Representative pure risk estimation by using data from epidemiologic studies, surveys, and registries: estimating risks for minority subgroups.

The article focuses on improving representative risk estimation for minority subgroups in clinical and epidemiologic studies by proposing a two-step pseudoweighting method called post-KW.S. This method combines individual-level cohort data, a probability-based reference survey, and national disease registry summary statistics to produce robust and efficient estimates of cause-specific absolute risk (pure risk) without relying on the generalisability of hazard ratios from cohorts to the target finite population. Simulation studies demonstrate that post-KW.S reduces bias and variance in hazard ratio and risk estimates, especially when cohort participation is informative or the propensity model is misspecified. Applied to all-cause mortality risk modeling using the NIH-AARP cohort, the 1997 National Health Interview Survey (NHIS), and CDC mortality registries, the method corrected biases in minority hazard ratio estimates and improved calibration of risk predictions compared to naïve cohort or survey-based estimates. The study highlights challenges in generalising minority risk estimates from volunteer cohorts and small survey samples, emphasizing the importance of post-stratification to registry event rates for reliable minority risk estimation.