Expanding the Capacity of General Practitioners in Sub‐Saharan Africa With Artificial Intelligence

Abstract Low‐ and middle‐income countries face substantial healthcare delivery challenges due to specialist shortages. This study assessed POE (Q‐Exp‐Claude‐3.5‐2kk), a multimodal large language model, for diagnostic accuracy and clinical utility in otolaryngology across 63 consecutive patients in rural Kenya. Patients underwent evaluation by both an otolaryngologist and primary care practitioner (PCP), with clinical data and micro‐otoscopic images submitted to POE's interface. POE's primary diagnoses demonstrated 79.4% concordance with otolaryngologist findings, while management recommendations aligned in 96.8% of cases. Differential diagnoses were judged plausible and correct in 73.0% of cases. POE demonstrated significantly superior diagnostic accuracy compared to PCPs (79.4% vs 50.8%, P = .001). Among cases where PCPs failed to establish a diagnosis, POE correctly identified 61.3%. With examination images, POE accurately identified conditions as primary (50.0%) or potential (35.7%) diagnoses, demonstrating value as a clinical decision support tool in resource‐limited settings.