The MERAGE score—encompassing Model for End-Stage Liver Disease (MELD), Ethnicity, Race, Age, Gender, and Education—achieved an area under the receiver operating characteristic curve (AUROC) of 0.69 in the validation cohort compared with 0.54 for the High-Risk Alcohol Relapse (HRAR) scale in patients hospitalized for alcohol-associated hepatitis (AH), according to research presented at the annual meeting of the American Association for the Study of Liver Diseases.
"MERAGE has superior performance for predicting return to drinking (RTD) compared to HRAR in AH, as demonstrated through independent validation," the researchers reported. "Including age, race/ethnicity, education, and MELD significantly improve the discriminant power."
The multicenter prospective observational cohort analyzed 347 patients with AH diagnosed per National Institute on Alcohol Abuse and Alcoholism consensus criteria, of whom 98 (28%) returned to drinking within 200 days of follow-up. The mean age was 45 years, with 43% female patients and 84% non-Hispanic White patients. Mean baseline MELD score was 25.
Patients who returned to drinking demonstrated greater numbers of daily drinks (12.7 vs 9.7), more drinking days (21.3 vs 13.6), and greater total drinks in the past 30 days (242.4 vs 133.6) compared with those who maintained abstinence. Baseline MELD scores were slightly lower in the return-to-drinking group (23.3 vs 24.9).
The HRAR scale, originally developed for liver transplantation candidates, incorporates duration of heavy drinking, number of daily drinks, and prior admissions to alcohol use disorder treatment programs. However, its utility in AH patients not undergoing transplantation remained unvalidated.
Researchers employed generalized additive models to capture nonlinear and interactive effects of drinking variables, with typical numbers of daily drinks and drinking days in the past 30 days included as a bivariate function. The bivariate analysis revealed return-to-drinking risk increased substantially in patients with more than 20 drinking days, particularly among those with higher daily intake. Data were split 70% to 30% for model training and validation.
Using an optimal threshold of 0.242 determined by Youden Index, MERAGE demonstrated sensitivity of 0.71, specificity of 0.60, positive predictive value of 0.34, and negative predictive value of 0.88 in the validation sample. By comparison, HRAR (optimal threshold=1.5) showed sensitivity of 0.38, specificity of 0.65, positive predictive value of 0.24, and negative predictive value of 0.78.
MERAGE also demonstrated improved overall predictive accuracy (62% vs 59%) compared with HRAR in the validation cohort. In the training sample, MERAGE showed even stronger performance with AUROC of 0.78 vs 0.55 for HRAR.
"RTD results from multiple clinical and behavioral factors whose nonlinear and interacting influences may determine the level of risk," noted Richard Sterling, MD, of Virginia Commonwealth University, and colleagues.
The researchers acknowledged that independent validation in broader patient populations is needed to establish practical utility.
Author disclosures were not provided in the conference abstract.
Source: AASLD