Leveraging Deep Generative Model For Causal Effect Estimation in Healthcare

Abstract

Deep generative models have risen to prominence in diverse domains, including healthcare. In particular, their application in causal effect estimation has the potential to drive significant advancements in personalized medicine. In this study, we conducted an empirical analysis to investigate the impact of selection bias on continuous treatment effect estimation using deep generative models. Our results demonstrate that the presence of selection bias can lead to estimation performance disparities of up to 8 to 9 times. Such significant deviations pose severe risks in medical applications, where accurate treatment effect estimation is crucial for patient outcomes. Therefore, this paper underscores the criticality of addressing these challenges and proposes directions for future research to ensure the robust application of AI in healthcare.