Prediction and evaluation of the risk of Stroke-associated pneumonia using an artificial neural network model

Objective This study developed a predictive model for the risk of stroke-associated pneumonia (SAP) based on an advanced artificial neural network (ANN) model. Methods Using clinical and laboratory data collected from a cohort of 456 hospital patients from July 2017 to April 2022, we constructed ANN and logistic regression (LR) models. The models were trained on a randomly selected group of 292 patients, and subsequent model validation and testing were carried out on two separate sets of 82 patients each. The predictive performances of both models were evaluated using a comprehensive range of statistical indices. Results During dataset partitioning, the 24 variables across the training, validation, and test sets displayed no significant discrepancies. The prediction performance of the ANN model was better than that of the LR model. When applied to the test cohort, the ANN model had a sensitivity of 83.53% and a specificity of 85.18%. Comparative analysis revealed discernible discrepancies between the performance indexes of the ANN and LR models. Based on the receiver operating characteristic curve, the ANN model showed robust ability to identify SAP, with an area under the curve value of 0.920. The principal independent predictors in the model were serum albumin, activities of daily living score, hemoglobin level, and hypersensitive C-reactive protein level. Conclusions The developed ANN model demonstrates promising predictive capability for assessing the risk of SAP. However, further verification with larger and more diverse datasets is needed to confirm its utility as a tool for clinical prediction.