Elderly Surgical Patients Multi-Infection Prediction: Machine Learning Model Development & Validation With SHAP Analysis
Elderly Surgical Patients Multi-Infection Prediction: Machine Learning Model Development & Validation With SHAP Analysis
Utilizing machine learning techniques, investigators developed the geriatric infection assessment model, leveraging domestic databases to predict multiple postoperative infections in elderly patients. The model addresses the current gap in predictive tools tailored for elderly surgical patients in China, offering insights into both overall and specific infection risks.
Backgrounds:
Postoperative infections are a leading cause of adverse perioperative outcomes, particularly for elderly patients. Given the varied diagnostic presentations of infection, there is a significant gap in the use of predictive tools to identify those at high risk of developing such complications.
Objective:
Investigators aimed at developing machine learning models to predict various postoperative infection risks in elderly patients, facilitating early detection and intervention.
Methods:
A retrospective analysis was conducted on 42,540 elderly patients who underwent non-cardiac surgery at the First Medical Center of the Chinese PLA General Hospital between January 2012 and August 2018, forming the Training set. From this, a 30% subset was randomly designated as the Test set. The models incorporated 51 variables including key infection-related factors. Three machine learning techniques-Logistic Regression (LR), Random Forest (RF), and Gradient Boosting Machines (GBM)-were utilized to develop predictive models for overall and specific postoperative infections, categorized according to the European Perioperative Clinical Outcome (EPCO) definitions. Model performance was gauged by metrics such as the Area Under the Receiver Operating Characteristic (ROC) Curve (AUC), accuracy, and precision. To enhance model interpretability, investigators employed the RF model's Variable Importance (VIMP) and Shapley Additive Explanations (SHAP) algorithm. For a demonstrable prediction of specific infection types, data of randomly selected 5 patients were fed into the model with the resulting probabilities depicted in a radar chart.
Inclusion Criteria:
Exclusion Criteria: