Learning Cost-Effective and Interpretable Regimes for Treatment Recommendation

Decision makers, such as doctors, make crucial decisions su ch as recommending treatments to patients on a daily basis. Such decisions typically involve careful assessment of the subject's condition, analyzing the costs associated with the possible actions, and the nature of the consequent outcomes. Further, there might be costs associated with the assessmen t of the subject's condition itself (e.g., physical pain endured during medical tests, monetary costs etc.). Decision makers often leverage personal experience to make decisions in these contexts, wi thout considering data, even if massive amounts of it exist. Machine learning models could be of immense help in such scenarios - but these models would need to consider all three aspects discussed ab ove: predictions of counterfactuals, costs of gathering information, and costs of treatments. Fu rther, these models must be interpretable in order to create any reasonable chance of a human decision m aker actually using them. In this work, we address the problem of learning such cost-effectiv e, interpretable treatment regimes from observational data. Prior research addresses various aspects of the problem at h and in isolation. For instance, there exists a large body of literature on estimating treatment ef fects [5, 12, 4], recommending optimal treatments [1, 15, 6], and learning intelligible models for prediction [9, 7, 10, 2].