Application of ChatGPT for clinical decision-making in patients with community-acquired pneumonia.

<bold>Background:</bold> ChatGPT is an accessible artificial intelligence solution that can be used by any physician as clinical decision support assisting in diagnosing and choosing treatment options. <bold>Objective:</bold> To evaluate the accuracy of ChatGPT for diagnosis and management of community-acquired pneumonia. <bold>Methods:</bold> We created 30 clinical cases based on real medical records of patients with a diagnosis of community-acquired pneumonia. Each case was presented to ChatGPT (version 3.5) using a PC. The accuracy of the diagnosis, treatment, recommendations for patient management and CRB-65 score was evaluated. <bold>Results:</bold> The nosological diagnosis was correct in 50% of cases (n=15) and partially correct in 46.67% of cases (n=14). The proposed treatment strategies were correct in 66.67% of cases (n=20), partially correct in 30% (n=9), and incorrect in 1 case (3.33%). General recommendations for patient management were correct in 80% of cases (n=24), partially correct in 16.67% (n=5), and incorrect in 1 case (3.33%). The CRB-65 score was correctly calculated in 60% of cases (n=18), and incorrect in 40% (n=12). <bold>Conclusion:</bold> ChatGPT proved to be a fairly accurate tool in the diagnosis of community-acquired pneumonia, as the diagnosis was completely correct or partially correct in 29 out of 30 cases. The same accuracy was found in the choice of treatment strategy and recommendations for patient management. ChatGPT was less accurate in calculating the CRB-65 score – correct calculations in slightly more than half of the cases.