Machine learning algorithms for the early detection of bloodstream infection in children with osteoarticular infections.

Background: Bloodstream infection (BSI) poses a significant life-threatening risk in pediatric patients with osteoarticular infections. Timely identification of BSI is crucial for effective management and improved patient outcomes. This study aimed to develop a machine learning (ML) model for the early identification of BSI in children with osteoarticular infections. Materials and methods: A retrospective analysis was conducted on pediatric patients diagnosed with osteoarticular infections admitted to three hospitals in China between January 2012 and January 2023. All patients underwent blood and puncture fluid bacterial cultures. Sixteen early available variables were selected, and eight different ML algorithms were applied to construct the model by training on these data. The accuracy and the area under the receiver operating characteristic (ROC) curve (AUC) were used to evaluate the performance of these models. The Shapley Additive Explanation (SHAP) values were utilized to explain the predictive value of each variable on the output of the model. Results: The study comprised 181 patients in the BSI group and 420 in the non-BSI group. Random Forest exhibited the best performance, with an AUC of 0.947 ± 0.016. The model demonstrated an accuracy of 0.895 ± 0.023, a sensitivity of 0.847 ± 0.071, a specificity of 0.917 ± 0.007, a precision of 0.813 ± 0.023, and an F1 score of 0.828 ± 0.040. The four most significant variables in both the feature importance matrix plot of the Random Forest model and the SHAP summary plot were procalcitonin (PCT), neutrophil count (N), leukocyte count (WBC), and fever days. Conclusions: The Random Forest model proved to be effective in early and timely identification of BSI in children with osteoarticular infections. Its application could aid in clinical decision-making and potentially mitigate the risk associated with delayed or inaccurate blood culture results.