Nationally Representative Individualized Risk Estimation Combining Individual Data from Epidemiologic Studies and Representative Surveys with Summary Statistics from Disease Re
Estimating individualized absolute risks is fundamental to clinical decision-making but are often based on data that does not represent the target population. Current methods improve external validity by including data from population registries but require transportability assumptions of model parameters (relative risks and/or population attributable risks) from epidemiologic studies to the population. We propose a two-step weighting procedure to estimate absolute risk of an event (in the absence of competing events) in the target population without transportability assumptions. The first step improves external-validity for the cohort by creating "pseudoweights" for the cohort using a scaled propensity-based kernel-weighting method, which fractionally distributes sample weights from external probability reference survey units to cohort units, according to their kernel smoothed distance in propensity score. The second step poststratifies the pseudoweighted events in the cohort to a population disease registry by variables available in the registry. Our approach produces design-consistent absolute risks under correct specification of the propensity model. Poststratification improves efficiency and further reduces bias of risk estimates overall and by demographic variables available in the registry when the true propensity model is unknown. We apply our methods to develop a nationally representative all-cause mortality risk model for potential clinical use.
READ FULL TEXT