Estimating the Treatment Effects of Multiple Drug Combinations on Multiple Outcomes in Hypertension

Hypertension poses a significant global health challenge, and its management is often complicated by the complexity of treatment strategies involving multiple drug combinations and the need to consider multiple outcomes. Traditional treatment effect estimation (TEE) methods struggle to address this complexity, as they typically focus on binary treatments and binary outcomes. To overcome these limitations, we introduce METO, a novel framework designed for TEE in the context of multiple drug combinations and multiple outcomes. METO employs a multi-treatment encoding mechanism to effectively handle various drug combinations and their sequences, and differentiates between effectiveness and safety outcomes by explicitly learning the outcome type when predicting the treatment outcomes. Furthermore, to address confounding bias in outcome prediction, we employ an inverse probability weighting method tailored for multiple treatments, assigning each patient a balance weight derived from their propensity score against different drug combinations. Our comprehensive evaluation using a real-world patient dataset demonstrates that METO outperforms existing TEE methods, with an average improvement of 5.0% in area under the precision-recall curve and 6.4% in influence function-based precision of estimating heterogeneous effects. A case study demonstrates that our method successfully identifies personalized optimal antihypertensive dual regimens, achieving maximal efficacy and minimal drug-related safety risks. This showcases its potential for improving treatment strategies and outcomes in hypertension management.