Keywords: Cost-effectiveness perspective, Decision-tree-based learning algorithm, Direct optimization method, Individual-level heterogeneity, Incremental net monetary benefit, Individualized treatment rule.
Precision medicine highly advocates tailoring healthcare services to the significant individual-level heterogeneity. Individualized treatment rules (ITR), as powerful tools in personalized decision-making, provide customized treatment suggestions based on individual characteristics so that the clinical benefit is maximized. From a health economic perspective, healthcare policy makers consider the trade-off between health gains and added cost as the key to advise medical resource allocation. However, most of the work so far has focused on maximizing the effectiveness of a treatment without considering the contribution of cost to the preferred decision. In this paper, we extend the idea of ITR to a composite-outcome setting and identify the cost-effective ITR that maximize the average net monetary benefit (NMB) through a direct optimization method. In details, we propose a decision-tree-based statistical learning algorithm to estimate the optimal regime in a data-driven way, and we further propose a new reward function using the estimates of NMB or incremental NMB; so, the impact of treatment decisions on cost and effectiveness are jointly considered. We employ several methods to estimate our reward function in the proposed learning process and perform simulation studies to compare the performance of each proposal. We further apply our top-performed algorithm to the NIH Systolic Blood Pressure Intervention Trial (SPRINT) and demonstrate the gains from considering personalized intensive blood pressure control strategies from a cost-effectiveness perspective.