A unified deep learning framework for prediction and evaluation of pediatric diseases

Advances in medical information have increased the complexity of clinical decision-making and introduced uncertainty in resource allocation. Automated prediction of pediatric diseases can improve resource allocation and guide clinical practice in pediatrics, addressing diagnostic uncertainty and complexity. We developed and evaluated a unified deep learning framework for disease classification and prediction using electronic medical records from 178,601 patients seen in the pediatric outpatient and pediatric emergency departments of the Second Affiliated Hospital of Shantou University Medical College between 2017 and 2023. The framework’s effectiveness was validated using commonly used neural network architectures, and a novel network was proposed to improve predictive performance by leveraging both contextual (local) and long-range information. Experimental results show that the proposed unified deep learning framework is effective for pediatric disease classification and prediction. Common networks achieve satisfactory performance within this framework, and the newly introduced network that captures both context and long-range dependencies further improves predictive ability. Overall, the framework demonstrates potential to support clinical decision-making in situations of diagnostic uncertainty and complexity. The unified deep learning framework and its enhanced network developed in this study can provide reliable clinical decision support from large-scale pediatric EMR data, helping to optimize clinical workflow and resource allocation and to mitigate challenges posed by diagnostic complexity.