Statistics Colloquium : Dr. Yei Eun Shin

Title:  Survey Calibration to Improve the Efficiency of Pure Absolute Risk Estimates from Case-control Samples Nested in a Cohort

Abstract:  Cohort studies provide information on relative and absolute risks of disease. For rare outcomes, large cohorts are needed to have sufficient numbers of events (“cases”), making it costly to obtain covariate information on all cohort members. We focus on nested case-control (NCC) designs that are used extensively in many studies. The full Cox regression model can be fitted to NCC data to estimate relative risks. Langholz and Borgan (1997) showed that absolute risks can also be estimated from NCC data by using information on the numbers at risk at each event time in the entire cohort. However, these approaches do not take advantage of some covariates that may be available on all cohort members. Breslow et al. (2009) suggested that the efficiency of relative risk estimation in case-cohort (CC) designs can be improved by applying survey calibration to sampling weights against “influences”, which show how sensitive regression coefficients are to removal of single observations. Our objective is to extend survey calibration approaches from CC designs to general NCC designs to improve precision of estimates of relative risks and absolute risks. We discovered that calibrating sample weights additionally against the numbers at risk in the entire cohort improves estimates of pure absolute risk. Efficiency improvements for relative risks for variables that are available on the entire cohort can be substantial, which translates to improved efficiency for absolute risks. The greatest improvements are for variables that are strongly correlated with covariates measured in the entire cohort.