Prediction of Imminent Peritoneal Dialysis-Associated Peritonitis Using Time-Updated Electronic Health Records and Machine Learning: A Temporal Validation Study.

OBJECTIVE

To develop and validate a robust machine learning (ML) model for the onset of peritoneal dialysis-associated peritonitis (PDAP) within three months using time-updated data from routine electronic health record (EHR).

METHODS

A retrospective cohort analysis of 1143 unique continuous ambulatory PD (CAPD) patients generating 25,710 quarterly assessments (patient-semesters) from 2017 to 2025 was randomly divided into training (n=8537 observations), internal validation (n=8538), and temporal validation (n=6635 observations, 2024-2025) sets. Thirty-one EHR variables were processed via low-variance filtering, correlation analysis, and Boruta selection. Nine ML models (including a Stacking ensemble model) were constructed with patient-level stratified 10-fold cross-validation, optimizing for recall to minimize missed diagnoses. The primary outcome was PDAP onset within three months after routine laboratory tests.

RESULTS

In internal validation cohort, the stacking model achieved good performance with area under the curve (AUC) of 0.811 (95% CI 0.792-0.830) and the highest recall of 0.794 (95% CI 0.769-0.819). In temporal validation cohort, it maintained robust good classification performance, achieving AUC of 0.795 (95% CI 0.771-0.819) and the highest recall of 0.833 (95% CI 0.792-0.874). The SHapley Additive exPlanation analysis identified several key features, supporting model interpretability and clinical utility for PDAP risk stratification.

CONCLUSION

Integrating time-updated EHR data with ML enables robust and clinically actionable PDAP risk stratification, facilitating timely interventions to optimize CAPD patient management and reduce peritonitis-related complications.