Performance of Machine Learning Algorithms for Predicting Disease Activity in Inflammatory Bowel Disease

Area of Science:

• Gastroenterology
• Medical Informatics
• Computational Biology

Background:

• Inflammatory Bowel Disease (IBD) management requires accurate disease activity prediction.
• Current prediction methods may lack precision, necessitating advanced analytical tools.

Purpose of the Study:

• To evaluate the efficacy of various machine learning (ML) models in predicting IBD disease activity.
• To identify key clinical and laboratory variables influencing disease activity prediction in IBD patients.

Main Methods:

• Retrospective analysis of 876 IBD patients (275 UC, 601 CD) from 2011-2019.
• Application of Least Absolute Shrinkage and Selection Operator (LASSO) for variable selection.
• Development and comparison of seven ML models (RFs, AdaBoost, KNNs, SVMs, NB, Ridge, XGBoost) for disease activity prediction.

Main Results:

• All seven ML models demonstrated strong predictive performance for both Crohn's Disease (CD) and Ulcerative Colitis (UC).
• Support Vector Machines (SVM) achieved the highest accuracy (0.933) for CD prediction (AUC 0.975).
• Adaptive Boosting (AdaBoost) showed the best performance for UC prediction (AUC 0.911, accuracy 0.855).

Conclusions:

• Machine learning algorithms are capable and effective tools for predicting IBD disease activity.
• ML models, utilizing clinical and laboratory data, show significant potential to aid physician decision-making in IBD care.