Adjunctive colposcopy technologies for assessing suspected cervical abnormalities: systematic reviews and economic evaluation

Additional searches

Owing to the lack of evidence found in the review of clinical effectiveness, additional pragmatic PubMed searches were conducted to identify recent systematic reviews reporting on the adverse effects of CIN treatments on fertility, pregnancy and neonatal outcomes.

Types of studies

Participants

Eligible studies included participants who were referred to colposcopy through a cervical screening programme as a result of a suspected abnormality identified via liquid-based cytology, Pap smear test or positive hrHPV test. People referred for colposcopy as a follow-up after a previous CIN diagnosis (including test of cure) were also eligible for inclusion.

Intervention

DySISmap (DySIS Medical) or ZedScan (Zilico Ltd) as an adjunct to binocular or video colposcopy used for the diagnosis of CIN or cervical cancer was the intervention of interest. Studies on all versions of these tools (including prototypes) were considered for inclusion.

Comparators

Standard colposcopy was the comparator of interest; however, data from standard colposcopy alone did not need to be reported in order for a paper to be eligible. Both binocular and video colposcopy were included.

Reference standard

The reference standard was histopathology based on excisional or treatment biopsies, used to classify samples into three CIN grades or cervical cancer. Studies that did not perform biopsies to confirm the absence of disease when colposcopic examination did not reveal any abnormalities were included.

Outcomes

The following outcomes were eligible for study inclusion:

• diagnostic accuracy – including sensitivity and specificity, or sufficient data to calculate these

• test failure rates (and reasons for test failure)

• number of biopsies (and type) performed

• diagnostic results of biopsies

• number of treatments and treatment type

• number of see-and-treat procedures

• duration of colposcopy examination

• number of people discharged from colposcopy.

Eligibility depended on the study reporting results from both the index test and the reference standard. Only studies that reported results in terms of graded CIN, differentiating between mild dysplasia or lower-grade dysplasia (≤ CIN 1, i.e. negative diagnostic result) and moderate dysplasia or worse (CIN 2 or higher-grade dysplasia, i.e. positive diagnostic result) were included.

The following clinical outcomes were also eligible for study inclusion:

• morbidity and mortality associated with treatment and biopsies conducted as part of the colposcopy examination (including subsequent obstetric outcomes, such as miscarriage and infertility)

• morbidity and mortality associated with cervical cancer (in studies of DySIS and ZedScan)

• health-related quality of life (HRQoL)

• pain and anxiety associated with colposcopic examination, biopsies, treatment and waiting for results

• any other adverse event that may have an impact on resource use or quality of life (e.g. infection, infertility, miscarriage).

Outcomes related to the implementation of the interventions of interest and related practical issues were eligible for study inclusion:

• acceptability of the adjunctive technologies (to clinicians and patients)

• patient satisfaction

• successful database and record management

• training requirements

• capacity to perform colposcopies

• uptake and compliance.

Additional data from manufacturers and study authors

For all studies, additional data on diagnostic accuracy were requested. Requests were made to the device manufacturers (DySIS Medical or Zilico) for studies in which they had direct involvement, or to the first author of the primary publication for those manufacturers that were not involved in the study.

Diagnostic accuracy data for both colposcopy and adjunctive colposcopy (with either DySISmap or ZedScan) were requested as a 5 × 5 table, with the results categorised as < CIN 1, CIN 1, CIN 2, CIN 3 and cancer. Also requested were 2 × 2 tables of diagnostic accuracy in the following participant subgroups:

• participants with hrHPV infection (HPV 16 or 18)

• participants with low-risk HPV or no HPV infection

• participants referred to colposcopy with high-grade dyskaryosis or worse

• participants referred to colposcopy with low-grade dyskaryosis or lower

• participants with a previous history of CIN or cervical cancer (including test of cure).

Statistical analyses

Estimates of sensitivity and specificity were calculated using diagnostic accuracy data from the constructed 2 × 2 tables or the 5 × 5 tables supplied by manufacturers, and presented both as forest plots and in the receiver operating characteristic (ROC) space to examine the within- and between-study variability of diagnostic test accuracy. PPVs and negative predictive values (NPVs) were also calculated, as were diagnostic odds ratios (DORs).

When equivalent clinical thresholds were used to diagnose CIN/cancer in three or more studies, the hierarchical bivariate model described by Reitsma et al. 41 was fitted, providing summary estimates of sensitivity and specificity, and associated 95% confidence intervals (CIs). The hierarchical summary ROC model42 was also fitted to provide summary ROC curves. As the bivariate model does not account for the fact that different diagnostic tests may be performed in the same study, a further logistic regression analysis43 was performed to meta-analyse sensitivity and specificity, accounting for the fact that standard colposcopy and adjunctive colposcopy were performed on the same participants.

Unless otherwise specified, all analyses used the threshold of CIN 2 or higher as the cut-off point for defining a positive diagnostic test. Data on other thresholds were generally too limited for analyses to be performed, and CIN 2 or higher is the standard threshold used in colposcopy in the NHS.

All statistical analyses assume that standard colposcopy and adjunctive colposcopy (DySISmap or ZedScan) are performed independently. This is not the case in practice, as the tests are performed together. It is therefore possible that knowledge of one test could bias interpretation of the other. For example, because it is known that a ‘second look’ will happen after standard colposcopy, the results may not be the same as if standard colposcopy had been performed without access to adjunctive tests. As this applies equally to all studies, it was not possible to investigate this possible bias further.

If at least two studies reported on the same clinical or implementation outcome, the results were pooled if reporting was consistent enough for feasible analysis; otherwise, the results were synthesised narratively. Meta-analyses were performed using standard random-effects DerSimonian and Laird methods.

Analyses were conducted in the R software package (The R Foundation for Statistical Computing, Vienna, Austria).

Narrative and qualitative syntheses

Qualitative synthesis was performed for outcomes pertaining to implementation. Summary information relating to implementation outcomes, the conclusions of these studies, the consequences of colposcopy, recommendations for practice and suggested needs for further research were tabulated and summarised.

Narrative summaries were also performed for outcomes for which meta-analyses or other statistical analyses were not deemed feasible. This included tabulation or plotting of results as reported, which were then narratively described and compared.

Dynamic Spectral Imaging System studies

Only one study was rated as being at a low risk of bias overall,60 and the remaining eight studies were rated as being at a high risk of bias. Significant levels of concern regarding applicability were raised for five of the nine DySIS studies. 50,60,79,88,111 The main source of bias in DySIS studies was related to verification bias. Only three studies conducted biopsies in all patients analysed. 60,79,88 (Confidential information has been removed.) The remaining two DySIS studies were conference abstracts and did not report sufficient data to assess the risk of verification bias. 80,111

The DySIS technology used in the earlier study by Soutter et al. 88 was a precommercial model (FPC-03). The study reported technical issues relating to the software, speculum and a batch of faulty disposable nozzles, leading to the exclusion of a large proportion of eligible participants (31%) from the analyses. Therefore, the applicability of the results of this study may be limited.

ZedScan studies

Both studies of ZedScan were rated as having a high risk of bias overall,94,103 and significant levels of concern regarding applicability were raised for both studies.

Neither study conducted biopsies in participants with a normal cervical transformation zone to confirm the absence of CIN, and so both were rated as having a high risk of verification bias. Both studies were rated as having a high risk of study selection bias, notably because of the exclusion of patients with transformation zone type 3, in whom colposcopy may be harder to perform. 94

(Confidential information has been removed.) However, one study94 did collect data on whether or not biopsy would have been taken with colposcopy alone regardless of the ZedScan result, and the diagnostic accuracy results for standard colposcopy were reported in a linked substudy. 95 Therefore, the ZedScan results were rated as having a high risk of reporting bias.

(Confidential information has been removed) and most patients included in the study by Tidy et al. 103 were examined in a single centre (Sheffield), and the extent to which the results of this study are applicable to other settings is uncertain.

Risk of bias associated with the reference standard

In all included studies, nearly all histology was performed based on samples collected from punch biopsies rather than from deeper treatment biopsies. Although it is obviously unethical to perform treatment biopsies when this is not clinically indicated, samples from punch biopsies may be less accurate. 37 Therefore, all studies were rated as having a high risk of bias associated with the reference standard.

Dynamic Spectral Imaging System forest plots of diagnostic accuracy

In this section we present diagnostic accuracy results from the studies of DySIS in the form of forest plots.

Figure 3 shows estimates of sensitivity and Figure 4 shows estimates of specificity. Colposcopy alone has moderate sensitivity (58.4%, 95% CI 50.3% to 66.5%) but high specificity (confidential information has been removed); colposcopy therefore misses many women who do have CIN of grade 2 or greater, but produces relatively few false-positive test results. DySISmap alone has similar performance (confidential information has been removed). For adjunctive DySIS use, the sensitivity rises to (confidential information has been removed), so using DySISmap in addition to colposcopy correctly identifies more CIN 2 cases, but with a higher false-positive rate, which may mean performing biopsies in a larger proportion of women who do not have CIN 2 (or greater).

Figure 17 in Appendix 6 presents the DORs for each study. The diagnostic ratio is a combination of sensitivity and specificity (formally, log-odds of sensitivity minus log-odds of specificity), which increases as the overall diagnostic accuracy of a test increases. The results show almost no difference between colposcopy and adjunctive DySIS (confidential information has been removed), suggesting that DySISmap does not improve the diagnostic accuracy of colposcopy when defined in terms of DORs.

Figures 18 and 19 (Appendix 6) show the studies’ PPVs and the NPVs, respectively. These are harder to interpret, as PPVs and NPVs vary with prevalence, which is different across the studies. The PPV for adjunctive colposcopy is lower than that for colposcopy alone (confidential information has been removed), so fewer than half of all women who receive a DySIS-guided biopsy will have high-grade CIN. The summary PPV, and the estimated PPV in most studies, is lower than the 65% level recommended by UK guidance. 15 The NPV is slightly higher with adjunctive DySIS (confidential information has been removed), so fewer high-grade CIN cases will be missed.

Heterogeneity was substantial in almost all meta-analyses. The I² values are summarised in Table 5. All but one analysis had an I² value above 60%.

Bivariate and regression models of diagnostic accuracy

The analyses presented so far have not accounted for the correlation between sensitivity and specificity. A formal bivariate meta-analysis of diagnostic accuracy should be used to account for this correlation. 41 The analyses have also not accounted for the fact that colposcopy and DySIS are performed in the same study on the same participants. Full individual-level data would be needed to properly account for the within-person correlation between test results. These were not available, but extensions to the bivariate model can account for the fact that the tests were compared within the same study. 43

Figure 5 shows the sensitivity and specificity for all included studies. It can be seen that, for all studies, adjunctive DySIS has higher sensitivity but lower specificity than using colposcopy alone. Using DySISmap alone generally falls somewhere between the two.

Table 6 shows the results of the bivariate meta-analyses; Appendix 6, Figure 20, shows the results in the ROC space, including the 95% confidence regions (the ellipses) and the summary ROC curves. The results are consistent with those seen in the forest plots of diagnostic accuracy (see Figures 3 and 4), and show that using adjunctive DySIS increases sensitivity when compared with colposcopy alone, but at the cost of reduced specificity. As only three studies reported the use of DySISmap alone, no bivariate model was fitted for that test.

The bivariate model analyses colposcopy and adjunctive DySIS separately, and does not account for the fact that these are measured in the same studies. To correct for this, we fitted logistic regression models, including study-level parameters, to account for a possible correlation between test results within studies (see Statistical synthesis of diagnostic accuracy for details).

The summary results for this regression model are shown in Table 7. The results are similar to the standard bivariate model in Table 6. This model also permits a direct comparison of colposcopy and adjunctive DySIS. This found evidence of a difference in specificity between the tests [difference in log-odds of specificity 1.33, standard error (SE) 0.33; p < 0.0001] but no evidence of a difference in diagnostic accuracy (difference in log-DORs 0.04, SE 0.20; p = 0.84). This suggests that using DySIS changes the test threshold for the diagnosis of CIN 2 such that more women go on to receive biopsy, but that it is not improving diagnostic accuracy (in terms of DOR) when compared with colposcopy alone.

To confirm this, we also fitted a regression model that constrains adjunctive DySIS and colposcopy to have the same diagnostic accuracy (but which permits differences in specificity). The Bayesian information criterion (BIC) is used when comparing regression models; generally, a lower BIC suggests a better-fitting and more parsimonious model. This new model had a BIC of 198.3, which is lower than the previous model, which had a BIC of 201.5. This confirms that assuming that DySIS and colposcopy have the same diagnostic accuracy is reasonable.

Test positive rates

Figure 21 in Appendix 6 shows the test positive rate (the proportion of women in whom colposcopy or adjunctive colposcopy suggests the presence of high-grade CIN) for each test in each study of DySIS. In every study, adjunctive use of DySIS increased the positive rate compared with colposcopy alone, often substantially. In the Louwers et al. 60 study, for example, the use of DySIS increased the positive rate from 33.1% to 55.5%. Hence, the use of DySIS will substantially increase the number of women who receive biopsies after colposcopy. The results for ZedScan are shown in Figure 22 in Appendix 6. These suggest that the positive rate for ZedScan is similar to that for colposcopy alone, regardless of the cut-off point used. The disease-present cut-off point, unsurprisingly, produces higher positive rates than the colposcopic impression cut-off point.

Subgroup analyses

Some of the studies provided diagnostic accuracy data for key identified subgroups, namely:

• high-grade and low-grade cytology referral

• high-risk HPV (including HPV 16) and low-risk HPV

• the ‘test-of-cure’ population.

We include here data on these subgroups provided by the manufacturers or study authors, data from primary publications and data from secondary reports of primary studies when those papers included subgroup data not reported in the original study publication. Appendix 6, Table 44, provides an overview of the subgroups reported in each study.

Figure 7 shows sensitivity and specificity forest plots by subgroup for colposcopy alone, Figure 8 shows the same analyses for adjunctive DySIS and Figure 9 shows the same analyses for adjunctive ZedScan. Summary meta-analyses are not presented because of the small number of studies in each subgroup.

Colposcopy appears to be less sensitive than average at detecting high-grade CIN in women with low-grade referrals. This may partly be a consequence of interpretation bias (the colposcopic results are analysed with more caution if a woman is known to have a high-grade cytology referral) or it may be because lesions are harder to detect in women with a low-grade cytology referral. The same applies to women who have low-grade referrals combined with hrHPV. This difference is not observed when using DySIS or ZedScan; in both cases, the sensitivity and the specificity are similar for both low- and high-grade referrals. This suggests that adjunctive colposcopy may improve the detection of high-grade CIN in women with a low-grade referral (i.e. mild dyskaryosis).

There is no convincing evidence that diagnostic accuracy differs between women who are infected with hrHPV and those who are not infected with hrHPV; however, data are limited. For women infected with hrHPV, both the sensitivity and the specificity are higher than average (see Figures 3 and 4) when using adjunctive DySIS, suggesting that high-grade CIN is easier to detect in women infected with hrHPV. The sensitivity appears to be higher among women with hrHPV when using adjunctive DySIS than with colposcopy alone. This suggests that adjunctive DySIS may improve the detection of high-grade CIN in women infected with hrHPV.

The results for ZedScan are more limited, with no apparent evidence of differences between subgroups.

We note that all of these conclusions are based on small subgroups, generally of only one to three studies, and so the results should be considered as speculative only.

Meta-analyses were generally not possible within subgroups, as only one or two studies reported data for each subgroup. For high- and low-grade referrals for DySIS, a logistic regression model (as shown in Table 7) was fitted in each subgroup to meta-analyse diagnostic accuracy within these subgroups, as there were two DySIS studies for each subgroup. The results of this analysis are shown in Table 9. These results suggest that, for colposcopy alone, the sensitivity to detect CIN 2 among low-grade referrals is much lower than that for high-grade referrals, for a similar specificity. This difference was not quite statistically significant (p = 0.072), owing to the limited data available. There was no evidence of any difference between types of referral when using adjunctive DySIS, suggesting that adjunctive DySIS may be preferable to colposcopy for women with a low-grade referral. Similar regression models did not find any evidence of differences between HPV subgroups.

Sensitivity analyses

The study by Roensbo et al. 79 differed from the other included studies, as it did not assess DySIS as an adjunct to colposcopy directly, but only whether a colposcopist agreed or disagreed with the DySISmap result. As noted in Characteristics of the included studies, the study also used the DySISmap more conservatively than other DySIS studies, by interpreting areas with weaker acetowhitening on the coloured map (including dark blue and green) as being ‘suspicious for high-grade disease’.

We performed a sensitivity analysis of the logistic regression model in Table 7 for sensitivity and specificity, excluding this study. The results are shown in Table 10. There is little change when excluding Roensbo et al. ,79 with all estimates being similar to when the study was included.

A particular concern identified during the quality assessment was that different studies had different practices, when neither colposcopy nor adjunctive DySIS or ZedScan identified any abnormal areas requiring biopsy. In four studies45,50,74,80 no biopsy was performed in those women, in two studies60,88 a single biopsy was typically performed at a randomly location and in one79 study multiple biopsies were performed. In the case of one study, 80 insufficient information was provided. Table 11 shows the results of the meta-analyses categorised by the type of biopsy performed in the DySIS studies. These results suggest that specificity and sensitivity tend to decline as more biopsies are performed on colposcopy-negative women. This was observed for colposcopy alone and adjunctive DySIS. Despite these differences, the comparison between colposcopy and adjunctive DySIS is unchanged; using DySISmap as an adjunct to colposcopy increases sensitivity but decreases specificity in all cases.

Comparison of Dynamic Spectral Imaging System and ZedScan

All analyses so far have considered DySIS versus colposcopy and ZedScan versus colposcopy. Here, we briefly compare all three technologies.

As no studies in the review included both DySIS and ZedScan, a direct comparison is not possible. Instead, we consider an indirect comparison of the technologies. This is less reliable than a direct comparison, as there are differences in the study populations and conduct that may alter diagnostic accuracy over and above the differences in the diagnostic technology used.

Table 12 presents what we consider to be the best estimates of diagnostic accuracy. For DySIS, these are sourced from the logistic regression model comparing adjunctive DySIS with colposcopy (see Table 7). (Confidential information has been removed.) When comparing ZedScan with colposcopy, the situation is more complex, as the most recent study did not report colposcopy diagnostic data. If we are willing to assume that binocular and video colposcopes have the same diagnostic performance, then the best evidence is that found from the DySIS studies. Only the ZedScan prototype study103 has reported diagnostic accuracy data for binocular colposcopy, so an alternative estimate is the ‘colposcopic impression’ cut-off point from that study, as that cut-off point is closest to that used to diagnose high-grade CIN.

These results show that both adjunctive DySIS and adjunctive ZedScan substantially increase sensitivity, but also have substantially reduced specificity when compared with colposcopy. It would appear that adjunctive ZedScan more strongly favours high sensitivity, with a corresponding loss of specificity, when compared with adjunctive DySIS, but this is an indirect comparison.

We also performed a further logistic regression model to indirectly compare all three diagnostic tests. This model included all diagnostic data from all DySIS and ZedScan studies, and accounted for the fact that tests were conducted in the same studies and for the differing cut-off points used in the ZedScan prototype study. Compared with colposcopy alone, DORs were not higher with adjunctive DySIS (difference in log-DOR 0.06; p = 0.74), but were increased by adjunctive ZedScan (difference in log-DOR 0.84; p = 0.003). Hence, ZedScan had a greater DOR than DySIS (difference in log-DOR 0.59; p = 0.003), suggesting that ZedScan could have better diagnostic accuracy than DySIS, but the exact benefit would depend on the choice of cut-off point and the corresponding sensitivity and specificity values.

Dynamic Spectral Imaging System

Overall, the results of the five DySIS studies included in this section confirm the results of the meta-analysis. Adjunctive DySIS improves sensitivity for detecting CIN 2+ compared with colposcopy alone, although this is associated with a reduction in specificity. (Confidential information has been removed.)

Louwers et al. 61 was a secondary analysis of Louwers et al. ,60 which aimed to reanalyse the performance of dynamic spectral imaging (DSI) and conventional colposcopy to determine the difference between low-grade cytology [borderline or mild dyskaryosis (BMD)] referrals and high-grade cytology referrals. The study also aimed to reanalyse the performance of DSI and conventional colposcopy by retrospectively assigning them to two referral strategies based on their initial cytology and hrHPV test results: (1) hrHPV testing as the primary screening test and cytology as triage (including patients with a positive hrHPV test and BMD or high-grade cytology), or (2) reflex hrHPV testing in patients with BMD cytology (including patients with BMD cytology and a hrHPV positive test or high-grade cytology, irrespective of the hrHPV test result). Compared with standard colposcopy, the sensitivity of adjunctive DySIS was higher and the specificity was lower in both referral strategies. Diagnostic accuracy estimates were similar between HPV primary and cytology primary referral strategies for adjunctive DySIS (sensitivity 81% vs. 80%, respectively; specificity 64% vs. 61%, respectively) and for colposcopy alone (sensitivity 53% vs. 54%, respectively; specificity 82% vs. 78%, respectively).

Natsis et al. 74 estimated the accuracy of adjunctive DySIS in a population of 287 hrHPV-positive patients with low-grade cytology. Initial colposcopy impression and potential biopsy sites were recorded before and after the appearance of the DySISmap. Colposcopy alone had low sensitivity (27%) but high specificity (91%) for detecting CIN 2+. Incorporation of DySISmap improved sensitivity (82%) but reduced specificity (36%).

Salter and Livingston80 reported some preliminary data from two colposcopy clinics as part of a large cohort of US community-based colposcopy clinics using adjunctive DySIS. Consistent with other studies, the addition of DySISmap increased sensitivity compared with colposcopy alone (83.9% with adjunctive DySIS compared with 61.3% for colposcopy alone), but was associated with a reduction in specificity (75.4% vs. 91.1%, respectively).

Tsetsa et al. 111 carried out a single-centre prospective diagnostic cohort study that assessed the performance of adjunctive DySIS in three different concentrations of acetic acid solution (3%, 4% and 5%). The study was reported only in conference abstracts and enrolled 57 patients with abnormal cytology, of whom 54 were analysed. Each patient was examined with DySIS colposcope and DySISmap in three successive examinations. Biopsies were collected from sites corresponding to the most atypical indications of the coloured map and sent for histology. The diagnostic performance of adjunctive DySIS was higher in examinations that used 3% acetic acid (sensitivity 86%, specificity 81%) than in those using 4% acetic acid (sensitivity 79%, specificity 77%) and 5% acetic acid (sensitivity 82%, specificity 77%); however, the study was small and it is not clear if these differences were statistically significant. The authors noted that morphological characteristics, such as mosaic pattern and atypical vessels, were better highlighted when 5% acetic acid was used.

ZedScan

(Confidential information has been removed.)

Macdonald et al. 95 evaluated the diagnostic accuracy of ZedScan in patients infected with known hrHPV and compared its performance between those with HPV 16 and those with other hrHPV genotypes. The study included 839 participants, of which 607 (72%) had abnormal cytology and were included in Tidy et al. 94 An additional 226 patients were included, of which most (82%) had a persistent HPV test and cytology-negative result. The sensitivity of adjunctive ZedScan was high (100%) regardless of the hrHPV genotype. The sensitivity of colposcopy alone was also high, and slightly higher in patients with the HPV 16 genotype (86.9%) than in those with other high-risk genotypes (79.7%), although this difference may not be significant as a result of overlapping CIs. Specificity estimates were not reported in this study.

Human papillomavirus primary screening

(Confidential information has been removed.)

Test failures

Appendix 7, Table 48, presents the rates of, and the reasons for, test failures. Reported rates of test failure varied widely across the studies. One study88 of a prototype version of DySIS reported a high rate of failure (31.4%), which was primarily attributable to unsatisfactory DySISmap image and issues with faulty disposable nozzles through which the acetic acid was delivered. Studies of a more recent version of DySIS reported lower failures rates, ranging from 2.9% to 16.7%, with a lack of/poor-quality imaging being the most common reason. The failure rates in the ZedScan and ZedScan prototype studies were (confidential information has been removed) and 13.4%, respectively. (Confidential information has been removed.)

Diagnostic and treatment biopsies

All included diagnostic accuracy studies reported some data on the diagnostic and treatment biopsies performed. However, owing to limited data, the impact of adjunctive technology on the rates of diagnostic biopsies and treatment (including unnecessary treatments) is uncertain.

Appendix 7, Table 49 presents the number of diagnostic biopsies and treatment biopsies performed in the diagnostic accuracy studies.

Three studies performed biopsies in all patients, as reported previously. 60,79,88 In other studies, the proportion of adjunctive colposcopy patients biopsied ranged from (confidential information has been removed). 80,94 The mean number of biopsies varied widely,60,83,88,103 (confidential information has been removed) to up to five in Roensbo et al. 79 when reported. Only four studies reported on the number of treatments performed during or after the examination, which were nearly always conducted as loop excisions. 60,79,94,103 See-and-treat cases ranged from 0 in Roensbo et al. 79 to (confidential information has been removed).

Only three studies provided data on the number of additional biopsies performed that were associated with the use of adjunctive colposcopy. However, these results are limited, as a result of the lack of randomised evidence comparing adjunctive colposcopy with standard colposcopy in parallel groups. (Confidential information has been removed.) Papagiannakis et al. 83 found that the addition of DySIS led to an increase of approximately one additional biopsy per five patients (from 1.03 biopsies per patient with standard colposcopy to 1.25 biopsies with adjunctive colposcopy), although this result is derived from a non-randomised comparison with a retrospective arm. Natsis et al. 74 reported that the proportion of patients undergoing biopsies was lower among patients undergoing colposcopy with adjunctive DySIS (80.8%) than in a parallel group of patients undergoing colposcopy alone (85.9%). (Confidential information has been removed.)

Kyrgiou et al. (2015)118

Kyrgiou et al. 118 assessed the effect of excisional or ablative CIN treatment on fertility and early pregnancy (< 24 weeks’ gestation)-related outcomes. The review included studies that compared fertility and early pregnancy (< 24 weeks’ gestation)-related outcomes in women with a history of CIN treatment with the same outcomes in women who had not received treatment. Any types of excisional and ablative treatments were included.

The review included 15 studies (2,223,592 participants, of whom 25,008 received treatment for CIN). All included studies were non-randomised and heterogeneity was high. The quality of the evidence for early pregnancy-related outcomes was low [Grading of Recommendations Assessment, Development and Evaluation (GRADE) classification]. The results of the meta-analysis showed that CIN treatment did not have an adverse effect on fertility outcomes. The overall pregnancy rate was higher for treated women than for those who were untreated [43% and 38%, respectively; relative risk (RR) 1.29, 95% CI 1.02 to 1.64; four studies]. There was no statistically significant difference in pregnancy rates in treated and untreated women with an intention to conceive and women taking more than 12 months to conceive. There was no increase in the total miscarriage rate or the first-trimester miscarriage rate between the treated and untreated groups; however, CIN treatment was associated with an increased risk of miscarriage in the second trimester (1.6% vs. 0.4%, RR 2.60, 95% CI 1.45 to 4.67; eight studies). Ectopic pregnancies (1.6% vs. 0.8%, RR 1.89, 95% CI 1.50 to 2.39; six studies) and terminations (12.2% vs. 7.4%, RR 1.71, 95% CI 1.31 to 2.22; seven studies) were also more frequent in treated women. The authors concluded that CIN treatment is unlikely to have an adverse effect on fertility and there was a slight increase in the risk of miscarriage in the second trimester associated with treatment; there was no stratification of risk by treatment type or magnitude. The authors also further emphasised the low quality of the evidence, but suggested that women should be advised that fertility is not compromised by treatment for CIN. The conclusions of this generally well-conducted review are likely to be reliable.

Kyrgiou et al. (2016)119

Kyrgiou et al. 119 focused on studies reporting obstetric (> 24 weeks’ gestation) and neonatal outcomes following local treatment for CIN or early-stage cervical cancer. The review included studies reporting on obstetric outcomes (> 24 weeks’ gestation) in women who had previously received local treatment for CIN or early-stage invasive cervical cancer compared with women with no history of treatment. All types of excisional and ablative treatments were included.

Outcomes pertaining to the risk of overall preterm birth were reported by 60 studies, with 25 reporting on ‘severe’ preterm birth (< 32–34 weeks’ gestation) and nine reporting on ‘extreme’ prematurity (< 28–30 weeks’ gestation). Other outcomes included the length of labour [precipitous or prolonged, use of analgesia (epidural, pethidine, other)], oxytocin use, cervical stenosis and haemorrhage. Neonatal outcomes were low birthweight, admission to neonatal intensive care, stillbirth, Apgar scores and perinatal mortality.

The review included 71 studies (6,338,982 participants, with 65,082 participants treated for CIN or early-stage invasive cancer). Nearly all of the studies were retrospective cohort studies and none was a randomised trial. Most studies were considered to be high-quality observational studies (Newcastle–Ottawa Scale scores 8–10 in all but two studies). For all treatment types, the risk of overall prematurity (< 37 weeks’ gestation) was increased in the treated group versus the untreated group (RR 1.78, 95% CI 1.60 to 1.98; 60 studies), as was severe prematurity (< 32–24 weeks’ gestation, RR 2.40, 95% CI 1.92 to 2.99; 25 studies) and extreme prematurity (< 28–30 weeks’ gestation, RR 2.54, 95% CI 1.77 to 3.63; nine studies). Compared with untreated women, all patients who were treated with LLETZ were at a higher risk of giving birth prematurely (RR 1.56, 95% CI 1.36 to 1.79; 26 studies), severely prematurely (RR 2.13, 95% CI 1.66 to 2.75; 11 studies) and extremely prematurely (RR 2.57, 95% CI 1.97 to 3.35; three studies). The meta-analysis showed that treatment for CIN increased the risk of preterm birth regardless of treatment method, with a higher magnitude of treatment effect associated with treatment techniques removing or ablating more tissue; deeper excisions were consistently associated with increased risk of preterm birth (≤ 10–12 mm: RR 1.54, 95% CI 1.09 to 2.18 mm; ≥ 10–12 mm: RR 1.93 mm, 95% CI 1.62 to 2.31 mm; ≥ 15–17 mm: RR 2.77 mm, 95% CI 1.95 to 3.93 mm; ≥ 20 mm: RR 4.91 mm, 95% CI 2.06 to 11.68 mm). There was also an association between the number of procedures undergone and the risk of preterm birth; pregnant women who had underone more than one treatment were significantly more likely to deliver prematurely (RR 3.78, 95% CI 2.65 to 5.39).

The risk of other adverse outcomes, such as spontaneous preterm birth, premature rupture of the membranes, chorioamnionitis, low birthweight, admission to neonatal intensive care and perinatal mortality, was also significantly increased after treatment. The authors concluded that any local treatment for preinvasive or early-stage invasive disease will increase the risk of preterm birth in a subsequent pregnancy and that the frequency and severity of adverse outcomes increases with greater cone depth and is higher for excision than for ablation. However, the authors noted that the risks associated with small excisions are likely to be smaller than those related to untreated CIN during pregnancy, which is itself linked to preterm birth (RR 1.24, 95% CI 1.14 to 1.35). The authors rightly noted the need to interpret the review results with caution, owing to the lack of prospective and randomised evidence, a high risk of confounding in the included studies and significant heterogeneity. Given this, the conclusions of this generally well-conducted review are likely to be reliable. However, as most of the studies included in these reviews were not recent UK cohort studies, the applicability of the conclusions of these reviews to the NHS context may be limited. 120–122

Acceptability to patients and patient satisfaction

Acceptability of the adjunctive technologies to clinicians

Training requirements

One study, which was conducted in a single centre in Sheffield that involved in-house colposcopists and was linked to Tidy et al. ,103 reported the time needed to train the colposcopists in using ZedScan for the first time. 96 The study reported that 5–10 minutes of additional time were needed for the initial training period and the colposcopists were able to complete the initial 10–20 ZedScan measurements within 2–3 minutes after examining 10–20 patients. No further details were reported.

The authors concluded that ZedScan as an adjunct to colposcopy has added minimal time to each appointment and exerted a negligible impact on the clinical output.

Other outcomes

No evidence was found for the following outcomes: successful database and record management, the capacity to perform colposcopies, and uptake and compliance.

Decision problem/objective

A decision-analytic model was developed to assess the cost-effectiveness of adjunctive colposcopy technologies for assessing suspected cervical abnormalities in women referred for colposcopy as part of the NHSCSP under the HPV triage screening protocol.

Three technologies were initially considered: DySIS, LuViva Advanced Cervical Scan and Niris Imaging System. Owing to the lack of reliable data, LuViva Advanced Cervical Scan and Niris Imaging System were subsequently excluded from the base-case analysis.

The model evaluated costs from the perspective of the NHS and Personal Social Services, expressed in Great British Pounds (GBP) at a 2011 price base. The outcomes in the model were expressed in terms of quality-adjusted life-years (QALYs). The model employed a lifetime (50-year) horizon, and costs and outcomes were discounted at a rate of 3.5% per annum.

Strategies/comparators

The base-case economic evaluation compared the costs and health outcomes of DySIS, alone and as a colposopic adjunct, with standard colposcopy. The base-case results were presented by reason for referral (borderline and HPV-positive, mild and HPV-positive, moderate, severe, possible invasion, possible neoplasia, three times inadequate) and, for the whole population, were based on a weighted average of the results of each reason for referral. A separate indicative analysis was undertaken to test the sensitivity needed for Niris and LuViva to be considered to be cost-effective, given their reported costs and an assumed specificity.

Model structure

The model incorporated two elements: first, a decision tree to represent the initial diagnostic and treatment pathways for patients referred to colposcopy from the NHSCSP (under the HPV triage screening protocol), and, second, a Markov model that simulated the natural history of patients and captured future cytological screening and referrals to capture the long-term costs and outcomes of the initial diagnostic and treatment pathways.

The decision tree was specifically developed for the appraisal informing NICE DG4,31 whereas the Markov model was based on a revised model previously used by Hadwin et al. ,125 and was referred to as the ‘Sheffield model’. The decision tree had three main components: (1) diagnostic outcome, (2) treatment decision and (3) treatment outcome.

The diagnostic outcome depends on the diagnostic accuracy in relation to a patient’s true underlying health state. Specifically, diagnostic accuracy is modelled as the probability of being diagnosed with health state h, conditional on the true underlying health state h.

Treatment decisions depend on diagnostic outcome and the reason for referral. A patient may either not receive treatment or be referred for a diagnostic biopsy, a treatment biopsy (LLETZ) or a cancer treatment.

Treatment outcome has an impact on both the true underlying health state and subsequent screening. The impact on a patient’s health state depends on the probability of being cured. If patients have not been treated, they enter the natural history Markov model with their initial true underlying health state. In the case of patients with precancerous lesions who have been treated with excision biopsy, there is a probability that this will be cured. If treatment has been successful, patients enter the natural history model Markov in the ‘clear’ state; if not, they enter the natural history model with their initial health state.

As previously stated, a state-transition (Markov) cohort model is used to capture the long-term costs and outcomes of the initial diagnostic and treatment pathways by simulating the natural history of the patient and incorporating future cytological screening and referrals. Although the use of a cohort model was appropriate for the specific decision problem, the model itself subsequently required hundreds of separate states to be included in order to capture the complexity of screening pathways and the heterogeneity in treatment decisions and outcomes. This complexity arose because the screening pathways and treatment decisions depended on a patient’s characteristics (e.g. age, health state) and previous history (e.g. screening results, treatment outcomes, follow-up), and these also had an impact on the transitions between health states in the natural history model. With a cohort approach, the only way to account for individual heterogeneity and to build sufficient ‘memory’ to capture the complexities is to increase the number of states. Although the complexity could have been potentially more efficiently reflected by using a patient-level simulation, the final model structure appeared to rely heavily on the use of an existing natural history cohort model from Sheffield, which may have constrained the authors in terms of their chosen modelling approach.

The treatment pathways incorporated in the model were based on NHSCSP guidelines,15 which describe good practice in treatment decisions and the follow-up of precancerous lesions and invasive cancer. However, the authors identified an important discrepancy between guidelines and clinical practice and chose to use evidence from the latter based on data from Gateshead to capture heterogeneity in treatment pathways. For instance, national guidelines state that a patient with low-grade referral with normal colposcopy should be discharged without any treatment and returned to routine screening. Observational data from the Gateshead clinic suggested that 73.5% of these patients would receive a diagnostic biopsy and only 10.7% would be sent back to routine screening.

Adjunctive technologies to colposcopy were assumed to be used for the initial referral and for subsequent colposcopy appointments (treatment, follow-up). The model also assumed that there is no loss to follow-up and, hence, all women were assumed to attend subsequent appointments for colposcopy and cytology.

Main sources of data

The model incorporated three main sources of data input: (1) diagnostic accuracy, (2) natural history of cervical cancer and (3) characteristics of the population referred for colposcopy.

Resource use and costs

The resource use and cost estimates included the acquisition costs of the alternative technologies (including maintenance and use of disposables) and further treatments. The number of patients managed under a single colposcopy device was provided by clinical advisors and was used to calculate the average cost per procedure for each technology. To capture the additional costs of a colposcopy visit (e.g. diagnostic and treatment biopsy), treatment costs from the TOMBOLA study130 were used. Cancer costs by stage were derived from a UK-based study by Wolstenholme et al. 131

The cost per patient of colposcopy and adjunctive technologies has two main components: (1) a cost per patient, common to all devices, which includes costs related to the examination itself, such as facilities and staff, and (2) an additional cost that depends on the device and includes the cost of acquisition, maintenance and disposables. The UK-based TOMBOLA study was used to estimate the cost of colposcopy alone (including the price of the device and maintenance) as well as the per-patient costs of biopsy and excision treatment (LLETZ). 130 The examination with DySIS was assumed to be equivalent to colposcopy alone in terms of staff resources (i.e. the same length of consultation and negligible staff training). The additional cost includes the acquisition cost, annual maintenance costs and disposables, all provided by the manufacturer. It was assumed that clinics have the choice between investing in a binocular colposcope or the DySIS device (colposcope plus digital map). Therefore, the purchase price of a binocular colposcope was included in the cost of colposcopy alone, but not for DySIS. Another important assumption is the number of patients examined per year per colposcope. This is critical when estimating the cost per patient, especially if technologies differ in terms of the initial investment and throughput. An estimate of 1229 patients per year was assumed based on clinical advice. This was assumed to be exogeneous to the performance of the device.

The management costs of cervical cancer by stage at the initial diagnosis were derived from Wolstenholme et al. 131 The study was based on an audit of resources and costs over 5 years on 261 women in the Trent region of central England in 1990. Although these costs appear to be based on the best evidence available at the time, there exists uncertainty regarding how representative these estimates are, given the historic nature of the study and the need to inflate costs over a significant time period.

Quality of life/utilities

Within the model, colposcopy and DySIS were assumed to have an impact on health outcomes in two ways: (1) through the disutility associated with the examination itself and with subsequent treatment and tests, and (2) through the disutility associated with the development of invasive cancer.

The direct disutility associated with colposcopy alone and DySIS was not assumed to be different. However, because diagnostic accuracy has an impact on the number of subsequent treatments and follow-up, QALY decrements associated with colposcopy examination, cytology test and biopsy were important for the assessment of cost-effectiveness.

The disutility associated with cytology examination was assumed to be 0.02 (Insinga et al. 132) over 1 month (i.e. –0.0016 per cytology exam). The disutility of colposcopy and biopsies was derived from a time trade-off analysis by Birch et al. 133 The study reported HRQoL for three different scenarios: ‘three repeat Pap Smears’ (0.958), ‘immediate colposcopy with no pathology’ (0.927) and ‘cone biopsy after immediate colposcopy’ (0.922). The difference of 0.031 between Pap smears and immediate colposcopy was used as a QALY decrement applied for each colposcopy examination. The difference of 0.005 between colposcopy with no pathology and cone biopsy was used as a QALY decrement for diagnostic and treatment biopsies. As noted by the authors, this rather small decrement, which was similar for diagnostic and treatment biopsies, is a strong assumption and may underestimate the impact of biopsies on health. As noted in Chapter 3, Systematic reviews of adverse outcomes of cervical intraepithelial neoplasia treatment, recent evidence exists that indicates that treatment biopsies may increase the risk of adverse obstetric outcomes, and specifically preterm delivery. However, the evidence at the time at which the model was conducted was reported to be scarce and inconclusive. Given this uncertainty, the QALY decrement of a treatment biopsy on health outcomes was further explored in a separate scenario analysis. The HPV, CIN 1 and CIN 2/3 states were assumed to be asymptomatic and hence were assigned the same utility as the clear state, reported as 0.91.

Health utilities associated with the four stages of invasive cancer were derived from Chuck et al. 134 It was assumed that invasive cancer would be detected only at an asymptomatic stage through routine screening. Patients who subsequently developed symptoms were assumed to be immediately referred for colposcopy and were appropriately diagnosed and treated. The model thus made an important distinction in the health utilities between undiagnosed and diagnosed cancer. Specifically, the model incorporated much lower HRQoL scores for untreated (and therefore undiagnosed) cancer than for treated cancer. These estimates were reported to be based on estimates reported by Chuck et al. 134

In our review, we identified some important uncertainties surrounding the source and subsequent application of the estimates within the model by Wade et al. 30 The estimates reported by Chuck et al. 134 are actually referenced to a separate study by Goldie et al. 5 Goldie et al. 5 distinguished ‘quality weights’ for detected invasive cancer, by stage and quality weights after treatment for invasive cancer. The use of the estimates from Goldie et al. 5 and their application in the model by Wade et al. 30 raises several potential issues. First, it is unclear whether or not ‘detected invasive cancer’ as described by Goldie et al. 5 can be interpreted as cancer ‘without treatment’. This appears to be inconsistent with the fact that undiagnosed cancer is assumed to be asymptomatic. Our interpretation of the estimates reported by Goldie et al. 5 is that the lower weights of detected cancer might more appropriately indicate that disutility is higher at the time of diagnosis and initial treatment than ‘after treatment’. 5 The ‘detected invasive cancer’ stage might therefore be considered to capture the disutility associated with the initial period after diagnosis. This would appear to be more consistent with the initial treatment burden that patients treated for cervical cancer may face after being diagnosed, but who may subsequently recover relatively quickly, particularly when cancer is detected at an early stage. Second, the method to estimate these ‘quality weights’ reported by Goldie et al. 5 cannot be identified and hence the extent to which these appropriately represent health utilities is unclear. 5 Finally, there appears to be a reporting error in the Chuck et al. 134 study, and this was subsequently reproduced in the model by Wade et al. :30 the score associated with stage 2 cancer ‘without treatment’ is reported as 0.67 (higher than stage 1) in Chuck et al. 134 and 0.56 in Goldie et al. 5

Main results (including of the base-case and key sensitivity analyses)

The results of the economic evaluation compared DySIS plus colposcopy with colposcopy alone, for each reason for referral and for the whole population. In most instances, colposcopy alone was found to be dominated by DySIS plus colposcopy. That is, colposcopy alone had higher costs and lower health outcomes than DySIS plus colposcopy (Table 16).

We identified the potentially counterintuitive result that an increase in a diagnostic device’s specificity resulted in worse outcomes. In particular, it appeared to be better to falsely identify patients with CIN 1 as having CIN 2/3 than to find that they truly have CIN 1. According to the authors, this result suggested that the treatment of CIN 1 is more cost-effective than watchful waiting, because of the low cost and low QALY decrements associated with treatment biopsy. Further analyses subsequently determined the threshold of each input at which an increase in specificity would improve outcomes. The QALY decrement of treatment biopsy needed to be increased from 0.005 to 0.13 or its cost needed to be increased from £97 to £2758.

Assessment of uncertainty

Several scenario analyses were considered, with the following variations in the base-case assumptions: the patient’s age, duration of the HRQoL decrements as a result of cancer, cancer treatment costs, perfect health for the clear and HPV states, the QALY decrement associated with treatment biopsies and cytological screening, the alternative costs of a colposcope, alternative treatment probabilities, and the assumption that patients who received negative test results through the use of colposcopy or adjunct technologies would be diagnosed as ‘clear’. Overall, for all of the sensitivity analyses undertaken, colposcopy alone was dominated by DySIS plus colposcopy.

A secondary analysis considered a higher cost and QALY decrement associated with treatment biopsy (–0.13 from 0.005 in the base case). At this value, a higher level of specificity is expected to generate improved outcomes. The secondary analysis was also combined with the sensitivity analyses previously described. The results suggested that colposcopy alone was still dominated by DySIS plus colposcopy (Table 17).

A probabilistic sensitivity analysis (PSA) was not conducted and all of the results were based on deterministic estimates.

Decision problem/objective

The aim of Whyte et al. 124 was to assess the cost-effectiveness of a prototype version of ZedScan as an adjunct to colposcopy relative to standard colposcopy alone from a NHS perspective. The population included in the analysis was women referred to colposcopy from the NHSCSP (under the HPV triage screening protocol).

The analysis focused on assessing the impact of ZedScan on initial colposcopy appointments and colposcopy follow-up appointments. Outcomes were predicted for three different types of colposcopy clinic [‘see-and-treat’, ‘treat-later’ (watchful waiting) and ‘triage’ clinics], which allowed for the optimal screening pathway to be established, as well as the cost-effectiveness of ZedScan in each type of clinic. The different types of clinic were formally defined as follows:

1. See-and-treat clinic – women may receive treatment at their initial colposcopy appointment if the diagnosis suggests that this is appropriate.

2. Treat-later clinic – no treatment at the initial colposcopy appointment. Biopsy confirmation before treatment at a later colposcopy appointment.

3. Triage clinic – high-grade referral patients are seen in a see-and-treat clinic, so women may receive diagnosis and treatment at the same colposcopy appointment. Low-grade referral patients are seen in a treat-later clinic.

The analysis used a diagnostic threshold for ZedScan, which set the sensitivity at a similar level to that of standard colposcopy, resulting in the main difference between the two devices being their specificity rates. The analysis assessed the cost-effectiveness of ZedScan by assessing several outcomes, which included the number of colposcopy appointments, the number of biopsies conducted, the costs related to colposcopy and adverse events, and overtreatment rates. In addition, the analysis also considered the treatment rates of CIN and associated adverse event rates, as well as whether or not an increased sensitivity level with ZedScan might increase clinicians’ confidence to offer more ‘see-and-treat’ appointments as a result of reductions in the numbers of unnecessary treatments and the associated cost-effectiveness of such changes.

The time horizon of the evaluation was 3 years and was justified on the basis that most patients would be returned to routine screening within this time period. Costs and outcomes were discounted at a rate of 3.5% and a price year was not stated.

Strategies/comparators

Using ZedScan as an adjunct to colposcopy was compared with colposcopy alone. No attempt was made to compare ZedScan with alternative diagnostic tools that can be used as an adjunct to colposcopy.

Model structure

The economic evaluation modelled the natural history of women at risk of developing cervical cancer and the colposcopy diagnostic pathway. A patient-level model was constructed, allowing the model to track the underlying health state of patients and their location within the colposcopy pathway. After an initial colposcopy in the first cycle, patients were located in one of the following discrete locations within the colposcopy pathway: ‘low-grade CIN 1 follow-up’, ‘high-grade follow-up pathway’, ‘Test of cure’, ‘routine screening’, ‘refer to colposcopy following fail at test of cure’, and ‘cancer’. Following the diagnostic pathway, the model tracked patients’ natural history, with women able to progress or regress through the CIN stages. The discrete health states included in the analysis were ‘clear’, ‘HPV’, ‘CIN 1’, ‘CIN 2’, ‘CIN 3’ and ‘cancer’, with cancer divided into International Federation of Gynecology and Obstetrics (FIGO) stages of severity from I to IV. A cycle length of 6 months was selected, as this is the shortest length of time between repeat colposcopies.

The model estimated the number of colposcopy appointments and biopsies conducted at each of the clinic types, as well as the number of women who received treatment for CIN for each type of colposcopy. This allowed for the estimation of costs associated with each scenario, as well as the average utility decrements associated with colposcopy, biopsy and treatment. The model simulated 1,000,000 women for each diagnostic tool and each clinic type.

The distribution of the colposcopy population on the basis of the women’s health states and cytology grades was estimated using data on the outcome of colposcopy given the referral cytology. After patients received their initial cytology examination, there were assumed to be seven possible outcomes. If the result was negative, then patients were returned to routine screening, a mild or borderline outcome results in a HPV test being conducted, with a positive HPV test result leading to a low-grade referral to colposcopy, and a negative result leading to patients being returned to routine screening. If the cytology result was ‘moderate’, ‘severe’ or ‘invasive cancer’, then patients were sent to colposcopy as a high-grade referral. If the outcome of cytology was inadequate, then patients were sent for a repeat cytology test in 3 months’ time, with three consecutive inadequate results leading to a low-grade colposcopy referral. Patients’ referral type was tracked, as it was assumed that ZedScan would be used with a different threshold for low-grade and high-grade referrals, resulting in different sensitivities and specificities.

Whether or not patients received treatment and the timing of treatment was dependent on their referral grade, their colposcopy result and the clinic type. The three types of clinics that were considered in the analysis were ‘see-and-treat’, ‘treat-later’ and ‘triage’ clinics. At the ‘see-and-treat’ clinics, women could receive treatment at their initial colposcopy appointment if the diagnosis was clear that this was appropriate. ‘Treat-later’ clinics do not provide treatment at the initial colposcopy appointment; instead, patients receive a biopsy confirmation and are treated at a future appointment. ‘Triage clinics’ split patients by their referral grade, with patients with high-grade referrals being managed in a ‘see-and-treat’ clinic and those with low-grade referrals being managed in a ‘treat-later’ clinic. Women treated for CIN are assumed to be treated with LLETZ, although in practice a range of treatment options are available.

After colposcopy, patients follow one of six pathways. First, women with a low-grade referral who are identified as clear are sent back to routine screening in 3 years’ time. Second, patients who are identified with cancer are sent to oncology regardless of their referral grade. The third pathway results in women with a low-grade referral identified as having CIN 1 receiving a further colposcopy in 12 months’ time. If they are found to have CIN 1 at the follow-up colposcopy then they are sent for a further colposcopy 1 year later, but after a third CIN 1 result, they are treated. At the second or third colposcopy those who are identified as clear return to routine screening, women found to have CIN 2+ are sent for treatment and those identified with cancer are referred to oncology. The fourth pathway involves women with a high-grade referral who are identified as clear or having CIN 1 being referred to colposcopy in 6 months’ time, with a subsequent result of CIN 1+ resulting in treatment and a result of cancer leading to an oncology referral. Those who are identified as clear in the repeat colposcopy are sent back for a colposcopy in a further 6 months, with five consecutive clear colposcopy results resulting in patients being sent back to routine screening. The fifth pathway results in patients who are treated for CIN going on to receive a test of cure, which consists of a cytology test at 6 months and a subsequent HPV test if the cytology result is negative. The final pathway involves referring patients who receive a test of cure and have a positive cytology or a positive HPV test then patients to colposcopy.

Following completion of the screening phase, patients enter a natural history model. This model assumes that patients start with a high risk of HPV infection, which can either clear or progress to CIN, with patients starting with mild changes (CIN 1) and either progressing to more severe stages (CIN 2 and CIN 3) or regressing back to clear. If patients progress to cancer, then it is assumed that they can regress back to a non-cancer state with or without treatment. Progression and regression rates of CIN 2/3 to and from cancer were assumed to be similar to the rates from CIN 1 to CIN 2/3 and clear.

Owing to the analysis focusing on a time horizon of only 3 years, the natural history model is unable to capture the long-term effects of failing to identify and treat CIN caused by ZedScan having a lower sensitivity than colposcopy for low-grade referrals. In the short term, failing to treat those with CIN will reduce costs; however, there will be longer-term treatment cost, HRQoL and mortality impacts that are not captured within the time horizon.

Main sources of data

The methods used to identify appropriate parameter estimates were not transparently reported. The study makes reference to a review that was conducted in order to identify studies with information useful for understanding the natural history of HPV. However, the details of the search were not reported. No review was explicitly reported for other parameters.

Resource use and costs

The costs of colposcopy, biopsy and treatment were estimated using data from Sheffield Teaching Hospitals, and, therefore, it is unclear whether or not they are generalisable to the wider UK population. These costs were calculated using the price of components and staff time, with a fixed cost included for colposcopy.

The cost per use of the ZedScan device was assumed to be £31, which was stated to include the disposable tip, cost of device, training and maintenance. However, no calculations were reported to demonstrate how this cost had been derived. The ZedScan device was assumed to be used for all diagnostic colposcopies, but not for treatment colposcopies following biopsy confirmation of disease.

The costs of HPV testing and cytology were taken from the HPV sentinel sites. Adverse event costs were included for women experiencing severe bleeding or discharge (the TOMBOLA Group). 144

Quality of life/utilities

Quality-of-life decrements were included for bleeding, pain and discharge, with data on adverse event frequency and the duration of each event taken from the TOMBOLA trial, a large, multicentre, UK-based study. 144 Utility decrements were also included to capture patients’ preferences for the follow-up they receive after an abnormal cytology result, which were estimated for colposcopy, colposcopy with biopsy and colposcopy with LLETZ using the time trade-off method. 133 The potential cost and quality-of-life impact of the increased risk of preterm birth as a result of treatment or biopsy was not included in the model. This was justified based on conflicting results from the identified studies.

Main results (including base-case and key sensitivity analyses)

The base-case results reported a lower frequency of biopsy, as well as lower total costs and a lower cost per woman with CIN 2/3 treated for those diagnosed using ZedScan than for those treated with standard colposcopy for each clinic type. ZedScan was also reported to lower the rates of overtreatment, with 12% of the level of overtreatment occurring in ‘see-and-treat’ clinics, and 17% of the level of overtreatment occurring in ‘triage-by-cytology-result’ clinics compared with the level of overtreatment in women who received standard colposcopy.

The lower costs reported for ZedScan were also reported to be attributable to a reduction in the number of follow-up appointments for CIN 1. The authors concluded that using the disease present method resulted in fewer biopsies being taken following ZedScan and more women being followed up with a cytology test rather than a repeat colposcopy appointment. Importantly, the lower sensitivity of ZedScan for low-grade referrals also resulted in a lower number of women with CIN 2+ being diagnosed using ZedScan, as well as a minor reduction in the number of cancers identified at colposcopy across all clinic types. Hence, some of the reduction in total costs for ZedScan was as a result of undertreatment for CIN 2+.

The base-case results also found that ‘treat-later’ clinics had the lowest cost per woman treated for CIN 2/3, for both ZedScan and standard colposcopy. Based on their findings, the authors concluded that the results did not appear to support a move from a ‘treat-later’ clinic to a ‘see-and-treat’ clinic using ZedScan. This was because the cost per woman treated for CIN 2/3 was lower for a ‘treat-later’ clinic using standard colposcopy than for a ‘see-and-treat’ clinic using ZedScan.

The authors noted the limitations arising from restricting the scope of the model to initial colposcopy and follow-up for up to 3 years. Specifically, the long-term costs and consequences of failing to identify and treat CIN 2+ or the impact of identifying and treating CIN or cancer at an earlier stage were not captured. The authors noted that care should thus be taken regarding the interpretation of the results, as a reduction in the treatment of CIN 2+ or cancer will appear to be beneficial in the short time horizon of the model (reducing treatment costs), but would actually lead to higher costs and lower outcomes over a longer time horizon.

From a policy perspective, these limitations mean that it is not possible to determine whether or not the short-term benefits of ZedScan arising from a reduction in unnecessary treatments offset any potential reduction in benefits over a longer time horizon and, hence, whether over a more appropriate horizon in which this could be reflected ZedScan would be cost-effective or not.

Assessment of uncertainty

A range of sensitivity analyses were conducted in order to assess the uncertainty around particular parameters. For underlying disease prevalence, the proportion of CIN 2+ patients was increased by 10%, and for costs, the values used in a previous diagnostics assessment report (Wade et al. 30) were used as an alternative (with the authors noting an important disparity in the cost of biopsy). In addition, the threshold at which a biopsy was taken was lowered for ZedScan and colposcopy. Different values for the frequency of biopsy for patients who were diagnosed using ZedScan were also used: a low estimate that assumed that only one biopsy was required based on clinical guidance and a high estimate that assumed that clinicians could take additional biopsies. An analysis was also conducted in which ZedScan was assumed to have the same specificity as colposcopy, and therefore the only difference between the devices was their sensitivity.

A PSA was also conducted to assess the joint uncertainty of the parameter inputs. Distributions were assigned to the test characteristic parameters, the initial distribution of health states, utility decrements and transition probabilities. However, it appears that an arbitrary estimate of uncertainty (± 5% the base-case value) was used to inform the distributions for both the natural history transition probabilities and the utility decrements. No details were reported in terms of the number of simulations undertaken within the PSA.

The results of the sensitivity analyses demonstrated that the findings appeared to be robust to a range of alternative assumptions. However, the findings were reported to be particularly sensitive to assumptions surrounding the costs of colposcopy. When the estimates reported in Wade et al. 30 were applied to the model, standard colposcopy appeared to be cheaper than ZedScan, in terms of both total costs and the cost per woman with CIN 2/3 treated. The authors highlighted that the costs reported in Wade et al. 30 for both cervical biopsy and LLETZ appeared to be markedly lower than those based on the estimates derived from the Sheffield Teaching Hospitals.

The sensitivity analysis also found that setting the threshold in a way that resulted in the specificity of ZedScan being equal to that of colposcopy also increased the sensitivity and reduced the specificity of ZedScan. This resulted in an increase in the cost of using ZedScan and the number of patients treated, which reduced the benefits of the device in terms of cost per woman with CIN 2/3 treated. However, ZedScan was still reported to be cheaper than using colposcopy alone.

Using the higher estimate of the number of biopsies taken when patients were diagnosed using ZedScan had a small impact on the cost per woman with CIN 2/3 treated, but not enough to make ZedScan the more expensive option. Adjusting the disease prevalence data and the length of colposcopy appointments had a minimal impact on the results. The PSA results were reported to be comparable to the deterministic outcomes, demonstrating linearity in the parameter values.

Patient-level Monte Carlo simulation

The model simulates individuals’ experiences with a Monte Carlo simulation and can be formally defined as a state-transition, patient-level Monte Carlo simulation. A patient-level model estimates the mean costs and benefits for a group of patients by considering the costs and benefits of each individual within the group. Each individual has specific characteristics that both affect and depend on the occurrence of events and associated transitions. In the present model, these characteristics are age, health state (clear, HPV, CIN 1, CIN 2/3, cancer), reason for referral to colposcopy (high-grade or low-grade dyskaryosis), next scheduled screening (routine call, 6-month cytology, 6-month colposcopy, test of cure, CIN 1 follow-up), time elapsed since last screening and type of clinic visited by the patient (see and treat or watchful waiting).

A cycle length of 6 months was selected, as this is the shortest time between repeat colposcopies. The cytology retesting (when inadequate) is the only event in the model that occurs after 3 months and can easily be taken into account in the 6-month probability of having a twice-inadequate cytology result. A 6-month cycle also avoids the computational burden of having to double the total number of cycles per individual simulation.

The decision-analytic model simulates for each patient the occurrence of uncertain events, such as disease progression, diagnostic results or treatment outcomes with a random walk (i.e. a series of uniform, pseudo-random numbers). A large number of simulations ensure that the proportion of patients in each state equals the individual probability. It is important to notice that a large number of simulations will appropriately characterise first-order uncertainty (i.e. the variability in the simulated experiences between patients), but not second-order uncertainty, which is linked to uncertainty around parameter values.

Implementation and schematics

The model is implemented with the TreeAge Pro 2016 software (TreeAge Software, Inc., Williamstown, MA, USA) and was run to simulate 500,000 women who were referred for an initial colposcopy appointment. The model has a decision tree structure, which represents events occurring sequentially as a pathway that is followed by individual patients. ‘Logic nodes’ (circles with an ‘L’ in the schematics) are used when the occurrence of the event is certain and depends on a patient’s characteristics only. ‘Chance nodes’ (an empty circle in the schematics) are used when the occurrence of the event is uncertain and depends on a probability. ‘Clones’ refer to subparts of the model that are common to different pathways (such as the natural history subpart) and therefore do not appear on the schematics for the sake of clarity. A triangle marks the end of a cycle; if the patient is still alive at the end of the cycle, they enter the model again; if they die, from cancer or other causes, they exit the model. The model structure is identical for the three strategies (colposcopy alone, DySIS and ZedScan), and only the input parameters vary (diagnostic accuracy and costs).

Natural history model

The natural history model (see Appendix 10, Figure 23) captures the progression of cervical cancer from the ‘clear’ state to stage 3 (distant) of invasive cancer. A patient enters the model with an initial health state and faces a 6-month transition probability to stay in the same state, progress or regress. The probability will determine the patient’s subsequent health state at the beginning of the next cycle.

The natural history model incorporates eight health states: clear, HPV infection without precancerous lesion, CIN 1, CIN 2/3, invasive cancer (local, regional and distant) and death. The structure of the natural history model is derived from Kulasingam et al. ,147 an update of Myers et al. ,128 which was used for the previous cost-effectiveness model by Wade et al. 30 Consequently, it relies on similar structural assumptions: HPV infection is a precondition to developing precancerous lesions, CIN 2 and CIN 3 are modelled as a single combined health state and patients are assumed to be cured from cancer if they survive for 5 years after treatment. The only update to the previous model is that HPV patients are allowed to develop CIN 2/3 within 6 months based on new evidence reported by Kulasingam et al. 147 on the possibility that women can progress in a short amount of time from being infected with HPV to developing CIN 2/3.

The modelling of cancer progression relies on several assumptions; most of these are similar to those in the previous model by Wade et al. :30 cancer cannot be cured without treatment and a patient progresses across stages until they receive treatment (i.e. while cancer remains undetected). Although Wade et al. 30 modelled cancer progression with four stages, our model considers only three stages (local, regional and distant). Indeed, the progression and mortality rates have been updated, based on a study by Campos et al. ,148 which relied on a three-stage model. Consistently with Wade et al. ,30 a patient with undetected cancer faces a 6-month probability to develop symptoms; the model assumes that a patient with symptoms is immediately referred for colposcopy, identified and treated. 30 When asymptomatic cancer is detected by the NHSCSP, it is assumed that the patient is in stage 1 of the model (local). A patient who survives for 5 years after treatment is assumed to be cured and is back to the ‘clear’ health state.

Screening and treatment pathways

At the beginning of each cycle, the patient follows one of four main screening and treatment pathways (which all end with the natural history model):

1. No screening – the patient directly enters the natural history model.

2. Colposcopy pathway – the patient is directly referred for colposcopy (first cycle and cancer symptoms).

3. Routine screening – the patient is recalled to routine screening every 3 or 5 years after their last test, depending on their age.

4. Follow-up pathways – after treatment, diagnosis or an inadequate result, the patient may be referred to follow-up, such as a test of cure (6 months after treatment), cytology and/or colposcopy (6 months after the initial test) or CIN 1 follow-up (12 months after diagnosis).

During the first cycle, all patients have a colposcopy examination with different outcomes depending on their health state, the reason for referral and the diagnostic accuracy of the colposcopy technologies. After the first cycle, subsequent examinations (follow-up or routine screening) depend on the patient’s characteristics and history.

Routine screening

The NHSCSP invites women aged 25–64 years to have a cervical screening test. Women aged < 50 years are invited 3 years after their last test, and women aged > 50 years are invited 5 years after their last test. Women who have been treated for precancerous lesions are invited 3 years after their last test, regardless of age. The model therefore keeps track of a patient’s age as well as the number of cycles elapsed since their last screening.

Cervical cytology test results are graded depending on the degree of abnormality. To be consistent with the data used for the diagnostic accuracy of cytology, the model refers to the previous NHS system terminology (British Society for Clinical Cytology 1986) and considers six possible cytology results: (1) negative, (2) inadequate, (3) borderline change, (4) mild dyskaryosis, (5) moderate dyskaryosis and (6) severe dyskaryosis. The current NHS system terminology (ABC3) is used in a second step to characterise patients referred for colposcopy: borderline change and mild dyskaryosis are defined as being low grade, whereas moderate or severe dyskaryosis is defined as being high grade.

The current management protocol is described in the NHSCSP colposcopy and programme management and referred to as ‘HPV triage’ (Figure 12). 15 An alternative protocol, known as ‘HPV primary’, has been implemented in pilot sites across England and is to be rolled out in the future within the NHSCSP (Figure 13). The cost-effectiveness analysis considers these two protocols using separate models.

FIGURE 12.

Human papillomavirus triage: routine screening and follow-up.

FIGURE 13.

Human papillomavirus primary: routine screening and follow-up.

Human papillomavirus triage

Under HPV triage, the patient is first tested with cytology. The diagnostic outcome is modelled as a probability of having a cytology result, r, given the underlying health state h. This probability depends on the cytology test performance. If the cytology is negative, the patient is sent back to routine screening and will be invited again 3 or 5 years later. When the cytology shows moderate or severe dyskaryosis, the patient is referred for colposcopy as a high-grade referral. When the cytology shows borderline changes or mild dyskaryosis, the sample collected during the cervical screening is used for hrHPV testing. The probability of being HPV-positive depends on the cytology result, the patient’s health state and the patient’s age. When HPV is detected, the patient is referred for colposcopy as a low-grade referral. Otherwise, she is sent back to routine screening. When the cytology is inadequate, the patient is tested again 3 months later. Because the cycle length is 6 months in the model, the probability of having a first inadequate result is included in the probability of having a negative, borderline, mild, moderate or severe cytology result. In the case of two consecutive inadequate results, the patient is invited 3 months later for another cytology test (i.e. 6 months after her initial cytology test). The protocol is similar to routine screening (under HPV triage), except that a patient with a third inadequate cytology result is referred for colposcopy. Although a patient with three inadequate cytology results is not defined as having low-grade dyskaryosis based on the ABC3 reporting terminology, clinical experts confirm that management after colposcopy is similar to that in low-grade dyskaryosis patients, hence the low-grade referral used in the model.

Human papillomavirus primary screening

Under HPV primary screening, the patient is first tested for hrHPV. If the HPV test is negative, the patient is sent back to routine screening. If the HPV test is positive, a cytology test is used as a triage to refer patients for colposcopy. A patient with a borderline/mild or moderate/severe cytology result is immediately referred for colposcopy, as a low-grade referral or a high-grade referral, respectively. Similarly to the HPV triage, patients are tested again 3 months after a first inadequate cytology result. Consequently, in the model, the probability of having a first inadequate result is included in the probability of having a negative, borderline, mild, moderate or severe cytology result. If the cytology is inadequate or negative twice, the patient is rescreened 12 months later (‘HPV primary rescreen at 12 months’ in the model).The 12-month follow-up protocol is similar to the initial routine test, except that a patient with inadequate cytology is referred for colposcopy. Patients with a second consecutive HPV-positive/cytology-negative result are rescreened 12 months later (i.e. 24 months after the initial test (‘HPV primary rescreen at 24 months’ in the model). In this case, a HPV-positive test result is sufficient for a patient to be referred for colposcopy. A patient with a HPV-negative test result is sent back to routine screening. Because HPV primary screening has not been implemented yet, there are uncertainties regarding the post-colposcopy management of patients with two consecutive negative cytology results. Indeed, under the HPV triage protocol, these patients were not referred to colposcopy clinics. According to clinical experts, it is likely that these patients would be managed in the same way as those with low-grade referrals.

Adverse obstetric outcomes

The adverse obstetric outcomes model (see Appendix 10, Figure 28) captures the impact of treatment for CIN on the adverse obstetric outcomes. The findings from two recent systematic reviews were previously summarised and discussed in Chapter 3, Systematic reviews of adverse outcomes of cervical intraepithelial neoplasia treatment. The two reviews reported results across a range of adverse fertility, pregnancy and obstetric outcomes. In the absence of robust evidence indicating an adverse treatment effect on fertility and early pregnancy (< 24 weeks’ gestation) outcomes, the adverse obstetric outcome model was limited to capturing the impact of an increased risk of preterm birth reported by Kyrgiou et al. 119

Following treatment for CIN, the adverse obstetric outcomes model captures the excess risk of treatment on preterm birth (< 37 weeks’ gestation) rates and applies a payoff to capture the associated costs and QALY decrements. As the model attempts to characterise the excess treatment risk only, only women who receive treatment for CIN enter the adverse obstetric outcome model. Within the model, a tracker variable is assigned to an individual patient following treatment with CIN. For each 12-month period following treatment (and up to the age of 45 years), the model captures the excess risk of preterm birth (< 37 weeks’ gestation) based on age-specific conception rates (adjusted for legal abortion), the risk of preterm birth for untreated women and the higher RR reported with treatment. The cost and QALY decrements capture the additional initial management costs associated with preterm birth and the increased probability of neonatal mortality, as well as the QALY decrements associated with higher risks of disability among survivors.

Colposcopy and adjunctive technologies

The sensitivity and specificity of colposcopy alone and the adjunctive use of DySIS and ZedScan are based on the best-evidence estimates of diagnostic accuracy reported in Chapter 3, Comparison of Dynamic Spectral Imaging System and Zedscan. An assumption is made that the binocular and video colposcopes have the same diagnostic performance and hence the diagnostic accuracy of colposcopy is based on the evidence reported in the DySIS studies. From the estimates of diagnostic accuracy, we derive the probability of a positive colposcopy result (i.e. CIN 2 or greater), given a patient’s true underlying health state. A key assumption is that the probability of a positive colposcopy result is similar for patients who are clear or have CIN 1 and for patients with CIN 2/3 or invasive cancer. A consequence of the CIN 2 threshold is that detecting CIN 1 during the colposcopy examination is not considered to be a positive colposcopy result. However, if the patient is referred for having high-grade dyskaryosis, a diagnostic biopsy will systematically reveal CIN 1 and the patient will be managed appropriately.

Based on the limitations identified in the clinical effectiveness review section, and particularly the lack of data on the diagnostic accuracy of colposcopy alone compared with the current version of ZedScan, separate pairwise analyses are presented comparing adjunctive DySIS with colposcopy alone, ZedScan with colposcopy alone and ZedScan with DySIS. Table 18 details the sensitivities and specificities used in the base-case analyses.

Cytology and human papillomavirus tests

The performance of cytology and HPV tests is also required for the model when patients are recalled for routine screening or follow-up (cytology at 6 months, test of cure, CIN 1 follow-up).

The performance of cytology test is modelled as the probability of having a certain cytology result (negative, borderline changes, mild dyskaryosis, moderate dyskaryosis, severe dyskaryosis), given a patient’s true underlying health state (Table 19). The diagnostic accuracy of cytology was derived from a published UK-based study on the cost-effectiveness of the NHSCSP [Hadwin et al. 125 – the probabilities were displayed in an unpublished document (S Eggington, personal communication)]. The probability of an inadequate cytology result was estimated to be 2.7%, based on Cervical Screening Programme, England – 2015–2016. 18 The probability of having an inadequate cytology result is assumed to be independent of a patient’s underlying health state and previous cytology results. These data were used for both HPV triage and the HPV primary screening protocol.

The performance of the HPV test is modelled as the probability of having a positive HPV test result given the patient’s true underlying health state, cytology result and age. The diagnostic accuracy of the HPV test under the HPV triage protocol (i.e. after a borderline or mild cytology result) was based on Cotton et al. 127 using data from the TOMBOLA trial. 130 The TOMBOLA trial aimed to compare different methods of management for women with low-grade cervical abnormalities under the NHSCSP in the UK. The study included 4439 women between 1999 and 2002, aged 20–59 years, with a cytology test showing borderline nuclear abnormalities or mild dyskaryosis. A cross-sectional analysis of trial data provided the sensitivity and specificity of a single HPV test in detecting CIN 2 or worse by cytology result and age (Table 20). Under the HPV primary screening protocol, the probabilities of a positive HPV test result by health state were derived from the ARTISTIC (A Randomised Trial in Screening to Improve Cytology) study, which based these estimates on extensive data in the literature on hrHPV testing [Kitchener et al. 149 (Table 21)].

Human papillomavirus triage

Table 55 in Appendix 11 reports the joint distribution of the health state and the reason for referral in a HPV triage protocol. Because the HPV triage protocol is currently implemented in the NHSCSP in England, the joint distribution of the health state and the reason for referral was based on the outcomes of colposcopy referrals published in Cervical Screening Programme, England – 2015–2016. 18 From April 2015 to June 2015, all 55 operating laboratories in England collected outcomes of colposcopy referrals and linked these results to the reason for referral. Consistently with the model structure, women who were referred after non-negative samples showing either BMD with a positive HPV test or persistent inadequate results were considered as having a low-grade referral. Women who were referred after a potentially significant abnormality, including high-grade dyskaryosis (moderate or severe); high-grade, invasive squamous carcinoma; or glandular neoplasia of the endocervical type, were considered as having a high-grade referral. In total, 31,114 samples were collected. We excluded 570 cases whose results were unknown or showed non-cervical cancer. Of the remaining 30,544 women, 68.6% had a low-grade referral and 31.4% had a high-grade referral. The outcomes of colposcopy referral (cervical cancer, CIN 3, CIN 2, CIN 1, HPV only, no CIN/no HPV) were confirmed by biopsy when the colposcopy revealed an abnormality. When no abnormalities were detected at the colposcopy examination, patients were considered to be ‘clear’. However, because colposcopy is not 100% sensitive, an unknown proportion of patients with no abnormality detected might have been misdiagnosed. Consequently, the distribution may underestimate the proportion of CIN 2 and worse in the population.

Human papillomavirus primary screening

Table 56 in Appendix 11 reports the joint distribution of the health state and the reason for referral in a HPV primary screening protocol. The data came from unpublished preliminary results collected in the HPV primary pilot sites and included a total of 15,781 women referred for colposcopy. A total of 12,626 women were referred after the first round of routine screening (HPV-positive result and cytology result of borderline dyskaryosis or worse), 2628 women were referred after the second round of screening (12-month repeat) and 527 women were referred after the third round of screening (24-month repeat). As expected, the proportion of low-grade referrals under the HPV primary screening protocol was slightly higher than that under the HPV triage protocol (70.56% vs. 68.6%).

However, these preliminary results must be interpreted with caution. First, data collection is still incomplete, especially for women referred after the third round of screening. Second, around 70% of the data on HPV primary screening (n = 11,139) came from laboratories that were implementing the HPV genotyping triage. During the second round of screening, women detected as being infected with HPV 16/18 were immediately referred for colposcopy, even if their cytology test result was negative. The impact of genotyping on the characteristics of the population referred for colposcopy is hard to predict, with potentially more low-grade referrals (women with HPV-positive and cytology-negative results are referred sooner), but also more severe cases resulting from the presence of HPV 16/18. Finally, because pilot sites were not randomly selected, we can expect selection issues, and especially variability, in the prevalence of HPV and CIN lesions compared with that in the general population, regardless of the impact of a change in the routine screening protocol. For the analysis on the HPV primary screening protocol, we used data reported by all pilot sites (with and without genotyping) as a base case (see Table 56) and provided sensitivity analyses around the characteristics of the initial population using data from pilot sites that are and are not implementing genotyping (see Underlying health state and reason for referral).

See-and-treat and watchful-waiting clinics

Treatment decisions after a positive colposcopy result (i.e. diagnostic biopsy or immediate treatment) vary considerably across England. Currently, for women referred with high-grade abnormalities, the most common treatment at first attendance is excision (53.2%), followed by diagnostic biopsy (35%). The use of excision at first attendance for women with high-grade referrals ranges from 11.6% in London to 65.4% in the North West of England (Cervical Screening Programme, England – 2015–2016). 18

Heterogeneity in treatment decisions after a positive colposcopy result is modelled as two different types of clinic a patient may visit, and is independent of health state or diagnostic accuracy. A patient may visit either a ‘see-and-treat’ clinic or a ‘watchful-waiting’ clinic. Because the NHSCSP guidance states that the PPV for CIN 2 or worse should be at least 90% to undergo ‘see and treat’, the model assumes that the use of excision at first attendance is possible for women with high-grade referrals with a positive colposcopy result only. When patients have a low-grade referral, CIN 2 or worse must be confirmed by the diagnostic biopsy result before being treated. This assumption has been validated by clinical experts.

As the cost-effectiveness of adjunctive colposcopy devices is likely to be driven by treatment practice, the results are presented separately for the two types of clinics. Indeed, a low specificity of colposcopy leads to overtreatment in a ‘see-and-treat’ clinic. In contrast, because biopsy is assumed to be 100% sensitive and specific, watchful-waiting clinics are assumed to never overtreat patients. Furthermore, watchful waiting requires two colposcopy examinations: a first one in which a diagnostic biopsy is performed and a second one in which CIN 2 lesions or worse are treated. However, ‘see-and-treat’ clinics require only one colposcopy.

Probability of cure from treatment biopsy

The probability of cure after treatment biopsy was derived from a study by Ghaem-Maghami et al. 150 that reported the failure rates for 2455 women treated for CIN for the first time between 1989 and 2004. Failure was measured by the detection of high-grade cervical disease after treatment, defined as the cytological findings of moderate dyskaryosis or more severe or histological findings of CIN 2 or worse. The median length of follow-up was 238 weeks. The authors reported that the cumulative failure rate at 10 years was 4.9% for CIN 1 (n = 570), 9.8% for CIN 2 (n = 886) and 10.3% for CIN 3 (n = 999). We calculated that the weighted excision failure rate of CIN 2/3 was 10.1% (see Appendix 11, Table 57).

Probabilities of adverse obstetric outcomes

For each 12-month period following treatment, the model captures the excess risk of preterm birth (< 37 weeks’ gestation) based on age-specific conception rates (adjusted for legal abortion), the risk of preterm birth for untreated women and the higher RR reported with treatment.

Age-specific conception rates were derived from national conception statistics reported by the Office for National Statistics (ONS). 151 The ONS statistics bring together records of birth registrations (including live births or stillbirths) and abortion notifications. Hence, the conception input data applied in the model do not include conceptions resulting in miscarriages or illegal abortions. The annual probability of conception (adjusted for legal abortion) applied in the model was derived from age-specific conception and legal abortion rates reported per 1000 women [England and Wales (see Appendix 11, Table 58)].

The excess risk of preterm birth was then estimated, based on applying a RR representing the additional risk following treatment for CIN to the probability of preterm birth without treatment. The increase in RR following treatment (LLETZ) was assumed to be 1.56. This was based on the results reported by Kyrgiou et al. 119 for all treatments in all preterm births (< 37 weeks’ gestation), when the external comparison includes the overall population.

The RR of preterm birth was then applied to the probability of preterm birth for untreated women. The probability of preterm birth for untreated women was assumed to be 7.3% based on data reported in the NICE guideline on preterm labour and birth. 152 Consequently, the excess risk of preterm delivery for women treated with LLETZ was estimated to be 4.09% [0.073 × (1.56 – 1); see Appendix 11, Table 59].

Precancerous lesions

The transition probabilities from the ‘clear’ state to CIN 2/3 were based on the transition probabilities reported by Kulasingam et al. ,147 an update of Myers et al. 128 used in Wade et al. 30 Kulasingam et al. 147 took into account recent evidence that, in young women (aged < 30 years), high-grade CIN may occur early in the course of a hrHPV infection. Recent studies also suggested that CIN may be more frequent in young women, but that progression to cancer from high-grade CIN is low. Compared with Myers et al. ,128 HPV and CIN incidence and regression estimates were higher, but progression rates between CIN states and from CIN 2/3 to cancer were lower. Transition probabilities were reported by the authors as annual probabilities and converted to 6-month probabilities in our model (see Appendix 11, Table 60).

Invasive cancer

Table 61 in Appendix 11 reports the parameters used to estimate the likelihood of symptoms of cervical cancer, progression across stages and cancer-related mortality. Women with undetected cancer may progress to a more severe stage (from local to regional to distant), with an increasing probability of developing symptoms. Once cancer is detected, the model assumes that patients will receive stage-specific treatment and face an excess risk of mortality as a result of having cervical cancer. Cancer-specific mortality depends on stage and decreases by time since diagnosis. After 5 years, excess mortality as a result of having cancer is assumed to be zero and patients are assumed to be cured. We assume that women with undetected cancer continually face the year 1 probability of cancer mortality until diagnosis.

Our estimates were based on Campos et al. ,148 which reported the monthly probability of symptoms, progression and mortality by stage (local, regional, distant). Cancer-specific mortality in Campos et al. 148 was derived from the Surveillance, Epidemiology, and End Results (SEER) Programme registry data from 2000 to 2009. 149 Data reported by Campos et al. ,148 which were based on US registry data, appeared to be the most recent, complete and consistent data on cancer progression and mortality. Transition probabilities (stage progression and mortality) when cancer is not detected should not be affected by differences in care between the USA and the UK. 148

All-cause mortality

Mortality rates from causes other than cervical cancer were calculated using data from the ONS in England and Wales in 2015. 153 Deaths resulting from cervical cancer were subtracted from the total number of deaths for each age group. The 2015 ONS data on the population for each age group were then used to calculate an annual mortality rate, converted to a 6-month probability in the model (see Appendix 11, Table 62).

Devices costs

The cost-effectiveness analysis requires an estimate of the average cost per procedure of each of the technologies being assessed. A colposcopy examination with DySIS or ZedScan is assumed to be equivalent to colposcopy alone in terms of staff resources and the length of consultation. The average cost of a procedure includes a set-up cost, annual recurring costs and per-patient costs. Information provided by the manufacturers has been used to estimate the costs of DySIS and ZedScan. For the purchase price and maintenance costs of colposcopy, we used estimates provided by clinical advisors for the previous diagnostics assessment report (Wade et al. 30) and inflated these to 2016 prices.

The set-up cost consists of the capital cost of the machine. The purchase price of each technology was annuitised over the expected lifetime of the technology. Consistent with Wade et al. ,30 the lifetime of a colposcopy was estimated to be 15 years. The lifetimes of DySIS and ZedScan were estimated to be 5 years each. The equivalent annual cost was calculated from the purchase price of the technology and the useful life of the equipment using the discount rate of costs of 3.5%.

The annual maintenance cost of the colposcope was estimated to be 10% of the purchase price and the cost of disposables was estimated to be equivalent to the cost of a speculum (£2.15). The annual maintenance costs and disposable costs of the adjunct technologies were provided by the manufacturers. For DySIS, the annual maintenance costs included the DySIS viewer licence registration and renewal as well as a 5-year service and maintenance plan. The price of a DySIS disposable speculum (£3.50) was added to the per-patient cost. The ZedScan manufacturer claimed no routine maintenance costs. However, a single-use EIS sensor (£30.00) is required for each patient examined. To estimate the total cost per patient, it is necessary to estimate the number of patients expected to be treated each year. We assumed that this number was independent of the type of devices and used the previous estimate of 1229 patients per device per year (see Appendix 11, Table 63). Sensitivity analyses were undertaken to test this particular assumption (see Underlying health state and reason for referral). As ZedScan uses a binocular colposcope to guide the probe or to confirm diagnosis, the cost of a colposcope was also added to its total cost. DySIS devices already include a colposcope and therefore do not require this additional cost.

In addition to the cost related to the device itself, the costs of a colposcopy visit, diagnostic biopsy and treatment (LLETZ) were estimated from NHS Reference Costs 2015 to 2016. 154 The NHS Reference Costs reported the costs (2016 prices) of a ‘diagnostic colposcopy’, a ‘diagnostic colposcopy with biopsy’ and a ‘therapeutic colposcopy’ (see Appendix 11, Table 64). The uncertainty surrounding whether or not the NHS reference costs accurately reflect resource use, and in particular if these include histology/pathology costs, is explored in a sensitivity analysis (see Underlying health state and reason for referral).

In the base case, we estimated the cost of a colposcopy visit to be £175 with a binocular colposcope, £180.49 with DySIS and £205.52 with ZedScan. The additional costs of a diagnostic biopsy and treatment were estimated to be £47 and £63, respectively. The costs of a cytology test and a HPV test were derived from the TOMBOLA study,130 inflated to 2016 prices, and were estimated to be £37.19 and £29.66, respectively (see Appendix 11, Table 65). 132

Cancer costs

Cancer costs by stage were taken from a UK-based study by Martin-Hirsch et al. ,155 which estimated the costs associated with the management of women with abnormal cervical cytology. The unit costs for cancer treatment (including chemotherapy, radiotherapy and inpatient care) were obtained by the authors from the national reference costs, the British National Formulary156 and from personal communications with the purchasing departments of clinics included in the study (six centres in England and Wales). The average costs per cancer treatment were reported in pounds sterling at 2006 prices by cancer stage, using the FIGO grading (stages I–IV). We assumed that the average cost of stage I and stage II refers to ‘local’ stage, the average cost of stage III refers to ‘regional’ stage and the average cost of ‘stage IV’ refers to ‘distant’ stage. In the model, all costs were inflated to 2016 prices (see Appendix 11, Table 66).

Costs associated with adverse obstetric outcomes

The additional costs associated with preterm birth were derived from the same source157 as the QALY decrements. The study by Lomas et al. 157 reported an expected incremental (discounted) lifetime cost of £24,071 per birth (inflated to £24,610 in 2016 prices). This estimate incorporates an estimate of the initial inpatient neonatal care and ongoing costs over the following 18 years of life in survivors, associated with higher rates of disability.

Screening disutility

The disutility associated with screening and treatment was based on a recent study, especially designed to estimate utility values for HPV testing and cytology-based screening states among women targeted for cervical screening (Simonella et al. ). 158 A total of 43 women (mean age 49 years), living in Sydney, Australia, participated in the study. The participants were asked to state their preferences (rank and utility scores) for hypothetical states relating to cytology and HPV screening and precancerous lesions. The utility values were estimated via a two-stage standard gamble. The model uses utility values reported by Simonella et al. 158 for four types of screening episodes: (1) a routine screening episode with a normal cytology result, (2) a false-positive referral to colposcopy, (3) a colposcopy referral that leads to confirmed but not treated CIN 1 and (4) a colposcopy referral that leads to the treatment of CIN lesions. Each scenario was described in a narrative format in Simonella et al. 158 to characterise the screening process and possible adverse outcomes associated with examination and treatment. Consequently, this set of values captures the disutility associated with the screening process, from the experience of being screened (even if the test result is negative) to the possible short-term adverse outcomes of colposcopy and treatment.

Simonella et al. 158 reported the mean standard gamble utility values over a 12-month period. In the model, we converted these scores into QALY decrements (1 – mean utility value) of undergoing a screening episode (initial referral for colposcopy, follow-ups or routine screening). A screening episode with cytology and/or HPV test that did not result in a referral for colposcopy induced a QALY decrement of 0.0062. The QALY decrement associated with a false-positive referral for colposcopy (the cytology test and/or HPV test are positive, but the colposcopy or histopathology are negative) or a confirmed but not treated CIN 1 lesion was estimated to be 0.0276. Finally, a positive diagnosis followed by excision treatment of the CIN lesion induced a QALY decrement of 0.0296 (see Appendix 11, Table 67).

Health-related quality of life of underlying true health states

As HPV, CIN 1, CIN 2/3 and undetected cancer are considered to be asymptomatic, we applied age- and gender-specific utilities from the Measurement and Valuation of Health survey, a nationally representative interview survey of 3395 men and women living in the UK conducted in 1993. The objective of the survey was to collect data on the health state valuations using a time-trade-off procedure and on individuals’ self-reported health status using the EuroQol-5 Dimensions descriptive classification system (Kind et al. )159 (see Appendix 11, Table 68). The possible disutility that patients can experience once CIN lesions are identified is subsequently captured by screening disutility as outlined previously.

Quality-adjusted life-year decrements associated with invasive cancer were obtained from a published study (Goldie et al. 5). The authors reported the HRQoL for ‘detected invasive cancer’ and ‘after treatment for invasive cancer’ by stage (local, regional and distant). These weights were constructed based on expert elicitation. We considered the first set of HRQoL to be a utility score associated with the first year post diagnosis, during which period patients undergo treatment. We used the second set for the remaining 4 years, during which time patients are not yet considered to be cured from cancer but do not receive further treatments (see Appendix 11, Table 69).

Quality-adjusted life-year decrement associated with adverse obstetric outcomes

A QALY decrement was applied to capture the HRQoL and mortality consequences of the increased risk of preterm birth following treatment for CIN. We did not attempt to identify evidence for this parameter systemically. Instead, we restricted our search to the evidence reported in the NICE guideline on preterm labour and birth (NG25);152 although this guideline reports utility and QALY estimates, none of these directly provided the required estimates for our model (i.e. the QALY decrement associated with preterm birth at < 37 weeks’ gestation). Instead, we sourced estimates based on discussions with colleagues and identified a recent study by Lomas et al. 157 that reported cost and QALY decrements that matched the requirements of the model. The QALY decrement reported in Lomas et al. 157 was 1.3 QALYs and was derived from calculations based on the QALY loss associated with neonatal mortality and the discounted QALY loss associated with increased disability rates reported among survivors.

Uncertainty around diagnostic accuracy

Uncertainty around costs

Population referred for colposcopy under the human papillomavirus primary screening protocol

Scenario analyses

Scenario analyses were undertaken to consider the impact of three key structural assumptions.

Human papillomavirus triage protocol: base-case results

The results for the base-case analysis under the HPV triage protocol are summarised in Figure 14 and presented in more detail in Tables 29–31 (costs and QALYs) and in Table 73 in Appendix 12 (secondary outcomes).

FIGURE 14.

Base-case analysis cost-effectiveness results: HPV triage protocol. (a) DySIS vs. colposcopy alone; (b) ZedScan vs. colposcopy alone; and (c) ZedScan vs. DySIS. HG, high grade; LG, low grade.

The main results of the base-case analysis under the HPV triage protocol can be summarised as follows:

• Dynamic Spectral Imaging System routinely dominated colposcopy alone, regardless of the type of clinic or the reason for referral (Table 29). The only exception was for referrals for high-grade abnormalities in watchful-waiting clinic settings, in which DySIS was more costly and more effective, with an associated ICER of £675 per QALY.

• ZedScan also dominated colposcopy alone in see-and-treat clinics (Table 30). However, in watchful-waiting clinics, ZedScan was always more effective than colposcopy alone, but was also more costly. The ICER for ZedScan in watchful-waiting clinics ranged from £272 (referrals for low-grade abnormalities) to £4070 per QALY (referrals for high-grade abnormalities).

• The higher sensitivity of DySIS and ZedScan resulted in increased QALYs compared with conventional colposcopy alone in all sets of results. In addition, the incremental gain in QALYs always appeared to be higher for referrals for low-grade abnormalities. The higher incremental gain for referrals for low-grade abnormalities is attributable to the assumption that a diagnostic biopsy will not be performed if the colposcopy result is negative. Hence, false-negative results for referrals for low-grade abnormalities will therefore potentially be missed, making higher sensitivity a more critical consideration for referrals for low-grade abnormalities.

• The impact on total cost appears to depend on the technology and clinic practice. Both adjunctive technologies generally decrease the average cost per patient in see-and-treat clinics. In watchful-waiting clinics, however, the average cost generally decreases to a smaller extent with DySIS and appears to increase with ZedScan compared with colposcopy alone.

• The indirect comparison between ZedScan and DySIS showed that ZedScan routinely appears to be more effective but also more costly than DySIS (Table 31). The ICER for ZedScan ranged from £109 per QALY for referrals for high-grade abnormalities in see-and-treat clinics to £9918 per QALY for referrals for high-grade abnormalities in watchful-waiting clinics.

• The secondary outcomes from the simulations show that a higher specificity (colposcopy alone) limits the number of unnecessary treatments and biopsies and consequently reduces the number of adverse obstetric outcomes (see Appendix 12, Table 73). In contrast, a higher sensitivity (adjunctive technologies) reduces the number of undetected CIN 2+, new cancer cases and deaths as a result of developing cancer.

In Figure 14, the incremental costs and QALYs of DySIS versus colposcopy alone, ZedScan versus colposcopy alone and ZedScan versus DySIS are represented visually using separate cost-effectiveness planes. The horizontal axis divides the plane according to the incremental cost (positive above and negative below) and the vertical axis divides the plane according to the incremental QALYs (positive to the right and negative to the left). The cost-effectiveness plane is thus divided into four quadrants, with different implications for decision-making. If an intervention falls in the south-east quadrant, then it dominates the comparator technology (i.e. is less costly and more effective). Similarly, the intervention would be dominated by the comparator in the north-west quadrant. When non-dominance exists (north-east and south-west quadrants), the resulting ICER can be compared against the conventional cost-effectiveness threshold (£20,000–30,000 per QALY) to determine whether or not the intervention is cost-effective. In Figure 14, a £20,000 threshold is represented by the straight line that further divides the north-east and south-west quadrants. Points above and below the line indicate that the ICER of the intervention is higher or lower than the threshold, respectively. The results by type of clinic and the reason for referral are represented in each plane by distinct markers.

Human papillomavirus primary protocol: base-case results

The results for the base-case analysis under the HPV primary screening protocol are summarised in Figure 15 and presented in more detail in Tables 32–34 (costs and QALYs) and Appendix 12, Table 74 (secondary outcomes). The interpretation of the cost-effectiveness planes in Figure 15 is the same as for Figure 14.

FIGURE 15.

Base-case analysis cost-effectiveness results: HPV primary screening protocol. (a) DySIS vs. colposcopy alone; (b) ZedScan vs. colposcopy alone; and (c) ZedScan vs. DySIS. HG, high grade; LG, low grade.

With regard to the cost-effectiveness of adjunctive technologies, the conclusions were quite similar under the HPV primary screening and HPV triage protocols:

• In most instances, DySIS dominated colposcopy alone except for referrals for high-grade abnormalities in watchful-waiting clinics, for which the ICER was estimated to be £1095 per QALY (Table 32).

• The results for ZedScan were more varied. ZedScan dominated colposcopy alone only for referrals for high-grade abnormalities in a see-and-treat clinic. In all other cases, ZedScan was more effective but also more costly than colposcopy alone. The ICER ranged from £417 per QALY for referrals for low-grade abnormalities in see-and-treat clinics to £4922 per QALY for referrals for high-grade abnormalities in watchful-waiting clinics (Table 33).

• ZedScan was always more effective but also more costly than DySIS. The ICER ranged from £426 per QALY for referrals for high-grade abnormalities in see-and-treat clinics to £8190 per QALY for referrals for high-grade abnormalities in watchful-waiting clinics (Table 34).

• Consistent with the findings reported in the ARTISTIC study149 simulations under the HPV primary screening protocol predicted higher health outcomes and a lower average cost per patient than under the HPV triage protocol. With regard to the cost-effectiveness of adjunctive technologies, the most significant impact of the HPV primary protocol was to reduce the incremental effect of the adjunctive technologies on health outcomes. Because HPV primary routine screening presents a higher sensitivity overall, cases of patients with CIN 2+ that were missed at the initial colposcopy appointment have a higher probability of being diagnosed 3 years later during routine screening, which would avoid the subsequent development of cancer. The lower sensitivity of colposcopy alone than that of adjunctive technologies therefore appears to be less critical in this context.

Uncertainty around diagnostic accuracy

Uncertainty around the diagnostic accuracy of colposcopy alone, DySIS and ZedScan was assessed with sensitivity analyses 1–4.4 (see 5.4.2.1 for a detailed description). Because the sensitivity analyses are different for DySIS and ZedScan, the results are presented separately in Figures 29 and 30 in Appendix 13. Detailed results for average and incremental costs and QALYs are presented in Appendix 13, Tables 75–88) for each sensitivity analysis, under the HPV triage and HPV primary screening protocols.

For DySIS compared with colposcopy alone, the results were globally unchanged compared with the base-case analysis:

• Dynamic Spectral Imaging System dominated colposcopy alone in most instances except for referrals for high-grade abnormalities in watchful-waiting settings, in which the ICER ranged from £188 [sensitivity analysis 2 under the HPV triage protocol (see Appendix 13, Table 77)] to £1633 [sensitivity analysis 2 under the HPV primary screening protocol (see Appendix 13, Table 78)].

The results of sensitivity analyses comparing ZedScan with colposcopy alone were more varied:

• The lower and upper bounds of ZedScan specificity and sensitivity (sensitivity analysis 4.3 and sensitivity analysis 4.4) had little impact on the results. ZedScan still dominated colposcopy alone in see-and-treat clinics and was more effective but also more costly than colposcopy alone in watchful-waiting clinics, under both the HPV triage and HPV primary screening protocols.

• When ZedScan was compared with colposcopy alone based on diagnostic accuracy data that stemmed from a similar setting (sensitivity analysis 3), the incremental cost increased and the incremental QALYs decreased. Overall, ZedScan no longer dominated colposcopy alone in see-and-treat clinics, with an ICER ranging from £590 under the HPV triage protocol to £1457 under the HPV primary screening protocol. In watchful-waiting clinics, the ICER increased from £418 in the base case to £1910 in sensitivity analysis 3 under the HPV triage protocol and from £988 to £4023 under the HPV primary screening protocol. Under HPV primary screening, the ICER exceeded £20,000 per QALY for referrals for high-grade abnormalities in watchful-waiting clinics (see Appendix 13, Tables 79 and 80).

Uncertainty around costs

Sensitivity analyses results exploring the uncertainty around the costs of devices (sensitivity analyses 5.1 and 5.2) and the costs of diagnostic biopsy and LLETZ (sensitivity analysis 6) are summarised in Appendix 13, Figure 31, for the HPV triage protocol and Appendix 13, Figure 32, for the HPV primary screening protocol. As only the cost parameters were varied, the incremental QALYs are identical between the sensitivity analyses and the base case for each type of simulation. Note that, in Chapter 3, Figures 5 and 6, the scale of the vertical axis of the cost-effectiveness planes has been altered in order to represent higher incremental costs. Detailed results are presented in Appendix 13, Tables 89–106.

Logically, sensitivity analyses 5.1 and 6 increased the incremental cost for both technologies, but they did not appear to alter the base-case conclusions:

• Owing to the higher purchase price, the results for DySIS appeared to be more sensitive to a decrease in the number of patients per colposcope per year (sensitivity analysis 5.1). DySIS no longer dominated colposcopy alone in watchful-waiting clinics under the HPV primary screening protocol, with an ICER of £270 for all referrals (see Appendix 13, Table 92).

• Owing to the lower purchase price of ZedScan, the results did not appear to be sensitive to the assumed variation in throughput (sensitivity analyses 5.1 and 5.2). However, because of its high sensitivity and low specificity, the higher cost estimates for diagnostic biopsies and LLETZ (sensitivity analysis 6) had an impact on the results for ZedScan more significantly than the results for DySIS, especially for referrals for low-grade abnormalities in watchful-waiting clinics. ZedScan no longer dominated colposcopy alone, regardless of the type of clinics or the routine screening protocol. The ICER increased to £6709 for referrals for high-grade abnormalities in watchful-waiting clinics under the HPV primary screening protocol (see Appendix 13, Table 105).

• The results on the indirect comparison between ZedScan and DySIS were unchanged.

Population referred for colposcopy under the human papillomavirus primary protocol

Sensitivity analyses 7.1 and 7.2 used the variation observed between pilot sites without and with HPV 16/18 genotyping, respectively, to explore the potential impact of uncertainty around the characteristics of the population referred for colposcopy under the HPV primary screening protocol. The results are summarised in Appendix 13, Figure 33, and presented in more detail in Appendix 13, Tables 107–112. Overall, the results are unchanged compared with those of the base case and the impact of a variation in population characteristics appears to be relatively small.

Scenario 1: time horizon of 3 years

The model was run for only 3 years to capture the cost and health outcomes of colposcopy and adjunctive technologies in the short term. Figures 31 and 32 in Appendix 13 summarise the results under the HPV triage and HPV primary screening protocols. The average and incremental costs and QALYs, as well as the secondary outcomes, are provided in Appendix 13 (see Tables 113–115).

The results were dramatically different when long-term costs and health outcomes were not taken into account in the evaluation:

• In the short term, colposcopy alone routinely dominated DySIS and ZedScan (i.e. it was less costly and more effective). The higher specificity of colposcopy alone limited the number of treatments and therefore reduced the average cost compared with DySIS or ZedScan. Meanwhile, the lower specificity of colposcopy alone was less penalised, as most individuals with untreated CIN 2+ would not have developed cancer or died from cancer 3 years after their initial examination (see Appendix 13, Tables 113 and 114).

• Colposcopy was not found to be dominant only for referrals for high-grade abnormalities in see-and-treat clinics. DySIS and ZedScan were less costly, but also less effective, than colposcopy alone, with ICERs, respectively, of £236,692 and £85,045 per QALY under the HPV triage protocol (see Appendix 13, Tables 113 and 114).

Scenario 2: adverse obstetric outcomes were excluded

Scenario 2 excluded from the analysis the adverse consequences of CIN treatment on obstetric outcomes. The results are summarised in Appendix 13, Figures 36 and 37 and presented in more detail in Tables 116–118.

When adverse obstetric outcomes were excluded, all technologies were found to be less costly than in the base-case scenario.

The cost-effectiveness results for DySIS and ZedScan compared with colposcopy alone were unchanged under both the HPV triage protocol and the HPV primary screening protocol:

• Dynamic Spectral Imaging System routinely dominated colposcopy alone, regardless of the type of clinic, the reason for referral or the routine screening protocol.

• ZedScan also dominated colposcopy alone in see-and-treat clinics. However, ZedScan was more effective but also routinely more costly than colposcopy alone in watchful-waiting clinics.

As in the base case, ZedScan was routinely more effective and more costly than DySIS:

• Because ZedScan presents a lower specificity than DySIS, the ICER was lower when adverse obstetric outcomes were excluded than in the base-case scenario: £427 per QALY compared with £980 per QALY for see-and-treat clinics under the HPV triage protocol (all referrals) and £1476 per QALY compared with £1746 per QALY under the HPV primary screening protocol (all referrals).

Scenario 3: ZedScan was used alongside colposcopy at all appointments

Scenario 3 assumed that ZedScan was used alongside colposcopy at all appointments, including therapeutic colposcopies, after confirmation by histology results. As the costs and QALYs were not altered for colposcopy alone and DySIS compared with the base case, Appendix 13, Figure 38, presents the results for ZedScan compared with colposcopy alone, and for ZedScan compared with DySIS only, under the HPV triage and HPV primary screening protocols. Detailed results are given in Tables 119–122 in Appendix 13.

As no additional benefit is associated with the use of ZedScan during therapeutic colposcopies, the impact of scenario 3 is only an increase in the incremental cost of ZedScan compared with colposcopy alone and DySIS:

• Overall, the conclusions remain unchanged compared with those of the base case.

However, the impact of using ZedScan alongside colposcopy at all appointments varies depending on the reason for referral and the type of clinic:

• The increase in cost was higher for referrals for high-grade abnormalities in watchful-waiting clinics – when ZedScan was compared with colposcopy alone, the ICER was £7270 in the HPV triage protocol (see Appendix 13, Table 119) and £8557 under the HPV primary screening protocol (see Appendix 13, Table 121); when ZedScan was compared with DySIS, the ICER reached £18,628 under the HPV triage protocol (see Appendix 13, Table 120) and £14,928 under the HPV primary screening protocol (see Appendix 13, Table 122).

Key

• / = indexing term [medical subject heading (MeSH)]

• exp = exploded indexing term (MeSH)

• $ = truncation

• ti,ab = terms in either title or abstract fields

• adj2 = terms within two words of each other (any order).

Key

• MeSH descriptor = indexing term (MeSH)

• * = truncation

• ti,ab,kw = terms in either title or abstract or keyword fields

• near/2 = terms within two words of each other (any order)

• next = terms are next to each other.

Key

• MH = indexing term (CINAHL heading)

• * = truncation

• TI = terms in the title

• AB = terms in the abstract

• N2 = terms within two words of each other (any order).

Key

• MeSH DESCRIPTOR = indexing term (MeSH)

• * = truncation

• NEAR2 = terms within two words of each other (order specified).

Key

• / = indexing term (Emtree heading)

• exp = exploded indexing term (Emtree heading)

• $ = truncation

• ti,ab = terms in either title or abstract fields

• dm = terms in device manufacturer field

• dv = terms in the device trade name field

• adj2 = terms within two words of each other (any order).

Key

• / = indexing term

• $ = truncation

• ti,ab = terms in either title or abstract fields

• adj2 = terms within two words of each other (any order).

Key

• [Mesh] = exploded indexing term (MeSH)

• [Mesh:noexp] = indexing term (MeSH) not exploded

• * = truncation

• ‘‘ ’’ = phrase search

• [Title/Abstract]) = terms in either title or abstract fields.

Key

• TS = topic tag; searches terms in title, abstract, author keywords and keywords plus fields

• * = truncation

• ‘‘ ’’ = phrase search

• NEAR/2 = terms within 2 words of each other (any order).

Search strategy

A total of 169 studies found for (Cervix OR cervical) AND (Colposcopy OR spectroscopy OR spectrometry OR ‘spectrum analysis’).

Four studies found for dysis OR dysismap OR ‘dynamic spectral imaging’ OR Zilico OR ZedScan OR Zed Scan OR ‘APX 100’ OR APX100 OR epitheliometer OR MKIII.

Search strategy

1. #30. #28 AND #8 (62)

2. Indexes=CPCI-S Timespan=2000-2017

3. #29. #28 AND #8 (67)

4. #28. #27 OR #26 OR #25 OR #24 OR #23 OR #22 OR #21 OR #20 OR #19 OR #18 OR #17 OR #16 OR #15 OR #14 OR #13 OR #12 OR #11 OR #10 OR #9 (20,223)

5. #27. TS=MKIII (27)

6. #26. TS=epitheliometer* (0)

7. #25. TS=EIS (3376)

8. #24. TS=(“APX 100” or APX100) (0)

9. #23. TS=(ZedScan or “Zed Scan”) (0)

10. #22. TS=Zilico (0)

11. #21. TS=“dynamic spectral imaging” (4)

12. #20. TS=(dysis or dysismap) (32)

13. #19. TS=(microcolposcop* or micro-colposcop*) (0)

14. #18. TS=((point or pencil or impedance) NEAR/2 probe*) (1606)

15. #17. TS=(optical NEAR/2 spectroscop*) (8132)

16. #16. TS=(telecolposcop* or tele-colposcop*) (3)

17. #15. TS=(Dielectric NEAR/2 “Spectrum analys*”) (3)

18. #14. TS=(impedance NEAR/2 “spectrum analys*”) (15)

19. #13. TS=(Dielectric NEAR/2 Spectrometr*) (19)

20. #12. TS=(impedance NEAR/2 spectrometr*) (51)

21. #11. TS=(Dielectric NEAR/2 Spectroscop*) (2063)

22. #10. TS=(impedance NEAR/2 spectroscop*) (7234)

23. #9. TS=(colposcop* NEAR/4 (adjunct* or digital* or DSI or computer* or video* or alternative* or conventional*)) (38)

24. #8. #7 OR #6 OR #5 OR #4 OR #3 OR #2 OR #1 (16,544)

25. #7. TS=“transformation zone*” (135)

26. #6. TS=((squamocolumnar or squamo-columnar) NEAR/2 junction) (60)

27. #5. TS=(ectocervic* or ecto-cervic*) (36)

28. #4. TS=(ectocervix or ecto-cervix) (20)

29. #3. TS=(endocervic* or endo-cervic*) (400)

30. #2. TS=(endocervix or endo-cervix) (54)

31. #1. TS=(cervix or cervic*) (16,171)

Search strategy

1. 15 result(s) found for (Cervix OR cervical) AND (Colposcopy OR spectroscopy OR spectrometry OR “spectrum analysis”).

2. Dysis OR dysismap OR “dynamic spectral imaging” – 0 results.

3. Zilico OR ZedScan OR “Zed Scan” OR “APX 100” OR APX100 OR epitheliometer OR MKIII – 0 results.

Search strategy

1. #1. MeSH DESCRIPTOR Cervix Uteri (10)

2. #2. cervix OR cervic* (399)

3. #3. endocervix OR endo-cervix (1)

4. #4. endocervic* OR endo-cervic* (4)

5. #5. ectocervix OR ecto-cervix (0)

6. #6. ectocervic$or ecto-cervic$ (0)

7. #7. ectocervic* OR ecto-cervic* (0)

8. #8. squamocolumnar OR squamo-columnar (1)

9. #9. transformation zone* (3)

10. #10. #1 OR #2 OR #3 OR #4 OR #5 OR #7 OR #8 OR #9 (400)

11. #11. MeSH DESCRIPTOR colposcopy (2)

12. #12. MeSH DESCRIPTOR Colposcopes (0)

13. #13. MeSH DESCRIPTOR Spectrum Analysis (1)

14. #14. MeSH DESCRIPTOR Dielectric Spectroscopy (0)

15. #15. (colposcop* AND (adjunct* OR digital* OR DSI OR computer* OR video* OR alternative* OR conventional*)) (3)

16. #16. ((impedance OR Dielectric) AND (spectroscop* OR spectrometr* OR spectrum analys*)) (1)

17. #17. telecolposcop* OR tele-colposcop* (0)

18. #18. telecolposcop* OR tele-colposcop* (0)

19. #19. optical AND spectroscop* (5)

20. #20. ((point OR pencil OR impedance) AND probe*) (16)

21. #21. microcolposcop* OR micro-colposcop* (1)

22. #22. dysis OR dysismap (2)

23. #23. dynamic spectral imaging (1)

24. #24. Zilico (1)

25. #25. ZedScan OR Zed Scan (0)

26. #26. APX 100 OR APX100 (1)

27. #27. EIS (1)

28. #28. epitheliometer* (0)

29. #29. MKIII (0)

30. #30. #11 OR #12 OR #13 OR #14 OR #15 OR #16 OR #17 OR #19 OR #20 OR #21 OR #18 OR #22 OR #23 OR #24 OR #25 OR #26 OR #27 OR #28 OR #29 (26)

31. #31. #30 AND #10 (4)

Search strategy

1. cervix OR cervical (Condition field) AND Colposcopy OR spectroscopy OR spectrometry OR spectrum analysis (Intervention field) – 16 records retrieved.

2. dysis OR dysismap OR dynamic spectral imaging OR Zilico OR ZedScan OR Zed Scan OR APX 100 OR APX100 OR epitheliometer OR MKIII (Intervention field) – 1 record retrieved.

Scottish Intercollegiate Guidelines Network (www.sign.ac.uk)

Searched the website using the terms ‘colposcopy’, ‘DySIS’, ‘ZedScan’, ‘Zed Scan’. Also browsed all guidelines. No new guidance found.

National Institute for Health and Clinical Excellence (www.nice.org.uk)

Searched the website using the terms ‘colposcopy’, ‘DySIS’, ‘ZedScan’, ‘Zed Scan’. Also browsed documents within the cervical cancer guidance section. Four relevant guidance documents were found.

National Guideline Clearinghouse (www.guideline.gov)

Searched using the terms ‘colposcopy OR DySIS OR ZedScan OR Zed Scan’, limited to publications from 2011 to 2017. A total of 19 results were browsed for relevance. Eight relevant guidelines were found.

NHS Evidence (www.evidence.nhs.uk)

Searched using the terms ‘colposcopy OR dysis OR ZedScan OR Zed Scan’. Filtered results by guidance and by date (1 January 2011 to 10 January 2017). A total of 40 records were retrieved and downloaded.

Turning Research into Practice database (www.tripdatabase.com)

Searched using the terms ‘colposcopy OR dysis OR ZedScan OR Zed Scan’. Filtered results by guidelines; 48 records were retrieved and browsed for relevance. One relevant record was found after duplicates were removed.

Public Health England (www.gov.uk/search)

Searched the website using the terms ‘colposcopy’, ‘DySIS’, ‘ZedScan’ and ‘Zed Scan’. Filtered by Public Health England. Nine results were retrieved and browsed for relevance. Seven relevant documents were found.

Royal College of Obstetricians and Gynaecologists (www.rcog.org.uk/en/guidelines-research-services/guidelines)

Searched all guidelines using the terms ‘colposcopy’, ‘DySIS’, ‘ZedScan’ and ‘Zed Scan’. Eight records were retrieved and browsed for relevance. One relevant report was found.

The British Society for Colposcopy and Cervical Pathology (www.bsccp.org.uk)

Searched the website using the terms ‘colposcopy’, ‘DySIS’, ‘ZedScan’ and ‘Zed Scan’, using the website general search box. A total of 110 results were returned and browsed for relevance. No guidelines were found.

Search strategy

1. Uterine Cervical Neoplasms/ (67,631)

2. Cervical Intraepithelial Neoplasia/ (8926)

3. exp Uterine Cervical Dysplasia/ (4058)

4. Cervix Uteri/ (25,999)

5. ((cervix or cervic*) adj3 (cancer* or neoplas* or carcinoma* or adenocarcinoma* or tumour* or tumor* dysplas* or dyskaryo* or precancer* or pre-cancer*)).ti,ab. (65,985)

6. ((cervix or cervic*) adj3 (abnormal* or lesion* or atypical or squamous)).ti,ab. (13,430)

7. (cervix or cervic*).ti,ab. (227,836)

8. (LSIL or HSIL or ASCUS or ASC-US or ASC-H or ASC).ti,ab. (8776)

9. ((intraepithelial or intra-epithelial) adj2 lesion$).ti,ab. (4844)

10. (atypical adj2 squamous).ti,ab. (1989)

11. 8 or 9 or 10 (12,051)

12. 7 and 11 (5015)

13. (CIN or CIN1* or CIN2* or CIN3* or CIN 1* or CIN 2* or CIN 3* or CIN I* or CIN II* or CIN III* or CINI* or CINII* or CINIII* or CGIN*).ti,ab. (10,256)

14. 1 or 2 or 3 or 4 or 5 or 6 or 12 or 13 (114,337)

15. exp Papillomavirus Infections/ (28,658)

16. Papillomaviridae/ (21,795)

17. exp Alphapapillomavirus/ (6098)

18. (human* adj2 (papillomavirus* or papillomaviridae or papilloma virus*)).ti,ab. (33,953)

19. (HPV* or hrHPV* or hr-HPV*).ti,ab. (35,059)

20. or/15-19 (53,740)

21. Vaginal Smears/ (21,441)

22. Papanicolaou Test/ (5902)

23. Cytological Techniques/ (10,199)

24. Cytodiagnosis/ (15,335)

25. Mass Screening/ (92,108)

26. “Early Detection of Cancer”/ (15,713)

27. DNA Probes, HPV/ (1080)

28. Human Papillomavirus DNA Tests/ (340)

29. ((vagina* or pap or papanicolaou) adj2 smear*).ti,ab. (9876)

30. ((pap or papanicolaou) adj2 (test* or analys* or screen*)).ti,ab. (4980)

31. cytolog*.ti,ab. (85,141)

32. or/21-31 (214,406)

33. 14 and 32 (26,028)

34. 20 and 32 (10,837)

35. (screen* adj3 (cervic* or cervix)).ti,ab. (9587)

36. ((cervic* or cervix) adj2 smear$).ti,ab. (4102)

37. ((HPV* or hrHPV* or hr-HPV*) adj4 (screen* or test* or detect* or triage*)).ti,ab. (11,312)

38. (human* adj2 (papillomavirus* or papillomaviridae or papilloma virus*) adj2 (screen* or test* or detect* or triage*)).ti,ab. (3300)

39. 35 or 36 or 37 or 38 (22,413)

40. 33 or 34 or 39 (36,627)

41. systematic$ review$.ti,ab. (103,638)

42. meta-analysis as topic/ (15,759)

43. meta-analytic$.ti,ab. (5302)

44. meta-analysis.ti,ab,pt. (116,352)

45. metanalysis.ti,ab. (157)

46. metaanalysis.ti,ab. (1389)

47. meta analysis.ti,ab. (94,402)

48. meta-synthesis.ti,ab. (524)

49. metasynthesis.ti,ab. (231)

50. meta synthesis.ti,ab. (524)

51. meta-regression.ti,ab. (4549)

52. metaregression.ti,ab. (442)

53. meta regression.ti,ab. (4549)

54. (synthes$ adj3 literature).ti,ab. (2221)

55. (synthes$ adj3 evidence).ti,ab. (6509)

56. integrative review.ti,ab. (1692)

57. data synthesis.ti,ab. (9772)

58. (research synthesis or narrative synthesis).ti,ab. (1591)

59. (systematic study or systematic studies).ti,ab. (9808)

60. (systematic comparison$ or systematic overview$).ti,ab. (2592)

61. evidence based review.ti,ab. (1694)

62. comprehensive review.ti,ab. (10,438)

63. critical review.ti,ab. (13,445)

64. quantitative review.ti,ab. (583)

65. structured review.ti,ab. (641)

66. realist review.ti,ab. (158)

67. realist synthesis.ti,ab. (118)

68. or/41-67 (239,126)

69. review.pt. (2,263,518)

70. medline.ab. (84,579)

71. pubmed.ab. (65,446)

72. cochrane.ab. (51,941)

73. embase.ab. (54,367)

74. cinahl.ab. (17,283)

75. psyc?lit.ab. (937)

76. psyc?info.ab. (17,111)

77. (literature adj3 search$).ab. (41,423)

78. (database$ adj3 search$).ab. (39,606)

79. (bibliographic adj3 search$).ab. (1816)

80. (electronic adj3 search$).ab. (14,770)

81. (electronic adj3 database$).ab. (18,516)

82. (computeri?ed adj3 search$).ab. (3229)

83. (internet adj3 search$).ab. (2468)

84. included studies.ab. (13,602)

85. (inclusion adj3 studies).ab. (10,824)

86. inclusion criteria.ab. (57,533)

87. selection criteria.ab. (25,429)

88. predefined criteria.ab. (1537)

89. predetermined criteria.ab. (904)

90. (assess$ adj3 (quality or validity)).ab. (58,471)

91. (select$ adj3 (study or studies)).ab. (51,767)

92. (data adj3 extract$).ab. (43,710)

93. extracted data.ab. (10,058)

94. (data adj2 abstracted).ab. (4252)

95. (data adj3 abstraction).ab. (1226)

96. published intervention$.ab. (143)

97. ((study or studies) adj2 evaluat$).ab. (145,298)

98. (intervention$ adj2 evaluat$).ab. (8561)

99. confidence interval$.ab. (314,381)

100. heterogeneity.ab. (125,402)

101. pooled.ab. (65,443)

102. pooling.ab. (9876)

103. odds ratio$.ab. (205,883)

104. (Jadad or coding).ab. (150,343)

105. or/70-104 (1,105,052)

106. 69 and 105 (179,404)

107. review.ti. (35,4575)

108. 107 and 105 (85,749)

109. (review$ adj4 (papers or trials or studies or evidence or intervention$ or evaluation$)).ti,ab. (142,763)

110. 68 or 106 or 108 or 109 (418,128)

111. letter.pt. (964,951)

112. editorial.pt. (434,093)

113. comment.pt. (685,970)

114. 111 or 112 or 113 (1,570,204)

115. 110 not 114 (408,039)

116. exp animals/ not humans/ (4,364,879)

117. 115 not 116 (396,895)

118. 40 and 117 (921)

119. limit 118 to yr=“2014 -Current” (267)

Search strategy

1. #1. MeSH descriptor: [Uterine Cervical Neoplasms] this term only (1975)

2. #2. MeSH descriptor: [Cervical Intraepithelial Neoplasia] this term only (518)

3. #3. MeSH descriptor: [Uterine Cervical Dysplasia] explode all trees (129)

4. #4. MeSH descriptor: [Cervix Uteri] this term only (1045)

5. #5. ((cervix or cervic*) near/3 (cancer* or neoplas* or carcinoma* or adenocarcinoma* or tumour* or tumor* dysplas* or dyskaryo* or precancer* or pre-cancer*)):ti,ab,kw (3701)

6. #6. ((cervix or cervic*) near/3 (abnormal* or lesion* or atypical or squamous)):ti,ab,kw (725)

7. #7. (cervix or cervic*):ti,ab,kw (13509)

8. #8. (LSIL or HSIL or ASCUS or ASC-US or ASC-H or ASC):ti,ab,kw (383)

9. #9. ((intraepithelial or intra-epithelial) near/2 lesion*):ti,ab,kw (217)

10. #10. (atypical near/2 squamous):ti,ab,kw (118)

11. #11. #8 or #9 or #10 (521)

12. #12. #7 and #11 (270)

13. #13. (CIN or CIN1* or CIN2* or CIN3* or CIN 1* or CIN 2* or CIN 3* or CIN I* or CIN II* or CIN III* or CINI* or CINII* or CINIII* or CGIN*):ti,ab,kw (1165)

14. #14. #1 or #2 or #3 or #4 or #5 or #6 or #12 or #13 (5623)

15. #15. MeSH descriptor: [Papillomavirus Infections] explode all trees (1107)

16. #16. MeSH descriptor: [Papillomaviridae] this term only (419)

17. #17. MeSH descriptor: [Alphapapillomavirus] explode all trees (220)

18. #18. (human* near/2 (papillomavirus* or papillomaviridae or papilloma next virus*)):ti,ab,kw (1353)

19. #19. (HPV* or hrHPV* or hr-HPV*):ti,ab,kw (1438)

20. #20. #15 or #16 or #17 or #18 or #19 (2137)

21. #21. MeSH descriptor: [Vaginal Smears] explode all trees (802)

22. #22. MeSH descriptor: [Papanicolaou Test] this term only (228)

23. #23. MeSH descriptor: [Cytological Techniques] this term only (82)

24. #24. MeSH descriptor: [Cytodiagnosis] this term only (120)

25. #25. MeSH descriptor: [Mass Screening] this term only (4758)

26. #26. MeSH descriptor: [Early Detection of Cancer] this term only (955)

27. #27. MeSH descriptor: [DNA Probes, HPV] this term only (16)

28. #28. MeSH descriptor: [Human Papillomavirus DNA Tests] this term only (8)

29. #29. ((vagina* or pap or papanicolaou) near/2 (smear*)):ti,ab,kw (1139)

30. #30. ((pap or papanicolaou) near/2 (test* or analys* or screen*)):ti,ab,kw (581)

31. #31. cytolog*:ti,ab,kw (2751)

32. #32. #21 or #22 or #23 or #24 or #25 or #26 or #27 or #28 or #29 or #30 or #31 (8679)

33. #33. #14 and #32 (1330)

34. #34. #20 and #32 (666)

35. #35. (screen* near/2 (cervic* or cervix)):ti,ab,kw (730)

36. #36. ((cervic* or cervix) near/2 smear*):ti,ab,kw (189)

37. #37. ((HPV* or hrHPV* or hr-HPV*) near/4 (screen* or test* or detect* or triage*)):ti,ab,kw (519)

38. #38. (human* near/2 (papillomavirus* or papillomaviridae or papilloma next virus*) near/2 (screen* or test* or detect* or triage*)):ti,ab,kw (240)

39. #39. #35 or #36 or #37 or #38 (1211)

40. #40. #33 or #34 or #39 (1740)

Search strategy

1. MeSH DESCRIPTOR Uterine Cervical Neoplasms (540)

2. MeSH DESCRIPTOR Cervical Intraepithelial Neoplasia (136)

3. MeSH DESCRIPTOR Uterine Cervical Dysplasia EXPLODE ALL TREES (22)

4. MeSH DESCRIPTOR Cervix Uteri (89)

5. (((cervix or cervic*) ADJ3 (cancer* or neoplas* or carcinoma* or adenocarcinoma* or tumour* or tumor* dysplas* or dyskaryo* or precancer* or pre-cancer*))) (693)

6. (((cancer* or neoplas* or carcinoma* or adenocarcinoma* or tumour* or tumor* dysplas* or dyskaryo* or precancer* or pre-cancer*) ADJ3 (cervix or cervic*))) (168)

7. (((cervix or cervic*) ADJ3 (abnormal* or lesion* or atypical or squamous))) (60)

8. (((abnormal* or lesion* or atypical or squamous) ADJ3 (cervix or cervic*))) (43)

9. ((cervix or cervic*)) AND (LSIL or HSIL or ASCUS or ASC-US or ASC-H or ASC) (41)

10. ((cervix or cervic*)) (1481)

11. (((intraepithelial or intra-epithelial) ADJ2 lesion*)) (66)

12. ((lesion* ADJ2 (intraepithelial or intra-epithelial))) (0)

13. ((atypical ADJ2 squamous)) (36)

14. ((squamous ADJ2 atypical)) (1)

15. #11 OR #12 OR #13 OR #14 (76)

16. #10 AND #15 (73)

17. ((CIN or CIN1* or CIN2* or CIN3* or CIN 1* or CIN 2* or CIN 3* or CIN I* or CIN II* or CIN III* or CINI* or CINII* or CINIII* or CGIN*)) (111)

18. #1 OR #2 OR #3 OR #4 OR #5 OR #6 OR #7 OR #8 OR #9 OR #16 OR #17 (801)

19. MeSH DESCRIPTOR Papillomavirus Infections EXPLODE ALL TREES (283)

20. MeSH DESCRIPTOR Papillomaviridae (114)

21. MeSH DESCRIPTOR Alphapapillomavirus EXPLODE ALL TREES (54)

22. ((human* ADJ2 (papillomavirus* or papillomaviridae or papilloma virus*))) (304)

23. (((papillomavirus* or papillomaviridae or papilloma virus*) ADJ2 human*)) (35)

24. (HPV* or hrHPV* or hr-HPV*) (259)

25. #19 OR #20 OR #21 OR #22 OR #23 OR #24 (409)

26. MeSH DESCRIPTOR Vaginal Smears EXPLODE ALL TREES (213)

27. MeSH DESCRIPTOR Papanicolaou Test (56)

28. MeSH DESCRIPTOR Cytological Techniques (34)

29. MeSH DESCRIPTOR Cytodiagnosis (40)

30. MeSH DESCRIPTOR Mass Screening (2100)

31. MeSH DESCRIPTOR Early Detection of Cancer (273)

32. MeSH DESCRIPTOR DNA Probes, HPV (6)

33. MeSH DESCRIPTOR Human Papillomavirus DNA Tests (6)

34. (((vagina* or pap or papanicolaou) ADJ2 smear*)) (258)

35. ((smear* ADJ2 (vagina* or pap or papanicolaou))) (10)

36. (((pap or papanicolaou) ADJ2 (test* or analys* or screen*))) (128)

37. (((test* or analys* or screen*) ADJ2 (pap or papanicolaou))) (63)

38. (cytolog*) (483)

39. #26 OR #27 OR #28 OR #29 OR #30 OR #31 OR #32 OR #33 OR #34 OR #35 OR #36 OR #37 OR #38 (2727)

40. #18 AND #39 (393)

41. #25 AND #39 (233)

42. ((screen* ADJ3 (cervic* or cervix))) (142)

43. (((cervic* or cervix) ADJ3 screen*)) (264)

44. (((cervic* or cervix) ADJ3 smear*)) (81)

45. ((smear* ADJ2 (cervic* or cervix))) (18)

46. (((HPV* or hrHPV* or hr-HPV*) ADJ4 (screen* or test* or detect* or triage*))) (135)

47. (((screen* or test* or detect* or triage*) ADJ4 (HPV* or hrHPV* or hr-HPV*))) (92)

48. ((papillomavirus* or papillomaviridae or papilloma virus*) ADJ2 (screen* or test* or detect* or triage*)) (115)

49. (((screen* or test* or detect* or triage*) ADJ2 (papillomavirus* or papillomaviridae or papilloma virus*))) (75)

50. (human) (3164)

51. #48 OR #49 (144)

52. #50 AND #51 (123)

53. #42 OR #43 OR #44 OR #45 OR #46 OR #47 OR #52 (392)

54. #40 OR #41 OR #53 (472)

55. (*) IN DARE (45,418)

56. #54 AND #55 (128)

57. (*) IN HTA (16,846)

58. #54 AND #57 (108)

Search strategy

1. exp uterine cervix tumor/ (101,917)

2. exp uterine cervix dysplasia/ (5017)

3. exp uterine cervix/ (28,075)

4. ((cervix or cervic*) adj3 (cancer* or neoplas* or carcinoma* or adenocarcinoma* or tumour* or tumor* dysplas* or dyskaryo* or precancer* or pre-cancer*)).ti,ab. (81,511)

5. ((cervix or cervic*) adj3 (abnormal* or lesion* or atypical or squamous)).ti,ab. (17,047)

6. (cervix or cervic*).ti,ab. (279,094)

7. (LSIL or HSIL or ASCUS or ASC-US or ASC-H or ASC).ti,ab. (12,691)

8. ((intraepithelial or intra-epithelial) adj2 lesion$).ti,ab. (6060)

9. (atypical adj2 squamous).ti,ab. (2491)

10. 7 or 8 or 9 (16,616)

11. 6 and 10 (6886)

12. (CIN or CIN1* or CIN2* or CIN3* or CIN 1* or CIN 2* or CIN 3* or CIN I* or CIN II* or CIN III* or CINI* or CINII* or CINIII* or CGIN*).ti,ab. (14,461)

13. 1 or 2 or 3 or 4 or 5 or 11 or 12 (147,667)

14. exp papillomavirus infection/ (25,629)

15. papillomaviridae/ (816)

16. exp alphapapillomavirus/ (12,821)

17. wart virus/ (37,172)

18. (human* adj2 (papillomavirus* or papillomaviridae or papilloma virus*)).ti,ab. (39,987)

19. (HPV* or hrHPV* or hr-HPV*).ti,ab. (45,330)

20. 14 or 15 or 16 or 17 or 18 or 19 (71,133)

21. vagina smear/ (10,593)

22. papanicolaou test/ (15,851)

23. uterine cervix cytology/ (12,091)

24. cytology/ (382,601)

25. cancer screening/ (66,243)

26. early cancer diagnosis/ (1191)

27. Human papillomavirus DNA test/ (1518)

28. DNA probe/ (27,048)

29. screening test/ (65,290)

30. diagnostic accuracy/ (217,693)

31. diagnostic test accuracy study/ (75,141)

32. ((vagina* or pap or papanicolaou) adj2 smear*).ti,ab. (11,672)

33. ((pap or papanicolaou) adj2 (test* or analys* or screen*)).ti,ab. (6474)

34. cytolog*.ti,ab. (105,880)

35. 21 or 22 or 23 or 24 or 25 or 26 or 27 or 28 or 29 or 30 or 31 or 32 or 33 or 34 (854,789)

36. 13 and 35 (36,343)

37. 20 and 35 (17,008)

38. (screen* adj3 (cervic* or cervix)).ti,ab. (11,964)

39. ((cervic* or cervix) adj2 smear$).ti,ab. (4794)

40. ((HPV* or hrHPV* or hr-HPV*) adj4 (screen* or test* or detect* or triage*)).ti,ab. (14,930)

41. (human* adj2 (papillomavirus* or papillomaviridae or papilloma virus*) adj2 (screen* or test* or detect* or triage*)).ti,ab. (3861)

42. 38 or 39 or 40 or 41 (28,042)

43. 36 or 37 or 42 (49,894)

44. systematic$ review$.ti,ab. (127,491)

45. systematic$ literature review$.ti,ab. (9255)

46. “systematic review”/ (159,479)

47. “systematic review (topic)”/ (28,244)

48. meta analysis/ (161,820)

49. “meta analysis (topic)”/ (39,256)

50. meta-analytic$.ti,ab. (5990)

51. meta-analysis.ti,ab. (121,194)

52. metanalysis.ti,ab. (390)

53. metaanalysis.ti,ab. (5712)

54. meta analysis.ti,ab. (121,194)

55. meta-synthesis.ti,ab. (482)

56. metasynthesis.ti,ab. (226)

57. meta synthesis.ti,ab. (482)

58. meta-regression.ti,ab. (5717)

59. metaregression.ti,ab. (735)

60. meta regression.ti,ab. (5717)

61. (synthes$ adj3 literature).ti,ab. (2538)

62. (synthes$ adj3 evidence).ti,ab. (7290)

63. (synthes$ adj2 qualitative).ti,ab. (1370)

64. integrative review.ti,ab. (1388)

65. data synthesis.ti,ab. (11,134)

66. (research synthesis or narrative synthesis).ti,ab. (1581)

67. (systematic study or systematic studies).ti,ab. (10,580)

68. (systematic comparison$ or systematic overview$).ti,ab. (2810)

69. (systematic adj2 search$).ti,ab. (19,687)

70. systematic$ literature research$.ti,ab. (218)

71. (review adj3 scientific literature).ti,ab. (1419)

72. (literature review adj2 side effect$).ti,ab. (12)

73. (literature review adj2 adverse effect$).ti,ab. (2)

74. (literature review adj2 adverse event$).ti,ab. (12)

75. (evidence-based adj2 review).ti,ab. (3042)

76. comprehensive review.ti,ab. (12,025)

77. critical review.ti,ab. (14,564)

78. critical analysis.ti,ab. (7278)

79. quantitative review.ti,ab. (653)

80. structured review.ti,ab. (841)

81. realist review.ti,ab. (141)

82. realist synthesis.ti,ab. (95)

83. (pooled adj2 analysis).ti,ab. (13,908)

84. (pooled data adj6 (studies or trials)).ti,ab. (2176)

85. (medline and (inclusion adj3 criteria)).ti,ab. (17,624)

86. (search adj (strateg$ or term$)).ti,ab. (27,662)

87. or/44-86 (397,542)

88. medline.ab. (101,252)

89. pubmed.ab. (83,024)

90. cochrane.ab. (65,587)

91. embase.ab. (67,292)

92. cinahl.ab. (19,062)

93. psyc?lit.ab. (977)

94. psyc?info.ab. (15,707)

95. lilacs.ab. (5248)

96. (literature adj3 search$).ab. (51,495)

97. (database$ adj3 search$).ab. (48,637)

98. (bibliographic adj3 search$).ab. (2080)

99. (electronic adj3 search$).ab. (17,339)

100. (electronic adj3 database$).ab. (24,395)

101. (computeri?ed adj3 search$).ab. (3688)

102. (internet adj3 search$).ab. (3184)

103. included studies.ab. (16,745)

104. (inclusion adj3 studies).ab. (13,046)

105. inclusion criteria.ab. (94,570)

106. selection criteria.ab. (27,646)

107. predefined criteria.ab. (2021)

108. predetermined criteria.ab. (1098)

109. (assess$ adj3 (quality or validity)).ab. (74,945)

110. (select$ adj3 (study or studies)).ab. (65,709)

111. (data adj3 extract$).ab. (57,099)

112. extracted data.ab. (12,500)

113. (data adj2 abstracted).ab. (6653)

114. (data adj3 abstraction).ab. (1741)

115. published intervention$.ab. (167)

116. ((study or studies) adj2 evaluat$).ab. (198,404)

117. (intervention$ adj2 evaluat$).ab. (11,277)

118. confidence interval$.ab. (367,946)

119. heterogeneity.ab. (154,328)

120. pooled.ab. (88,249)

121. pooling.ab. (12,583)

122. odds ratio$.ab. (252,412)

123. (Jadad or coding).ab. (172,491)

124. evidence-based.ti,ab. (104,203)

125. or/88-124 (1,482,382)

126. review.pt. (2,263,944)

127. 125 and 126 (181,834)

128. review.ti. (404,112)

129. 125 and 128 (105,785)

130. (review$ adj10 (papers or trials or trial data or studies or evidence or intervention$ or evaluation$ or outcome$ or findings)).ti,ab. (413,152)

131. (retriev$ adj10 (papers or trials or studies or evidence or intervention$ or evaluation$ or outcome$ or findings)).ti,ab. (21,339)

132. 87 or 127 or 129 or 130 or 131 (776,417)

133. letter.pt. (983,216)

134. editorial.pt. (537,824)

135. 133 or 134 (1,521,040)

136. 132 not 135 (761,183)

137. (animal/ or nonhuman/) not exp human/ (5,103,479)

138. 136 not 137 (735,678)

139. 43 and 138 (2214)

140. limit 139 to yr=“2014 -Current” (676)