Comprehensive Sepsis Risk Prediction in Leukemia Using a Random Forest Model and Restricted Cubic Spline Analysis.

Area of Science:

• Hematology
• Infectious Diseases
• Medical Informatics

Background:

• Sepsis is a critical complication in leukemia patients, associated with high mortality.
• Early identification of sepsis predictors is vital for prompt clinical intervention.
• Machine learning offers potential for developing predictive models in this high-risk population.

Purpose of the Study:

• To develop and validate a machine learning-based predictive model for sepsis risk in leukemia patients.
• To identify key clinical and laboratory predictors of sepsis in this cohort.
• To assess the performance of various machine learning models for sepsis prediction.

Main Methods:

• Retrospective analysis of 4310 leukemia patients' data (2005-2024).
• Feature selection using univariate logistic regression, LASSO, and Boruta algorithms.
• Development and evaluation of seven machine learning models, including random forest, using ROC curves and DCA; SHAP and RCS for interpretation.

Main Results:

• The random forest model demonstrated superior performance (AUC 0.765 training, 0.700 validation).
• Key predictors identified: C-reactive protein (CRP), procalcitonin (PCT), neutrophil count (Neut), lymphocyte count (Lymph), thrombin time (TT), red blood cell count (RBC), total bile acid (TBA), and systolic blood pressure (SBP).
• Non-linear relationships and interactions among predictors were significant.

Conclusions:

• The random forest model provides a robust tool for early sepsis risk assessment in leukemia patients.
• This predictive capability can aid clinicians in optimizing treatment strategies.
• Identifying key predictors facilitates targeted monitoring and intervention.