Study on the Prognostic Prediction Model and Clinical Application Value of Machine Learning-based Approach for Septic Children in PICU

Abstract

Objective: To explore the application value of a machine learning-based prediction model in assessing the prognosis of septic children in the pediatric intensive care unit (PICU) and provide data support for clinical decision-making. Methods: A total of 180 septic children admitted to the PICU of a tertiary hospital from January 2020 to December 2024 were selected. They were divided into a control group (90 cases, using traditional scoring methods to predict prognosis) and an observation group (90 cases, using a multivariable model based on machine learning algorithms to predict prognosis) according to the random number table method. General information, laboratory indicators, and clinical interventions were collected. Various models such as Random Forest (RF), Support Vector Machine (SVM), and Logistic Regression (LR) were established. The model performance was evaluated using ROC curve, AUC value, accuracy, sensitivity, and specificity. Results: The machine learning models performed better than traditional scoring methods in predicting the 28-day mortality rate of septic children. Among them, the RF model achieved an AUC value of 0.921, a sensitivity of 85.6%, and a specificity of 88.1%, which were significantly higher than the PIM3 score (AUC 0.762). The prediction accuracy and timeliness of clinical intervention in the observation group were significantly improved, leading to a shortened hospital stay and reduced mortality rate (p < 0.05). Conclusion: The prediction model based on machine learning can more accurately assess the prognostic risk of septic children in PICU, showing good clinical application prospects and providing references for individualized treatment and optimal resource allocation.