Neural network-based modeling of monkeypox patient status using clinical and demographic characteristics

Background Monkeypox (Mpox) has emerged as a growing public health concern, requiring effective tools for early patient risk assessment and clinical decision-making. This study aimed to develop an artificial intelligence (AI)-based machine learning framework using a multilayer perceptron (MLP) neural network to analyze a Mpox dataset collected from 11 non-endemic countries, identify the most significant clinical symptoms, investigate the relationships among travel history, hospitalization, isolation, and infection status, and predict the hospitalization and isolation status of Mpox patients. Methods A total of 365 non-severe Mpox cases from 11 non-endemic countries were included in the analysis provided in publicly available World Health Organization (WHO) reports from May 6 to May 29, 2022. Patients were classified into two groups: Hospitalized and isolated patients. Clinical data were converted to binary dummy variables and demographic and travel-history data were added as features. The data were analyzed using cross-tabulation and descriptive visualization for symptom patterns and relationships between clinical and demographic variables. A supervised AI-based MLP neural network was then developed to predict the status of hospitalization and isolation, and the performance of the model was assessed by accuracy, recall, false omission rate (FOR) and F1 score. Results The symptoms most commonly reported in the analysis included genital ulcers, fever, and lesions. The eleven countries included in the study reported varying symptom patterns, with Argentina showing the highest symptom diversity, with up to four symptoms recorded per case.. The overall training accuracy of the proposed MLP neural network for the prediction of hospitalization was 94.7% and the overall testing accuracy was 90.7%. The model showed good predictive performance for each patient status classification task with an accuracy of 94.2% at training phase and 90.1% at test phase for isolation prediction. Conclusion The proposed AI-based multilayer perceptron (MLP) neural network provides an effective approach for predicting the hospitalization and isolation status of Mpox patients using clinical and demographic variables. Symptom distributions varied across countries, indicating regional heterogeneity in disease presentation.