Primary Submission Category: Machine Learning and Causal Inference
Efficient generalizability and transportability of survival causal effects
Authors: Axel Martin, Ivan Diaz, Michele Santacatterina,
Presenting Author: Axel Martin*
Randomized clinical trials (RCTs) are widely regarded as the gold standard for estimating survival causal effects. However, the subpopulation included in RCTs rarely reflects real-world populations. As a result, the generalization or transportability of results from RCTs to those populations of interest is limited. Real-world observational data, on the other hand, are generally widely available and contain a substantial amount of information regarding the population of interest.
Recent works have proposed leveraging observational data to generalize and transport causal effects. However, few of these methods focus on survival causal effects, have been widely understudied in the literature. We contribute to this important literature by proposing a flexible and robust doubly-robust estimator that incorporates machine learning techniques to transport and generalize survival causal effects from RCTs to real-world target populations. Additionally, we propose more efficient estimators by leveraging treatment effect modification. We demonstrate their large sample properties, evaluate their finite sample performance in simulations, and apply them using the Women’s Health Initiative RCT and observational studies. Finally, we have developed an open-source R package, ‘dmlSurv,’ to implement these estimators, facilitating their accessibility and utilization.