ABSTRACT
Jordan C. Brooks, Alan S. Go, Daniel E. Singer, and Mark J. van der Laan
Safety analysis involves assessing the causal effect of a treatment or exposure regimen on one or more outcomes of interest based on observing a sample of subjects over time. The observed data might constitute an observational study or a randomized controlled trial or even a collection of observational or randomized studies. Since randomized trials are typically not powered
CONTENTS
10.1 Introduction and Motivation .................................................................... 173 10.2 Roadmap for Safety Analysis ................................................................... 177
10.2.1 Observed Data Structure .............................................................. 177 10.2.2 Structural Causal Model ............................................................... 177 10.2.3 Causal Quantities ........................................................................... 178 10.2.4 Identiability Result ...................................................................... 179 10.2.5 Statistical Model, Statistical Target Parameter,
and Estimation Problem ............................................................... 179 10.2.6 Inuence Curve for Inference ...................................................... 180 10.2.7 Multiple Testing Based on Joint Distribution ............................ 180 10.2.8 Sensitivity Analysis ....................................................................... 181
10.3 TMLE ........................................................................................................... 181 10.3.1 Super Learning ............................................................................... 182
10.4 TMLE of Intervention-Specic Mean ...................................................... 183 10.5 Safety Analysis of Warfarin: The Causal Effect on Nonstroke
Death in Atrial Fibrillation ....................................................................... 185 10.5.1 ATRIA-1 Cohort Data .................................................................... 186 10.5.2 Causal Effect of Warfarin Parameter .......................................... 187 10.5.3 TMLE Implementation .................................................................. 187 10.5.4 Results ............................................................................................. 187
10.6 Concluding Remark ................................................................................... 189 References ............................................................................................................. 189
for safety outcomes, it is not uncommon that one combines various randomized studies, possibly augmented with observational studies, in order to evaluate safety signals in the data.
