B0853
Title: Weighted analyses of survival outcomes under complex study designs: An R implementation
Authors: Yilin Ning - National University of Singapore (Singapore) [presenting]
Abstract: The nested case-control (NCC) design is widely used in survival studies when full cohort analysis is not feasible. NCC samples are often analyzed using conditional logistic regression models, but weighted analyses of sampled subjects after breaking the matching bring additional benefits, e.g., improved statistical efficiency, and estimation of absolute risk and effects of matching factors (if any). Incorporation of sampling weights in analyses becomes essential to the unbiased and efficient estimation of exposure effects under some more complex study designs, e.g., counter-matching for studying rare exposures and extreme case-control design that maximizes information given limited sample size by focusing on early deaths and longest survivors. Benefits of these designs and weighted analyses have been demonstrated in clinical examples, but the uptake of these approaches remains low in practice, partly due to the lack of software tools to conveniently compute the required weights. The SamplingDesignTool R package (https://github.com/nyilin/SamplingDesignTools) closes the gap between methodological developments and practical applications by providing simple commands for drawing samples and computing sampling weights under aforementioned designs, supporting external approximations to risk set sizes when the full cohort is unavailable. The use of the package and the benefit of weighted analyses under the various designs will be illustrated using simulated data.
