Artificial Intelligence in Identifying Patients With Undiagnosed Nonalcoholic Steatohepatitis

Area of Science:

• Hepatology
• Data Science
• Public Health

Background:

• Nonalcoholic steatohepatitis (NASH) prevalence is rising, yet diagnosis remains challenging in clinical settings.
• Undiagnosed NASH poses significant risks for disease progression and patient outcomes.

Purpose of the Study:

• To develop and validate a machine learning algorithm for identifying patients with probable undiagnosed NASH within the Veterans Affairs (VA) health system.
• To leverage claims data for efficient and scalable NASH risk stratification.

Main Methods:

• A machine learning model was trained and validated on a large VA dataset (over 4.2 million patients).
• The study employed a cross-validation technique, comparing gradient-boosted classification trees, naïve Bayes, and random forest models.
• Performance was evaluated using receiver operator characteristics, area under the curve, and accuracy metrics.

Main Results:

• The algorithm identified 514,997 (12%) patients as likely having NASH from the at-risk cohort.
• Key predictors for NASH probability included age, obesity, and abnormal liver function tests.
• The best-performing model demonstrated robust accuracy in identifying at-risk individuals.

Conclusions:

• Machine learning offers a powerful tool for the early recognition of undiagnosed NASH.
• This predictive capability facilitates timely interventions and personalized treatment strategies.
• The developed algorithm can serve as an effective initial screening tool for NASH.