Microscopic verification of cervical cancer diagnoses in the Mashhad population-based cancer registry, 2017–2021: variation by age, place and period

Abstract Background Microscopic verification (MV) is a core indicator of cancer registry data quality, but its variation within registries is rarely examined. We described MV and the proportion of death-certificate-only (DCO) registrations for cervical cancer in the Mashhad University of Medical Sciences (MUMS) population-based cancer registry, estimated registered incidence, and examined which registry-available information is associated with non-verification. Methods We analysed all 284 cervical cancer records reported in Persian years 1396–1399 (2017–2021), extracted on [extraction date]; one woman registered twice was counted once (283 women). MV was a histopathological or cytological diagnosis. Fields created by the diagnostic process were excluded to avoid leakage. We estimated MV and DCO% by year, half-year, age and residence; registered incidence per 100,000 women (age-standardized, Segi; gamma confidence intervals; Poisson trend); and adjusted odds ratios for MV. Four prediction algorithms were compared as a benchmark (repeated cross-validation, calibration, temporal validation, permutation test). Results MV was 67.5% (95% CI 61.8–72.7) and DCO 7.1% (4.6–10.7). MV was lowest in 1396 (55.3%), among women aged ≥ 70 years (46.7%) and in MUMS counties other than Mashhad (40.5%), and was 100% in 16 women with unknown city. Adjusted odds of MV were lower with older age (OR 0.68 per 10 years) and in other MUMS counties (0.38; 0.62, 0.24–1.56 without Taybad). Registered incidence among residents of MUMS counties was 2.20 per 100,000 women (age-standardized 2.26, 95% CI 1.96–2.60), falling by about 15% per year (rate ratio 0.85, 0.76–0.96; 0.87, 0.73–1.05 without 1399), not separable from incomplete or late registration. Cross-validated AUC was about 0.69 for all four algorithms (0.67–0.68 without year); temporal validation gave 0.58–0.71 and the permutation test p ≤ 0.005. Conclusions About one third of registered cervical cancers lacked microscopic verification, with gradients by age, area and period. Registry-available variables predicted non-verification only modestly, and algorithms of very different complexity performed alike, so the choice of algorithm is immaterial. The findings support routine audit of MV and DCO by county and period rather than individual prediction. Registered incidence needs a completeness assessment.