Stratified Case-Cohort Analysis of General Cohort Sampling Designs

It is shown that variance estimates for regression coefficients in exposure-stratified case-cohort studies (Borgan "et al"., "Lifetime Data Anal.," 6, 2000, 39-58) can easily be obtained from influence terms routinely calculated in the standard software for Cox regression. By allowing for post-stratification on outcome we also place the estimators proposed by Chen ("J. R. Statist. Soc. Ser. B", 63, 2001, 791-809) for a general class of cohort sampling designs within the Borgan "et al."'s framework, facilitating simple variance estimation for these designs. Finally, the Chen approach is extended to accommodate stratified designs with surrogate variables available for all cohort members, such as stratified case-cohort and counter-matching designs. Copyright 2007 Board of the Foundation of the Scandinavian Journal of Statistics..