Type: article
Study design: Clinical Trial
Abstract: Patients increasingly bring generative AI (GenAI) diagnoses into clinical encounters, yet little is known about how physicians use such patient-provided information. We conduct a randomized field experiment on an online medical consultation platform using 436 physicians and standardized dermatology cases with confirmed diagnoses. Physicians receive no external diagnosis, a correct diagnosis attributed to GenAI, or the identical diagnosis attributed to a physician. Correct GenAI-attributed diagnoses improve both benign/malignant and exact-diagnosis accuracy, but physicians still distinguish benign from malignant conditions in only about two-thirds of accepted consultations and identify the exact diagnosis in less than one-third. The GenAI gains are nearly eliminated for uncommon cases and are accompanied by more hedging and recommendations for further testing. Across the main outcomes, source attribution itself matters little. These findings highlight that highly useful patient-provided AI diagnoses can improve physician accuracy and shape clinical decisions, yet physicians make only partial use of the information provided.
Source: AIS Electronic Library (AISeL)
Authors: Chen, Zhichen, Burtch, Gordon, Gu, Bin, Liu, Yixuan
Institutions: Boston University, China Europe International Business School
Language: English
Subfield: Health Informatics
Field: Medicine
Domain: Health Sciences
Open Access status: green
OpenAlex ID: W7222988470
