Identificación de prácticas deshonestas e inseguras en médicos latinoamericanos mediante modelos de inteligencia artificial

Introduction: Physicians’ professional conduct is crucial to ensuring quality and trust in health systems. Objective: To explore the potential of multilayer perceptron artificial intelligence models to predict dishonest and unsafe practices among Latin American physicians. Materials and Methods: An analytical study was conducted using a database of 424 Paraguayan, Peruvian, and Cuban physicians. A multilayer perceptron artificial neural network was implemented. Forty-three variables were used to predict four key unethical behaviors: “recording physical findings that were not examined,” “concealing one’s own medical error,” “overlooking colleagues’ medical errors,” and “neglecting continuing education in one’s Knowledge and skills.” Performance was assessed using the area under the curve (AUC) and classification tables. Results: The model showed high-accuracy performance for “recording physical findings that were not examined” and “concealing one’s own medical error,” with training values of 80.60% and 81.30%, respectively, and testing values of 77.90% for both variables. For “overlooking colleagues’ medical errors” and “neglecting continuing education,” testing accuracy was 66.20% for both outcomes. The AUC showed excellent discriminatory capacity for concealing one’s own medical errors (AUC = 0.863) and recording physical findings that were not examined (AUC = 0.840), as well as moderate capacity to identify overlooking colleagues’ medical errors (AUC = 0.769) and neglecting continuing education (AUC = 0.763). Conclusions: The multilayer perceptron artificial neural network showed potential for predicting and classifying aspects related to dishonesty and unsafe practices in patient care among Latin American physicians.