Machine learning algorithms for the early prediction of sepsis in children with bone and joint infections

Abstract

Abstract Objectives Early detection of sepsis is crucial in pediatric patients. This study employed machine learning algorithms to develop an artificial intelligence model for the early identification of sepsis in children with bone and joint infections. Materials and methods This retrospective analysis utilized case data from pediatric patients with septic osteoarticular infections admitted to three hospitals in China between January 2012 and January 2023. All patients underwent blood bacterial cultures and puncture fluid bacterial cultures. Seventeen early-available independent variables were selected, and eight different machine learning algorithms were applied to construct the model by training on these data. Results The study included 183 patients in the sepsis group and 422 patients in the no-sepsis group. Among the machine learning algorithms, RandomForest exhibited the best performance with an AUC of 0.946 ± 0.025. The model demonstrated an accuracy of 0.919 ± 0.021, sensitivity of 0.825 ± 0.056, specificity of 0.957 ± 0.012, precision of 0.888 ± 0.044, and an F1 score of 0.855 ± 0.047. In terms of characteristic importance, the seventeen variables ranked in order were: maximum heating time, procalcitonin (PCT), duration of symptoms, platelet, weight, age, peak temperature, fever days, neutrophil, hemoglobin, recent diseases, symptoms of other systems, gender, bone damage, leukocyte, erythrocyte sedimentation rate (ESR), and C-reactive protein (CRP). Conclusions The model can effectively predict the risk of sepsis in children with septic osteoarthritic infections early and timely, which assists in clinical decision-making and reduces the risks and consequences of delayed test results.