A Deep Learning Approach for Classifying Nonalcoholic Steatohepatitis Patients from Nonalcoholic Fatty Liver Disease Patients Using Electronic Medical Records

摘要

Nonalcoholic Steatohepatitis (NASH), an advanced stage of Nonalcoholic Fatty Liver Disease (NAFLD) causes liver inflammation and can lead to cirrhosis. In this paper, we present a deep learning approach to identify patients at risk of developing NASH, given that they are suffering from NAFLD. For this, we created two sub cohorts within NASH (NASH suspected (NASH-S) and NASH biopsy-confirmed (NASH-B)) based on the availability of liver biopsy tests. We utilized medical codes from patient electronic medical records and augmented it with patient demographics to build a long short-term memory based NASH vs. NAFLD classifier. The model was trained and tested using five-fold cross-validation and compared with baseline models including XGBoost, random forest and logistic regression. An out-of-sample area under the precision-recall curve (AUPRC) of 0.61 was achieved in classifying NASH patients from NAFLD. When the same model was used to classify out-of-sample NASH-B cohort from NAFLD patients, a highest AUPRC of 0.53 was achieved which was better than other baseline methods.