Send us Fan Mail

Jörn Schattenberg discusses the issues that drove the project to develop a Machine Learning model that can identify patients likely to have NASH in clinical settings, and discusses key elements of study and model design.

Prof. Schattenberg and colleagues built the NASHmap model from a NIDDK database and validated it using the Optum de-identified EHR dataset. The model includes 14 variables, some obvious, others less so, that produce an AUC of 0.82 in the NIDDK database and 0.79 in the Optum database. This conversation focuses on why the collaborators undertook this project and how the model was built.