We used the Policy1-Cervix platform, an extensively validated dynamic model of HPV transmission, vaccination, HPV type-specific natural history, cancer survival, screening, diagnosis and treatment,5,8,14−23 to predict outcomes for each strategy across the lifetime of females aged 10–84 who turn 30 in 2030 (born 2000) across all 78 LMIC (Appendix pp 25–26). The Policy1-Cervix model was one of three models used by the CCEMC to evaluate the impact of cervical cancer elimination targets in 78 LMIC and was reviewed and endorsed by the WHO Advisory Committee on Immunization and Vaccines related Implementation Research (IVIR-AC) for the use in modelling elimination targets for WHO. Details of the modelled approach on the calibration to 78 LMICs is described in detail in the earlier CCEMC publications.4,5 A list of each of the 78 countries included along with their GDP per-capita is described in the Appendix pp 45–48. Reporting is performed according to HPV-FRAME standards for models evaluating HPV vaccination and cervical screening46 (Appendix pp 49–51).
To ensure adequate communication between the different expert groups involved in informing the update of cervical screening and treatment guidelines, weekly meetings were held between the modelling team, representatives from the WHO Secretariat and representatives from the systematic review and costing teams. Regular meetings were also held between GDG members and the systematic review, modelling and costing teams to present and discuss the priority management algorithms and evaluation results. The modelled evaluation was performed over a three-stage process, which is described in detail in Appendix pp 25–27.
Screening strategies
We considered the benefits, harms and cost-effectiveness of seven priority screening algorithms as identified by the GDG, compared to no-screening: Primary VIA, Primary cytology with HPV DNA triage (ASC-US referral), Primary HPV DNA without triage (assessment of eligibility for ablative treatment), Primary HPV DNA with HPV16/18 triage, VIA triage, cytology triage, and colposcopy triage. Detailed management for each of these screening scenarios, including downstream management for women in follow-up, at colposcopy and after precancer treatment are described Appendix pp 41–45. Screening ages and frequencies considered for this analysis are shown in Table 2. Variations in age-ranges and frequencies considered generate a total of 19 scenarios.
Test performance
Based on updated systematic review evidence on cross-sectional sensitivity and specificity (Appendix pp 33–39), as well as test performance rates from the literature these reviews, we assumed a CIN2 + sensitivity of 94% for primary HPV DNA testing and 70% for primary cytology testing (we focussed on CIN2 + rather than CIN3 + because women with histological CIN2 + are treated in the algorithms modelled). These studies include a range of validated HPV DNA testing assays, which may target slightly different groups of HPV types, though they overlap on the most oncogenic ones.37 For VIA, test performance was based on a combination of evidence from cross-sectional studies and larger scale population-level longitudinal studies (Appendix pp 33–36). Based on this combined evidence, the GDG agreed that for VIA we would assume 40% sensitivity to CIN2 + for the base-case analysis but also consider 60% sensitivity to CIN2 + as a favourable upper bound (‘high sens’).
Screening adherence
In this normative analysis across countries, we made relatively favourable assumptions about screening and follow-up attendance, in order to predict the relative impact and cost-effectiveness of screen-and-treat strategies in LMIC in the ‘realistic best case scenario’, understanding that especially at the inception of new programs, participation is unlikely to be this high in all settings. For the base-case analysis, we assumed that 70% of women attend each routine screening visit, but that 10% would be never screeners (so the 70% are selected from the 90% of ever-screeners). We made the favourable assumption that women referred for follow-up or treatment would attend at 90% adherence if the follow-up was to occur on a later day. If same-day treatment could be offered after an HPV positive result – for instance, primary HPV with VIA triage or primary HPV without triage – we assume that a point-of-care HPV test is used 50% of the time and that 100% compliance with follow-up is achieved when the point-of-care-test is used. This results in an average of 95% of women complying with same-day treatment after primary HPV with VIA triage or primary HPV without triage. We assumed that same-day test and treatment would be available for all primary VIA scenarios and therefore made the favourable assumption that 100% adherence would be achieved in women eligible for same-day treatment after primary VIA. We assumed 90% of screen-detected cervical cancer cases would receive adequate treatment, however access to cancer treatment for symptomatically detected cancers would remain unchanged from the status-quo (rates vary by country, averaging up to 33% across all 78 LMICs). In sensitivity analysis we considered a favourable scenario in which 90% of both screen-detected and symptomatically detected cervical cancers receive adequate treatment.
Outcomes assessed
For each strategy we report on outcomes over the lifetime of unvaccinated women who would turn 30 in 2030, the first cohort to be fully impacted by scale-up of cervical screening to 70% coverage by 2030. Outcomes assessed include the lifetime number of cervical cancer cases and deaths and age-standardised incidence and mortality rates as a measure of the benefits. We assessed the number of precancer treatments needed to avert a cervical cancer death (‘NNT’) and preterm delivery events due directly to precancer treatment (‘additional preterm delivery events’) as a measure of the harms associated with screening. We also report on resource utilisation events including the lifetime number of VIA, cytology and HPV tests, ablation and excisional treatment events, and colposcopy and biopsy events. We report on the cost and cost-effectiveness of each strategy as a cost per HALY saved, assuming 0% discounting for effects and 3% discounting for costs as recommended by WHO for health economic evaluation of vaccination programs 47 , and assuming discounting starts from age 30. We presented results at a population-weighted average across 78 LMICs which we refer to as a ‘normative approach’, using 2015 population structure for population-weighted contribution of each country. There is no defined willingness-to-pay (WTP) when presenting cost-effectiveness at this multi-country average level; however, as a reference point for a potential WTP threshold in this population, the population-weighted average GDP-per-capita (pc) for 2019 across the 78 LMICs is US$2,093, and 71 of 78 [91%] LMICs had a GDP-per-capita equal to or above US$518, considering countries GDP per-capita being related to the countries willingness-to-pay. 48 We identified strategies that appear on, or near, the cost-effectiveness frontier as being the strategies with the best balance of costs and effects.
Model of obstetric complications
To evaluate adverse obstetric outcomes due to precancer treatment, we developed a Monte Carlo individual-based simulation model which incorporates country-specific and age-specific fertility rates, as well as precancer treatment outcomes by mode of treatment, and explicitly model additional preterm delivery events as a result of ablation and excisional treatments for 78-LMICs. This was adapted from a previous component of Policy1-Cervix which has been used to simulate adverse obstetric outcomes after screening in high-income countries.49 Combining systematic review evidence on the risk of preterm delivery after excision (excision versus no treatment: 11.2% versus 5.5%, RR 1.87, 95% CI 1.64 to 2.12)30 with a detailed model of cervical cancer screening and precancer treatment for Australia,50 estimated preterm delivery events for Australia49 and Australian fertility data, we estimated that women with a history of excisional treatment have an excess probability of preterm delivery of 4.8% for each subsequent pregnancy. Systematic reviews indicate that the risk of preterm delivery after ablation is lower than that after excision (ablation versus no treatment: 7.7% versus 4.6%, RR 1.35, 95% CI 1.20 to 1.52).30 We therefore estimated that the additional probability of preterm delivery per pregnancy in women with a history of ablation without excision is (1.35-1)/(1.87-1)*4.8%=1.9%. We obtained national age-specific fertility rates for each of the 78 LMICs from the United Nations (2019),51 and performed a population-weighted average to generate fertility rates for all 78 LMIC. We conservatively assumed that multiple treatments of the same type do not generate any additional risk of adverse pregnancy outcomes. In sensitivity analysis, we also considered a scenario in which ablative treatments did not increase the probability of preterm deliveries for subsequent pregnancies.
Costs and HALYs
Costs for each screening event were provided separately for each of the 78 LMICs by WHO.52,53 We present the population-weighted aggregate cost (weightings for ages 30–49) of each event across the 78 LMICs, shown in Table 3. Ranges considered in sensitivity analysis are also shown.
Disability weights for cancer states were estimated by the Global Burden of Disease study 2010,54 and were applied to cancer based on stage and time since diagnosis. The disability weights used to evaluate HALYs are shown in Appendix Table 4 (pp 46–51).
Supplementary analysis – alternative follow-up management
Management of HPV-positive and triage-negative
For the base case we assumed women who tested HPV positive and triage-negative would return in 12 months for an HPV test; if negative at this visit, women are then referred for their next routine screening visit or discharged from screening. As a supplementary analysis, we considered two alternative management options for this group based on discussions with the GDG: one was a less-aggressive management option in which triage-negative women return in 24 months for the follow-up HPV test (assuming 10% loss-to-follow-up for the return visit at 24 months, but also considering an supplementary analysis of 30% loss-to-follow-up), and another more aggressive management option in which women return at both 12 and 24 months, with 10% loss-to-follow-up assumed for each visit; in this more aggressive scenario, women are returned to routine screening or discharged from screening after testing negative at both visits.
Management of women after treatment for precancer (and did not have CIN3 detected by histology)
For the base case, we assumed that women who have been treated for cervical precancer and did not have a histological diagnosis of CIN3 would return in 12 months for an HPV test and are returned to routine screening (or discharged if outside of the age range) if negative at this visit. In the supplementary analysis, we considered alternative management scenarios as informed by discussion with the GDG. One was the option in which these women would return in 24 months for an HPV test and assumed a 30% loss-to-follow-up at this extended timeframe. The other was an option in which these women return at 12 months for an HPV and cytology co-test, with a 10% loss-to-follow-up assumed at this visit; women are returned to routine screening (or discharged if outside of the age-range) after testing negative with both tests.
Sensitivity analysis
A range of sensitivity analyses were considered. These are summarised in Appendix Table 4 (Appendix pp 46–51). A lower screening adherence scenario, in which we assumed 50% adherence with routine attendance (30% of women never attend, 50% selected from the pool of ever-screeners) and 75% for adherence with treatment or follow-up visits (100% for same-day eligibility) was explored for all screening approaches. We also perform sensitivity analysis on primary test performance assumptions, including a lower bound CIN2 + sensitivity assumption of 30% for VIA, 46.8% for cytology and 88% for HPV testing and an upper bound CIN2 + sensitivity assumption of 60% for VIA, 80% for cytology and 95.7% for HPV. We considered a scenario in which 90% of symptomatically-detected cancers received adequate treatment in addition to the screen-detected cases, and a scenario in which both symptomatic and screen-detected cancers received treatment at current access rates (33% across all 78 LMICs). We also performed one-way sensitivity analysis assuming a 3% discount rate for both costs and effects, and considering life-years instead of HALYs.
PSA was also performed to explore uncertainties in costs. We generated 10,000 cost parameter sets based on the upper and lower ranges for each parameter as described in Table 3 (these ranges were discussed with the WHO GDG). To generate the sets, we divided cost values into five independent groups of variables, namely (1) cancer diagnosis, staging and treatment costs; (2) pre-cancer treatment costs; (3) HPV test costs; (4) VIA test costs; and (5) cytology test costs, and generated 10,000 samples with Latin hypercube sampling. Acceptability curves were generated for a range of WTP values from US$100-$2,000/HALY saved.