Interferon gamma release assays for Diagnostic Evaluation of Active tuberculosis (IDEA): test accuracy study and economic evaluation

Article history

The research reported in this issue of the journal was funded by the HTA programme as project number 08/106/02. The contractual start date was in March 2011. The draft report began editorial review in May 2016 and was accepted for publication in September 2016. The authors have been wholly responsible for all data collection, analysis and interpretation, and for writing up their work. The HTA editors and publisher have tried to ensure the accuracy of the authors’ report and would like to thank the reviewers for their constructive comments on the draft document. However, they do not accept liability for damages or losses arising from material published in this report.

Declared competing interests of authors

Ajit Lalvani is the named inventor for several patents underpinning T-cell-based diagnosis including interferon gamma, enzyme-linked immunospot assay, ESAT-6, CFP-10, Rv3615c, Rv3873 and Rv3879c. He has royalty entitlements from the University of Oxford spin-out company (Oxford Immunotec plc), in which he has held a minority share of equity and he is a member of the Efficacy and Mechanism Evaluation Board. Jonathan J Deeks is a member of the Health Technology Assessment (HTA) Commissioning Board and the HTA Efficient Study Designs Board. Onn Min Kon is chairperson of the UK Joint Tuberculosis Committee. Peter White has received research funding from Otsuka SA for a retrospective study of multidrug-resistant tuberculosis treatment in several eastern European countries outside the submitted work. He received grants from the Medical Research Council during the conduct of the study.

Copyright statement

© Queen’s Printer and Controller of HMSO 2019. This work was produced by Takwoingi et al. under the terms of a commissioning contract issued by the Secretary of State for Health and Social Care. This issue may be freely reproduced for the purposes of private research and study and extracts (or indeed, the full report) may be included in professional journals provided that suitable acknowledgement is made and the reproduction is not associated with any form of advertising. Applications for commercial reproduction should be addressed to: NIHR Journals Library, National Institute for Health Research, Evaluation, Trials and Studies Coordinating Centre, Alpha House, University of Southampton Science Park, Southampton SO16 7NS, UK.

2019 Queen’s Printer and Controller of HMSO

Background

In 2014, globally, an estimated 9.6 million cases of tuberculosis (TB) and 1.5 million deaths caused by the disease were reported to the World Health Organization. 1 Co-infection with TB and human immunodeficiency virus (HIV) accounts for a significant proportion of cases globally (12%) and together these infections are the biggest infectious causes of death. Health inequalities are exacerbated by the burden that TB places on the most vulnerable and poor around the world. Over the past 20 years, worldwide TB incidence and mortality have declined. However, despite this, the prevalence of TB still remains unacceptably high. Furthermore, there is not yet evidence of a reduction in the number of cases in England, particularly in major cities such as London and Birmingham.

In England, 6520 cases of TB were reported in 2014, with an incidence rate of 12.0 per 100,000. 2 Of these, 2572 cases were in London, where the incidence rate was 30.1 per 100,000. 2 Other urban high-incidence areas include Leicester, Birmingham, Luton, Manchester and Coventry. The majority of cases in England (72%) occurred in new entrants who were born outside the UK. 2

Tuberculosis is caused by active infection with Mycobacterium tuberculosis (Mtb). Upon initial infection with Mtb, a state of asymptomatic dormancy typically occurs. In most infected individuals this results in a prolonged and perhaps lifelong latent TB infection (LTBI). In others, however, a breakdown of immune control and activation of infection results in symptomatic disease, which can be fatal without treatment. Co-infection with HIV dramatically increases the likelihood of progression from latent to active disease.

Active TB can manifest with pulmonary or extrapulmonary phenotypes. In its active pulmonary form, TB is highly contagious. The diagnosis of active TB is central to preventing the spread of disease and thus controlling the TB epidemic. 3 However, the slow speed and poor sensitivity of existing diagnostic tools often lead to delays diagnosis and treatment of the disease. 4

Diagnosis of tuberculosis

Conventional methods of diagnosing active TB rely primarily on the identification of Mtb bacilli, as well as imaging affected areas. Smear microscopy is quick and inexpensive, but typically lacks sensitivity. Although cell culture is considered the ‘gold standard’ for diagnosis of active TB because of its higher sensitivity, it is often slow, taking up to 6 or 8 weeks to give a result. Polymerase chain reaction (PCR)-based tests for Mtb, such as the new GeneXpert Mtb/RIF^® (Sunnyvale, Cepheid, CA, USA), can be quick and sensitive, but they are typically expensive and require a high standard of infrastructure. Imaging techniques such as chest radiography and computed tomography (CT) are quick and usually sensitive, but are not specific. Furthermore, they can be expensive, and, again, require a high standard of infrastructure (for example in the case of CT). All of these tests tend to be less accurate in cases of active TB with HIV co-infection, and also in cases of extrapulmonary TB.

Currently available tests for LTBI include the tuberculin skin test (TST) and interferon gamma release assays (IGRAs). The TST measures the in vivo delayed-type hypersensitivity response to intradermal inoculation of a crude mixture of mycobacterial antigens. Because this mixture contains antigens also present in bacillus Calmette–Guérin (BCG), the test can be confounded by prior BCG vaccination. IGRAs, on the other hand, detect ex vivo interferon gamma (IFN-γ) release from T cells (lymphocytes that play a key role in cell-mediated immunity) in response to Mtb-specific antigens, early secretory antigenic 6 kDa (ESAT-6) and culture filtrate protein 10 (CFP-10). These antigens are absent from the BCG vaccine and most environmental mycobacteria, and thus IGRAs tend to be more specific than the TST. 5 The two types of commercially available IGRAs are QuantiFERON GOLD In-Tube (QFT-GIT; Cellestis, Carnegie, VIC, Australia), a whole-blood enzyme-linked immunosorbent assay (ELISA), and T-SPOT. TB^® (Oxford Immunotec, Abingdon, UK), an enzyme-linked immunospot assay (ELISpot); both IGRAs utilise peripheral blood mononuclear cells (PBMCs). These tests have been recommended by the UK, European and North American guidelines for diagnosis of LTBI. 6

Although typically used in the diagnosis of LTBI, TSTs and IGRAs actually detect Mtb infection in its entirety (i.e. active or latent). The role of the tests within published guidelines for diagnostic evaluation of suspected active TB to date has been limited because of their low specificity for the disease: they could never confirm a diagnosis of active TB because they cannot differentiate latent and active infection. However, because Mtb infection is a pre-requisite for TB disease, reliable determination of infection status could accelerate diagnostic assessment by enabling rapid exclusion of TB (within 24 hours) when the result is negative.

In order for the IGRA or TST to reliably rule out a diagnosis of Mtb infection and thus TB disease, the sensitivity of the test must be very high (> 95%). The sensitivity and specificity of IGRAs compared with the TST in active TB have been examined in a number of studies, varying in size and quality. 7 IGRAs are typically more specific than the TST for diagnosing Mtb infection and T-SPOT. TB is more sensitive than the TST for diagnosing TB. However, the diagnostic accuracy of T-SPOT. TB and QFT-GIT has not been compared directly head to head in suspected active TB in the UK, nor comprehensively assessed in immunosuppressed patients. Therefore, there is uncertainty in the role and clinical utility of IGRAs in the diagnostic workup of suspected TB, as well as their cost-effectiveness in UK NHS practice.

Aim and study objectives

Overview of the study design

This prospective multicentre study comparing the accuracy of IGRAs was conducted in routine clinical practice in the UK. Adults presenting with suspected active TB at NHS outpatient or inpatient services to participating hospitals in London, Slough, Oxford, Leicester and Birmingham were recruited. We used a within-patient design to compare test accuracy by performing all IGRAs on blood samples from each patient with the presence or absence of active TB verified using the reference standard. This design minimises between-patient variability while also allowing estimation of the accuracy of combinations of IGRAs. Blood samples for IGRA testing were collected from patients at baseline and follow-up (2 and 6 months). If necessary, and when available, TST results were used as part of the composite reference standard for verifying the final diagnosis of patients. The TST results were obtained from routine clinical care and so the availability of TST results reflects local practice in participating hospitals.

Participants

Setting

Patients were recruited at the point of diagnostic workup from 14 hospitals in 10 NHS trusts in the UK.

Recruitment process

An overview of the recruitment process is shown in Figure 1. Potential participants presenting to participating NHS centres were referred to a TB research nurse by the attending clinician. The nurse then screened participants to ensure that they were eligible for the study according to the inclusion/exclusion criteria stated above. Each potential participant was provided with an information sheet and a verbal description of the study. Participants were included in the study if they were willing and informed consent was obtained.

FIGURE 1.

Overview of recruitment process.

Data collection and management

Sample collection

Index tests

Interferon gamma release assays

Two types of commercially available IGRAs, QFT-GIT and T-SPOT. TB, were evaluated. In addition, a new ELISpot-based assay utilising novel antigens (Rv3615c, Rv2654, Rv3879c and Rv3873) was evaluated. The performance of each antigen was evaluated individually and in combinations that included either ESAT-6 and CFP-10, the two antigens that constitute T-SPOT. TB, or both. IGRA testing is not standard practice for HIV-negative patients suspected of having TB in the hospitals of our consortium and is not currently recommended for HIV-positive patients suspected of having TB. However, if IGRAs were used locally at participating hospitals as part of the routine diagnostic workup of patients, we recorded the tests done but we did not analyse the test results as study results for the IDEA study. Thus, only from IGRAs performed in our research laboratory specifically for this study were recorded and assessed. Laboratory staff performing the IGRAs and recording the test results were blinded to clinical information and reference standard results.

Analysis of T-SPOT.TB and novel antigens

Peripheral blood mononuclear cells were isolated from heparinised whole blood using the Ficoll-Paque™ density centrifugation method (GE Healthcare Bio-Science, Uppsala, Sweden), as described by Whitworth et al. 6 In brief, whole blood was diluted in Roswell Park Memorial Institute (RPMI) 1640 medium (Sigma-Aldrich, Dorset, UK) and layered onto Ficoll-Paque™ Plus at a ratio of 2 : 1 in 50-ml Falcon^® centrifuge tubes (Corning Science Mexico S.A. de C.V., Reynosa, USA). The tubes were centrifuged for 20 minutes at 18–25 °C and the cloudy PBMC layer was aspirated into fresh RPMI 1640 medium. Cells were washed with fresh RPMI 1640 medium and counted using trypan blue stain for use with the Countess^® Automated Cell counter (Life Technologies, Eugene, OR, USA). T-SPOT. TB was applied to the freshly isolated PBMCs as per the manufacturer’s instructions9 and as described by Whitworth et al. 6

Cells were resuspended in AIM-V^® Serum-Free Medium (Gibco by Life Technologies, Thermo Fisher Scientific, Waltham, MA, USA) at a concentration of 2.5 million cells/ml and 250,000 cells per well were incubated overnight (18 hours) at 37 °C with Mtb-specific antigens (ESAT-6, CFP-10, Rv3615c, Rv2654, Rv3879c and Rv3873) individually, and positive (phytohaemagglutinin) and negative (RPMI 1640 medium) controls in a 96-well plate, pre-coated with IFN-γ-specific monoclonal capture antibodies (included in T-SPOT. TB kit). Thus, a total of eight wells were used per patient and samples from 12 patients were included on a plate. Overnight incubation of the cells with antigens allows for IFN-γ secretion from activated Mtb-specific effector T cells present in the culture. Secreted IFN-γ binded to the pre-coated IFN-γ-specific monoclonal capture antibodies on the membrane of each well. After incubation, wells were washed with phosphate-buffered saline and an alkaline phosphatase-conjugated secondary IFN-γ-specific monoclonal antibody was added to bind to any captured IFN-γ. For a visible representation of the spots on the membrane, an alkaline phosphatase chromogen substrate was added.

Each spot formed on the membrane signifies IFN-γ release by a single activated Mtb antigen-specific T cell. Spot-forming cells (SFCs) are expected to be detected in positive control wells and absent in negative control wells. SFCs in TB antigen-stimulated wells indicate infection.

Spot-forming cells were counted using an automated ELISpot plate reader (AID ELISpot read system ELRIFL04; Advanced Imaging Devices GmbH, Strasbourg, Germany), with saturation level set at 60%. For individual antigens, test results were classified as negative, positive, borderline (equivocal) or indeterminate (invalid) by subtracting the spot count in the negative control well from the spot count in each panel, according to the algorithm illustrated in Figure 2 (based on the package insert for T-SPOT. TB). 9 Panels A, B, C, D, E and F correspond to ESAT-6, CFP-10, Rv3615c, Rv2654c, Rv3879c and Rv3873, respectively.

FIGURE 2.

Algorithm for defining test positivity of antigens. Panel X indicates one of the six panels, A to F, which correspond to the following antigens: A, ESAT-6; B, CFP-10; C, Rv3615c; D, Rv2654; E, Rv3879c; and F, Rv3873. TNTC, too numerous to count.

For T-SPOT. TB as well as other antigen combinations, if the positive control spot count was < 20, the result was deemed indeterminate, unless the response to one of the Mtb antigens was positive (or borderline), in which case the test result for the combination was deemed positive (or borderline). Thus, we applied an ‘OR’ rule (at least one antigen spot count deemed positive) for antigen combinations. For example, for T-SPOT. TB, a result was positive if the negative-control spot count was ≤ 10 and either ‘panel A minus negative control’ or ‘panel B minus negative control’ was ≥ 8 spots. This implies the T-SPOT. TB result was negative if both ‘panel A minus negative control’ and ‘panel B minus negative control’ were negative (≤ 4 spots). In the IDEA study, borderline test results (5–7 spots) were considered as positives.

Analysis of QuantiFERON GOLD In-Tube

The QFT-GIT assay was performed in two stages as per the manufacturer’s instructions10 and as described in Whitworth et al. 6 First, whole blood was collected from each participant into three QFT-GIT tubes containing a negative control, mitogen-positive control and Mtb antigens [ESAT-6, CFP-10 and TB7.7 (also known as Rv2654c, a possible PhiRv2 prophage protein) combined], as provided by the manufacturer. 10 The tubes were incubated at 37 °C for 16–24 hours to allow IFN-γ secretion from antigen-specific effector T cells into the extracellular fluid (plasma). After incubation, tubes were centrifuged and 150 µl of plasma was collected and stored in a 96-well plate for up to 4 weeks at 2–8 °C prior to performing the remainder of the assay.

To perform the ELISA step, 50 µl of plasma from each of the QFT-GIT tubes (i.e. containing a mitogen control, negative control and Mtb antigens) was transferred to wells of another 96-well plate pre-coated with IFN-γ-specific monoclonal capture antibodies and incubated with a conjugate [an IFN-γ-specific antibody conjugated to horseradish peroxidase (included in T-SPOT. TB kit)] for 2 hours at room temperature (22 °C). Plasma samples in each well were mixed thoroughly using a microplate shaker (PMS-1000i Microplate Shaker; Grant Instruments Ltd, Shepreth, UK) for 1 minute to ensure that any IFN-γ was evenly distributed throughout the sample. Secreted IFN-γ in the plasma will be sandwiched between the two antibodies.

After incubation and thorough washing with detergent, a photosensitive chromogen substrate solution (3,3’,5,5’-tetramethylbenzidine; included in QFT-GIT kit) was added, which converted the sample to a detectable form (blue colour signal). The reaction was stopped with a substrate stopping solution (sulphuric acid; included in QFT-GIT kit). The intensity of the colour is directly proportional to the levels of IFN-γ present in the plasma after activation of TB-specific T cells by Mtb antigens. Colour should develop in the positive control well and not in the negative control well. Colour in the TB antigen well indicates infection.

The optical density of each well was measured using a microplate reader (Elx800 Absorbance Reader; BioTek, Carnegie, VIC, Australia) with a 450-nm filter and a 620- to 650-nm reference filter. The concentration (IU/ml) of IFN-γ for the plasma sample from each of the three tubes (negative, mitogen and Mtb antigen) was determined against a series of standard concentrations (the standard curve). The test result (negative, positive or indeterminate) was calculated from the concentration values using a US Food and Drug Administration-approved algorithm (Figure 3) run on QuantiFERON-TB Gold In-Tube Analysis Software version 2.62 (Cellestis) in accordance with the manufacturer’s instructions. 10

FIGURE 3.

Determination of the test results for QFT-GIT. Reprinted from Methods, Vol. 61, Whitworth HS, Scott M, Connell DW, Dongés B, Lalvani A. IGRAs – the gateway to T cell based TB diagnosis, pp. 52–62. © 2013, with permission from Elsevier. 6

Reference standard

Participating hospitals followed the minimum set of tests defined within the NICE guideline11 for diagnosing active TB and in accordance with local routine practice. However, the final diagnosis of participants was verified using the composite reference standard defined in Appendix 2. The reference standard was applied by a panel of clinicians blinded to local (routine) and study IGRA results. The role of the clinical panel was to assess anonymised patient clinical data without knowledge of the IGRA results in order to confirm the diagnostic category of all study participants. The composition of the panel and the assessment process is described in Composition of the clinical panel and Assessment of final diagnosis.

Outcomes

Sensitivity, specificity, PPVs and NPVs and likelihood ratios for each test and combinations of tests were calculated to determine their diagnostic accuracy and clinical utility. The likely primary clinical utility of immune-based testing is to exclude TB. Thus, when interpreting the analyses and drawing conclusions, the focus was primarily on the sensitivities and NPVs. For test comparisons, relative test performance was assessed by comparing the sensitivity and specificity of one test with those of another test. These results were presented as relative sensitivities and relative specificities.

Statistical analyses

Patient and public involvement

Ms Nisha Karnani was our patient and public representative for the duration of the study. She was consulted at key points during the study and was invited to SSC meetings and the IDEA study presentation at the end of the study.

Study oversight and management arrangements

Ethics arrangements and regulatory approvals

Recruitment of participants into main study cohort

A total of 1074 participants, including 177 (16.5%) who were HIV positive, were recruited from 10 NHS trusts into the main study between 25 November 2011 and 31 August 2013. The number of patients recruited at each centre is shown in Table 1. Over half of the study participants were recruited from two trusts: Imperial College Healthcare NHS Trust (26.4%) and London North West Healthcare NHS Trust (27.8%).

The flow of patients through the study is shown in Figure 4. Of the 1074 patients recruited, 845 were included in the analyses. Reasons for exclusion are shown in Figure 4. Forty patients from Frimley Health NHS Foundation Trust were excluded because patients all had diagnoses of confirmed or highly probable TB (categories 1 and 2) due to an error of implementation of recruitment criteria at this site (i.e. the natural spectrum of patients with suspected TB was not being recruited). The decision to exclude these patients was approved by the SSC following a SSC meeting during which the diagnosis of patients recruited at each centre was reviewed.

FIGURE 4.

Study flow diagram of patients with suspected active TB. The final four boxes show the number of patients with available IGRA results. a, Patients with previously diagnosed TB were excluded from analyses because IGRA results cannot be reliably interpreted in previously treated patients. The decision to exclude was taken by the expert diagnostic panel and study management group in consultation with the independent Study Steering Committee before unblinding of IGRA and next-generation IGRA results. b, On advice from the Study Steering Committee, and following consultation between the study management group and data management groups, 80 patients were excluded from analyses. Patients recruited from Frimley Health NHS Foundation Trust (n = 40) were excluded because they all had diagnoses of confirmed or highly probable TB (categories 1 and 2) due to an error of implementation of the recruitment criteria at this site, i.e. the natural spectrum of patients with suspected TB was not being recruited. A further subset of patients (n = 27) were, on review, considered by the expert diagnostic panel to be ineligible (before unblinding IGRA results) on the basis that they were being investigated for TB (due to an incidental abnormal chest X-ray, known contact with active TB, or screening for anti-TNF treatment), but did not present with symptoms or signs suggestive of TB. An additional 13 patients were excluded due to invalid consent forms. Republished with permission of Elsevier Science and Technology Journals, from Clinical utility of existing and second-generation interferon-γ release assays for diagnostic evaluation of tuberculosis: an observational cohort study, The Lancet Infectious Diseases, Whitworth HS, Badhan A, Boakye AA, Takwoingi Y, Rees-Roberts M, Partlett C, et al. , vol. 19, pp. 193–202, copyright 2019;27 permission conveyed through Copyright Clearance Center, Inc.

Republished with permission of Elsevier Science and Technology Journals, from Clinical utility of existing and second-generation interferon-γ release assays for diagnostic evaluation of tuberculosis: an observational cohort study, The Lancet Infectious Diseases, Whitworth HS, Badhan A, Boakye AA, Takwoingi Y, Rees-Roberts M, Partlett C, et al. , vol. 19, pp. 193–202, copyright 2019;27 permission conveyed through Copyright Clearance Center, Inc.

Table 2 shows the number of patients assigned to each diagnostic category. There were 43 (5.1%) patients with a clinically indeterminate (category 3) diagnosis. Of the remaining 802 patients, there were 363 (45.3%) cases of active TB [based on those with culture-confirmed (category 1) and highly probable (category 2) TB] and 439 (54.7%) in whom active TB was excluded (categories 4A to 4D). Of the 439 non-active TB cases, 117 (26.7%) were category 4D.

Baseline characteristics of participants

The main demographic characteristics of the 845 patients are given in Table 3. Most patients (64.4%) were recruited from an outpatient setting, and the remaining were recruited from an inpatient setting. The median age of patients was 38 (range 16–86) years and most (59.3%) of the patients were male. Almost half (48.2%) of the study population were of Indian origin. Altogether, 75 countries of birth were represented in the study; 16 of the countries had at least 10 participants, with India (26.7%) and the UK (22.8%) accounting for almost half of the study population. The complete list of countries is given in Appendix 4, Table 54.

Table 4 shows the clinical characteristics of the patients. Of the 845 patients, 135 (16.0%) were HIV positive. Out of the 135 patients, there were two (1.5%) with a clinically indeterminate diagnosis. Of the remaining 133 patients, 25 (18.8%) were active TB cases and 108 (81.2%) were non-active TB cases.

Of the 845 patients, over half had other comorbidities: 300 (35.5%) patients had a single comorbidity, 127 (15.0%) had multiple comorbidities and the remaining 418 (49.5%) had none. There were 88 (10.4%) patients with pre-existing diabetes mellitus, 12 (1.4%) patients with chronic/end-stage renal failure and 105 (12.4%) patients were on immunosuppressive therapy (Table 5). These were the three key subgroups that we had planned to investigate in the subgroup analyses. The thresholds used to categorise vitamin D status varied between hospital trusts, as detailed in Appendix 5, Table 55. Although vitamin D measurements were missing for a large number of patients (49.1%), when the results were available, many patients were categorised as either vitamin D deficient (26.5%) or insufficient (16.9%), with few (7.5%) having normal results (see Table 4).

The social history of participants is outlined in Table 6. Smoking history was missing for two patients; about two-thirds of the patients had never smoked, and the remaining patients were current or ex-smokers. Most (58.6%) patients also had no history of alcohol use. Almost all patients (97.6%) had no history of homelessness and a few patients (27/845, 3.2%) had a history of imprisonment.

Table 7 summarises the frequency of presenting symptoms for 827 patients. The main symptoms recorded were cough, fever, night sweats, weight loss, haemoptysis and lethargy. Patients generally presented with multiple symptoms, but a cough was often present (576/827, 69.6%). The median number of symptoms was four (range 1–10).

Final diagnosis

Table 8 shows the diagnostic tests performed during the diagnostic workup of patients. Chest radiography and culture were often performed (in 89.4% and 86.5% of patients, respectively), but cerebrospinal fluid testing and magnetic resonance imaging (MRI) were uncommon (in 3.6% and 12.0%, respectively). The number of T-SPOT. TB, QFT-GIT and TST tests performed as part of routine care at each centre is shown in Appendix 6, Table 56. However, for the purpose of the IDEA study, IGRAs were not used in determining the final diagnosis of patients. TSTs were performed in only 336 patients across the nine centres and results were available for 322 patients. Most of these 336 patients were recruited at London North West Healthcare NHS Trust (57%) and Imperial College Healthcare NHS Trust (26%).

The final diagnosis of active TB patients is detailed in Table 9. Of the 363 patients with active TB, 237 (65.3%) had smear-negative TB. Forty-five (12.4%) active TB cases had both pulmonary and extrapulmonary TB. Approximately half (189/363, 52.1%) had only extrapulmonary TB. This occurred more often among those diagnosed as having highly probable TB (75/102, 73.5%) relative to culture-confirmed cases (114/261, 43.7%). In 129 (35.5%) active TB cases, patients had only pulmonary TB. In contrast to extrapulmonary TB, pulmonary TB was more common among culture-confirmed cases (110/261, 42.15%) than in highly probable TB cases (19/102, 18.6%). The most common sites of TB infection were the lungs (174/363, 47.9%) and lymph nodes (154/363, 42.4%). The drug sensitivity profile shows that of the 351 culture tests performed, 239 (65.8%) were fully sensitive and 22 (6.3%) were drug resistant.

Table 10 shows the final diagnosis of non-active TB patients. A patient may have multiple conditions. Of the seven conditions listed in the table, pneumonia was the most frequent diagnosis, with 104 of 439 (23.7%) patients having the condition. A higher proportion of inpatients were diagnosed with cancer (14.1%) or pneumonia (38.8%) than outpatients (4.5% and 14.1%, respectively). In contrast, a higher proportion of outpatients were diagnosed with chest infections, latent TB infection and sarcoidosis.

Overview

Estimates of the accuracy of IGRAs presented in this chapter are for the main cohort of patients (including HIV-positive patients recruited prior to the extension period). Estimates of the accuracy of T-SPOT. TB and QFT-GIT are presented individually, followed by comparisons of test accuracy (T-SPOT. TB vs. QFT-GIT). The results of subgroup analyses are then presented for T-SPOT. TB and QFT-GIT. Finally, test accuracy estimates are provided for ESAT-6, CFP-10 and second-generation IGRAs (Rv3615c, Rv2654, Rv3879c and Rv3873) individually and in combinations; the diagnostic accuracy of the test combinations were then compared with that of T-SPOT. TB. For completeness, we also briefly address the accuracy of the TST.

Completeness of interferon gamma release assay results

Of the 845 patients, T-SPOT. TB and QFT-GIT results were available for 809 (96%) and 820 (97%) patients, respectively. All 809 patients with T-SPOT. TB results also had results for the second-generation IGRAs (Rv3615c, Rv2654, Rv3879c and Rv3873). Table 11 shows the results for T-SPOT. TB and QFT-GIT by diagnostic category. For 805 patients, results were available for both IGRAs; reasons for missing T-SPOT. TB and QFT-GIT results are shown in Table 12. The cross-classified results of the two tests are given in Appendix 7 (see Table 57) for the 805 patients.

Diagnostic accuracy of T-SPOT.TB and QuantiFERON GOLD In-Tube

Table 13 shows the cross-tabulation of T-SPOT. TB and QFT-GIT results against active TB status (i.e. active TB or not). The proportion of indeterminate test results was 7.0% (57/809) for T-SPOT. TB and 9.6% (79/820) for QFT-GIT. The difference between the two proportions was 2.6% (95% CI –0.1% to 5.3%; p = 0.06). Based on all culture-confirmed and highly probable active TB cases and excluding indeterminate IGRA results, sensitivity was 82.3% (95% CI 77.8% to 86.1%) for T-SPOT. TB and 67.3% (95% CI 62.0% to 72.1%) for QFT-GIT. Among those in whom active TB was excluded, specificity was 82.6% (95% CI 78.6% to 86.1%) for T-SPOT. TB and 80.4% (95% CI 76.1% to 84.1%) for QFT-GIT (Table 14). The PPVs for T-SPOT. TB and QFT-GIT were 80.1% (95% CI 75.5% to 84.0%) and 74.8% (95% CI 69.6% to 79.5%), respectively, and the NPVs were 84.6% (95% CI 80.6% to 87.9%) and 74.0% (95% CI 69.5% to 78.0%), respectively. For T-SPOT. TB, 4.1% (33/809) of the test results were borderline (see Table 13). When these borderline results were excluded from the T-SPOT. TB analysis, sensitivity (95% CI) was 81.4% (76.6% to 85.3%), specificity (95% CI) was 86.2% (82.3% to 89.4%), PPV (95% CI) was 83.2% (78.6% to 87.0%) and NPV (95% CI) was 84.6% (80.6% to 87.9%). The full results of the analysis are shown in Appendix 7, Table 58.

Using only culture-confirmed active TB cases, the sensitivities of T-SPOT. TB and QFT-GIT increased slightly to 85.9% (95% CI 80.9% to 89.8%) and 70.6% (95% CI 64.4% to 76.1%), respectively. The sensitivity of T-SPOT. TB was 7.0% lower in patients diagnosed with pulmonary TB than in those with extrapulmonary TB [77.0% (95% CI 68.4% to 83.8%) vs. 84.0% (95% CI 77.9% to 88.7%)]. In contrast, although the difference was small (2.3%), the sensitivity of QFT-GIT was higher in patients who had pulmonary TB (68.4%, 95% CI 59.4% to 76.2%) than in those with extrapulmonary TB (66.1%, 95% CI 58.7% to 72.8%). Using only category 4D patients without active TB gave much higher specificities than using all patients without active TB (see Table 14).

In the sensitivity analyses with indeterminate test results included as test positives, the sensitivity of T-SPOT. TB was 83.2% (95% CI 78.9% to 86.8%) and 69.7% (95% CI 64.7% to 74.2%) for QFT-GIT. The specificity of T-SPOT. TB was 75.4% (95% CI 71.1% to 79.3%) and 71.5% (95% CI 67.1% to 75.6%) for QFT-GIT. Full results are provided in Appendix 7, Table 59.

Sensitivity, specificity and predictive values are presented as percentages. For TSPOT. TB, there were 69 test positives out of 94 highly probable TB cases, with sensitivity (95% CI) of 73.4% (63.7% to 81.3%). For QFT-GIT, there were 57 test positives out of 96 highly probable TB cases, with sensitivity (95% CI) of 59.4% (49.4% to 68.7%). Note that in the primary analyses of T-SPOT. TB, borderline test results were included as test positives. Sensitivity analyses were performed with borderline T-SPOT. TB results excluded and the results are shown in Appendix 7, Table 58. Indeterminate IGRA results were excluded from all analyses. See Appendix 7 for sensitivity analyses using all IGRA results with indeterminates included as test positives.

Comparison of diagnostic accuracy of T-SPOT.TB and QuantiFERON GOLD In-Tube

Excluding indeterminate IGRA results, there were 714 T-SPOT. TB results and 705 QFT-GIT results. Based on analyses using GEE models, the sensitivity of T-SPOT. TB was superior to that of QFT-GIT (relative sensitivity 1.22, 95% CI 1.14 to 1.31; p < 0.001), but there was no statistical evidence of a difference in specificity (relative specificity 1.02, 95% CI 0.97 to 1.08; p = 0.3) (Table 15). Excluding the 33 borderline T-SPOT. TB results (see Appendix 7, Table 60), the results for relative sensitivity were similar to those from the main analysis but there was statistical evidence of a difference in specificity; the relative sensitivity (95% CI) was 1.20 (1.12 to 1.29; p < 0.001) and the relative specificity (95% CI) was 1.07 (1.02 to 1.12; p = 0.004). When all IGRA results were analysed and indeterminate IGRA results were included as test positives in a sensitivity analysis, the analysis included 768 T-SPOT. TB results and 778 QFT-GIT results. Similar to the primary analysis, there was statistical evidence of a difference in sensitivity (relative sensitivity 1.19, 95% CI 1.12 to 1.28; p < 0.001) and no evidence of a difference in specificity (relative specificity 1.05 95% CI 0.98 to 1.13; p = 0.1) (see Appendix 7, Table 61).

Subgroup analyses for T-SPOT.TB and QuantiFERON GOLD In-Tube

Diagnostic accuracy of second-generation interferon gamma release assays

The diagnostic accuracy of ESAT-6, CFP-10 and four second-generation IGRAs (Rv3615c, Rv2654, Rv3879c and Rv3873) is presented in the following sections as individual tests and as test combinations. The accuracy of different test combinations was compared with that of T-SPOT. TB.

Evaluation of the tuberculin skin test

The TST was not performed as part of the IDEA study. We did not aim to evaluate the accuracy of the TST and so we did not impose a standard protocol for the conduct of TSTs; each centre followed local trust policy. Five of the nine hospital trusts included in the analyses performed TSTs, but only on a subset of patients, ranging between 20% and 74% of those recruited. We do not know why certain patients were selected for the TST and others were not but the reasons may involve a mix of patient- and clinician-specific factors. Altogether, the TST results were available for only 38% of the study population. The TST was mainly performed at two centres (London North West Healthcare NHS Trust and Imperial College Healthcare NHS Trust), and these centres together accounted for 83% (266/322) of the available TST results. Furthermore, practice varied across the centres in the IDEA study.

Nonetheless, for completeness we estimated the diagnostic accuracy of the TST alone. In addition, because one of our primary objectives was to develop an evidence-based optimal testing algorithm that defines the role of IGRAs in the diagnostic workup of suspected active TB, we considered combinations with T-SPOT. TB or QFT-GIT. The results are presented in Appendix 13.

The sensitivity of the TST was evaluated at each of the three prespecified thresholds (≥ 5 mm, ≥ 10 mm and the stratified threshold). The findings presented in Appendix 13 should be interpreted cautiously given the limited number of data and variation in practice between centres. Moreover, the findings may be subject to bias. The final diagnosis of patients was verified using a composite reference standard applied by a panel of clinicians to anonymised patient clinical data. The panel were blinded to routine and study IGRA results. However, if TSTs were performed, then the results were available thus creating a potential for incorporation bias in the evaluation of the diagnostic accuracy of the TST.

Discussion

This large prospective study of the diagnostic accuracy of T-SPOT. TB, QFT-GIT, TSTs and second-generation IGRAs was conducted in secondary care in a population representative of clinical practice. The differences observed between inpatients and outpatients reflect the case mix of acute admissions, in which, for instance, pneumonia-like illnesses and advanced malignancy presentations are more common. However, the final TB diagnosis in these two groups was broadly similar and allows for generalisability of our findings. Figure 5 summarises the sensitivities of T-SPOT. TB, QFT-GIT, TSTs, and combinations of TSTs and IGRAs. Figure 5 also shows the sensitivities of combinations of novel antigens. T-SPOT. TB showed higher sensitivity than QFT-GIT, a relative increase of 22% (95% CI 14% to 31%), thus indicating greater utility as a rule-out test.

FIGURE 5.

Summary of sensitivities of IGRAs. The sensitivities of the six tests are sorted on the plot in increasing order.

Literature searches in PubMed were regularly conducted during the IDEA study to update the SSC on developments in the field. Appendix 11, Table 74, gives a summary of 31 IGRA studies and one systematic review published between 1 January 2013 and 16 March 2016. To the authors’ knowledge, the IDEA study is the largest prospective and most recent comparative evaluation of the accuracy of T-SPOT. TB and QFT-GIT for diagnosis of active TB in a high-income setting. None of the studies in Appendix 11 compared T-SPOT. TB and QFT-GIT either prospectively or retrospectively. QFT-GIT was compared with the TST in 14 (45.2%) studies and T-SPOT. TB was compared with the TST in four (12.9%) studies; the remaining 13 (41.9%) studies evaluated only T-SPOT. TB or QFT-GIT. According to the systematic review of pleural TB published in 2015,28 there was only one good-quality study out of the 21 studies and two small studies compared QFT-GIT and T-SPOT. TB. The authors of the review pooled results across all QuantiFERON and T-SPOT. TB assays and so results are not comparable with the findings of the IDEA study.

The rate of indeterminate IGRA results was higher for QFT-GIT (9.6%) than for T-SPOT. TB (7.0%). For both IGRAs, most of the indeterminate results were among patients without active TB, with 59% (47/79) and 65% (37/57) occurring for QFT-GIT and T-SPOT. TB. Appendix 12 (see Table 75) summarises some key characteristics of patients with indeterminate IGRA results. Indeterminates were excluded from the main analyses and the impact on findings was investigated in sensitivity analyses by including indeterminate IGRA results as test positives, that is, as true positives for those with culture-confirmed or highly probable active TB and as false positives for those without active TB. Based on this rule, an increase in sensitivities, a decrease in specificities and no change in NPVs was expected. Generally, small increases were seen in sensitivity accompanied by decreases in specificity.

Human immunodeficiency virus-positive patients were a key subgroup in which analyses of T-SPOT. TB and QFT-GIT were planned. However, the number of patients (n = 135) recruited in the main cohort was small. The IDEA study was extended to facilitate recruitment of additional patients. The results for the entire HIV cohort (i.e. including patients recruited during the extension period) are presented in Chapter 5. For the main study cohort, the sensitivities of T-SPOT. TB and QFT-GIT were lower in those with HIV co-infection than the HIV-negative subgroup. Similarly, the two IGRAs had lower sensitivities in those patients with diabetes mellitus than those without. Specificity was higher in HIV-positive patients than HIV-negative patients but was lower in those with diabetes mellitus than those without diabetes mellitus. Although there appeared to be differences in test performance between subgroups, there was no statistical evidence of an effect of HIV infection status or diabetes mellitus on relative test performance. The findings from these analyses should be taken with caution, as the number of test results in some subgroups was small. Subgroup analyses were not possible for the other two key subgroups: those with end-stage renal failure and those on immune suppressants.

The most promising novel antigen was RV3615c, with a sensitivity that was higher than that of the other five antigens, including CFP-10 and ESAT-6 (the two antigens that constitute T-SPOT. TB). Test combinations including Rv3615c showed higher sensitivity than T-SPOT. TB, with limited gains in sensitivity when more antigens were added to a combination. Thus, the two-antigen combination of CFP-10 and Rv3615c or the three-antigen combination of ESAT-6, CFP-10 and Rv3615c seem promising. The latter combination had a sensitivity and specificity of 90.5% (95% CI 86.9% to 93.2%) and 70.0% (95% CI 65.4% to 74.2%), compared with 83.2% (95% CI 78.9% to 86.8%) and 75.4% (95% CI 71.1% to 79.3%) for T-SPOT. TB. The added value of Rv3615c to T-SPOT. TB was a 9% (95% CI 5% to 12%) relative increase in sensitivity at the expense of specificity with a relative decrease of 7% (95% CI 4% to 10%). The incremental gain in sensitivity to 90% is likely to be clinically useful and warrants further investigation.

Recruitment of human immunodeficiency virus-positive patients

Following the closure of the main study to recruitment, two additional NHS trusts (King’s College Healthcare NHS Foundation Trust and Barts Health NHS Trust) were invited to participate in the study to facilitate the recruitment of HIV-positive patients. A total of 263 patients were recruited from 12 NHS trusts between 25 November 2011 and 19 December 2014. The number of patients recruited at each centre is shown in Table 27. Over half of the patients were recruited from two trusts: Royal Free London NHS Foundation Trust (29.3%) and Imperial College Healthcare NHS Trust (25.9%).

The flow of patients through the study is shown in Figure 6. Of the 263 HIV-positive patients recruited, 201 were included in the analyses. Reasons for exclusion are shown in Figure 6. Two patients from Frimley Health NHS Foundation Trust were excluded, as mentioned earlier in Chapter 3 (see Recruitment of participants into main study cohort).

FIGURE 6.

Study flow diagram of HIV-positive patients with suspected active TB. The final four boxes show the number of patients with available IGRA results. a, On advice from the Study Steering Committee, and following consultation between the study management group and data management groups, nine patients were excluded from analyses. Two patients recruited from Frimley Health NHS Foundation Trust were excluded because all patients recruited from that site had diagnoses of confirmed or highly probable TB (categories 1 and 2) due an error of implementation of recruitment criteria, i.e. the natural spectrum of patients with suspected TB was not being recruited. Two patients were, on review, considered by the expert diagnostic panel to be ineligible (before unblinding IGRA results) on the basis that they were being investigated for TB (due to an incidental abnormal chest X-ray, known contact with active TB, or screening for anti-TNF treatment), but did not present with symptoms or signs suggestive of TB. An additional five patients were excluded due to invalid consent forms.

The final diagnosis of four (2.0%) patients was clinically indeterminate (Figure 6 and Table 28). Among the remaining 197 patients, there were 32 (16.2%) cases of active TB and 165 (83.8%) cases of non-active TB. Of the 165 non-active TB cases, 68 (41.2%) were category 4D.

Baseline characteristics of the human immunodeficiency virus-positive cohort

The demographic characteristics of the 201 patients are given in Table 29. The median age of patients was 43 years (range 18–79 years) and the majority (67.7%) were male. A substantial number of HIV-positive patients were of black (45.3%) or white (37.8%) ethnicity. A total of 53 countries of birth were represented; countries with at least three participants are shown in Table 29. Many patients (97/201, 48.3%) were in paid employment.

The clinical characteristics of the patients are presented in Table 30. Of the 201 patients, 71 (35.3%) had no other comorbidities. There were 10 (5.0%) patients with pre-existing diabetes mellitus, four (2.0%) with chronic/end-stage renal failure and none was on immunosuppressive therapy. Vitamin D measurement was unknown for most (85.1%) patients (see Appendix 5, Table 55, for thresholds used to categorise vitamin D status). Table 30 shows CD4 (cluster of differentiation 4) count grouped into four categories. CD4 count was missing for eight patients. Most of the 193 patients had a CD4 count of ≥ 200 cells/µl (46.8%). The median CD4 count was 285 cells/µl (range 0–1228 cells/µl). The distribution of CD4 count in the cohort is shown in Figure 7.

FIGURE 7.

Distribution of CD4 counts in HIV-positive substudy cohort.

Table 31 gives the social history of patients. Smoking history was missing for one patient; almost half (48.8%) of the patients had never smoked, whereas the remaining were current (31.3%) or ex-smokers (19.4%). Fourteen (7.0%) patients had a history of alcohol misuse and the history of recreational drug use was unknown for many (64.7%) patients. Few patients (6.0%) had a history of homelessness and 7.5% had a history of imprisonment.

The frequency of symptoms is shown in Table 32 based on data from 197 patients. The four other patients were recruited on the basis of abnormal clinical signs rather than symptoms. The main symptoms were cough, fever, night sweats, weight loss and lethargy. Patients generally presented with multiple symptoms; the median number of symptoms was three (range 1–10). Cough, as a symptom, was often present (68.5%) in patients.

Final diagnosis in human immunodeficiency virus-positive patients

The diagnostic tests performed during the diagnostic workup of patients are shown in Table 33. Chest radiography and culture were the most common tests performed (with each performed in 90.6% of patients). The number of T-SPOT. TB, QFT-GIT and TST tests performed as part of routine care, at each centre, is shown in Appendix 14 (see Table 78). These routine IGRA results were not used in the final diagnosis of patients in the IDEA study. TSTs were performed in only 12 patients at 3 of the 11 centres; TST results were available for all 12 patients.

The final diagnoses of active TB patients are summarised in Table 34. Of the 32 patients with active TB, 19 (59.4%) had smear-negative TB. A total of 13 patients (40.6%) had pulmonary TB, 15 (46.9%) patients had extrapulmonary TB, and the remaining four (12.5%) patients had both forms of TB. The most common sites of infection were the lungs (53.1%) and lymph nodes (31.1%). Of the 31 culture tests performed, 20 (64.5%) had no drug resistance. Table 35 shows the final diagnosis of non-active TB patients. A patient may have multiple conditions, but pneumonia was the most frequent diagnosis (40.0%).

Test results for human immunodeficiency virus-positive patients

Of the 201 patients, 194 (96.5%) had results for both T-SPOT. TB and QFT-GIT; reasons for missing T-SPOT. TB and QFT-GIT results are shown in Table 36. The cross-classified results of the two tests are given in Appendix 14, Table 79. Table 37 shows the results for T-SPOT. TB, QFT-GIT and the TST according to the four diagnostic categories. Table 38 shows the cross-classification of the two tests for those with (categories 1 and 2) and without active TB (category 4).

Diagnostic accuracy of T-SPOT.TB and QuantiFERON GOLD In-Tube in human immunodeficiency virus-positive cohort

The proportion of indeterminate test results was 23.1% (45/195) for T-SPOT. TB and 19.5% (39/200) for QFT-GIT. The difference between the two proportions was 3.6% (95% CI –4.5% to 11.6%; p = 0.4). Sensitivity was 68.0% (95% CI 48.4% to 82.8%) for T-SPOT. TB and 56.7% (95% CI 39.2% to 72.6%) for QFT-GIT (Table 39). The specificities were 83.5% (95% CI 75.8% to 89.0%) and 91.4% (95% CI 85.3% to 95.1%). The PPVs for T-SPOT. TB and QFT-GIT were 46.0% (95% CI 31.0% to 61.6%) and 60.7% (95% CI 42.4% to 76.4%), respectively, and the NPVs were 92.7% (95% CI 86.2% to 96.2%) and 90.0% (95% CI 83.6% to 94.1%), respectively.

Using only culture-confirmed active TB cases, the sensitivity of T-SPOT. TB decreased slightly to 66.7% (95% CI 41.7% to 84.8%), whereas that of QFT-GIT increased to 61.1% (95% CI 38.6% to 79.7%). When the analyses were restricted to category 4D patients without active TB, specificities were higher than using all patients without active TB (see Table 39). In sensitivity analyses with indeterminate test results included as test positives, the sensitivity of T-SPOT. TB was 74.2% (95% CI 56.8% to 86.3%) and 59.4% (95% CI 42.3% to 74.5%) for QFT-GIT. The specificity of T-SPOT. TB was 63.1% (95% CI 55.4% to 70.2%) and 71.3% (95% CI 64.0% to 77.7%) for QFT-GIT. See Appendix 14, Table 80, for full results.

The diagnostic performance of the two IGRAs is shown stratified by CD4 count in Table 40. The estimates within each stratum give results comparable to the entire cohort in terms of the higher sensitivity of T-SPOT. TB and higher specificity of QFT-GIT. However, the estimates are based on very small numbers of active TB and non-active TB cases and are presented solely for illustrative purposes.

Comparison of diagnostic accuracy of T-SPOT.TB and QuantiFERON GOLD In-Tube in human immunodeficiency virus-positive cohort

Excluding indeterminate IGRA results, there were 146 T-SPOT. TB results and 158 QFT-GIT results. The sensitivity of T-SPOT. TB was higher than that of QFT-GIT, with a relative sensitivity of 1.12 (95% CI 0.87 to 1.44). There was no statistical evidence of a difference (p = 0.4). In contrast, the specificity of T-SPOT. TB was significantly lower than that of QFT-GIT, with a relative specificity of 0.91 (95% CI 0.84 to 0.99) (Table 41). When indeterminate IGRA results were included as test positives in a sensitivity analysis, the analysis included 191 T-SPOT. TB results and 196 QFT-GIT results. Unlike the primary analysis, there was no statistical evidence of a difference in specificity (see Appendix 14, Table 81).

Discussion

Of the 263 patients recruited, 201 patients were included in the analyses. This is well below the target sample size of 390. Although T-SPOT. TB showed higher sensitivity than QFT-GIT, a relative increase of 12%, there was no statistical evidence of a difference in sensitivity. The sensitivity of QFT-GIT in the IDEA study was lower than that in other recent studies29–31 (see Appendix 11, Table 74).

The indeterminate rate was lower for QFT-GIT (19.5%) than for T-SPOT. TB (23.1%). The indeterminate results were mainly in those without active TB, with 92.3% (36/39) for QFT-GIT and 86.7% (39/45) for T-SPOT. TB. The impact of indeterminate results was explored in sensitivity analyses by including indeterminate results as test positives. There was a small increase in the sensitivity of QFT-GIT but a large increase in the sensitivity of T-SPOT. TB. This is because of the higher indeterminate rate for T-SPOT. TB (18.2%) among category 1 and 2 active TB cases than for QFT-GIT (5.9%).

Decision tree model

We developed a decision tree model to calculate the incremental costs and incremental health utilities (quality-adjusted life-year; QALYs) of changing from current practice to using an IGRA as an initial rule-out test. Current practice was determined by the analysis of patient records. The model structure representing current practice is shown in Figure 7. Adding a rule-out test to the diagnostic pathway introduces additional delay in the diagnosis of active TB in those patients who have the disease, as it introduces an additional step in the pathway. Patients who were not initially diagnosed with active TB have a follow-up consultation after approximately 2 months; those who had a false-negative rule-out test result, that is, they had TB incorrectly ruled out, can have TB identified at this point. The final diagnostic outcomes were the four categories described in Dosanjh et al. ,3 herein referred to as ‘Dosanjh categories’ (see Appendix 2, Table 52). The health economic analysis was undertaken from a NHS perspective. No discounting was required, as the diagnostic process occurs over a relatively short time period.

The model contained two levels of uncertainty.

1. Individual-level uncertainty: patient records revealed variation in the number and type of tests used for TB diagnosis and time to diagnosis.

2. Parameter uncertainty: uncertainty in the costs of tests and procedures, and the sensitivity and specificity of IGRAs.

A (balanced) bootstrap sample of TB and non-TB patients was created for each simulation of the decision tree, which retained the subsample sizes. Individual patient costs and time to diagnosis were jointly sampled to preserve the dependency between them. A Bayesian forward sampling approach generated values from all of these distributions to simulate a particular model outcome. This was then repeated in a Monte Carlo framework to obtain a sample of several thousand model runs that encapsulated the first-order (patient variability) and second-order (parameter) uncertainty in the model outputs. Table 42 summarises the model parameters. The model was implemented in the statistical programming language R (version R-3.3.0; R Foundation for Statistical Computing, Vienna, Austria).

Introduction

This chapter presents the results of the economic evaluation. First, a data analysis with a health economics focus is given. This demonstrates the patterns and variability in times and costs between patients and also indicates why a simple direct estimation of the relevant summary statistics is not appropriate. Second, the modelling results are given and, finally, the main base-case scenario results are presented.

Results

The idealised diagnostic pathway representing current practice is shown in Figure 8. Patient records were analysed to determine what proportion of patients followed the pathways in this diagram. For each Dosanjh category, the particular tests expected to be performed and their corresponding results were determined and then compared with patient records. Tables 47 and 48 present the frequencies of different tests performed, stratified by final diagnostic outcome. Importantly, the process of TB diagnosis rarely followed the idealised diagnostic pathways. Although culture, IGRAs (QFT-GIT and T-SPOT. TB), TSTs, sputum smear microscopy and chest radiography were frequently used, there was substantial variation between patients. These results are presented graphically in Figure 9, in which the number of patients who traverse each branch are indicated by the width of the branch. From the individual-level patient data, we used the empirical distributions of time to diagnosis and total test costs in the probabilistic sensitivity analysis in the following health economic evaluation.

FIGURE 8.

Idealised TB diagnostic pathway.