Longitudinal ALT Trajectories are Generally Stable Among Patients with Non-Alcoholic Fatty Liver Disease (NAFLD): An Investigation Using Artificial Recurrent Neural Networks.

Challenge

ALT is widely used as a liver injury biomarker in NAFLD trials, but the natural longitudinal variability of ALT in real-world patients—and the patient characteristics that predict ALT trajectory—had not been modeled at scale, limiting confidence in ALT-based endpoint interpretation.

Solution

The TARGET-NASH cohort was analyzed using a recurrent neural network to model longitudinal ALT transitions across four categories, with a multivariable logistic model identifying predictors of ALT category increase, providing the first machine learning-based real-world ALT trajectory characterization in NAFLD.

Impact

Demonstrating that ALT trajectories are generally stable and characterizing predictors of change provides drug developers and regulators with empirical noise estimates for ALT-based endpoints, directly informing endpoint variability assumptions and responder threshold design in NAFLD trials.