Using novel biomarkers to triage young adult women with minor cervical lesions: a cost‐effectiveness analysis

To evaluate the short‐term consequences and cost‐effectiveness associated with the use of novel biomarkers to triage young adult women with minor cervical cytological lesions.


Introduction
Recent advances in cervical cancer screening technologies have prompted decision makers worldwide to evaluate the use of novel screening diagnostics to reduce the incidence of cervical cancer. Following a decade of randomised controlled trials, several countries have begun to shift primary screening from Pap smears (i.e. cytology) to testing for the presence of human papillomavirus (HPV), 1,2 the necessary cause of cervical cancer. 3 As HPV testing may be less appropriate for younger women because of the high prevalence of transient HPV infections, 4 many countries (including Norway) will continue to use cytology as the primary screening method (every 3 years), particularly for younger women not vaccinated against HPV. A challenge within the current cytology-based screening programme is the This article includes Author Insights, a video abstract available at https://vimeo.com/rcog/authorinsights14135. management of women with minor cervical cytological lesions [i.e. atypical squamous cells of undetermined significance (ASC-US) or low-grade intraepithelial lesion (LSIL)], who are at an elevated risk of progressing to highgrade precancer [i.e. cervical intraepithelial neoplasia grade 2 or worse (CIN2+)] within a short period of time. 5,6 In many settings, however, the detection of minor cervical cytological lesions is not deemed sufficient risk to justify a direct referral to diagnostic colposcopy and biopsy, a procedure that may cause anxiety, pain, bleeding, or discharge. 7 Therefore, intensified surveillance (i.e. triage, often involving repeat cytology and/or HPV DNA testing) is recommended for the appropriate clinical management of these women.
In line with international cervical cancer screening guidelines, 8 a triage algorithm should balance the detection of cervical precancers, resource use, and potential harms to women following diagnostic work-up. Candidate biomarkers to triage women with minor cervical cytological lesions include HPV DNA tests, HPV-16/HPV-18 DNA genotyping (herein referred to as genotyping), HPV viral messenger ribonucleic acid (mRNA) transcripts of E6/E7 proteins (herein referred to as HPV mRNA tests), and p16/Ki67 dual staining (herein referred to as dual staining). 9 These biomarkers vary in diagnostic accuracy (i.e. the sensitivity and specificity for detecting high-grade precancer), and directly influence the number of false-positive and falsenegative results, yet the trade-offs associated with each biomarker remains unknown. Our objective was to quantify the short-term (i.e. within one screening round) health and economic outcomes associated with a set of candidate strategies involving novel biomarkers to triage younger women (aged 25-33 years) with primary ASC-US/LSIL results, in order to inform key stakeholders in Norway about the trade-offs associated with improving cervical cancer screening for these women.

Analytic overview
We expanded a previously developed decision-tree model to include triage screening strategies that involved novel biomarkers (i.e. HPV mRNA testing, genotyping, and dual staining) in addition to the existing strategies that involved HPV DNA testing (Tables S1 and S2 in Appendix S1). 10 The model simulates a cohort of women with an index cytology result of ASC-US or LSIL through a single screening round, including follow-up (i.e. a total of 3 years). We restricted our analysis to women aged 25-33 years, as this group of younger women is unlikely to be recommended primary HPV DNA testing under the new Norwegian screening guidelines. 11 Similar age-specific changes are expected in other countries. 2 Model development, assumptions, calibration, and validation have been described elsewhere. 10 We compared the health and economic outcomes of 13 alternative triage strategies that varied with respect to: the triage test (i.e. HPV DNA testing with and without genotyping, HPV mRNA testing for five and 14 genotypes, and dual staining), immediate or 12month delayed triage, and criteria for returning a woman to a routine screening schedule. Model outcomes included the number of detected CIN2+, colposcopy-directed biopsies, physician consultations, and total cost per woman (aged 25-33 years) detected with an ASC-US/LSIL result at the primary screening test.
In addition to the model capturing cervical cancerrelated outcomes, we developed a companion submodel that projects the impact of detecting a precancer (i.e. CIN2 and CIN3) on changes in preterm birth rates before 28 weeks of gestation following the loop electrosurgical excision procedure (LEEP). To project changes in preterm births, we used data from the Medical Birth Registry of Norway on the current proportion of preterm births in the total population (i.e. 0.44% of all births), 12 and assumed this proportion reflects the preterm birth rate expected under the current screening algorithm. To calculate the baseline risk of preterm birth in the absence of screening, we adjusted for the prevalence of women with previous conisation in Norway, resulting in a risk of preterm births among unexposed women of 0.40%. We applied the relative risk of having a preterm birth following LEEP from a recently published meta-analysis of 2.33 (95% confidence interval: 1.84-2.94; details are available in Appendix S1, section 5.4). 13 To identify cost-efficient screening algorithms we estimated the incremental cost-effectiveness ratio (ICER), calculated as the additional monetary costs required to detect one additional precancer compared with the next most costly strategy. Strategies were considered cost-efficient if either more precancers were detected at lower costs or more precancers were detected and the ICER was favourable compared with more effective strategies. We excluded strategies that provided less health benefits while simultaneously requiring higher costs or that had a higher ICER. For the cost-efficient strategies, we quantified the changes in physician consultations and colposcopy referrals in order to identify capacity requirements and the burden placed on women attending screening. Whether strategies provide 'good value' for resource use (i.e. are cost-effective) should be based on the societal willingness to pay for additional health benefits. There is no threshold value for what constitutes 'good value' for resource use in terms of CIN2+ detection in Norway; therefore, we used the threshold value projected from the current Norwegian strategy as a proxy for the amount that society is willing to pay to detect a CIN2+. We considered the triage strategy that provided a cost per additional detected precancer just under that of current Norwegian guidelines to be cost-effective, but acknowledge that this may underestimate society's true willingness to pay. In sensitivity analysis, we identified efficient triage strategies in terms of non-monetary resource use (i.e. additional physician consultations or colposcopy referrals per additional CIN2+ detected).
Data to inform epidemiological and cost inputs to the model were provided by the Cancer Registry of Norway, 14 the University Hospital of North Norway (Sveinung W. Sørbye, personal communication), Norwegian fee schedules, 15,16 as well as primary data on HPV mRNA testing for five genotypes in Norway for the years 2003-2004 (Finn Egil Skjeldestad, personal communication). To account for the societal costs associated with alternative algorithms, we included patient time and travel costs. Due to the short analytic time frame, we did not discount costs or health benefits; however, we varied this assumption in one-way sensitivity analysis (i.e. changing one model input at a time while holding all other inputs constant, see details in Appendix S1). We also assessed uncertainty around all model inputs simultaneously (i.e. probabilistic sensitivity analysis, see details in Appendix S1).

Screening strategies
The Norwegian Cervical Cancer Screening Program recently altered the triage algorithm for women with an ASC-US/ LSIL result; therefore, we compared two Norwegian screening triage approaches (i.e. the current and former triage strategies) with 11 candidate triage strategies for younger women with ASC-US/LSIL ( Figure 1; additional details are available in Table S1). The current Norwegian strategy, implemented in July 2014, involves reflex HPV DNA testing with delayed repeat co-testing (cytology and HPV DNA testing) at 12 months for women who are HPV DNA-positive (strategy 1a, herein referred to as 'current guidelines'). The former screening algorithm (strategy 1b, herein referred to as 'former guidelines'), for which most empirical data are available, involves co-testing at 12 months for women with ASC-US/LSIL, with additional follow-up determined by the co-test result. Candidate strategies (strategies 2-12) included a range of alternative algorithms that were broadly classified by triage test method: reflex HPV DNA testing (strategies 2-6), reflex HPV mRNA testing (strategies 7-9), reflex HPV DNA testing with sequence HPV mRNA testing for women who tested positive for HPV DNA (strategy 10), reflex HPV DNA testing with genotyping of HPV-16 and HPV-18 (strategy 11), and dual staining (strategy 12).
Strategies 3-6 were outlined in collaboration with the Cancer Registry of Norway, which manages the Norwegian Cervical Cancer Screening Program, and were included in our previous analysis. 10 For the current analysis, we included seven additional triage strategies that involved additional biomarkers (e.g. genotyping, HPV mRNA testing, and dual staining). These strategies were defined in collaboration with clinicians and are based on recommendations from published sources. [17][18][19] For HPV mRNA testing, we evaluated the performance of two HPV mRNA tests that differ according to the number of high-risk HPV genotypes included: five HPV genotypes (strategies 7 and 8) or 14 HPV genotypes (strategy 9). We assumed imperfect screening compliance (i.e. a proportion of women fail to attend recommended follow-up within the 3-year interval), based on observed Norwegian data. 20

Epidemiological and cost data
We used primary epidemiological data from Norway to inform key model inputs, 14,20 and supplemented model inputs with published literature and expert opinion when empirical data were unavailable (Tables 1, S3, and S6). [21][22][23] We expressed the underlying natural history of disease by allowing high-grade precancers to regress to cervical intraepithelial neoplasia grade 1 (CIN1) or negative for intraepithelial lesion or malignancy (NILM) at a monthly probability of 2%. 33 Because of the short time frame of the analysis, the model captures the composite outcome CIN2+ and does not differentiate between unique HPV genotypes. We conducted additional one-way sensitivity analyses by varying compliance rates (70 versus 100% compliance), the baseline prevalence of CIN2+ (base case plus or minus 20%), and the CIN2+ regression rate (50% slower or higher regression). We also explored the potential impact of verification bias in estimates of diagnostic accuracy using Figure 1. Screening strategies to triage women with an ASC-US or LSIL index result. The current and former cervical cancer screening strategies in Norway (strategies 1a and 1b) were compared with 11 alternative strategies (strategies 2-12) that varied the triage management for women aged 25-33 years with primary minor cervical lesions (i.e. atypical squamous cells of undetermined significance or low-grade squamous intraepithelial lesion: ASC-US/LSIL). Strategies 3-6 involved reflex HPV DNA testing and one of four combinations of pathways: a, b, c, and d. Strategy 3 = ac, strategy 4 = ad, strategy 5 = bc, and strategy 6 = bd. Strategies 7-9 involved reflex HPV mRNA testing, with either repeat cytology in 12 months for women testing negative for HPV mRNA (strategy 7, pathway a) or a return to screening (strategies 8 and 9, pathway b). The HPV mRNA test varied with respect to the number of high-risk HPV genotypes detected: either five HPV genotypes (strategies 7 and 8) or 14 HPV genotypes (strategy 9). For all strategies, we assumed that women would receive the same strategy-specific triage test to follow-up after a negative biopsy. Circles denote a chance node (dependent on diagnostic test result), and squares denote a decision node (i.e. subsequent management is conditioned on the test result). Strategies are described in detail in Table S1 in Appendix S1. ASC-H, atypical squamous cells -cannot exclude high-grade lesion; ASC-US, atypical squamous cells of undetermined significance; CIN2+, cervical intraepithelial neoplasia grade 2 or worse; HPV, human papillomavirus; HSIL, high-grade intraepithelial lesion; LSIL, lowgrade intraepithelial lesion; mRNA, messenger ribonucleic acid; neg, negative; pos, positive.
the lower bounds of test characteristics reported in metaanalyses and independent studies. 17,[28][29][30][31][32] Based on identified cost inputs from a recently published cost analysis of the Norwegian Cervical Cancer Screening Program and a cost-effectiveness study within the Norwegian context, 34,35 we updated cost input values to reflect 2014 estimates (details are available in Appendix S1, Tables S4 and S5). The unit costs of analysing a test sample for each diagnostic test were valued based on actual resource use in Norwegian pathology laboratories as well as on information from the relevant distributers (Table 1). We valued all costs in 2014 Norwegian Kroner (NOK) and converted this to Euros (€1.00 = NOK8.35). 36 To capture uncertainty in the cost estimates, we varied the mean cost estimates by plus or minus 20% in the probabilistic sensitivity analysis. In addition, we varied the patient time and travel costs by plus or minus 50% of the base-case estimate in the one-way sensitivity analysis.

Health and economic trade-offs
The current and former Norwegian triage strategies detected fewer precancers and were more costly compared with the candidate triage algorithms (i.e. were strongly dominated); the strategies were subsequently excluded from further analysis. Five out of the remaining 11 strategies were identified as efficient (ICER range: €2978-33 095), including: two strategies involving reflex HPV mRNA testing with five or 14 genotypes (strategies 8 and 9); two strategies that involved reflex HPV DNA testing with immediate colposcopy for women testing positive for HPV; and one strategy that involved the differential management of ASC-US/ LSIL, referring all women with an LSIL result directly to colposcopy (strategy 2) ( Figure S1; Table 2). When we used the cost per CIN2+ detected, calculated using the current Norwegian triage strategy, as a proxy for value for money (i.e. €4195), the preferred strategy, i.e. the strategy that fell just below the willingness-to-pay threshold, involved triaging younger women with the HPV mRNA 14 genotype test (strategy 9). For a moderate increase in the willingness-topay threshold, the preferred strategy involved HPV DNA testing with direct colposcopy referral for all women testing positive for HPV (i.e. €5522 per additional CIN2+, strategy 4), whereas referring all women with an LSIL result directly to colposcopy (strategy 2) would require a willingness to pay of more than seven times what is currently considered as good value for money.
Compared with the current Norwegian guidelines (strategy 1a), the five strategies identified as cost-efficient were projected to increase CIN2+ detection, but would also require additional resources, particularly in terms of the number of colposcopy referrals (Figure 2). For example, our model projected that strategy 8 (HPV mRNA testing for five genotypes) would detect 18% more cases of CIN2+ at a 16% lower monetary cost than the current strategy, but would increase colposcopy referrals by 14%, compared with the current Norwegian triage algorithm. Similarly, the HPV mRNA testing for 14 genotypes (strategy 9), the preferred strategy given the current willingness-to-pay threshold, was also more effective (44% increased CIN2+ detection), but increased costs by just 3% compared with current guidelines. Importantly, both HPV mRNA strategies required fewer physician consultations compared with current guidelines. In comparison, strategies involving HPV DNA testing yielded the most benefit in terms of CIN2+ detection, but would increase the monetary costs, number of physician consultations, and colposcopy referrals, compared with current levels. For example, strategy 4, involving immediate colposcopy for women testing positive for HPV and returning women testing negative for HPV to routine screening, detected 52% more precancers, but increased monetary costs by 13%, the number of consultations by 8%, and the number of colposcopies by 88%. When we allowed for the differential management of ASC-US and LSIL (strategy 2), we found this strategy to be the most effective in terms of CIN2+ detection, but it also demanded the most resources. Using a simplified approach to extrapolate the consequences of changes in the precancer detection rate beyond the initial screening round, we found that the cost-efficient strategies may be associated with an increase in the number of preterm births before 28 weeks of gestation. For example, we projected that approximately 1.8-10.3 additional preterm births per 10 000 women with an index result of ASC-US or LSIL are expected, compared with current levels (Table S12).

Sensitivity analysis
In one-and multi-way sensitivity analyses, the rank order of the efficient strategies was moderately influenced by reasonable changes in the diagnostic accuracy of the different tests, as well as the regression of precancers (Tables S10 and S11 in Appendix S1). For example, strategy 9 (i.e. HPV mRNA testing for 14 genotypes) was no longer considered cost-efficient when we reduced the sensitivity of the HPV mRNA 14 genotype test for detecting CIN2+, or when we simultaneously reduced the diagnostic sensitivity of all tests. Among strategies that were not originally identified as being cost-efficient in our base-case analysis, only strategy 12 (i.e. dual staining) became cost-efficient under sensitivity analysis scenarios [i.e. when we reduced the test sensitivity of all biomarkers simultaneously (including dual staining), or reduced the test sensitivity of the HPV mRNA 14 genotype test]. When we identified efficient strategies using non-monetary resource use (i.e. additional physician consultations or colposcopy referrals per additional CIN2+ detected), strategies 4 and 9 were no longer considered efficient (Tables S7 and S8 in Appendix S1). The values of the ICERs were reasonably robust to discounting (Table S9) and changes in model input values (Tables S10 and S11). In all scenarios, strategy 9 was consistently below the current willingness-to-pay threshold, whereas strategy 2 would never be considered cost-effective unless the willingness-to-pay per additional CIN2+ detected was at least three times greater than the current level. Finally, strategy 9 had the highest probability of being cost-effective for a range of plausible willingness-to-pay thresholds (i.e. €3500-7000 per CIN2+ detected), whereas strategy 3 was most likely to be cost-effective when the willingness-to-pay threshold exceeded €7000 per additional CIN2+ detected ( Figure S2).

Main findings
In primary cytology-based cervical cancer screening, incorporating novel biomarkers to triage young adult women with ASC-US/LSIL have the potential to improve the current screening algorithm in Norway. We used a decisionanalytic model to identify five efficient strategies that were more effective in terms of precancer detection, and provided greater value for both monetary and non-monetary resource use, than current Norwegian guidelines. Two of the efficient strategies involved HPV mRNA testing (detecting either five or 14 genotypes), two involved HPV DNA testing with immediate colposcopy for women testing positive for HPV, and one involved the differential management of ASC-US and LSIL. The strategies involving HPV mRNA testing reflect 'good value for money' in terms of the cost per additional CIN2+ detected, compared with what Norwegian decision-makers are currently willing to pay. In comparison, HPV DNA-based strategies detected the most precancers, but required a moderate increase in resource use, especially in the number of colposcopy referrals.

Strengths and limitations
The model used in this analysis projects the number of detected precancers and does not consider long-term end points, yet the optimal strategy in terms of preventing cervical cancer from developing will depend on the degree to which precancers progress to cancer or spontaneously regress. However, CIN2+ is the treatment threshold and is therefore an important endpoint to consider, both in terms of resource requirements as well as potential harms to women. The exact capacity constraints within the Norwegian healthcare system were not available, thus we cannot know whether a given strategy will exceed the *The incremental cost-effectiveness ratio (ICER) represents the additional costs per additional CIN2+ detected of a strategy compared with the next most effective strategy. Strategies requiring higher incremental costs for lower incremental CIN2+ detected were classified as 'strongly dominated', whereas strategies requiring a higher ICER than the next, more effective strategy, were classified as 'weakly dominated', and were excluded from further consideration. Strategies in bold represent cost-efficient strategies. available resources in the short term. Analyses that can quantify expected changes in resource use associated with a given triage strategy may help to inform the availability of required resources as well as the implementation process. Despite the uncertainty surrounding the potential for progression of a detected precancer, quantifying the number of diagnostic procedures required to detect a precancer, the treatment threshold, highlights an important trade-off in screening triage. The model does not distinguish between HPV genotypes; therefore, the regression rate of precancers is independent of underlying HPV status. When we investigated the impact of CIN2+ regression in sensitivity analyses, our results were only moderately influenced. Furthermore, the estimates used to inform model inputs are uncertain, particularly because of the limited number of empirical trials for HPV mRNA tests and dual staining, verification bias, and the unobserved underlying prevalence of CIN2+. We have conducted extensive literature reviews for model input values (see Appendix S1, section 4), however, and have explored uncertainty in probabilistic and deterministic sensitivity analyses, in which the ranking of strategies were robust to reasonable changes in input values. Finally, decisionmakers have yet to decide the primary screening method for vaccinated women, thus we did not explore the impact of vaccination against HPV on optimal triage management.
To our knowledge, this is among the first analyses directly comparing a wide array of alternative biomarkers and algorithms to triage women with an ASC-US/LSIL result, thus limiting our ability for cross-model validation. However, there was one recently published study that investigated the use of reflex HPV mRNA testing (five genotypes) in sequence of HPV DNA-positive, compared with reflex HPV DNA testing alone, in triage for women of all ages with an ASC-US/LSIL result. 18 This study concluded that adding the HPV mRNA test could reduce colposcopy referrals by 50% compared with HPV DNA testing alone, without significantly reducing the sensitivity of the triage algorithm. When we compared just Reflex HPV indicates re-testing a woman's initial cytology sample for HPV using an HPV test (either DNA or mRNA, depending on the strategy). Percentage change in the detection of CIN2+, total costs (EUR, €1 = NOK8.35), the number of physician consultations, and the number of colposcopies (including biopsy) performed, of each efficient strategy (with respect to monetary cost per CIN2+ detected), compared with current guidelines in Norway (strategy 1a). ASC-US, atypical squamous cells of undetermined significance; CIN2+, cervical intraepithelial neoplasia grade 2 or worse; gt, genotypes; HPV, human papillomavirus; LSIL, low-grade intraepithelial lesion; mRNA, messenger ribonucleic acid; neg, negative. these two strategies (i.e. strategies 4 and 10), our model predicted similar findings: namely, that using the HPV mRNA five genotype test in sequence for women testing positive for HPV DNA could reduce colposcopy referrals by 29%, while reducing the number of CIN2+ detected by just 9%, compared with HPV DNA testing alone. A second recently published study compared the cost-effectiveness of two candidate biomarker strategies (i.e. HPV DNA testing and HPV mRNA testing for 14 genotypes) for women of all ages within the context of the USA. 37 Their results support our findings that HPV mRNA testing may provide health benefits more efficiently than HPV DNA testing, in terms of both costs and colposcopy referrals.

Interpretation
Whereas a test with a high probability of a positive test result in women with underlying CIN2+ (i.e. sensitive tests) is beneficial in primary screening to ensure that women who are at an elevated risk of cancer are offered additional surveillance, a more specific test may be advantageous in triage to avoid unnecessary follow-up. HPV DNA tests have been characterised as so-called presence or absence tests with high diagnostic sensitivity for CIN2+, 28,38 which may result in high colposcopy referral rates if all women testing positive for HPV DNA are advised immediate colposcopy. In a triage setting, HPV mRNA tests have been reported to be more specific but less sensitive than HPV DNA tests, and may thus detect high-grade precancers and simultaneously require fewer follow-up referrals and lower costs compared with HPV DNA tests. 5,17,39 Our results indicate that strategies involving HPV mRNA testing detect fewer cases of CIN2+ than HPV DNA strategies, yet detect precancers for lower monetary and non-monetary costs. As such, a more specific screening test may benefit society and individual women alike by increasing CIN2+ detection and reducing monetary costs, without substantially increasing colposcopy referrals.
Decision analysis represents a useful tool to narrow down the number of alternative courses of action that policy makers need to consider. For this analysis, we identified efficient strategies using the total monetary cost per additional CIN2+ detected. Total monetary costs reflect the market value of all societal resources required for a given strategy; however, physical resource constraints (e.g. the physicians, gynaecologists, lab technicians, and laboratory equipment available) are also likely to play a key role for the decision makers. In our analysis, all efficient strategies required increasing the number of colposcopy referrals, yet the number of Norwegian gynaecologists is limited and reflects a physical capacity constraint in Norway. A health intervention identified as both efficient and feasible is only preferred if decision makers (and individual women alike) are willing to pay the associated monetary and non-monetary costs. We used the average cost per CIN2+ detected associated with current guidelines as a benchmark for how much decision makers are currently willing to pay for each CIN2+ detected, yet this may not represent how much society and women are in fact willing to pay to detect a case of CIN2+.

Conclusion
Incorporating HPV mRNA or DNA testing to triage younger women with minor cervical cytological lesions has the potential to improve the accuracy and effectiveness of cervical cancer screening, yet the candidate biomarkers differ substantially in the burden placed on women attending screening in terms of colposcopy referrals. Whereas triage strategies that involved HPV mRNA testing detected additional high-grade cervical precancers with just a moderate increase in colposcopy referrals, HPV DNA-based strategies (and/or referring women with LSIL directly to colposcopy) are expected to increase colposcopy referrals substantially. Additional analytic studies will be required to assess how these triage algorithms will impact the long-term outcomes, such as lifetime reduction in cervical cancer risk and lifeyears gained. Although society may be willing to pay the additional resources required for the identified cost-efficient strategies to be optimal, a better understanding of women's perceived discomforts associated with screening consultations and colposcopic examinations, including biopsy, may help to identify strategies that are optimal from the perspective of individual women.

Disclosure of interests
None declared. Completed disclosure of interests form available to view online as Supporting Information.

Contribution to authorship
All authors contributed to the conception, planning, and carrying out of the research. KP conducted the analyses. KP drafted the article, and all authors assisted with revisions. All authors accept responsibility for the article as published.

Details of ethics approval
Not applicable.

Funding
This work was funded by the University of Oslo and the Research Council of Norway (ref. no. 238042). The funders had no role in the study design, data collection and analysis, decision to publish, or preparation of the article.