Assessing the impact of AI on physician decision-making for mental health treatment in primary care.

Abstract

AI models may soon be poised to recommend mental health treatments or referrals in primary care, yet little is known regarding their impact on physician decision-making. In this web-based study, primary care physicians (n = 420) were presented with a clinical scenario describing a patient with psychiatric symptoms, an AI tool for referring or prescribing, and the recommendation of the AI. A sequentially randomized vignette method was used to test the impact of initial assessments and AI output on physician decision-making patterns. Physicians were significantly more likely to change their decisions when the AI recommendation was misaligned with their initial assessment, especially when AI recommended treatment. There was no difference between the change-in-decision rate of physicians who received an AI recommendation to not treat, indicating that the direction of AI recommendations may influence physician decision-making, and raising important considerations for how physician decisions may be anticipated in the context of AI.