Primary Submission Category: Applications in Health and Biology
Assessing Principal Causal Effects with Outcome-dependent Sampling Using Principal Score Methods: An Application to the E3N Cohort
Authors: Lisa Braito, Fabrizia Mealli, Vittorio Perduca, Gianluca Severi,
Presenting Author: Lisa Braito*
Outcome-dependent sampling designs, commonly known as case-control studies, are widely used in epidemiology to estimate the impact of an exposure on a binary outcome. These designs involve sampling individuals from a population conditional on observed outcomes. Conventional methods like logistic regression, though prevalent, may be limited for valid causal inference, especially when the goal is to understand causal mechanisms through intermediate variables. We show that directly applying principal stratification methods to case-control designs can yield biased estimates. To address this, we propose a method for estimating principal causal effects in such designs, building on principal score methods developed for observational studies and incorporating external auxiliary information about the target population. This study investigates the relationship between menopausal hormone therapy (MHT), mammographic density, and breast cancer (BC) risk using a principal stratification approach. Evidence suggests MHT increases BC risk, partially mediated by mammographic density. Data are from a nested case-control study within the French E3N cohort. To validate our methodology, we use a secondary intermediate variable, body mass index (BMI), measured across the cohort, to simulate case-control samples and compare results to full-cohort analyses. Our findings refine causal inference methods for complex sampling designs, offering a robust framework for epidemiological research.