The Diagnostic and Triage Accuracy of the GPT-3 Artificial Intelligence Model

Abstract

Importance: Artificial intelligence (AI) applications in health care have been effective in many areas of medicine, but they are often trained for a single task using labeled data, making deployment and generalizability challenging. Whether a general-purpose AI language model can perform diagnosis and triage is unknown. Objective: Compare the general-purpose Generative Pre-trained Transformer 3 (GPT-3) AI model's diagnostic and triage performance to attending physicians and lay adults who use the Internet. Design: We compared the accuracy of GPT-3's diagnostic and triage ability for 48 validated case vignettes of both common (e.g., viral illness) and severe (e.g., heart attack) conditions to lay people and practicing physicians. Finally, we examined how well calibrated GPT-3's confidence was for diagnosis and triage. Setting and Participants: The GPT-3 model, a nationally representative sample of lay people, and practicing physicians. Exposure: Validated case vignettes (<60 words; <6th grade reading level). Main Outcomes and Measures: Correct diagnosis, correct triage. Results: Among all cases, GPT-3 replied with the correct diagnosis in its top 3 for 88% (95% CI, 75% to 94%) of cases, compared to 54% (95% CI, 53% to 55%) for lay individuals (p<0.001) and 96% (95% CI, 94% to 97%) for physicians (p=0.0354). GPT-3 triaged (71% correct; 95% CI, 57% to 82%) similarly to lay individuals (74%; 95% CI, 73% to 75%; p=0.73); both were significantly worse than physicians (91%; 95% CI, 89% to 93%; p<0.001). As measured by the Brier score, GPT-3 confidence in its top prediction was reasonably well-calibrated for diagnosis (Brier score = 0.18) and triage (Brier score = 0.22). Conclusions and Relevance: A general-purpose AI language model without any content-specific training could perform diagnosis at levels close to, but below physicians and better than lay individuals. The model was performed less well on triage, where its performance was closer to that of lay individuals.

Diagnostic Accuracy of Differential-Diagnosis Lists Generated by Generative Pretrained Transformer 3 Chatbot for Clinical Vignettes with Common Chief Complaints: A Pilot Study

“…The diagnostic accuracy of the GPT-3 model is considerably limited. A preprint article revealed the correct diagnosis to be 88% within the three differential-diagnosis lists [ 30 ]. Therefore, the diagnostic accuracy of the differential-diagnosis lists generated by AI chatbots, including ChatGPT-3, is unknown.…”

Assessing the Utility of ChatGPT Throughout the Entire Clinical Workflow

“…15 Recently, there has been great interest in utilizing the nascent but powerful chatbot for clinical decision support. 16–18…”

Assessment of ChatGPT in the preclinical management of ophthalmological emergencies – an analysis of ten fictional case vignettes

“…A very encouraging triage accuracy of 87.2% in our study stands in contrast to recent results on the non-ophthalmological general medical domain published in a preprint by Levine and colleagues, who found a triage accuracy of 71% for GPT-3 and 96% for physicians. [7] Whether this contrast is due to the different testing domains, wording of the individual vignettes or an improvement between the different GPT versions remains unclear. Moreover, we must point out, that a technically high triage accuracy does not imply a great utility of the information on urgency: In our study ChatGPT frequently recommended to consult a physician "as soon as possible", which was judged to be appropriate for the urgency levels "emergency" and "same day".…”

“…Triage accuracy however was slightly and insignificantly lower compared to laypersons but by far and significantly lower compared to physicians. [7] For "can't-miss diagnoses", the aforementioned study from the Wills Eye emergency department showed a diagnostic accuracy of triaging ophthalmology staff to be as high as 97.2%. [11] We therefore clearly recommend contacting established providers of ophthalmological emergency services in case of acute symptoms.…”

“…determining the timespan in that he should be referred to an ophthalmologist) and initiation of appropriate preclinical first-aid measures. [4] Encouraging results with regards to differential diagnosis in the general medical domain have been published, [6,7] and ChatGPT has been found useful for simplifying access to information on cardiopulmonary resuscitation. [8] Despite these encouraging results, ChatGPT can also produce wrong information and has been reported to give potentially harmful advice in the ophthalmological and wider medical domains.…”