Machine learning models to predict disease progression among veterans with hepatitis C virus

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Chronic Disease Management

Background:

• Chronic hepatitis C virus (CHC) infection poses challenges for clinical risk prediction due to non-linear disease progression.
• Machine learning (ML) offers robust methods for building accurate prediction models using longitudinal data and numerous variables.

Purpose of the Study:

• To develop and compare machine learning algorithms for predicting cirrhosis development in a large CHC-infected cohort.
• To evaluate the efficacy of longitudinal versus cross-sectional data in predicting cirrhosis in CHC patients.

Main Methods:

• Utilized national Veterans Health Administration (VHA) data from 2000-2016 to identify CHC patients.
• Defined cirrhosis development by two consecutive AST-to-platelet ratio indexes (APRIs) > 2.
• Compared cross-sectional (CS) models with longitudinal models incorporating summary variables (max, min, slope, variation) and ML algorithms (Cox, boosted-survival-tree).

Main Results:

• Analyzed 72,683 CHC patients; 16% developed cirrhosis over a mean 7-year follow-up.
• Longitudinal boosted-survival-tree models showed superior predictive performance (concordance 0.774) compared to CS Cox models (concordance 0.746).
• Longitudinal models demonstrated higher accuracy (AuROC) at 1, 3, and 5 years post-time zero compared to CS models.

Conclusions:

• Boosted-survival-tree models leveraging longitudinal data are statistically superior for predicting cirrhosis in CHC.
• These ML approaches can be adapted for predicting outcomes in other non-linear chronic diseases.
• All models demonstrated high accuracy, highlighting the utility of ML in CHC risk stratification.