Receptiveness of physicians towards artificial intelligence-driven drug prescription: a nationwide survey

Using artificial intelligence (AI) to prescribe drugs has advanced slowly. Whether a "doctor-in-the-loop" design would increase acceptance of drug-prescribing AI is unknown, as are settings where physicians envision AI-driven drug prescription most likely to be implemented. We surveyed a stratified sample of 2708 physicians throughout China to interrogate their opinions on drug-prescribing AI. Most respondents (78%) are receptive to using drug-prescribing AI and anticipate doing so within 5 years. Respondents suggested initial settings for AI-driven drug prescribing include situations where there are standard guidelines (74%), where the decision is whether to continue a current prescription in someone (55%), and where prescribing decisions rely on high-complexity clinical data (44%). Many (66%) indicated a preference for conditional to fully autonomous drug-prescribing AI. Clustering analysis identified 2 psychological profile-types, "optimists" and "pragmatists", who have different standards for model efficacy, expediency, explainability, and governance/stewardship for drug-prescribing AI. A high level of using medical AI is the strongest predictor for being an optimist (OR = 2.98 [2.53, 3.51]; P < 0.0001). In conclusion, our data point to the wide acceptability of conditional autonomous drug-prescribing AI among Chinese physicians. Moreover, disparity in optimism about drug-prescribing AI is caused by disparity in prior exposure to medical AI.