Development of Time-Aggregated Machine Learning Model for Relapse Prediction in Pediatric Crohn's Disease

Sooyoung Jang1, JaeYong Yu2,3, Sowon Park4

  • 1Department of Biomedical Systems Informatics, Yonsei University College of Medicine, Seoul, Republic of Korea.

Insights

A new model effectively predicts pediatric Crohn's disease (CD) relapses using C-reactive protein and lymphocyte fraction. This tool aids in timely clinical decisions for managing pediatric CD activity.

Area of Science:

  • Gastroenterology
  • Pediatric Medicine
  • Clinical Prediction Modeling

Background:

  • Pediatric Crohn's disease (CD) has a higher tendency for active disease progression compared to adults.
  • Predicting and minimizing CD relapses in children is crucial for effective management.
  • Current prediction models for pediatric CD relapse at various time points are understudied.

Purpose of the Study:

  • To develop a real-time aggregated model for predicting pediatric CD relapse.
  • To identify optimal time points (TPs) and time windows (TWs) for relapse prediction.
  • To determine key variables influencing pediatric CD relapse.

Main Methods:

  • Retrospective study of 180 children diagnosed with CD (2015-2022).
  • Data collection included laboratory results and demographics starting 3 months post-diagnosis.
  • Cohorts formed at 6 TPs with 1-month intervals; relapse predicted using a 3-month TW at a 3-month TP.

Main Results:

  • An optimal TP of 3 months and a 3-month TW achieved a high prediction accuracy (AUC=0.89).
  • Key predictive variables identified include C-reactive protein levels and lymphocyte fraction.
  • The model demonstrated reliability in predicting pediatric CD relapse.

Conclusions:

  • A time-aggregated model was successfully developed to predict pediatric CD relapse across multiple TPs and TWs.
  • The model highlights critical variables for relapse prediction, aiding clinical decision-making.
  • This approach supports real-time management of pediatric Crohn's disease.
Abstract