[2505.20609] Comparisons between a Large Language Model-based Real-Time Compound Diagnostic Medical AI Interface and Physicians for Common Internal Medicine Cases using Simulated Patients

[Submitted on 27 May 2025]

View a PDF file from the paper entitled Comparatives between a diagnostic diagnostic diagnostic interface in a large vehicle in actual time and doctors for common internal medicine using simulation patients, by hyungjun park (1 and 18 other authors

PDF HTML (experimental) view

a summary:The goal of developing the AI ​​Medical Medical Complend interface based on LLM and conducted a clinical trial that compares this interface and doctors to joint internal medicine conditions based on USML CAS (CS). A non -random clinical trial methods were performed on August 20, 2024. We have employed one general doctor, a resident of internal medicine (second and third year), and five simulator patients. Clinical short articles were adapted from USMle Step 2 CS tests. We have developed 10 representative cases of internal medicine based on actual patients and included information available on the initial diagnostic evaluation. The initial result was the accuracy of the first differential diagnosis. Repetition was evaluated based on the percentage of agreement. The results ranged from the first differential diagnosis of doctors from 50 % to 70 %, while the diagnostic intelligence interface actually achieved 80 % accuracy. The ratio of the agreement on the first differential diagnosis was 0.7. The precision of the first and second differential diagnoses ranged from 70 % to 90 % for doctors, while the artificial intelligence interface achieved a 100 % accuracy rate. The average time for artificial intelligence (557 seconds) was 44.6 % of doctors (1006 seconds). The artificial intelligence interface ($ 0.08) reduced the costs by 98.1 % compared to the average doctors ($ 4.2). The patient’s satisfaction ranged from 4.2 to 4.3 for care by doctors and was 3.9 for the AI ​​interface, a LLM actual diagnostic interface showed diagnostic accuracy and the patient’s satisfaction similar to the doctor, while requires less time and reduce costs. These results indicate that artificial intelligence facades may have the ability to help primary care consulting for common internal medical medicine.