Weighing Patient Preferences: Lessons for a Patient Preference Predictor

Abstract: Proposed Patient Preference Predictors (PPPs)—algorithms capable of predicting a patient’s preferences on the basis of demographic or more personalized data—are promising tools for guiding the treatments of incapacitated patients. But there is a risk that utilizing PPPs might cause practitioners to myopically focus on predicted preferences at the expense of other ethically relevant considerations. In this paper, drawing on some illustrative cases, I will distinguish two kinds of preferences a PPP might predict—discrete preferences for treatments in hypothetical circumstances and more general preferences regarding quality of life or goals of care—and show that the reasons generated by such preferences, though important, must be weighed against sometimes competing reasons grounded in practitioners’ obligations to promote patient wellbeing and respect patient autonomy. In light of this, there are three important lessons that should inform the implementation of any PPP. First, we should avoid the temptation—one that often accompanies algorithmic assistance—to overly mechanize medical decision-making and thereby fail to be appropriately sensitive to all the ethically relevant considerations. Second, when training our PPP, we should take care to ensure that we are tracking the right kinds of preferences—namely those that are grounded in a patient’s deep and enduring commitments and values. Finally, we should not restrict our use of PPPs to just incapacitated patients—the kinds of preferences a PPP might track can be relevant and helpful when treating capacitated and incapacitated patients alike.