NV Day Biostatistics Seminar



Target Population Risk Prediction with Time-to-Event Outcome

Dandan Liu, PhD
Vanderbilt University Medical Center

Most existing risk prediction tools for time-to-event outcomes were developed using large epidemiologic cohorts. Although these data sources are often of high quality, their external generalizability to a much broader population, termed target population in this paper, may be limited. Many external validation studies of existing risk prediction tools have shown adequate discrimination but poor calibration which is largely driven by the differences in the baseline hazard between the sourcing cohort and the target population. In this paper, we developed a statistical method to reconcile this discrepancy by incorporating the composite incidence of the target population in the development of risk prediction tools using the sourcing cohort. An individualized absolute risk prediction tool could then be improved using the proposed baseline hazard estimator. The asymptotic properties of the proposed absolute risk estimators were established. Simulation studies examining small sample properties of the proposed estimators were conducted. The proposed methods will be illustrated using the Womens Health Initiative Observational Study and Canary Prostate Active Surveillance Study.


MRBIII, Room 1220
31 October 2018
1:30pm


Topic revision: r1 - 23 Oct 2018, ThomasStewart
 

This site is powered by FoswikiCopyright © 2013-2017 by the contributing authors. All material on this collaboration platform is the property of the contributing authors.
Ideas, requests, problems regarding Vanderbilt Biostatistics Wiki? Send feedback