Background: Phenotypic methods such as elevated liver tests and/or the presence of obesity or hypertension are often used to identify patients with T2DM at risk for NASH. Machine learning (ML) models allow the prediction of outcomes with many interacting features and may be superior to phenotype in the context of NASH and T2DM. Here we present an intergroup comparison between patients with T2DM predicted as positive or negative for NASH by NASHmap, a ML model showing high prediction performance in real-word datasets.

Methods: We evaluated a subgroup of patients from a large T2DM clinical trial database (2002-2012) who had data for 14 NASHmap parameters. These include clinical observations and routine laboratory parameters but not HOMA-IR. We compared the profile of patients predicted by NASHmap (NASH positive or negative) using 17 parameters (14 NASHmap and 3 additional parameters). We compared HOMA-IR values between NASHmap positive and negative patients in the subgroup where the values were available.

Results: The clinical database had 23038 metabolically compensated T2DM patients without chronic liver disease of other etiologies. Data for all 14 NASHmap parameters were available in 4336 patients. NASHmap positive patients were younger, had higher HbA1c, liver enzymes, triglycerides, WBC, body mass index, platelets, and a higher rate for hypertension. Data for HOMA-IR were available in 1208 patients. Median HOMA-IR value was 4.2 in NASHmap positive group vs. 2.6 in NASHmap negative group. Patients in both groups had a similar median duration of diabetes.

Conclusion: Overall, the patients identified as at risk for NASH by NASHmap showed a metabolically unfavorable phenotype indicated by the higher HOMA-IR values in this group that the ML algorithm is also able to detect. ML models may be an important tool in effectively identifying patients at risk for NASH in electronic health records and clinical care.


A. Tietz: Employee; Self; Novartis AG, Stock/Shareholder; Self; Novartis AG. G. Bader: Employee; Self; Novartis AG. M. Docherty: Employee; Self; ZS Associates. B. Reinhart: Employee; Self; ZS Associates. M. Balp: Employee; Self; Novartis AG. M. Pedrosa: Employee; Self; Novartis AG. S. K. Acharya: Employee; Self; Novartis Healthcare Pvt. Ltd., Employee; Spouse/Partner; Novartis Healthcare Pvt. Ltd. J. Loeffler: Employee; Self; Novartis AG, Stock/Shareholder; Self; Novartis AG. J. M. Schattenberg: Advisory Panel; Self; Boehringer Ingelheim International GmbH, Consultant; Self; Boehringer Ingelheim, BMS, Echosens, Genfit, Gilead Sciences, Intercept Pharmaceuticals, Madrigal, Novartis, Novo Nordisk, Nordic Bioscience, Pfizer, Roche, Sanofi, Siemens Healthcare GmbH, Zydus, Bristol-Myers Squibb Company.

Readers may use this article as long as the work is properly cited, the use is educational and not for profit, and the work is not altered. More information is available at