The cost-effectiveness of implementing HPV testing for cervical cancer screening in El Salvador

Objective To assess the cost-effectiveness of HPV-based screening and management algorithms for HPV-positive women in phase 2 of the Cervical Cancer Prevention in El Salvador (CAPE) demonstration, relative to the status quo of Pap-based screening. Methods Data from phase 2 of the CAPE demonstration (n=8000 women) were used to inform a mathematical model of HPV infection and cervical cancer. The model was used to project the lifetime health and economic outcomes of HPV testing every 5 years (age 30–65 years), with referral to colposcopy for HPV-positive women; HPV testing every 5 years (age 30-65 years), with immediate cryotherapy for eligible HPV-positive women; and Pap testing every 2 years (age 20–65 years), with referral to colposcopy for Pap-positive women. Results Despite slight decreases in the proportion of HPV-positive women who received treatment relative to phase 1, the health impact of screening in phase 2 remained stable, reducing cancer risk by 58.5%. As in phase 1, HPV testing followed by cryotherapy for eligible HPV-positive women remained the least costly and most effective strategy (US$490 per year of life saved). Conclusion HPV-based screening followed by immediate cryotherapy in all eligible women would be very cost-effective in El Salvador.

data on the cost-effectiveness of these management strategies are needed.
In 2012, the Cervical Cancer Prevention in El Salvador (CAPE) project was launched to assess the feasibility and cost-effectiveness of incorporating low-cost HPV testing into the national cervical cancer screening program. The CAPE project is a demonstration project in three phases, conducted by the Salvadoran Ministry of Health (MINSAL) with technical support from the non-profit organization Basic Health International.
In phase 1-a pilot study of 2000 women aged 30-49 years screened at four health centers in the Paracentral region-women who tested positive for HPV received one of two management algorithms: colposcopy management (referral to colposcopy followed by treatment for women with cervical intraepithelial neoplasia), or screen and treat (immediate treatment for all eligible women). 3 Phase 2 scaled up HPV testing to 8000 women at eight health centers in order to compare the two management algorithms in a larger and more diverse screening population. 4 In phase 1, more women received recommended follow-up in the screen and treat cohort than in the colposcopy management cohort, and screen and treat was found to be very cost-effective for management of HPV-positive women in El Salvador. 3,5 The objective of the present analysis was to use data from CAPE phase 2 to evaluate the cost-effectiveness of the colposcopy management and screen and treat management algorithms (relative to Pap-based screening) as implementation and scale-up continues in the public sector.

| MATERIALS AND METHODS
The present study was conducted using an individual-based Monte Carlo simulation model of the natural history of HPV and cervical cancer, 6,7 as in phase 1. The model projects the lifetime health and economic outcomes associated with each screening strategy. As described in previous publications, individual girls enter the simulation model at age 9 years and transition between health states (including typespecific HPV infection status, histologic grade of precancer [CIN2 or 3], and stage of cancer) each month until death. Monthly transition probabilities may vary by age, HPV type, duration of infection or lesion status, and prior HPV infection. Death from all causes can occur from any health state, and excess mortality from cervical cancer can occur after its onset, depending upon the stage of cancer. The model tracks disease progression and regression, screening and treatment events, and healthcare costs over the lifetime of each woman. These outcomes are then aggregated over the population and used for analysis. 6,7 Details of the model parameterization and calibration process have been described elsewhere. [6][7][8] In brief, baseline parameter values were established for the natural history component of the model using longitudinal data for age-and type-specific HPV incidence, as well as typespecific and time-dependent rates of HPV clearance and progression. [9][10][11][12][13] To reflect differences in HPV incidence and burden between settings, in addition to uncertainty in the degree of natural immunity following initial infection and in progression and regression of precancer, plausible ranges were set around these input parameter values. Repeated natural history model simulations selected a single random value from the range for each uncertain parameter to form a unique natural history input parameter set. A goodness-of-fit score was then computed for each unique set by summing the log-likelihood of model-projected outcomes to represent the quality of fit to epidemiologic data (i.e., calibration targets), including the age-specific prevalence of oncogenic HPV among women aged 30-49 years in phase 2 of the CAPE project, 4 prevalence of HPV genotypes 16 and 18 in cervical cancer in South and Central America, 14 and age-specific cervical cancer incidence in El Salvador. 1 The 50 top good-fitting input parameter sets were selected for use in cost-effectiveness analysis, as a form of probabilistic sensitivity analysis ( Figures S1-S3). Results were reported as the mean and range of outcomes across these top 50 parameter sets.
As in the phase 1 cost-effectiveness analysis, Pap testing was compared with two HPV screening and management algorithms (colposcopy management and screen and treat) for women who tested positive for HPV. Following an initial screening visit at the clinic, women were scheduled to return for results. Women in the colposcopy management cohort who screened positive were then scheduled to receive colposcopy at the designated hospital, whereas women in the screen and treat cohort received visual assessment to determine eligibility for immediate cryotherapy (with ineligible women referred to colposcopy).
It was assumed that (1) the initial screening populations were identical for each strategy; (2) the proportion of women who attended visits to receive results, cryotherapy, colposcopy, and treatment were based on phase 2 data from the relevant cohort (colposcopy management or screen and treat) and complied with recommended follow-up within 6 months (Table 1)  HPV testing with provider collection of HPV specimens was assumed to take place every 5 years between ages 30 and 65 years (colposcopy management and screen and treat), while Pap testing with colposcopy management was assumed to take place every 2 years between ages 20 and 65 years (consistent with recommended screening ages in national guidelines).
In accordance with guidelines for cost-effectiveness analysis, a societal perspective was applied, including costs irrespective of the payor. 15 Cost data are presented in Table 1. Direct medical cost data were estimated in phase 1 using a microcosting methodology. 5 For phase 2, all costs were updated from 2012 to 2014 US$ using gross domestic product (GDP) deflators; 16 cost per cryotherapy was updated to reflect the cost per nitrous oxide tank refill and average number of patients treated per tank in phase 2; and the cost of fuel used to transport HPV and Pap specimens to the laboratory was added. Women's time spent traveling, waiting for, and receiving care was valued using national household income data; women's transportation costs to travel to healthcare facilities were estimated by in-country clinicians. 5,17 Reported model outcomes include lifetime risk of cervical cancer, expected total lifetime cost per woman, and life expectancy. After discounting future costs and life-years at a rate of 3% per year, incremental cost-effectiveness ratios (ICERs) were calculated. ICERs represent the additional cost of a strategy divided by its additional benefit relative to the next most costly strategy after eliminating strategies that For reduction in cancer risk, discounted lifetime costs, and discounted life expectancy from age 9 years, the mean value is reported across 50 input parameter sets; the reported ICER is the ratio of the mean costs divided by the mean effects of one strategy vs another across the 50 sets. b Relative to no screening. c Phase 1 results are different than previously published estimates 5 due to updates in the natural history model and calibration, updated test performance data, updating of costs to 2014 US$ and updated fuel and cryotherapy gas costs, and the start age for Pap screening beginning at age 20 y (rather than age 30 y, in the previous analysis). Phase 1 and 2 results differ only in visit compliance parameters, as indicated in Table 1. Sensitivity analyses were conducted to examine the impact of alternative values for model inputs (Table 1). Additionally, the analysis was repeated using visit compliance data from phase 1 to directly compare cost-effectiveness results for phase 1 with phase 2 using the updated model and holding other input parameters constant.

Variable [reference] Baseline value Sensitivity analyses
Ethics approval was not required owing to the retrospective design, as de-identified data were analyzed retrospectively as part of a previous analysis. All procedures conducted as part of CAPE were approved by the national ethics review board of El Salvador.

| RESULTS
In phase 2, HPV testing with screen and treat was the most effective screening strategy, predicted to reduce the absolute risk of cervical cancer by 58.5% (  respectively. Though Pap was slightly more effective than screen and treat in these instances, the ICER for Pap was prohibitively expensive.
Thus, screen and treat remained the most effective strategy with an ICER below per capita GDP in all sensitivity analyses.
Results comparing phases 1 and 2-using the updated model, costs, and input parameters, except for phase-specific visit compliance indicators-are presented in Table 2. Due to slightly higher visit compliance in the phase 1 screen and treat cohort, screen and treat was slightly more effective and had a lower ICER in phase 1 (US$470 per YLS vs US$490 per YLS in phase 2). Although Pap yielded lower reductions in cancer risk relative to screen and treat in phase 1, it was associated with slightly greater life expectancy gains because it detected a small number of early cancers between ages 20 and 30 years. However, these gains were very costly and yielded an ICER of US$26 900.

| DISCUSSION
In the present study, a microsimulation model of HPV infection and cervical cancer was updated to fit epidemiologic data from phase 2 of the CAPE project, in which HPV-based cervical cancer screening is being implemented in El Salvador's public sector. The cost-effectiveness of two management algorithms (colposcopy management and screen and treat) for HPV-positive women was estimated relative to Pap screening using phase 2 data for costs, visit compliance, and eligibility for cryotherapy.
The present study found that HPV testing followed by immediate treatment with cryotherapy for all eligible women (screen and treat) every   24 Because the ICER for screen and treat is well below 50% of per capita GDP, it is likely that the strategy would remain "cost-effective" even at a lower threshold.
Phase 3 of the CAPE project, commenced in May 2015, is currently underway with the goal of screening 20 000 women per year with screen and treat to reach 80% of screening-eligible women within 5 years. As scale-up continues, health promotors will target underscreened women, potentially reaching women at greater risk of cervical cancer. Cryotherapy will not be available at all screening facilities and, in some cases, screen-positive women will be referred to receive treatment elsewhere. Continued follow-up of treated women in phases 1 and 2 will also provide data on cryotherapy effectiveness.
As further data on implementation costs and health impact become available, cost-effectiveness estimates will continue to be updated to provide decision makers with information on one of the first national HPV-based screening programs in a lower-middle-income country.

AUTHOR CONTRIBUTIONS
NGC contributed to study conceptualization, study design, data collection, data analysis, data interpretation, and drafted the manuscript. MM contributed to study conceptualization, study design, data collection, data analysis, data interpretation, and revision of the manuscript. KA contributed to data collection, data analysis, and revision of the manuscript. JCG contributed to study design, data collection, data analysis, and revision of the manuscript. PEC contributed to study conceptualization, data interpretation, and revision of the manuscript. JCF contributed to data collection and revision of the manuscript. RM contributed to study design and revision of the manuscript. MC contributed to study conceptualization, study design, data interpretation, and revision of the manuscript. JJK contributed to study conceptualization, study design, data collection, data interpretation, and revision of the manuscript.

ACKNOWLEDGMENTS
The authors gratefully acknowledge the support of the Einhorn Family

SUPPORTING INFORMATION
Additional supporting information may be found online in the Supporting Information section at the end of the article.