Risk prediction of patients with perioperative pneumonia exacerbation being transferred to the ICU based on different...

To develop a prediction model for the risk of patients with perioperative pneumonia exacerbation being transferred to the intensive care unit (ICU) based on different machine learning algorithms and to evaluate its predictive efficacy.

A total of 309 surgical inpatients (2022-2025) were included; 80 (25.89%) were transferred to the ICU due to exacerbation of pneumonia. Patients were randomly divided in a 7:3 into training (217) and validation (92) sets. The least absolute shrinkage and selection operator (LASSO) regression was applied to identify predictors of ICU transfer following pneumonia exacerbation. Four models (RF, SVM, LR, and GBM) were constructed to predict ICU transfer in patients with perioperative pneumonia exacerbation. Performance was assessed using receiver operating characteristic (ROC) curves, area under the curve (AUC), calibration curves, and decision curve analysis (DCA). The Shapley additive explanation (SHAP) method was used to explain the contribution of the variables to the optimal model's predictions.

The LASSO selected surgery type, surgery duration, ASA class, and WBC count. Among the four models, the SVM model achieved the highest AUC in the validation set (0.817), with a training set AUC of 0.862. The calibration was good, and the DCA showed a net clinical benefit. SHAP confirmed that all four were key predictors. Two cases illustrated model use: an ICU-transferred patient had a SHAP score of 0.897, while a non-transferred patient scored 0.073.

In this study, four machine learning models were developed and validated to predict ICU transfer in patients with perioperative pneumonia exacerbation. The SVM model showed the highest validation AUC among the four models. WBC count, duration of surgery, ASA grade, and type of surgery were identified as key risk factors for progression from pneumonia exacerbation to ICU admission. However, as a single-center study with internal validation only, external validation is required before clinical application.