Comparative Age-Period-Cohort Analysis Reveals Divergent Temporal Trajectories of Breast and Colorectal Cancer Among Women in Singapore

Background: Breast cancer (BC) and colorectal cancer (CRC) are leading malignancies among women in Singapore, but their temporal drivers may differ despite occurring within the same healthcare system. We compared both cancers using a common analytical framework across major ethnic groups to identify ethnic differences that may be hidden in national trends. Objective: To disentangle age, period, and cohort effects on incidence trends among women aged 25–84 years by ethnicity (Chinese, Malay, Indian). Methods: We analysed population-based Singapore Cancer Registry data from 1968 to 2022. We calculated age-standardised incidence rates using the Segi world standard population and fitted age–period–cohort models separately by cancer and ethnicity. Nested models were compared using quasi-likelihood F tests with Benjamini–Hochberg-adjusted p-values. Overdispersion was assessed using Pearson χ2/df. For BC, quasi-Poisson regression models, which extend Poisson regression by allowing the variance to exceed the mean, tested whether cohort associations differed between women aged <49 and ≥49 years (age-band midpoint <49 versus ≥49). Sensitivity analyses examined the influence of imputing suppressed counts in sparse strata. Results: We analysed 53,514 BC and 25,658 CRC cases. BC incidence increased across ethnicities, mostly along a linear drift that cannot be attributed to period or cohort. Beyond this drift, non-linear cohort effects contributed more to model fit than non-linear period effects. Cohort curvature rose up to the Baby Boomer cohorts and fell thereafter, indicating a slower increase than the linear trend in later cohorts rather than a decline in absolute risk. The age-group by birth-cohort interaction significantly improved model fit overall and among Chinese women (p = 0.00339), but not among Indian (p = 0.0873) or Malay women (p = 0.672). Among Indian women, the age–period model had the lowest AIC, but nested-model comparisons depended on the value imputed for suppressed counts, so the relative contributions of period and cohort effects could not be determined reliably. Cohort effects were more strongly supported among Malay women, whereas both period and cohort effects contributed to model fit among Chinese women. Early incidence trends and nested-model comparisons among Indian women were sensitive to suppressed-count imputation. Conclusions: Beyond the linear drift, BC trends are explained more strongly by cohort curvature than by period curvature. This is consistent with a role for generational changes in risk exposures, which this study did not measure directly. CRC showed ethnicity-specific period and cohort associations, although their relative contributions among Indian women remained uncertain because of sensitivity to suppressed-count imputation.