Explainable machine learning model for predicting spontaneous bacterial peritonitis in cirrhotic patients with...

Area of Science:

• Medical Informatics
• Machine Learning in Medicine
• Hepatology

Background:

• Spontaneous bacterial peritonitis (SBP) is a severe complication in patients with cirrhosis.
• Early prediction and outcome interpretation of SBP are crucial for patient management.

Purpose of the Study:

• To develop an explainable machine learning model for early prediction and outcome interpretation of SBP.
• To identify key predictive variables for SBP in cirrhotic patients with ascites.

Main Methods:

• Utilized the CatBoost algorithm to construct two predictive models (MODEL-1 with 46 variables, MODEL-2 after dimensionality reduction).
• Employed SHAP (SHapley Additive exPlanations) for model interpretability.
• Evaluated model performance using AUROC, sensitivity, and negative predictive value (NPV).

Main Results:

• Both models demonstrated strong predictive capability for SBP risk (AUROC: 0.822).
• MODEL-2 showed improved sensitivity (0.927) and NPV (0.904) compared to MODEL-1 (0.894 and 0.885, respectively).
• Identified six key predictive variables: total protein, C-reactive protein, prothrombin activity, cholinesterase, lymphocyte ratio, and apolipoprotein A1.

Conclusions:

• The CatBoost-based machine learning models provide a practical and explainable approach for SBP risk prediction in cirrhotic patients with ascites.
• The identified variables offer valuable insights into the factors contributing to SBP development.