Prediction of Imminent Peritoneal Dialysis-Associated Peritonitis Using Time-Updated Electronic Health Records and...

Area of Science:

• Nephrology
• Data Science
• Medical Informatics

Background:

• Peritoneal dialysis-associated peritonitis (PDAP) is a significant complication in continuous ambulatory peritoneal dialysis (CAPD).
• Early detection and risk stratification of PDAP are crucial for effective patient management and preventing adverse outcomes.
• Leveraging electronic health record (EHR) data offers a promising avenue for developing predictive models.

Purpose of the Study:

• To develop and validate a robust machine learning (ML) model for predicting PDAP onset within three months.
• To utilize time-updated data from routine EHR for enhanced predictive accuracy.
• To assess the clinical utility of the ML model for PDAP risk stratification in CAPD patients.

Main Methods:

• Retrospective cohort analysis of 1143 CAPD patients with 25,710 quarterly assessments.
• Feature selection from 31 EHR variables using low-variance filtering, correlation analysis, and Boruta selection.
• Construction and evaluation of nine ML models, including a Stacking ensemble, with a focus on recall for minimizing missed diagnoses.

Main Results:

• The Stacking ensemble model demonstrated good performance in internal validation (AUC 0.811, recall 0.794) and temporal validation (AUC 0.795, recall 0.833).
• SHapley Additive exPlanation (SHAP) analysis identified key predictive features, enhancing model interpretability.
• The model effectively stratified PDAP risk using time-updated EHR data.

Conclusions:

• Integrating time-updated EHR data with ML provides a robust method for PDAP risk stratification.
• The developed ML model offers clinically actionable insights for timely interventions in CAPD patients.
• This approach can optimize CAPD patient management and mitigate peritonitis-related complications.