COMPARING MODEL-PROJECTED OUTCOMES OF CERVICAL CANCER SCREENING IN INDIA WITH EMPIRICAL DATA

Purpose: To assess the correspondence between projections from an empirically-calibrated model of human papillomavirus (HPV)-induced cervical carcinogenesis and results from a clinical trial of cervical cancer screening strategies in India.

Method: We used a microsimulation model of HPV and cervical cancer that had previously been calibrated using epidemiological data from India to evaluate cervical cancer prevention strategies, including HPV DNA testing, visual inspection with acetic acid (VIA), and cytology screening. After completion of the evaluation, results from a cluster-randomized trial in India became available, reporting the benefits from a single round of screening over eight years of follow-up. We modeled the study by simulating a cohort with similar characteristics including age distribution, screening history, and background mortality. Comparable outcomes included % screen positivity, disease prevalence at screening, and detected cancers after screening. Simulations were performed with 50 good-fitting parameter sets to produce the mean and range of model results, which were compared to the mean and 95% confidence intervals (CI) reported by the study.

Result: With a one-time screen, the model projected mean values (range) of 9.0% (6.4%-12.0%) screen positivity with HPV DNA testing, 17.3% (16.7%-18.7%) with VIA, and 6.3% (5.6%-7.9%) with cytology. The corresponding mean (95% CI) estimates from the study were 10.3% (10.0%-10.7%) with HPV, 13.9% (13.5%-14.4%) with VIA, and 7.0% (6.7%-7.3%) with cytology. Model projections of cross-sectional prevalence of high-grade cervical disease at time of screening were 1.1% (range, 0.6%-2.2%) with HPV, 1.1% (0.6%-2.0%) with VIA, and 1.0% (0.5%-1.9%) with cytology; study results ranged from 0.8%-1.0% with HPV, 0.6%-0.8% with VIA, and 0.9%-1.2% with cytology. For cancers per 100,000 person-years over an eight year period, the model projected mean values (range) of 59.0 (35.9-92.3) with HPV, 64.0 (39.6-97.3) with VIA, and 65.6 (40.4-99.2) with cytology; mean (95% CI) estimates from the study were 47.4 (39.2-55.6) with HPV, 58.7 (49.5-67.9) with VIA, and 60.7 (51.1-70.3) with cytology.

Conclusion: Mean outcomes projected from the empirically-calibrated model generally fell within the 95% CI of the study results, suggesting reasonable correspondence. Ex-post comparisons between model output and independent empirical data not used for model parameterization permit evaluation of model performance, can help identify influential heterogeneities, and can enhance transparency of complex models.