Development of a Claims-Based Computable Phenotype for Ulcerative Colitis Flares.

Area of Science:

• Gastroenterology
• Medical Informatics
• Machine Learning

Background:

• Acute severe ulcerative colitis (ASUC) lacks a unique diagnostic code, hindering research and tracking.
• Current methods cannot automatically identify ASUC hospital admissions from observational data.

Purpose of the Study:

• To develop an automated method for identifying hospital admissions for ASUC.
• To enable large-scale research on non-coded conditions like ASUC.

Main Methods:

• Retrospective cohort study of ulcerative colitis (UC) patients (2014-2019).
• Trained logistic regression, random forest (RF), and support vector machine (SVM) models on administrative claims data.
• Validated model performance using chart review data.

Main Results:

• The RF model achieved 95.5% classification accuracy and an AUROC of 0.96.
• Key predictive features included endoscopy findings, length of stay, age, and abdominal X-ray.
• The model demonstrated 81.5% sensitivity and 96.5% specificity for ASUC identification.

Conclusions:

• A machine learning model can reliably identify ASUC admissions from claims data.
• This automated approach facilitates the creation of large, accurate datasets for ASUC research.
• Improved identification of non-coded conditions can advance understanding and research in real-world data.