|
Abstract:
|
Cost-effectiveness (CE) analysis has a critical role in informing healthcare policy making. So far, most CE analyses evaluate health economic gains at a population level; however, linking rich clinical data from electronic health records and large claims databases enables us to assess individual CE benefits. There is currently a lack of statistical tools that consider the tradeoffs between health benefits and added costs when estimating individualized treatment rules (ITRs). Thus, we propose to use a composite outcome, net monetary benefit (NMB), for CE balanced rules. In this paper, we estimate the ITR as a function of patients' characteristics that optimizes the use of limited healthcare resources. We propose an NMB based classification algorithm, where a conditional forest is used to improve efficiency. Also, we propose two partitioned estimators to separately model the classification weights for health and cost outcomes and effectively incorporate data from censored subjects. Our simulation studies evaluate the finite sample performance of each method. We apply our top-performing algorithm to the NIH-funded SPRINT to illustrate the CE gains from assigning personalized regimes.
|
