Shared Decision-Making under Bounded Rationality: Why Personalization Doesn’t Always Help

Problem Definition: Shared decision-making processes, in which doctors and patients work together to choose among treatment options, have gained substantial support from clinicians, policymakers, and health systems. Shared decision-making personalizes care by combining health outcome predictions from doctors with preference-based input from patients. Despite this broad advocacy, important challenges remain: outcome predictions are noisy, patients often misinterpret trade-offs, and it is unclear when personalization improves or worsens outcomes.

Methodology/Results: We develop a stylized analytical model to characterize how personalizing outcome predictions and incorporating patient preferences affect patient utility under bounded rationality. We show that personalizing outcome predictions can backfire when doctors overweight noisy signals or when underlying health-outcome heterogeneity is low relative to doctor error. With patient participation, the interaction between doctor and patient errors becomes critical: when patients have strong preferences, this interaction reduces—and can even reverse—the value of personalizing outcome predictions; in contrast, when patients have weak preferences, the same interaction enhances its value. We also uncover a counterintuitive non-monotonic effect: utility losses from personalizing outcome predictions peak not at the lowest, but at moderate levels of health-outcome heterogeneity relative to doctor error. Finally, we show that preference-based personalization can reduce utility when treatments appear similar across patient types and patients misinterpret the associated trade-offs.

Managerial implications: Contrary to the common belief that advocates for increasing personalization uniformly, our results show the trade-offs between personalization and standardization that arise from limitations in prediction accuracy and human cognition. Our results describe conditions under which personalization can backfire.