OBJECTIVES : To identify and characterize (a) broad cohort of probable nonalcoholic steatohepatitis (NASH) patients and (b) characterize those with evidence of progression to more severe disease.

METHODS : The OM1 Data Cloud (OM1, Boston, MA) collects, links and leverages structured and unstructured data from EMR, claims and other sources in the US, in an ongoing and continuously updating manner. A combination of sophisticated artificial intelligence (AI) algorithms were used to characterize the likelihood that a patient who is not an ICD10 identified NASH patient is a NASH patient. Algorithms were initially applied to a sample of >44 million obese and/or diabetic patients [ROC out-of-sample 0.86], then applied to a larger unselected U.S. population of ~240 million. Methods were validated by systematic comparisons to clinically documented NASH patients and alignment of clinical characteristics. Cohort identified in data available as of May 2019 was explored to characterize patients with severe manifestations by demography, comorbidities and treatments.

RESULTS : A total of 971,903 high-likelihood NASH patients were identified (including >140,000 with relevant ICD10 codes) and approximately 11,500 patients under age 18. Mean age was 56 years (SD 14.1) and male to female ratio was 51:49. Cirrhosis was documented in 24.4% of patients (including >500 pediatric patients), portal hypertension in 7.8% and history of liver transplant in 1.8%. Of the over 237,000 calculated FIB-4 scores in 77,862 patients, 24% of the results were indicative of advanced fibrosis, the majority of which were in patients not identified by relevant diagnosis codes.

CONCLUSIONS : This AI-based approach is a robust and broadly applicable means of identifying probable NASH patients that employs data collected during routine clinical care by the wide variety of healthcare providers managing patients. Key applications include exploring potentially aggressive phenotypes (e.g., pediatric NASH, rapidly progressing adult NASH) with the greatest unmet clinical need.