Estimating a cost-effective individualized treatment rule (CE-ITR) based on machine learning

Abstract

Policy makers employ Cost-Effectiveness Analysis (CEA) to evaluate a new treatment based on its cost and effectiveness. ITR is the treatment recommendation based on patient’s characteristics. However, the recommends generated from ITR and CEA could mismatch, even opposite since their aim is different. Therefore, policy makers need a tool to trade-off between ITR and CEA. Traditionally, optimal ITR focus on the mean benefit on population level, not on individual level. In the era of precision medicine, an ideal intervention needs to be optimized based on individual level. ☐ Here a composite outcome, Net Monetary Benefit (NMB) which integrates the clinical benefits and corresponding cost, is adopted to address the optimization of the cost-effective ITR. ITR is taken as a function of patients’ characteristics that, when implemented, optimizes the allocation of limited healthcare resources by optimizing clinical benefits while minimizing treatment-related costs. Applying machine learning approach –conditional random forest and others(such as XGBoost) we can consider ITR and CEA jointly on individual level to estimate a Cost-Effective ITR(CE-ITR) and apply it to real world clinical data.