Birth-cohort effects and the limits of ecological evaluation of mammography screening: an age-period-cohort analysis of breast cancer mortality in Japan, 1958–2024

Abstract Background Ecological evaluations of mammography screening are confounded by birth-cohort effects and concurrent advances in treatment. We examined the age-period-cohort (APC) structure of breast cancer mortality in Japan and quantified how much screening exposure varies independently of that structure. Methods Deaths among Japanese women aged 30–84 years in 1960–2024 were tabulated by 5-year age group and period and analysed with Holford’s identifiable APC model and quasi-Poisson errors. Screening eligibility (≥ 50 years from 2000; ≥40 years from 2004) was added as an age × period interaction, as binary or cumulative eligibility with lags of 0, 5 and 10 years. Separability was quantified as the R² of exposure on the APC terms. Incidence for 1976–2020 was analysed on the same grid. Results Among 437,134 deaths, the age-cohort model fitted far better than the age-period model (deviance 389.2 vs 9,041.8). The cohort deviation from the log-linear trend peaked in women born around 1950 (rate ratio [RR] 1.51; 95% CI, 1.47–1.54). Incidence, including three high-quality regional registries, shared this peak less markedly than mortality (RR 1.22 vs 1.43, 1976–2020). APC terms explained 91% to > 99% of the variation in screening exposure. With a 10-year lag, eligibility was not associated with mortality (RR 1.005; 0.971–1.040); across definitions, RRs ranged from 0.955 to 1.196. Conclusions Non-linear changes in breast cancer mortality in Japan were aligned mainly with birth cohort. Less than 10% of the variation in screening exposure was independent of the APC structure, so its effect cannot be reliably estimated from national statistics; individual-level linked data are needed.