097/00042737-199706000-00002.
156. Feighery C, Weir DG, Whelan A, Willoughby R, Youngprapakorn S, Lynch S, et al. Diagnosis of gluten-sensitive enteropathy: is exclusive reliance on histology appropriate?. Eur J Gastroenterol Hepatol 1998;10:919-25. https://doi.org/10.1097/00042737-199811000-00004.
157. Kotze LM, Utiyama SR, Nisihara RM, de Camargo VF, Ioshii SO. IgA class anti-endomysial and anti-tissue transglutaminase antibodies in relation to duodenal mucosa changes in coeliac disease. Pathology 2003;35:56-60. https://doi.org/10.1097/01268031-200335010-00010.
158. Lau MS, Mooney PD, Rees MA, White WL, Marks LJ, Hadjivassiliou M, et al. Gluten free diet adherence assessment using CDAT and BIAGI questionnaires in patients with coeliac disease. Gut 2018;67:A160-A1. https://doi.org/10.1136/gutjnl-2018-BSGAbstracts.318.
159. McMillan SA, Haughton DJ, Biggart JD, Edgar JD, Porter KG, McNeill TA. Predictive value for coeliac disease of antibodies to gliadin, endomysium, and jejunum in patients attending for jejunal biopsy. BMJ 1991;303:1163-5. https://doi.org/10.1136/bmj.303.6811.1163.
160. Mooney PD, Kurien M, Evans KE, Chalkiadakis I, Hale MF, Kannan MZ, et al. Point-of-care testing for celiac disease has a low sensitivity in endoscopy. Gastrointest Endosc 2014;80:456-62. https://doi.org/10.1016/j.gie.2014.02.009.
161. Valdimarsson T, Franzen L, Grodzinsky E, Skogh T, Ström M. Is small bowel biopsy necessary in adults with suspected celiac disease and IgA anti-endomysium antibodies?. Dig Dis Sci 1996;41:83-7. https://doi.org/10.1007/BF02208588.
162. Vogelsang H, Genser D, Wyatt J, Lochs H, Ferenci P, Granditsch G, et al. Screening for celiac disease: a prospective study on the value of noninvasive tests. Am J Gastroenterol 1995;90:394-8.
163. Bhatnagar S, Gupta SD, Mathur M, Phillips AD, Kumar R, Knutton S, et al. Celiac disease with mild to moderate histologic changes is a common cause of chronic diarrhea in Indian children. J Pediatr Gastroenterol Nutr 2005;41:204-9. https://doi.org/10.1097/01.mpg.0000172261.24115.29.
164. Bürgin-Wolff A, Gaze H, Hadziselimovic F, Lentze M, Nusslé D, Hadziselimovic F, et al. Inflammatory Bowel Disease and Coeliac Disease in Children. Dordrecht: Springer; 1990.
165. Camarero C, Roldan B, Sebastian M, Barrio A, Alvarez I, Eiras P, et al. Predictive value of antigliadin, antireticulin, and antiendomysium antibodies in the diagnosis of gluten-associated enteropathy. Rev Esp Pediatr 1997;53:309-14.
166. Del Rosario MA, Fitzgerald JF, Chong SK, Croffie JM, Gupta SK. Further studies of anti-endomysium and anti-gliadin antibodies in patients with suspected celiac disease. J Pediatr Gastroenterol Nutr 1998;27:191-5. https://doi.org/10.1097/00005176-199808000-00012.
167. Fernández E, Riestra S, Rodrigo L, Blanco C, López-Vázquez A, Fuentes D, et al. Comparison of six human anti-transglutaminase ELISA-tests in the diagnosis of celiac disease in the Saharawi population. World J Gastroenterol 2005;11:3762-6. https://doi.org/10.3748/wjg.v11.i24.3762.
168. Gandolfi L, Catassi C, Garcia S, Modelli IC, Campos D, Pratesi R. Antiendomysial antibody test reliability in children with frequent diarrhea and malnutrition: is it celiac disease?. J Pediatr Gastroenterol Nutr 2001;33:483-7. https://doi.org/10.1097/00005176-200110000-00013.
169. Grodzinsky E, Jansson G, Skogh T, Stenhammar L, Fälth-Magnusson K. Anti-endomysium and anti-gliadin antibodies as serological markers for coeliac disease in childhood: a clinical study to develop a practical routine. Acta Paediatr 1995;84:294-8. https://doi.org/10.1111/j.1651-2227.1995.tb13631.x.
170. Lindquist BL, Rogozinski T, Moi H, Danielsson D, Olcén P. Endomysium and gliadin IgA antibodies in children with coeliac disease. Scand J Gastroenterol 1994;29:452-6. https://doi.org/10.3109/00365529409096837.
171. Mendez NI, Guastavino MA, Zacharzewski CL, Zapata PD, Sprang M. Autoantibodies: utility in coeliac disease. a study in Misiones province. Acta Bioquim Clin Latinoam 1998;32:55-61.
172. Russo PA, Chartrand LJ, Seidman E. Comparative analysis of serologic screening tests for the initial diagnosis of celiac disease. Pediatrics 1999;104:75-8. https://doi.org/10.1542/peds.104.1.75.
173. Vargas Pérez ML, Melero Ruiz J, Fernández de Mera J, González Roiz C, Catalina Fernández I, Romero Albillos A. Serological and genetic markers in the diagnosis and follow-up of coeliac disease. An Pediatr 2005;62:412-19. https://doi.org/10.1157/13074614.
174. Wildfang S, Knauss M, Stern M. IgA endomysium antibodies. Detection in children with celiac disease. Monatsschr Kinderheilkd 1992;140:639-45.
175. Kurien M, Johnston A, Avgerinos A, Sanders DS. Does IGA tissue transglutaminase antibody levels correlate with histological findings of coeliac disease (CD)?. Gut 2013;62:A26-A. https://doi.org/10.1136/gutjnl-2013-304907.059.
176. Mooney PD, Kurien M, Wong S, Sanders DS. Point of care testing for adult coeliac disease: a potential role in endoscopy. United European Gastroenterol J 2014;2.
177. Björck S, Brundin C, Lörinc E, Lynch KF, Agardh D. Screening detects a high proportion of celiac disease in young HLA-genotyped children. J Pediatr Gastroenterol Nutr 2010;50:49-53. https://doi.org/10.1097/MPG.0b013e3181b477a6.
178. Mäki M, Mustalahti K, Kokkonen J, Kulmala P, Haapalahti M, Karttunen T, et al. Prevalence of celiac disease among children in Finland. N Engl J Med 2003;348:2517-24. https://doi.org/10.1056/NEJMoa021687.
179. Sandström O, Rosén A, Lagerqvist C, Carlsson A, Hernell O, Högberg L, et al. Transglutaminase IgA antibodies in a celiac disease mass screening and the role of HLA-DQ genotyping and endomysial antibodies in sequential testing. J Pediatr Gastroenterol Nutr 2013;57:472-6. https://doi.org/10.1097/MPG.0b013e31829ef65d.
180. Wahab RJ, Beth SA, Derks IPM, Jansen PW, Moll HA, Kiefte-de Jong JC. Celiac disease autoimmunity and emotional and behavioral problems in childhood. Pediatrics 2019;144. https://doi.org/10.1542/peds.2018-3933.
181. Beth SA, Jansen MAE, Elfrink MEC, Kiefte-de Jong JC, Wolvius EB, Jaddoe VWV, et al. Generation R birth cohort study shows that specific enamel defects were not associated with elevated serum transglutaminase type 2 antibodies. Acta Paediatr 2016;105:e485-e91. https://doi.org/10.1111/apa.13533.
182. Jansen MA, Kiefte-de Jong JC, Gaillard R, Escher JC, Hofman A, Jaddoe VW, et al. Growth trajectories and bone mineral density in anti-tissue transglutaminase antibody-positive children: the Generation R Study. Clin Gastroenterol Hepatol 2015;13:913-20.e5. https://doi.org/10.1016/j.cgh.2014.09.032.
183. National Institute for Health and Care Research Applied Research Collaboration West . Coeliac Disease Survey, Video 1 n.d. https://youtube/M5tfW8-yLfM (accessed 7 February 2022).
184. National Institute for Health and Care Research Applied Research Collaboration West . Coeliac Disease Survey, Video 2 n.d. https://youtu.be/_PVLGDVSL3E (accessed 7 February 2022).
185. National Institute for Health and Care Research Applied Research Collaboration West . Coeliac Disease Survey, Video 3 n.d. https://youtu.be/GL8tzH4IxAk (accessed 7 February 2022).
186. International Organization for Standardization . ISO/IEC/27001:/Information/Security/Management n.d. www.iso.org/isoiec-27001-information-security.html (accessed 27 May 2022).
187. Information Commissioner’s Office . Guide to the UK General Data Protection Regulation (UK GDPR) n.d. https://ico.org.uk/for-organisations/guide-to-data-protection/guide-to-the-general-data-protection-regulation-gdpr/ (accessed 27 May 2022).
188. Dorn SD, Hernandez L, Minaya MT, Morris CB, Hu Y, Leserman J, et al. The development and validation of a new coeliac disease quality of life survey (CD-QOL). Aliment Pharmacol Ther 2010;31:666-75. https://doi.org/10.1111/j.1365-2036.2009.04220.x.
189. Scottish Government . Scottish Index of Multiple Deprivation 2020v2 Postcode Lookup File n.d. www.gov.scot/publications/scottish-index-of-multiple-deprivation-2020v2-postcode-look-up/ (accessed 27 May 2022).
190. Welsh Government . Welsh Index of Multiple Deprivation (WIMD) Index Guidance, Technical Information and Background Documentation n.d. https://gov.wales/welsh-index-multiple-deprivation-index-guidance (accessed 27 May 2022).
191. Northern Ireland Statistics and Research Agency . NI Multiple Deprivation Measures n.d. www.nisra.gov.uk/statistics/deprivation (accessed 27 May 2022).
192. Hsieh HF, Shannon SE. Three approaches to qualitative content analysis. Qual Health Res 2005;15:1277-88. https://doi.org/10.1177/1049732305276687.
193. National Institute for Health and Care Excellence (NICE) . Coeliac Disease. Full Health Economics Report 2015.
194. Mohseninejad L, Feenstra T, van der Horst HE, Woutersen-Koch H, Buskens E. Targeted screening for coeliac disease among irritable bowel syndrome patients: analysis of cost-effectiveness and value of information. Eur J Health Econ 2013;14:947-57. https://doi.org/10.1007/s10198-012-0441-4.
195. Harewood GC, Murray JA. Diagnostic approach to a patient with suspected celiac disease: a cost analysis. Dig Dis Sci 2001;46:2510-14. https://doi.org/10.1023/a:1012396408370.
196. Mein SM, Ladabaum U. Serological testing for coeliac disease in patients with symptoms of irritable bowel syndrome: a cost-effectiveness analysis. Aliment Pharmacol Ther 2004;19:1199-210. https://doi.org/10.1111/j.1365-2036.2004.01958.x.
197. Spiegel BM, DeRosa VP, Gralnek IM, Wang V, Dulai GS. Testing for celiac sprue in irritable bowel syndrome with predominant diarrhea: a cost-effectiveness analysis. Gastroenterology 2004;126:1721-32. https://doi.org/10.1053/j.gastro.2004.03.012.
198. Shamir R, Hernell O, Leshno M. Cost-effectiveness analysis of screening for celiac disease in the adult population. Med Decis Making 2006;26:282-93. https://doi.org/10.1177/0272989X06289012.
199. Swigonski NL, Kuhlenschmidt HL, Bull MJ, Corkins MR, Downs SM. Screening for celiac disease in asymptomatic children with Down syndrome: cost-effectiveness of preventing lymphoma. Pediatrics 2006;118:594-602. https://doi.org/10.1542/peds.2005-2123.
200. Dorn SD, Matchar DB. Cost-effectiveness analysis of strategies for diagnosing celiac disease. Dig Dis Sci 2008;53:680-8. https://doi.org/10.1007/s10620-007-9939-5.
201. Chang M, Green PH. Genetic testing before serologic screening in relatives of patients with celiac disease as a cost containment method. J Clin Gastroenterol 2009;43:43-50. https://doi.org/10.1097/MCG.0b013e318187311d.
202. Hershcovici T, Leshno M, Goldin E, Shamir R, Israeli E. Cost effectiveness of mass screening for coeliac disease is determined by time-delay to diagnosis and quality of life on a gluten-free diet. Aliment Pharmacol Ther 2010;31:901-10. https://doi.org/10.1111/j.1365-2036.2010.04242.x.
203. Park KT, Tsai R, Wang L, Khavari N, Bachrach L, Bass D. Cost-effectiveness of universal serologic screening to prevent nontraumatic hip and vertebral fractures in patients with celiac disease. Clin Gastroenterol Hepatol 2013;11:645-53. https://doi.org/10.1016/j.cgh.2012.12.037.
204. Jönsson B, Christiansen C, Johnell O, Hedbrandt J. Cost-effectiveness of fracture prevention in established osteoporosis. Osteoporos Int 1995;5:136-42. https://doi.org/10.1007/BF01623315.
205. Oleksik A, Lips P, Dawson A, Minshall ME, Shen W, Cooper C, et al. Health-related quality of life in postmenopausal women with low BMD with or without prevalent vertebral fractures. J Bone Miner Res 2009;15:1384-92. https://doi.org/10.1359/jbmr.2000.15.7.1384.
206. Yang JJ, Thanataveerat A, Green PH, Lebwohl B. Cost effectiveness of routine duodenal biopsy analysis for celiac disease during endoscopy for gastroesophageal reflux. Clin Gastroenterol Hepatol 2015;13:1437-43. https://doi.org/10.1016/j.cgh.2015.03.022.
207. Broide E, Matalon S, Kriger-Sharabi O, Richter V, Shirin H, Leshno M. Cost effectiveness of routine duodenal biopsies in iron deficiency anemia. World J Gastroenterol 2016;22:7813-23. https://doi.org/10.3748/wjg.v22.i34.7813.
208. Siebert U, Alagoz O, Bayoumi AM, Jahn B, Owens DK, Cohen DJ, et al. State-transition modelling: a report of the ISPOR-SMDM Modelling Good Research Practices Task Force-3. Value Health 2012;15:812-20. https://doi.org/10.1016/j.jval.2012.06.014.
209. Coeliac UK. Ongoing Symptoms and Complications n.d. www.coeliac.org.uk/healthcare-professionals/management/ongoing-symptoms-and-complications/ (accessed 25 January 2021).
210. Welton NJ, McAleenan A, Thom HH, Davies P, Hollingworth W, Higgins JP, et al. Screening strategies for atrial fibrillation: a systematic review and cost-effectiveness analysis. Health Technol Assess 2017;21. https://doi.org/10.3310/hta21290.
211. Iragorri N, Spackman E. Assessing the value of screening tools: reviewing the challenges and opportunities of cost-effectiveness analysis. Public Health Rev 2018;39. https://doi.org/10.1186/s40985-018-0093-8.
212. Briggs AH, Claxton K, Sculpher MJ. Decision Modelling for Health Economic Evaluation. Oxford: Oxford University Press; 2006.
213. Johnson B, Basson MD. Absence of complications after endoscopic mucosal biopsy. Dig Dis 2018;36:328-32. https://doi.org/10.1159/000489394.
214. Hall NJ, Rubin G, Charnock A. Systematic review: adherence to a gluten-free diet in adult patients with coeliac disease. Aliment Pharmacol Ther 2009;30:315-30. https://doi.org/10.1111/j.1365-2036.2009.04053.x.
215. Muhammad H, Reeves S, Jeanes YM. Identifying and improving adherence to the gluten-free diet in people with coeliac disease. Proc Nutr Soc 2019;78:418-25. https://doi.org/10.1017/S002966511800277X.
216. Dhalwani NN, West J, Sultan AA, Ban L, Tata LJ. Women with celiac disease present with fertility problems no more often than women in the general population. Gastroenterology 2014;147:1267-74.e1. https://doi.org/10.1053/j.gastro.2014.08.025.
217. Tata LJ, Card TR, Logan RF, Hubbard RB, Smith CJ, West J. Fertility and pregnancy-related events in women with celiac disease: a population-based cohort study. Gastroenterology 2005;128:849-55. https://doi.org/10.1053/j.gastro.2005.02.017.
218. Zugna D, Richiardi L, Akre O, Stephansson O, Ludvigsson JF. A nationwide population-based study to determine whether coeliac disease is associated with infertility. Gut 2010;59:1471-5. https://doi.org/10.1136/gut.2010.219030.
219. Hogen Esch CE, Van Rijssen MJ, Roos A, Koning F, Dekker FW, Mearin ML, et al. Screening for unrecognized coeliac disease in subfertile couples. Scand J Gastroenterol 2011;46:1423-8. https://doi.org/10.3109/00365521.2011.615858.
220. Fuchs V, Kurppa K, Huhtala H, Laurila K, Mäki M, Collin P, et al. Serology-based criteria for adult coeliac disease have excellent accuracy across the range of pre-test probabilities. Aliment Pharmacol Ther 2019;49:277-84. https://doi.org/10.1111/apt.15109.
221. Hill PG, Holmes GK. Coeliac disease: a biopsy is not always necessary for diagnosis. Aliment Pharmacol Ther 2008;27:572-7. https://doi.org/10.1111/j.1365-2036.2008.03609.x.
222. NHS . National Schedule of NHS Costs 2019 n.d. www.england.nhs.uk/national-cost-collection/ (accessed 5 October 2021).
223. Curtis L, Burns A. Unit Costs of Health and Social Care 2020. Canterbury: PSSRU, University of Kent; 2020.
224. Wang HI, Smith A, Aas E, Roman E, Crouch S, Burton C, et al. Treatment cost and life expectancy of diffuse large B-cell lymphoma (DLBCL): a discrete event simulation model on a UK population-based observational cohort. Eur J Health Econ 2017;18:255-67. https://doi.org/10.1007/s10198-016-0775-4.
225. Department of Health and Social Care . Payment by Results in the NHS: Tariff for 2013 to 2014 2013. www.gov.uk/government/publications/payment-by-results-pbr-operational-guidance-and-tariffs (accessed October 2021).
226. Curtis EM, van der Velde R, Moon RJ, van den Bergh JP, Geusens P, de Vries F, et al. Epidemiology of fractures in the United Kingdom 1988–2012: variation with age, sex, geography, ethnicity and socioeconomic status. Bone 2016;87:19-26. https://doi.org/10.1016/j.bone.2016.03.006.
227. Leal J, Gray AM, Prieto-Alhambra D, Arden NK, Cooper C, Javaid MK, et al. REFReSH study group. Impact of hip fracture on hospital care costs: a population-based study. Osteoporos Int 2016;27:549-58. https://doi.org/10.1007/s00198-015-3277-9.
228. Dolan P, Torgerson DJ. The cost of treating osteoporotic fractures in the United Kingdom female population. Osteoporos Int 1998;8:611-17. https://doi.org/10.1007/s001980050107.
229. Office for National Statistics . Consumer Price Inflation Tables n.d. www.ons.gov.uk/economy/inflationandpriceindices/datasets/consumerpriceinflation (accessed 11 November 2021).
230. Janssen B, Szende A, Szende A, Janssen B, Cabases J. Self-Reported Population Health: An International Perspective Based on EQ-5D 2014:19-30. https://doi.org/10.1007/978-94-007-7596-1_3.
231. Violato M, Gray A. The impact of diagnosis on health-related quality of life in people with coeliac disease: a UK population-based longitudinal perspective. BMC Gastroenterol 2019;19. https://doi.org/10.1186/s12876-019-0980-6.
232. Casellas F, Rodrigo L, Vivancos JL, Riestra S, Pantiga C, Baudet JS, et al. Factors that impact health-related quality of life in adults with celiac disease: a multicenter study. World J Gastroenterol 2008;14:46-52. https://doi.org/10.3748/wjg.14.46.
233. Si L, Winzenberg TM, de Graaff B, Palmer AJ. A systematic review and meta-analysis of utility-based quality of life for osteoporosis-related conditions. Osteoporos Int 2014;25:1987-97. https://doi.org/10.1007/s00198-014-2636-2.
234. Fargier E, Ranchon F, Huot L, Guerre P, Safar V, Dony A, et al. SMABcare study: subcutaneous monoclonal antibody in cancer care: cost–consequence analysis of subcutaneous rituximab in patients with follicular lymphoma. Ann Hematol 2018;97:123-31. https://doi.org/10.1007/s00277-017-3147-y.
235. Ludvigsson JF, Michaelsson K, Ekbom A, Montgomery SM. Coeliac disease and the risk of fractures – a general population-based cohort study. Aliment Pharmacol Ther 2007;25:273-85. https://doi.org/10.1111/j.1365-2036.2006.03203.x.
236. Godfrey JD, Brantner TL, Brinjikji W, Christensen KN, Brogan DL, Van Dyke CT, et al. Morbidity and mortality among older individuals with undiagnosed celiac disease. Gastroenterology 2010;139:763-9. https://doi.org/10.1053/j.gastro.2010.05.041.
237. Silano M, Volta U, Mecchia AM, Dessí M, Di Benedetto R, De Vincenzi M. Collaborating centers of the Italian registry of the complications of coeliac disease. Delayed diagnosis of coeliac disease increases cancer risk. BMC Gastroenterol 2007;7. https://doi.org/10.1186/1471-230X-7-8.
238. Goldacre MJ, Wotton CJ, Yeates D, Seagroatt V, Jewell D. Cancer in patients with ulcerative colitis, Crohn’s disease and coeliac disease: record linkage study. Eur J Gastroenterol Hepatol 2008;20:297-304. https://doi.org/10.1097/MEG.0b013e3282f2a5e2.
239. Fleurence RL, Hollenbeak CS. Rates and probabilities in economic modelling: transformation, translation and appropriate application. PharmacoEconomics 2007;25:3-6. https://doi.org/10.2165/00019053-200725010-00002.
240. Office for National Statistics . National Life Tables, United Kingdom, 1980–1982 to 2017–2019 n.d. www.ons.gov.uk/peoplepopulationandcommunity/birthsdeathsandmarriages/lifeexpectancies/datasets/nationallifetablesunitedkingdomreferencetables (accessed 10 November 2020).
241. Office for National Statistics . Cancer Survival in England: Adults Diagnosed Between 2013 and 2017 and Followed Up to 2018 2018.
242. Lunn D, Jackson C, Best N, Thomas A, Spiegelhalter DJ. The BUGS Book : A Practical Introduction to Bayesian Analysis. Boca Raton, FL: CRC Press; 2013.
243. Klop C, Welsing PM, Cooper C, Harvey NC, Elders PJ, Bijlsma JW, et al. Mortality in British hip fracture patients, 2000–2010: a population-based retrospective cohort study. Bone 2014;66:171-7. https://doi.org/10.1016/j.bone.2014.06.011.
244. Joint Formulary Committee . British National Formulary n.d. www.medicinescomplete.com (accessed 11 November 2021).
245. National Institute for Health and Care Excellence . Guide to the Methods of Technology Appraisal 2013 n.d. www.nice.org.uk/process/pmg9/chapter/foreword (accessed 27 May 2022).
246. Wilson ECF. Methodological note: reporting deterministic versus probabilistic results of Markov, partitioned survival and other non-linear models. Appl Health Econ Health Policy 2021;19:789-95. https://doi.org/10.1007/s40258-021-00664-2.
247. Paul SP, Harries SL, Basude D. Barriers to implementing the revised ESPGHAN guidelines for coeliac disease in children: a cross-sectional survey of coeliac screen reporting in laboratories in England. Arch Dis Child 2017;102:942-6. https://doi.org/10.1136/archdischild-2016-312027.
248. Barton GR, Briggs AH, Fenwick EA. Optimal cost-effectiveness decisions: the role of the cost-effectiveness acceptability curve (CEAC), the cost-effectiveness acceptability frontier (CEAF), and the expected value of perfection information (EVPI). Value Health 2008;11:886-97. https://doi.org/10.1111/j.1524-4733.2008.00358.x.
249. Fenwick E, Steuten L, Knies S, Ghabri S, Basu A, Murray JF, et al. Value of information analysis for research decisions – an introduction: report 1 of the ISPOR Value of Information Analysis Emerging Good Practices Task Force. Value Health 2020;23:139-50. https://doi.org/10.1016/j.jval.2020.01.001.
250. Rothery C, Strong M, Koffijberg HE, Basu A, Ghabri S, Knies S, et al. Value of information analytical methods: report 2 of the ISPOR Value of Information Analysis Emerging Good Practices Task Force. Value Health 2020;23:277-86. https://doi.org/10.1016/j.jval.2020.01.004.
251. Baio G, Dawid AP. Probabilistic sensitivity analysis in health economics. Stat Methods Med Res 2015;24:615-34. https://doi.org/10.1177/0962280211419832.
252. Strong M, Oakley JE, Brennan A. Estimating multiparameter partial expected value of perfect information from a probabilistic sensitivity analysis sample: a nonparametric regression approach. Med Decis Making 2014;34:311-26. https://doi.org/10.1177/0272989X13505910.
253. Strong M, Oakley JE, Brennan A, Breeze P. Estimating the expected value of sample information using the probabilistic sensitivity analysis sample: a fast, nonparametric regression-based method. Med Decis Making 2015;35:570-83. https://doi.org/10.1177/0272989X15575286.
254. Fang W, Wang Z, Giles MB, Jackson CH, Welton NJ, Andrieu C, et al. Multilevel and quasi Monte Carlo methods for the calculation of the expected value of partial perfect information. Med Decis Making 2022;42:168-81. https://doi.org/10.1177/0272989X211026305.
255. Office for National Statistics . Population Estimates n.d. www.ons.gov.uk/peoplepopulationandcommunity/populationandmigration/populationestimates (accessed 10 November 2021).
256. UK Government . UK Population by Ethnicity: Age Groups n.d. www.ethnicity-facts-figures.service.gov.uk/uk-population-by-ethnicity/demographics/age-groups/latest (accessed 27 May 2022).
257. Incerti D, Thom H, Baio G, Jansen JP. R You still using Excel? The advantages of modern software tools for health technology assessment. Value Health 2019;22:575-9. https://doi.org/10.1016/j.jval.2019.01.003.
258. Antiga E, Caproni M. The diagnosis and treatment of dermatitis herpetiformis. Clin Cosmet Investig Dermatol 2015;8:257-65. https://doi.org/10.2147/CCID.S69127.
259. Kunst N, Wilson ECF, Glynn D, Alarid-Escudero F, Baio G, Brennan A, et al. Computing the expected value of sample information efficiently: practical guidance and recommendations for four model-based methods. Value Health 2020;23:734-42. https://doi.org/10.1016/j.jval.2020.02.010.
260. Watson J, Nicholson BD, Hamilton W, Price S. Identifying clinical features in primary care electronic health record studies: methods for codelist development. BMJ Open 2017;7. https://doi.org/10.1136/bmjopen-2017-019637.
261. Davis S, Stevenson M, Tappenden P, Wailoo A. NICE DSU Technical Support Document 15: Cost-Effectiveness Modelling Using Patient-Level Simulation. London: National Institute for Health and Care Excellence; 2014.
262. Ravelli A, Villanacci V, Monfredini C, Martinazzi S, Grassi V, Manenti S. How patchy is patchy villous atrophy? Distribution pattern of histological lesions in the duodenum of children with celiac disease. Am J Gastroenterol 2010;105:2103-10. https://doi.org/10.1038/ajg.2010.153.
263. Lebwohl B, Kapel RC, Neugut AI, Green PH, Genta RM. Adherence to biopsy guidelines increases celiac disease diagnosis. Gastrointest Endosc 2011;74:103-9. https://doi.org/10.1016/j.gie.2011.03.1236.
264. Cohn A, Sofia AM, Kupfer SS. Type 1 diabetes and celiac disease: clinical overlap and new insights into disease pathogenesis. Curr Diab Rep 2014;14. https://doi.org/10.1007/s11892-014-0517-x.
265. Irvine AJ, Chey WD, Ford AC. Screening for celiac disease in irritable bowel syndrome: an updated systematic review and meta-analysis. Am J Gastroenterol 2017;112:65-76. https://doi.org/10.1038/ajg.2016.466.
266. Julian T, Hadjivassiliou M, Zis P. Gluten sensitivity and epilepsy: a systematic review. J Neurol 2019;266:1557-65. https://doi.org/10.1007/s00415-018-9025-2.
267. Singh P, Arora S, Lal S, Strand TA, Makharia GK. Celiac disease in women with infertility: a meta-analysis. J Clin Gastroenterol 2016;50:33-9. https://doi.org/10.1097/MCG.0000000000000285.
268. Roy A, Laszkowska M, Sundström J, Lebwohl B, Green PH, Kämpe O, et al. Prevalence of celiac disease in patients with autoimmune thyroid disease: a meta-analysis. Thyroid 2016;26:880-90. https://doi.org/10.1089/thy.2016.0108.
269. Sainsbury A, Sanders DS, Ford AC. Meta-analysis: coeliac disease and hypertransaminasaemia. Aliment Pharmacol Ther 2011;34:33-40. https://doi.org/10.1111/j.1365-2036.2011.04685.x.
270. Laszkowska M, Mahadev S, Sundström J, Lebwohl B, Green PHR, Michaelsson K, et al. Systematic review with meta-analysis: the prevalence of coeliac disease in patients with osteoporosis. Aliment Pharmacol Ther 2018;48:590-7. https://doi.org/10.1111/apt.14911.
271. Acharya P, Mathur M. Association between psoriasis and celiac disease: a systematic review and meta-analysis. J Am Acad Dermatol 2020;82:1376-85. https://doi.org/10.1016/j.jaad.2019.11.039.
272. Pinto-Sanchez MI, Seiler CL, Santesso N, Alaedini A, Semrad C, Lee AR, et al. Association between inflammatory bowel diseases and celiac disease: a systematic review and meta-analysis. Gastroenterology 2020;159:884-903.e31. https://doi.org/10.1053/j.gastro.2020.05.016.
273. Zis P, Julian T, Hadjivassiliou M. Headache associated with coeliac disease: a systematic review and meta-analysis. Nutrients 2018;10. https://doi.org/10.3390/nu10101445.
274. Gabrielli M, Cremonini F, Fiore G, Addolorato G, Padalino C, Candelli M, et al. Association between migraine and celiac disease: results from a preliminary case–control and therapeutic study. Am J Gastroenterol 2003;98:625-9. https://doi.org/10.1111/j.1572-0241.2003.07300.x.
275. Singh P, Arora S, Lal S, Strand TA, Makharia GK. Risk of celiac disease in the first- and second-degree relatives of patients with celiac disease: a systematic review and meta-analysis. Am J Gastroenterol 2015;110:1539-48. https://doi.org/10.1038/ajg.2015.296.
276. Yap TW, Chan WK, Leow AH, Azmi AN, Loke MF, Vadivelu J, et al. Prevalence of serum celiac antibodies in a multiracial Asian population – a first study in the young Asian adult population of Malaysia. PLOS ONE 2015;10. https://doi.org/10.1371/journal.pone.0121908.
277. Abu-Zeid YA, Jasem WS, Lebwohl B, Green PH, ElGhazali G. Seroprevalence of celiac disease among United Arab Emirates healthy adult nationals: a gender disparity. World J Gastroenterol 2014;20:15830-6. https://doi.org/10.3748/wjg.v20.i42.15830.
278. Romanos J, van Diemen CC, Nolte IM, Trynka G, Zhernakova A, Fu J, et al. Analysis of HLA and non-HLA alleles can identify individuals at high risk for celiac disease. Gastroenterology 2009;137. https://doi.org/10.1053/j.gastro.2009.05.040.
279. Romanos J, Rosén A, Kumar V, Trynka G, Franke L, Szperl A, et al. Improving coeliac disease risk prediction by testing non-HLA variants additional to HLA variants. Gut 2014;63:415-22. https://doi.org/10.1136/gutjnl-2012-304110.
280. Wilson PW, D’Agostino RB, Levy D, Belanger AM, Silbershatz H, Kannel WB. Prediction of coronary heart disease using risk factor categories. Circulation 1998;97:1837-47. https://doi.org/10.1161/01.cir.97.18.1837.
281. Hippisley-Cox J, Coupland C. Symptoms and risk factors to identify women with suspected cancer in primary care: derivation and validation of an algorithm. Br J Gen Pract 2013;63:e11-21. https://doi.org/10.3399/bjgp13X660733.
282. Hippisley-Cox J, Coupland C. Symptoms and risk factors to identify men with suspected cancer in primary care: derivation and validation of an algorithm. Br J Gen Pract 2013;63:e1-10. https://doi.org/10.3399/bjgp13X660724.
283. Díaz-Redondo A, Miranda-Bautista J, García-Lledó J, Gisbert JP, Menchén L. The potential usefulness of human leukocyte antigen typing for celiac disease screening: a systematic review and meta-analysis. Rev Esp Enferm Dig 2015;107:423-9. https://doi.org/10.17235/reed.2015.3758/2015.
284. UK National Screening Committee . Criteria for Appraising the Viability, Effectiveness and Appropriateness of a Screening Programme n.d. www.gov.uk/government/publications/evidence-review-criteria-national-screening-programmes/criteria-for-appraising-the-viability-effectiveness-and-appropriateness-of-a-screening-programme (accessed 27 May 2022).
285. Sarna VK, Lundin KEA, Mørkrid L, Qiao SW, Sollid LM, Christophersen A. HLA-DQ-gluten tetramer blood test accurately identifies patients with and without celiac disease in absence of gluten consumption. Gastroenterology 2018;154:886-96.e6. https://doi.org/10.1053/j.gastro.2017.11.006.
286. Hopper AD, Azmy IA, Rahman N, Hurlstone DP, Leeds JS, George RR, et al. Association of adult coeliac disease with surgical abdominal pain: a case controlled study in patients referred to secondary care. Gastroenterology 2005;128:A255-A.
287. Sanders DS, Hopper AD, Azmy IA, Rahman N, Hurlstone DP, Leeds JS, et al. Association of adult celiac disease with surgical abdominal pain: a case–control study in patients referred to secondary care. Ann Surg 2005;242:201-7. https://doi.org/10.1097/01.sla.0000171301.35513.cf.
288. Choung RS, Rubio-Tapia A, Lahr BD, Kyle RA, Camilleri MJ, Locke GR, et al. Evidence against routine testing of patients with functional gastrointestinal disorders for celiac disease: a population-based study. Clin Gastroenterol Hepatol 2015;13:1937-43. https://doi.org/10.1016/j.cgh.2015.05.014.
289. Oliveira RP, Sdepanian VL, Barreto JA, Cortez AJ, Carvalho FO, Bordin JO, et al. High prevalence of celiac disease in Brazilian blood donor volunteers based on screening by IgA antitissue transglutaminase antibody. Eur J Gastroenterol Hepatol 2007;19:43-9. https://doi.org/10.1097/01.meg.0000250586.61232.a3.
290. Hujoel IA, Van Dyke CT, Brantner T, Larson J, King KS, Sharma A, et al. Natural history and clinical detection of undiagnosed coeliac disease in a North American community. Aliment Pharmacol Ther 2018;47:1358-66. https://doi.org/10.1111/apt.14625.
291. Fitzpatrick KP, Sherman PM, Ipp M, Saunders N, Macarthur C. Screening for celiac disease in children with recurrent abdominal pain. J Pediatr Gastroenterol Nutr 2001;33:250-2. https://doi.org/10.1097/00005176-200109000-00004.
292. Dalgic B, Sari S, Ozcan B, Basturk B, Ensari A, Egritas O, et al. The evaluation of factors and symptoms related to celiac disease in Turkish children. Turk Pediatri Arsivi 2011;46:314-21.
293. Stahl MG, Geno Rasmussen C, Dong F, Waugh K, Norris JM, Baxter J, et al. Mass screening for celiac disease: the Autoimmunity Screening for Kids study. Am J Gastroenterol 2021;116:180-7. https://doi.org/10.14309/ajg.0000000000000751.
294. Lasa J, Spallone L, Gandara S, Chaar E, Berman S, Zagalsky D. Celiac disease prevalence is not increased in patients with functional dyspepsia. Arq Gastroenterol 2017;54:37-40. https://doi.org/10.1590/s0004-2803.2017v54n1-07.
295. Lecleire S, Di Fiore F, Antonietti M, Savoye G, Lerebours E, Ducrotte P. Endoscopic markers for screening of villous atrophy in patients with dyspepsia or high-risk of celiac disease: a prospective controlled study. Gastroenterology 2005;128:A257-A.
296. Katz KD, Rashtak S, Lahr BD, Melton LJ, Krause PK, Maggi K, et al. Screening for celiac disease in a North American population: sequential serology and gastrointestinal symptoms. Am J Gastroenterol 2011;106:1333-9. https://doi.org/10.1038/ajg.2011.21.
297. Lara-Carmona J, Amieva-Balmori M, Martinez-Conejo A, Jorge FJC, Garcia-Zermeno KR, Flores KH, et al. Prevalence of celiac disease (CD) in subjects with dyspeptic symptoms: a case–control study. Gastroenterology 2019;156:S-918. https://doi.org/10.1016/S0016-5085(19)39262-5.
298. Tikkakoski S, Savilahti E, Kolho KL. Undiagnosed coeliac disease and nutritional deficiencies in adults screened in primary health care. Scand J Gastroenterol 2007;42:60-5. https://doi.org/10.1080/00365520600789974.
299. Ludvigsson JF, Aro P, Walker MM, Vieth M, Agréus L, Talley NJ, et al. Celiac disease, eosinophilic esophagitis and gastroesophageal reflux disease, an adult population-based study. Scand J Gastroenterol 2013;48:808-14. https://doi.org/10.3109/00365521.2013.792389.
300. Locke GR, Murray JA, Zinsmeister AR, Melton LJ, Talley NJ. Celiac disease serology in irritable bowel syndrome and dyspepsia: a population-based case–control study. Mayo Clin Proc 2004;79:476-82. https://doi.org/10.4065/79.4.476.
301. Vivas S, Ruiz de Morales JM, Martinez J, González MC, Martín S, Martín J, et al. Human recombinant anti-transglutaminase antibody testing is useful in the diagnosis of silent coeliac disease in a selected group of at-risk patients. Eur J Gastroenterol Hepatol 2003;15:479-83. https://doi.org/10.1097/01.meg.0000059104.41030.1c.
302. Alexander TS, Oshilaja O, Define L. The incidence of celiac disease antibodies in plasma specimens with low hemoglobin levels. Ann N Y Acad Sci 2009;1173:186-9. https://doi.org/10.1111/j.1749-6632.2009.04755.x.
303. Javid G, Lone SN, Shoukat A, Khan BA, Yattoo GN, Shah A, et al. Prevalence of celiac disease in adult patients with iron-deficiency anemia of obscure origin in Kashmir (India). Indian J Gastroenterol 2015;34:314-19. https://doi.org/10.1007/s12664-015-0586-z.
304. Uçardağ D, Güliter S, Ceneli O, Yakaryilmaz F, Atasoy P, Cağlayan O. Celiac disease prevalence in patients with iron deficiency anemia of obscure origin. Turk J Gastroenterol 2009;20:266-70. https://doi.org/10.4318/tjg.2009.0024.
305. Lasa JS, Olivera P, Soifer L, Moore R. Iron-deficiency anemia as a subclinical celiac disease presentation in an Argentinian population. Rev Gastroenterol Mex 2017;82:270-3. https://doi.org/10.1016/j.rgmxen.2016.12.004.
306. Cikrikcioglu MA, Halac G, Hursitoglu M, Erkal H, Cakirca M, Kinas BE, et al. Prevalence of gluten sensitive enteropathy antibodies in restless legs syndrome. Acta Neurol Belg 2011;111:282-6.
307. Morawiec-Szymonik E, Foltyn W, Marek B, Głogowska-Szeląg J, Kos-Kudła B, Kajdaniuk D. Antibodies involved in the development of pernicious anemia and other autoimmune diseases. Pol Arch Intern Med 2020;130:31-7. https://doi.org/10.20452/pamw.15094.
308. Ransford RA, Hayes M, Palmer M, Hall MJ. A controlled, prospective screening study of celiac disease presenting as iron deficiency anemia. J Clin Gastroenterol 2002;35:228-33. https://doi.org/10.1097/00004836-200209000-00006.
309. Sanders DS, Patel D, Stephenson TJ, Ward AM, McCloskey EV, Hadjivassiliou M, et al. A primary care cross-sectional study of undiagnosed adult coeliac disease. Eur J Gastroenterol Hepatol 2003;15:407-13. https://doi.org/10.1097/00042737-200304000-00012.
310. Greco L, Veneziano A, Di Donato L, Zampella C, Pecoraro M, Paladini D, et al. Undiagnosed coeliac disease does not appear to be associated with unfavourable outcome of pregnancy. Gut 2004;53:149-51. https://doi.org/10.1136/gut.53.1.149.
311. Kalayci AG, Kanber Y, Birinci A, Yildiz L, Albayrak D. The prevalence of coeliac disease as detected by screening in children with iron deficiency anaemia. Acta Paediatr 2005;94:678-81. https://doi.org/10.1111/j.1651-2227.2005.tb01964.x.
312. Shahriari M, Honar N, Yousefi A, Javaherizadeh H. Association of potential celiac disease and refractory iron deficiency anemia in children and adolescents. Arq Gastroenterol 2018;55:78-81. https://doi.org/10.1590/s0004-2803.201800000-15.
313. Narang M, Natarajan R, Shah D, Puri AS, Manchanda V, Kotru M. Celiac disease in children with moderate-to-severe iron-deficiency anemia. Indian Pediatr 2018;55:31-4. https://doi.org/10.1007/s13312-018-1223-6.
314. Ertekin V, Selimoglu MA, Kardas F, Aktas E. Prevalence of celiac disease in Turkish children. J Clin Gastroenterol 2005;39:689-91. https://doi.org/10.1097/01.mcg.0000174026.26838.56.
315. Bizzaro N, Villalta D, Tonutti E, Doria A, Tampoia M, Bassetti D, et al. IgA and IgG tissue transglutaminase antibody prevalence and clinical significance in connective tissue diseases, inflammatory bowel disease, and primary biliary cirrhosis. Dig Dis Sci 2003;48:2360-5. https://doi.org/10.1023/b:ddas.0000007875.72256.e8.
316. Feighery L, Collins C, Feighery C, Mahmud N, Coughlan G, Willoughby R, et al. Anti-transglutaminase antibodies and the serological diagnosis of coeliac disease. Br J Biomed Sci 2003;60:14-8. https://doi.org/10.1080/09674845.2003.11783671.
317. Nisihara RM, Skare TL, Silva MB, Utiyama SR. Rheumatoid arthritis and anti-endomysial antibodies. Acta Reumatol Port 2007;32:163-67.
318. Picarelli A, Di Tola M, Sabbatella L, Vetrano S, Anania MC, Spadaro A, et al. Anti-tissue transglutaminase antibodies in arthritic patients: a disease-specific finding?. Clin Chem 2003;49:2091-4. https://doi.org/10.1373/clinchem.2003.023234.
319. Riente L, Chimenti D, Pratesi F, Delle Sedie A, Tommasi S, Tommasi C, et al. Antibodies to tissue transglutaminase and Saccharomyces cerevisiae in ankylosing spondylitis and psoriatic arthritis. J Rheumatol 2004;31:920-4.
320. Luft LM, Barr SG, Martin LO, Chan EK, Fritzler MJ. Autoantibodies to tissue transglutaminase in Sjögren’s syndrome and related rheumatic diseases. J Rheumatol 2003;30:2613-19.
321. Heikkilä K, Heliövaara M, Impivaara O, Kröger H, Knekt P, Rissanen H, et al. Celiac disease autoimmunity and hip fracture risk: findings from a prospective cohort study. J Bone Miner Res 2015;30:630-6. https://doi.org/10.1002/jbmr.2380.
322. Gheita TA, Fawzy SM, Nour El-Din AM, Gomaa HE. Asymptomatic celiac sprue in juvenile rheumatic diseases children. Int J Rheum Dis 2012;15:220-6. https://doi.org/10.1111/j.1756-185X.2011.01681.x.
323. Sahin Y, Sahin S, Barut K, Cokugras FC, Erkan T, Adrovic A, et al. The frequency of the celiac disease in patients with juvenile idiopathic arthritis. J Pediatr Gastroenterol Nutr 2017;64:222-3.
324. Skrabl-Baumgartner A, Christine Hauer A, Erwa W, Jahnel J. HLA genotyping as first-line screening tool for coeliac disease in children with juvenile idiopathic arthritis. Arch Dis Child 2017;102:607-11. https://doi.org/10.1136/archdischild-2016-311544.
325. Stagi S, Giani T, Simonini G, Falcini F. Thyroid function, autoimmune thyroiditis and coeliac disease in juvenile idiopathic arthritis. Rheumatology 2005;44:517-20. https://doi.org/10.1093/rheumatology/keh531.
326. Robazzi TC, Adan LF, Pimentel K, Guimarães I, Magalhães Filho J, Toralles MB, et al. Autoimmune endocrine disorders and coeliac disease in children and adolescents with juvenile idiopathic arthritis and rheumatic fever. Clin Exp Rheumatol 2013;31:310-17.
327. Taneja A, Prahalad S, Hersh AO, Ponder L, Chan LHK, Rouster-Stevens KA, et al. Prevalence of celiac antibodies and IgA deficiency in juvenile idiopathic arthritis. Arthritis Rheumatol 2017;69:22-3.
328. Toğrol RE, Nalbant S, Solmazgül E, Ozyurt M, Kaplan M, Kiralp MZ, et al. The significance of coeliac disease antibodies in patients with ankylosing spondylitis: a case–controlled study. J Int Med Res 2009;37:220-6. https://doi.org/10.1177/147323000903700127.
329. Nenna R, Tiberti C, Petrarca L, Lucantoni F, Mennini M, Luparia RP, et al. The celiac iceberg: characterization of the disease in primary schoolchildren. J Pediatr Gastroenterol Nutr 2013;56:416-21. https://doi.org/10.1097/MPG.0b013e31827b7f64.
330. Çakir M, Cezaroğlu S, Çobanoğlu Ü. Celiac disease in children with chronic constipation. Turk J Med Sci 2016;46:651-6. https://doi.org/10.3906/sag-1502-130.
331. Fifi AC, Velasco-Benitez C, Saps M. Celiac disease in children with functional constipation: a school-based multicity study. J Pediatr 2020;227:77-80. https://doi.org/10.1016/j.jpeds.2020.07.052.
332. Kumar V, Jarzabek-Chorzelska M, Sulej J, Rajadhyaksha M, Jablonska S. Tissue transglutaminase and endomysial antibodies-diagnostic markers of gluten-sensitive enteropathy in dermatitis herpetiformis. Clin Immunol 2001;98:378-82. https://doi.org/10.1006/clim.2000.4983.
333. Velikova T, Shahid M, Ivanova-Todorova E, Drenovska K, Tumangelova-Yuzeir K, Altankova I, et al. Celiac-related autoantibodies and IL-17A in Bulgarian patients with dermatitis herpetiformis: a cross-sectional study. Medicina 2019;55. https://doi.org/10.3390/medicina55050136.
334. Smecuol E, Sugai E, Niveloni S, Vázquez H, Pedreira S, Mazure R, et al. Permeability, zonulin production, and enteropathy in dermatitis herpetiformis. Clin Gastroenterol Hepatol 2005;3:335-41. https://doi.org/10.1016/S1542-3565(04)00778-5.
335. Imanzadeh F, Sayyari AA, Yaghoobi M, Akbari MR, Shafagh H, Farsar AR. Celiac disease in children with diarrhea is more frequent than previously suspected. J Pediatr Gastroenterol Nutr 2005;40:309-11. https://doi.org/10.1097/01.MPG.0000154012.10420.08.
336. Ranua J, Luoma K, Auvinen A, Mäki M, Haapala AM, Peltola J, et al. Celiac disease-related antibodies in an epilepsy cohort and matched reference population. Epilepsy Behav 2005;6:388-92. https://doi.org/10.1016/j.yebeh.2005.01.007.
337. Mavroudi A, Antigoni M, Xinias I, Papastavrou T, Theodouli P, Karatza E, et al. Increased prevalence of silent celiac disease among Greek epileptic children. Pediatr Neurol 2007;36:165-9. https://doi.org/10.1016/j.pediatrneurol.2006.11.011.
338. Dalgiç B, Dursun I, Serdaroğlu A, Dursun A. Latent and potential celiac disease in epileptic Turkish children. J Child Neurol 2006;21:6-7. https://doi.org/10.1177/08830738060210010301.
339. Dai AI, Akcali A, Varan C, Demiryürek AT. Prevalence of resistant occipital lobe epilepsy associated with celiac disease in children. Childs Nerv Syst 2014;30:1091-8. https://doi.org/10.1007/s00381-014-2387-6.
340. Djurić Z, Nagorni A, Jocić-Jakubi B, Dimić M, Novak M, Milićević R, et al. Celiac disease prevalence in epileptic children from Serbia. Turk J Pediatr 2012;54:247-50.
341. Giordano L, Valotti M, Bosetti A, Accorsi P, Caimi L, Imberti L. Celiac disease-related antibodies in Italian children with epilepsy. Pediatr Neurol 2009;41:34-6. https://doi.org/10.1016/j.pediatrneurol.2009.02.009.
342. Işikay S, Hizli Ş, Yilmaz K. Prevalence of celiac disease in Turkish children with idiopathic epilepsy. Iran J Pediatr 2014;24:280-4.
343. Işıkay S, Kocamaz H. Prevalence of celiac disease in children with idiopathic epilepsy in southeast Turkey. Pediatr Neurol 2014;50:479-81. https://doi.org/10.1016/j.pediatrneurol.2014.01.021.
344. Lahat E, Broide E, Leshem M, Evans S, Scapa E. Prevalence of celiac antibodies in children with neurologic disorders. Pediatr Neurol 2000;22:393-6. https://doi.org/10.1016/S0887-8994(00)00129-6.
345. Pratesi R, Gandolfi L, Martins RC, Tauil PL, Nobrega YK, Teixeira WA. Is the prevalence of celiac disease increased among epileptic patients?. Arq Neuropsiquiatr 2003;61:330-4. https://doi.org/10.1590/S0004-282X2003000300002.
346. Soni S, Agarwal A, Singh A, Gupta V, Khadgawat R, Chaturvedi PK, et al. Prevalence of thyroid autoimmunity in first-degree relatives of patients with celiac disease. Indian J Gastroenterol 2019;38:450-5. https://doi.org/10.1007/s12664-019-00990-3.
347. Choung RS, Horwath IE, Marietta E, Bublitz JT, Olson JE, Murray JA. Prevalence of celiac disease autoimmunity and associated morbid conditions in first-degree relatives of patients with celiac disease. Gastroenterology 2019;156:S-136. https://doi.org/10.1016/S0016-5085(19)37130-6.
348. Fasano A, Berti I, Gerarduzzi T, Not T, Colletti RB, Drago S, et al. Prevalence of celiac disease in at-risk and not-at-risk groups in the United States: a large multicenter study. Arch Intern Med 2003;163:286-92. https://doi.org/10.1001/archinte.163.3.286.
349. Beser OF, Gulluelli E, Cullu Cokugras F, Erkan T, Kutlu T, Yagci RV, et al. Prevalence and clinical features of celiac disease in healthy school-aged children. Dig Dis Sci 2019;64:173-81. https://doi.org/10.1007/s10620-018-5320-0.
350. Cintado A, Sorell L, Galvan JA, Martinez L, Castaneda C, Fragoso T, et al. HLA DQA1*0501 and DQB 1*02 in Cuban celiac patients. Hum Immunol 2006;67:639-42. https://doi.org/10.1016/j.humimm.2006.04.009.
351. Kotze LM, Utiyama SR, Nisihara RM, Zeni MP, de Sena MG, Amarante HM. Antiendomysium antibodies in Brazilian patients with celiac disease and their first-degree relatives. Arq Gastroenterol 2001;38:94-103. https://doi.org/10.1590/s0004-28032001000200004.
352. Nass FR, Kotze LM, Nisihara RM, de Messias-Reason IJ. Ramos da Rosa Utiyama S. Serological and clinical follow-up of relatives of celiac disease patients from southern Brazil. Digestion 2011;83:89-95. https://doi.org/10.1159/000320451.
353. Utiyama SR, Nass FR, Kotze LM, Nisihara RM, Ambrosio AR, Messias-Reason IT. Serological screening of relatives of celiac disease patients: antiendomysium antibodies, anti-tissue transglutaminase or both?. Arq Gastroenterol 2007;44:156-61. https://doi.org/10.1590/S0004-28032007000200014.
354. Hjelle AM, Apalset E, Mielnik P, Nilsen RM, Lundin KEA, Tell GS. Positive IgA against transglutaminase 2 in patients with distal radius and ankle fractures compared to community-based controls. Scand J Gastroenterol 2018;53:1212-16. https://doi.org/10.1080/00365521.2018.1509122.
355. Hjelle AM, Mielnik P, Apalset E, Tell GS. Celiac disease and positive IgA tissue transglutaminase in patients with distal radius or ankle fracture: interim analysis. Ann Rheum Dis 2014;73. https://doi.org/10.1136/annrheumdis-2014-eular.1238.
356. LeBoff MS, Cobb H, Gao LY, Hawkes W, Yu-Yahiro J, Kolatkar NS, et al. Celiac disease is not increased in women with hip fractures and low vitamin D levels. J Nutr Health Aging 2013;17:562-5. https://doi.org/10.1007/s12603-013-0017-8.
357. Potter MDE, Walker MM, Hancock S, Holliday E, Brogan G, Jones M, et al. A serological diagnosis of coeliac disease is associated with osteoporosis in older Australian adults. Nutrients 2018;10. https://doi.org/10.3390/nu10070849.
358. Agardh D, Björck S, Agardh CD, Lidfeldt J. Coeliac disease-specific tissue transglutaminase autoantibodies are associated with osteoporosis and related fractures in middle-aged women. Scand J Gastroenterol 2009;44:571-8. https://doi.org/10.1080/00365520902718929.
359. West J, Logan RF, Hill PG, Lloyd A, Lewis S, Hubbard R, et al. Seroprevalence, correlates, and characteristics of undetected coeliac disease in England. Gut 2003;52:960-5. https://doi.org/10.1136/gut.52.7.960.
360. Choung RS, Larson SA, Khaleghi S, Rubio-Tapia A, Ovsyannikova IG, King KS, et al. Prevalence and morbidity of undiagnosed celiac disease from a community-based study. Gastroenterology 2017;152:830-9.e5. https://doi.org/10.1053/j.gastro.2016.11.043.
361. Walker MM, Murray JA, Ronkainen J, Aro P, Storskrubb T, D’Amato M, et al. Detection of celiac disease and lymphocytic enteropathy by parallel serology and histopathology in a population-based study. Gastroenterology 2010;139:112-19. https://doi.org/10.1053/j.gastro.2010.04.007.
362. Khudher SN, Mohammed KAS, Ali NH. Influence of HLA-DQ on clinical and serological biomarkers in patients with celiac disease. Ann Trop Med Public Health 2020;23. https://doi.org/10.36295/ASRO.2020.231617.
363. Horoldt BS, Leeds JS, Sidhu R, Hopper AD, Robinson K, Toulson B, et al. Is there a relationship between coeliac disease and inflammatory bowel disease? A bidirectional prevalence study with controls. Gut 2006;55:A98-A.
364. Leeds JS, Höroldt BS, Sidhu R, Hopper AD, Robinson K, Toulson B, et al. Is there an association between coeliac disease and inflammatory bowel diseases? A study of relative prevalence in comparison with population controls. Scand J Gastroenterol 2007;42:1214-20. https://doi.org/10.1080/00365520701365112.
365. Watanabe C, Komoto S, Hokari R, Kurihara C, Okada Y, Hozumi H, et al. Prevalence of serum celiac antibody in patients with IBD in Japan. J Gastroenterol 2014;49:825-34. https://doi.org/10.1007/s00535-013-0838-6.
366. El-Matary W, Fedorak RN, Senthilselvan A, Spady D. Celiac disease and inflammatory bowel disease in children: is there a link?. Gastroenterology 2012;142:S370-S1. https://doi.org/10.1016/S0016-5085(12)61398-5.
367. Kull K, Uibo O, Salupere R, Metsküla K, Uibo R. High frequency of antigliadin antibodies and absence of antireticulin and antiendomysium antibodies in patients with ulcerative colitis. J Gastroenterol 1999;34:61-5. https://doi.org/10.1007/s005350050217.
368. Mehdi Z, Sakineh E, Mohammad F, Mansour R, Alireza A. Celiac disease: serologic prevalence in patients with irritable bowel syndrome. J Res Med Sci 2012;17:839-42.
369. Respondek W, Tomasiuk R, Jarosz M, Traczyk I, Mekus M. Is it reasonable to perform serological tests for celiac disease in patients with irritable bowel syndrome?. Prz Gastroenterol 2013;8:184-90. https://doi.org/10.5114/pg.2013.36333.
370. Almazar AE, Talley NJ, Larson JJ, Atkinson EJ, Murray JA, Saito YA. Celiac disease is uncommon in irritable bowel syndrome in the USA. Eur J Gastroenterol Hepatol 2018;30:149-54. https://doi.org/10.1097/MEG.0000000000001022.
371. Cash BD, Rubenstein JH, Young PE, Gentry A, Nojkov B, Lee D, et al. The prevalence of celiac disease among patients with nonconstipated irritable bowel syndrome is similar to controls. Gastroenterology 2011;141:1187-93. https://doi.org/10.1053/j.gastro.2011.06.084.
372. Domżał-Magrowska D, Kowalski MK, Szcześniak P, Bulska M, Orszulak-Michalak D, Małecka-Panas E. The prevalence of celiac disease in patients with irritable bowel syndrome and its subtypes. Prz Gastroenterol 2016;11:276-81. https://doi.org/10.5114/pg.2016.57941.
373. Kou GJ, Guo J, Zuo XL, Li CQ, Liu C, Ji R, et al. Prevalence of celiac disease in adult Chinese patients with diarrhea-predominant irritable bowel syndrome: a prospective, controlled, cohort study. J Dig Dis 2018;19:136-43. https://doi.org/10.1111/1751-2980.12587.
374. Saito-Loftus Y, Brantner T, Zimmerman J, Talley N, Murray J. The prevalence of positive serologic tests for celiac sprue does not differ between irritable bowel syndrome (IBS) patients compared with controls. Am J Gastroenterol 2008;103. https://doi.org/10.14309/00000434-200809001-01208.
375. Sánchez-Vargas LA, Thomas-Dupont P, Torres-Aguilera M, Azamar-Jacome AA, Ramírez-Ceervanes KL, Aedo-Garcés MR, et al. Prevalence of celiac disease and related antibodies in patients diagnosed with irritable bowel syndrome according to the Rome III criteria. A case–control study. Neurogastroenterol Motil 2016;28:994-1000. https://doi.org/10.1111/nmo.12799.
376. Sanders DS, Carter MJ, Hurlstone DP, Pearce A, Ward AM, McAlindon ME, et al. Association of adult coeliac disease with irritable bowel syndrome: a case–control study in patients fulfilling ROME II criteria referred to secondary care. Lancet 2001;358:1504-8. https://doi.org/10.1016/S0140-6736(01)06581-3.
377. Vargas LAS, Garces M, Dupont PT, Meixueiro A, Jacome AAA, Roesch FB, et al. Prevalence of antibodies related to celiac disease (CD) in patients with irritable bowel syndrome (IBS) according to ROME III criteria. A case–control study. Gastroenterology 2013;144. https://doi.org/10.1016/S0016-5085(13)60894-X.
378. Wang H, Zhou G, Luo L, Crusius JBA, Yuan A, Kou J, et al. Serological screening for celiac disease in adult Chinese patients with diarrhea predominant irritable bowel syndrome. Medicine 2015;94. https://doi.org/10.1097/MD.0000000000001779.
379. Khayyat YM. Undiagnosed celiac disease in patients presenting with irritable bowel syndrome symptoms. J Ayub Med Coll Abbottabad 2020;32:18-23.
380. Olen O, Sandstrom O, Myleus A, Rosen A, Carlsson A, Hogberg L, et al. Functional gastrointestinal disorders and the association to celiac disease-a population based screening study in children. Gastroenterology 2014;146:S-173. https://doi.org/10.1016/S0016-5085(14)60611-9.
381. Chatzicostas C, Roussomoustakaki M, Drygiannakis D, Niniraki M, Tzardi M, Koulentaki M, et al. Primary biliary cirrhosis and autoimmune cholangitis are not associated with coeliac disease in Crete. BMC Gastroenterology 2002;2. https://doi.org/10.1186/1471-230X-2-5.
382. Durante-Mangoni E, Iardino P, Resse M, Cesaro G, Sica A, Farzati B, et al. Silent celiac disease in chronic hepatitis C: impact of interferon treatment on the disease onset and clinical outcome. J Clin Gastroenterol 2004;38:901-5. https://doi.org/10.1097/00004836-200411000-00014.
383. Hernandez L, Johnson TC, Naiyer AJ, Kryszak D, Ciaccio EJ, Min A, et al. Chronic hepatitis C virus and celiac disease, is there an association?. Dig Dis Sci 2008;53:256-61. https://doi.org/10.1007/s10620-007-9851-z.
384. Sjöberg K, Lindgren S, Eriksson S. Frequent occurrence of non-specific gliadin antibodies in chronic liver disease. Endomysial but not gliadin antibodies predict coeliac disease in patients with chronic liver disease. Scand J Gastroenterol 1997;32:1162-7. https://doi.org/10.3109/00365529709002997.
385. Yuan J, Gao J, Yao Y, Chen H. Serologic testing for celiac disease in young people with elevated transaminases. Turk J Med Sci 2015;45:668-73. https://doi.org/10.3906/sag-1403-127.
386. El-Shabrawi M, El-Karaksy H, Mohsen N, Isa M, Al-Biltagi M, El-Ansari M. Celiac disease in children and adolescents with autoimmune hepatitis: a single-centre experience. J Trop Pediatr 2011;57:104-8. https://doi.org/10.1093/tropej/fmq057.
387. Oana B, Laura O, Ioan S, Tamara M, Otilia M. Celiac disease prevalence among children with autoimmune hepatitis and autoimmune thyroid disorders. A 10-years single centre experience. J Gastrointestin Liver Dis 2018;27.
388. Germenis AE, Yiannaki EE, Zachou K, Roka V, Barbanis S, Liaskos C, et al. Prevalence and clinical significance of immunoglobulin A antibodies against tissue transglutaminase in patients with diverse chronic liver diseases. Clin Diagn Lab Immunol 2005;12:941-8. https://doi.org/10.1128/CDLI.12.8.941-948.2005.
389. Villalta D, Girolami D, Bidoli E, Bizzaro N, Tampoia M, Liguori M, et al. High prevalence of celiac disease in autoimmune hepatitis detected by anti-tissue tranglutaminase autoantibodies. J Clin Lab Anal 2005;19:6-10. https://doi.org/10.1002/jcla.20047.
390. Balci O, Yilmaz D, Sezer T, Hizli S. Is celiac disease an etiological factor in children with migraine?. J Child Neurol 2016;31:929-31. https://doi.org/10.1177/0883073816630088.
391. Alehan F, Ozçay F, Erol I, Canan O, Cemil T. Increased risk for coeliac disease in paediatric patients with migraine. Cephalalgia 2008;28:945-9. https://doi.org/10.1111/j.1468-2982.2008.01630.x.
392. Inaloo S, Dehghani SM, Farzadi F, Haghighat M, Imanieh MH. A comparative study of celiac disease in children with migraine headache and a normal control group. Turk J Gastroenterol 2011;22:32-5. https://doi.org/10.4318/tjg.2011.0153.
393. Abolfazli R, Mirbagheri A, Rabbani Anari M, Samadzadeh S. The association of celiac disease with multiple sclerosis. J Neurol Sci 2009;285. https://doi.org/10.1016/S0022-510X(09)70816-0.
394. Nicoletti A, Patti F, Lo Fermo S, Sciacca A, Laisa P, Liberto A, et al. Frequency of celiac disease is not increased among multiple sclerosis patients. Mult Scler 2008;14:698-700. https://doi.org/10.1177/1352458507087268.
395. Rodrigo L, Hernández-Lahoz C, Fuentes D, Alvarez N, López-Vázquez A, González S. Prevalence of celiac disease in multiple sclerosis. BMC Neurol 2011;11. https://doi.org/10.1186/1471-2377-11-31.
396. Roth EB, Theander E, Londos E, Sandberg-Wollheim M, Larsson A, Sjoberg K, et al. Pathogenesis of autoimmune diseases: antibodies against transglutaminase, peptidylarginine deiminase and protein-bound citrulline in primary Sjögren’s syndrome, multiple sclerosis and Alzheimer’s disease. Scand J Immunol 2008;67:626-31. https://doi.org/10.1111/j.1365-3083.2008.02115.x.
397. Khoshbaten M, Farhoudi M, Nikanfar M, Ayromlou H, Shaafi S, Sadreddini SA, et al. Celiac disease and multiple sclerosis in the northwest of Iran. Bratisl Lek Listy 2012;113:495-7. https://doi.org/10.4149/bll_2012_109.
398. Gusso L, Simoes MC, Skare TL, Nisihara R, Burkiewicz CC, Utiyama S. Celiac disease screening in Brazilian patients with osteoporosis. Arq Bras Endocrinol Metabol 2014;58:270-3. https://doi.org/10.1590/0004-2730000002919.
399. Shahbazkhani B, Aletaha N, Khonche A, Farahvash B, Malekzadeh R. Is it necessary to screen for celiac disease in adult idiopathic osteoporosis?. Gastroenterol Hepatol Bed Bench 2015;8:140-5.
400. Stenson WF, Newberry R, Lorenz R, Baldus C, Civitelli R. Increased prevalence of celiac disease and need for routine screening among patients with osteoporosis. Arch Intern Med 2005;165:393-9. https://doi.org/10.1001/archinte.165.4.393.
401. Vanciková Z, Chlumecký V, Sokol D, Horáková D, Hamsíková E, Fucíková T, et al. The serologic screening for celiac disease in the general population (blood donors) and in some high-risk groups of adults (patients with autoimmune diseases, osteoporosis and infertility) in the Czech republic. Folia Microbiol 2002;47:753-8. https://doi.org/10.1007/BF02818684.
402. Shen Y, Wang M, Martinez D. Celiac disease and low femoral bone density in a multiethnic US national survey. Osteoporos Int 2014;25.
403. De Bastiani R, Gabrielli M, Lora L, Napoli L, Tosetti C, Pirrotta E, et al. Association between coeliac disease and psoriasis: Italian primary care multicentre study. Dermatology 2015;230:156-60. https://doi.org/10.1159/000369615.
404. Akbulut S, Gür G, Topal F, Senel E, Topal FE, Alli N, et al. Coeliac disease-associated antibodies in psoriasis. Ann Dermatol 2013;25:298-303. https://doi.org/10.5021/ad.2013.25.3.298.
405. Dhattarwal N, Mahajan VK, Mehta KS, Chauhan PS, Yadav RS, Sharma SB, et al. The association of anti-gliadin and anti-transglutaminase antibodies and chronic plaque psoriasis in Indian patients: preliminary results of a descriptive cross-sectional study. Australas J Dermatol 2020;61:e378-e382. https://doi.org/10.1111/ajd.13308.
406. Montesu MA, Dessí-Fulgheri C, Pattaro C, Ventura V, Satta R, Cottoni F. Association between psoriasis and coeliac disease? A case–control study. Acta Derm Venereol 2011;91:92-3. https://doi.org/10.2340/00015555-0960.
407. Nagui N, El Nabarawy E, Mahgoub D, Mashaly HM, Saad NE, El-Deeb DF. Estimation of (IgA) anti-gliadin, anti-endomysium and tissue transglutaminase in the serum of patients with psoriasis. Clin Exp Dermatol 2011;36:302-4. https://doi.org/10.1111/j.1365-2230.2010.03980.x.
408. Singh S, Sonkar GK, Usha Singh S. Celiac disease-associated antibodies in patients with psoriasis and correlation with HLA Cw6. J Clin Lab Anal 2010;24:269-72. https://doi.org/10.1002/jcla.20398.
409. Marai I, Shoenfeld Y, Bizzaro N, Villalta D, Doria A, Tonutti E, et al. IgA and IgG tissue transglutaminase antibodies in systemic lupus erythematosus. Lupus 2004;13:241-4. https://doi.org/10.1191/0961203304lu1004oa.
410. Picceli VF, Skare TL, Nisihara R, Kotze L, Messias-Reason I, Utiyama SR. Spectrum of autoantibodies for gastrointestinal autoimmune diseases in systemic lupus erythematosus patients. Lupus 2013;22:1150-5. https://doi.org/10.1177/0961203313503911.
411. Sahin Y, Sahin S, Adrovic A, Erkan T, Kutlu T, Barut K, et al. The frequency of celiac disease in Turkish children with systemic lupus erythematosus. J Pediatr Gastroenterol Nutr 2017;64.
412. Cedíková M, Ulčová-Gallová Z, Bibková K, Mičanová Z. The incidence of latent asymptomatic celiac disease in women with decreased fertility. Ceska Gynekol 2013;78:247-51.
413. Remes-Troche JM, Vargas LAS, Meixueiro A, Patino ED, Gonzalez-Sicilia E, Abreu JA, et al. Celiac disease screening in patients previously diagnosed with infertility. A prospective study in Mexican population. Gastroenterology 2014;146. https://doi.org/10.1016/S0016-5085(14)61677-2.
414. Herraiz-Nicuesa L, Tejera-Alhambra M, Garcia-Segovia A, Ramos-Medina R, Alonso B, Gil-Pulido J, et al. Increasing prevalence of undiagnosed celiac disease in Spanish women with recurrent reproductive failure. Hum Reprod 2013;28:i139-49. https://doi.org/10.1093/humrep/det209.
415. Kumar A, Meena M, Begum N, Kumar N, Gupta RK, Aggarwal S, et al. Latent celiac disease in reproductive performance of women. Fertil Steril 2011;95:922-7. https://doi.org/10.1016/j.fertnstert.2010.11.005.
416. Kutteh MA, Abiad M, Norman GL, Kutteh WH. Comparison of celiac disease markers in women with early recurrent pregnancy loss and normal controls. Am J Reprod Immunol 2019;82. https://doi.org/10.1111/aji.13127.
417. Sarikaya E, Tokmak A, Aksoy RT, Pekcan MK, Alisik M, Alkan A. The association between serological markers of celiac disease and idiopathic recurrent pregnancy loss. Fetal Pediatr Pathol 2017;36:373-9. https://doi.org/10.1080/15513815.2017.1346018.
418. Shamaly H, Mahameed A, Sharony A, Shamir R. Infertility and celiac disease: do we need more than one serological marker?. Acta Obstet Gynecol Scand 2004;83:1184-8. https://doi.org/10.1111/j.0001-6349.2004.00592.x.
419. Sharshiner R, Romero S, Silver R, Branch DW. Celiac disease serum markers and recurrent pregnancy loss. Am J Obstet Gynecol 2013;208:S75-S6. https://doi.org/10.1016/j.ajog.2012.10.316.
420. Tiboni GM, de Vita MG, Faricelli R, Giampietro F, Liberati M. Serological testing for celiac disease in women undergoing assisted reproduction techniques. Hum Reprod 2006;21:376-9. https://doi.org/10.1093/humrep/dei314.
421. Zahmatkeshan M. Prevalence of coeliac disease in infertile women. J Pediatr Gastroenterol Nutr 2019;68.
422. Celdir MG, Choung RS, Rostamkolaei SK, Jansson-Knodell CL, King KS, Larson JJ, et al. Reproductive characteristics and pregnancy outcomes in hidden celiac disease autoimmunity. Am J Gastroenterol 2021;116:593-9. https://doi.org/10.14309/ajg.0000000000001148.
423. Martinelli P, Troncone R, Paparo F, Torre P, Trapanese E, Fasano C, et al. Coeliac disease and unfavourable outcome of pregnancy. Gut 2000;46:332-5. https://doi.org/10.1136/gut.46.3.332.
424. Mohammed MA, Elrabbat AM, Omar NM, Shebl AM, Mansour AH, Elmasry E, et al. Celiac disease prevalence and its HLA-genotypic profile in Egyptian patients with type 1 diabetes mellitus. Trends Medical Res 2014;9:81-97. https://doi.org/10.3923/tmr.2014.81.97.
425. Picarelli A, Sabbatella L, Di Tola M, Vetrano S, Casale C, Anania MC, et al. Anti-endomysial antibody of IgG1 isotype detection strongly increases the prevalence of coeliac disease in patients affected by type I diabetes mellitus. Clin Exp Immunol 2005;142:111-15. https://doi.org/10.1111/j.1365-2249.2005.02866.x.
426. Dagdelen S, Hascelik G, Bayraktar M. Simultaneous triple organ specific autoantibody profiling in adult patients with type 1 diabetes mellitus and their first-degree relatives. Int J Clin Pract 2009;63:449-56. https://doi.org/10.1111/j.1742-1241.2007.01619.x.
427. Güvenç S, Kaymakoğlu S, Gürel N, Karşidağ K, Demir K, Dinçer D, et al. The prevalence of manifest and latent celiac disease in type 1 diabetes mellitus. Turk J Gastroenterol 2002;13:103-7.
428. Hanukoglu A, Mizrachi A, Dalal I, Admoni O, Rakover Y, Bistritzer Z, et al. Extrapancreatic autoimmune manifestations in type 1 diabetes patients and their first-degree relatives: a multicenter study. Diabetes Care 2003;26:1235-40. https://doi.org/10.2337/diacare.26.4.1235.
429. Kurien M, Leeds JS, Hopper AD, Wild G, Egner W, Tesfaye S, et al. Serological testing for coeliac disease in type 1 diabetes mellitus: is immunoglobulin A level measurement necessary?. Diabet Med 2013;30:840-5. https://doi.org/10.1111/dme.12163.
430. Shivaprasad C, Kolly A, Pulikkal A, Kumar KMP. High prevalence of organ specific autoantibodies in Indian type 1 diabetic patients. J Pediatr Endocrinol Metab 2017;30:707-12. https://doi.org/10.1515/jpem-2017-0011.
431. Zhao Z, Zou J, Zhao L, Cheng Y, Cai H, Li M, et al. Celiac disease autoimmunity in patients with autoimmune diabetes and thyroid disease among Chinese population. PLOS ONE 2016;11. https://doi.org/10.1371/journal.pone.0157510.
432. Abu-Zekry M, Kryszak D, Diab M, Catassi C, Fasano A. Prevalence of celiac disease in Egyptian children disputes the east–west agriculture-dependent spread of the disease. J Pediatr Gastroenterol Nutr 2008;47:136-40. https://doi.org/10.1097/MPG.0b013e31815ce5d1.
433. Adlercreutz EH, Svensson J, Hansen D, Buschard K, Lernmark Å, Mortensen HB, et al. Prevalence of celiac disease autoimmunity in children with type 1 diabetes: regional variations across the Øresund strait between Denmark and southernmost Sweden. Pediatr Diabetes 2015;16:504-9. https://doi.org/10.1111/pedi.12200.
434. Aktay AN, Lee PC, Kumar V, Parton E, Wyatt DT, Werlin SL. The prevalence and clinical characteristics of celiac disease in juvenile diabetes in Wisconsin. J Pediatr Gastroenterol Nutr 2001;33:462-5. https://doi.org/10.1097/00005176-200110000-00008.
435. Baptista ML, Koda YKL, Mitsunori R, Nisihara, Ioshii SO. Prevalence of celiac disease in Brazilian children and adolescents with type 1 diabetes mellitus. J Pediatr Gastroenterol Nutr 2005;41:621-4. https://doi.org/10.1097/01.mpg.0000181400.57884.c3.
436. Djurić Z, Stamenković H, Stanković T, Milićević R, Branković L, Cirić V, et al. Celiac disease prevalence in children and adolescents with type 1 diabetes from Serbia. Pediatr Int 2010;52:579-83. https://doi.org/10.1111/j.1442-200X.2010.03085.x.
437. Frohnert BI, Simmons K, Liu E, Taki I, Klingensmith GJ, McFann K, et al. Poor glycemic control and celiac disease autoimmunity are independently associated with reduced bone mineral density in children with type 1 diabetes. Diabetes 2015;64.
438. Gurau G, Dobre M, Nechita A. Anti-tissue transglutaminase antibodies in patients with anti-glutamate dehydrogenase positive type 1 diabetes mellitus. Rev Romana De Medicina De Lab 2012;20:39-46.
439. Hansson T, Dahlbom I, Tuvemo T, Frisk G. Silent coeliac disease is over-represented in children with type 1 diabetes and their siblings. Acta Paediatr 2015;104:185-91. https://doi.org/10.1111/apa.12823.
440. Krause I, Anaya JM, Fraser A, Barzilai O, Ram M, Abad V, et al. Anti-infectious antibodies and autoimmune-associated autoantibodies in patients with type I diabetes mellitus and their close family members. Ann N Y Acad Sci 2009;1173:633-9. https://doi.org/10.1111/j.1749-6632.2009.04619.x.
441. Soyucen E, Yilmaz S, Çeltık C, Vatansever U, Öner N, Karasalıhoğlu S. Seroprevalence of autoimmune thyroiditis and celiac disease in children with insulin-dependent diabetes mellitus in the Thrace region of Turkey. Turk J Gastroenterol 2010;21:231-5. https://doi.org/10.4318/tjg.2010.0093.
442. Velasco C, Ortiz C, Ruiz A. HLA typing in healthy and diabetic children with celiac disease in Cali, Colombia. J Pediatr Gastroenterol Nutr 2016;63.
443. Not T, Tommasini A, Tonini G, Buratti E, Pocecco M, Tortul C, et al. Undiagnosed coeliac disease and risk of autoimmune disorders in subjects with type I diabetes mellitus. Diabetologia 2001;44:151-5. https://doi.org/10.1007/s001250051593.
444. Premawardhana LD, Wijeyaratne CN, Chen S, Wijesuriya M, Illangasekera U, Brooking H, et al. Islet cell, thyroid, adrenal and celiac disease related autoantibodies in patients with type 1 diabetes from Sri Lanka. J Endocrinol Invest 2006;29:968-74. https://doi.org/10.1007/BF03349209.
445. Jaeger C, Hatziagelaki E, Petzoldt R, Bretzel RG. Comparative analysis of organ-specific autoantibodies and celiac disease – associated antibodies in type 1 diabetic patients, their first-degree relatives, and healthy control subjects. Diabetes Care 2001;24:27-32. https://doi.org/10.2337/diacare.24.1.27.
446. Kanungo A, Shtauvere-Brameus A, Samal KC, Sanjeevi CB. Autoantibodies to tissue transglutaminase in patients from eastern India with malnutrition-modulated diabetes mellitus, insulin-dependent diabetes mellitus, and non-insulin-dependent diabetes mellitus. Ann N Y Acad Sci 2002;958:232-4. https://doi.org/10.1111/j.1749-6632.2002.tb02976.x.
447. Sari S, Yeşilkaya E, Eğritaş O, Bideci A, Cinaz P, Dalgiç B. Prevalence of celiac disease in Turkish children with type 1 diabetes mellitus and their non-diabetic first-degree relatives. Turk J Gastroenterol 2010;21:34-8. https://doi.org/10.4318/tjg.2010.0045.
448. Lampasona V, Bonfanti R, Bazzigaluppi E, Venerando A, Chiumello G, Bosi E, et al. Antibodies to tissue transglutaminase C in type I diabetes. Diabetologia 1999;42:1195-8. https://doi.org/10.1007/s001250051291.
449. Sharifi N, Khoshbaten M, Aliasgarzade A, Bahrami A. Celiac disease in patients with type-1 diabetes mellitus screened by tissue transglutaminase antibodies in northwest of Iran. Int J Diabetes Dev Ctries 2008;28:95-9. https://doi.org/10.4103/0973-3930.44081.
450. Sheikholeslami H, Boostani K, Hashemipoor S, Hadjmanoochehri F, Ziaee A. Prevalence of celiac disease in type1 diabetic patients comparing with nondiabetic healthy individuals. J Iran Diabetes Lipid 2005;4:49-55.
451. Kizilgul M, Ozcelik O, Beysel S, Akinci H, Kan S, Ucan B, et al. Screening for celiac disease in poorly controlled type 2 diabetes mellitus: worth it or not?. BMC Endocr Disord 2017;17. https://doi.org/10.1186/s12902-017-0212-4.
452. Szepietowska B, Wawrusiewicz-Kurylonek N, Krętowski A, Górska M, Szelachowska M. Endocrine autoimmunity in patients with latent autoimmune diabetes in adults (LADA) — association with HLA genotype. Endokrynol Pol 2016;67:197-201. https://doi.org/10.5603/EP.a2016.0017.
453. Ravaglia G, Forti P, Maioli F, Volta U, Arnone G, Pantieri G, et al. Increased prevalence of coeliac disease in autoimmune thyroiditis is restricted to aged patients. Exp Gerontol 2003;38:589-95. https://doi.org/10.1016/S0531-5565(03)00037-8.
454. Riseh SH, Farhang MA, Mobasseri M, Jafarabadi MA. The relationship between thyroid hormones, antithyroid antibodies, anti-tissue transglutaminase and anti-gliadin antibodies in patients with Hashimoto’s thyroiditis. Acta Endocrinol 2017;13:174-9. https://doi.org/10.4183/aeb.2017.174.
455. Miskiewicz P, Gos-Zajac A, Kurylowicz A, Plazinska TM, Franaszczyk M, Bartoszewicz Z, et al. HLA DQ2 haplotype, early onset of Graves’ disease, and positive family history of autoimmune disorders are risk factors for developing celiac disease in patients with Graves’ disease. Endocr Pract 2015;21:993-1000. https://doi.org/10.4158/EP15700.OR.
456. Berti I, Trevisiol C, Tommasini A, Città A, Neri E, Geatti O, et al. Usefulness of screening program for celiac disease in autoimmune thyroiditis. Dig Dis Sci 2000;45:403-6. https://doi.org/10.1023/a:1005441400107.
457. Ch’ng CL, Biswas M, Benton A, Jones MK, Kingham JG. Prospective screening for coeliac disease in patients with Graves’ hyperthyroidism using anti-gliadin and tissue transglutaminase antibodies. Clin Endocrinol 2005;62:303-6. https://doi.org/10.1111/j.1365-2265.2005.02214.x.
458. Guliter S, Yakaryilmaz F, Ozkurt Z, Ersoy R, Ucardag D, Caglayan O, et al. Prevalence of coeliac disease in patients with autoimmune thyroiditis in a Turkish population. World J Gastroenterol 2007;13:1599-601. https://doi.org/10.3748/wjg.v13.i10.1599.
459. Volta U, Ravaglia G, Granito A, Forti P, Maioli F, Petrolini N, et al. Coeliac disease in patients with autoimmune thyroiditis. Digestion 2001;64:61-5. https://doi.org/10.1159/000048840.
460. Metzger MH, Heier M, Mäki M, Bravi E, Schneider A, Löwel H, et al. Mortality excess in individuals with elevated IgA anti-transglutaminase antibodies: the KORA/MONICA Augsburg cohort study 1989–1998. Eur J Epidemiol 2006;21:359-65. https://doi.org/10.1007/s10654-006-9002-4.
461. Marwaha RK, Garg MK, Tandon N, Kanwar R, Narang A, Sastry A, et al. Glutamic acid decarboxylase (anti-GAD) &amp; tissue transglutaminase (anti-TTG) antibodies in patients with thyroid autoimmunity. Indian J Med Res 2013;137:82-6.
462. Sahin Y, Evliyaoglu O, Erkan T, Cokugras FC, Ercan O, Kutlu T. The frequency of celiac disease in children with autoimmune thyroiditis. Acta Gastroenterol Belg 2018;81.
463. Sari S, Yesilkaya E, Egritas O, Bideci A, Dalgic B. Prevalence of celiac disease in Turkish children with autoimmune thyroiditis. Dig Dis Sci 2009;54:830-2. https://doi.org/10.1007/s10620-008-0437-1.
464. van der Pals M, Ivarsson A, Norström F, Högberg L, Svensson J, Carlsson A. Prevalence of thyroid autoimmunity in children with celiac disease compared to healthy 12-year olds. Autoimmune Dis 2014;2014. https://doi.org/10.1155/2014/417356.
465. Risan FA. Association between autoimmune thyroiditis with gluten sensitive enteropathy patients in Baghdad. J Pharm Sci Res 2018;10:3307-9.
466. de Melo FM, Cavalcanti MSM, dos Santos Severino B, Lopes AKBF, de Oliveira FAA. Association between serum markers for celiac and thyroid autoimmune diseases. Arq Bras Endocrinol Metabol 2005;49:542-7. https://doi.org/10.1590/S0004-27302005000400012.
467. National Institute for Health and Care Excellence . Fertility Problems: Assessment and Treatment n.d. www.nice.org.uk/guidance/CG156 (accessed 27 June 2022).