ROC analysis of alcoholism markers – 100% specificity

Abstract

A combination of 4 so-called markers of alcoholism, i. e. methanol, aceton + 2-propanol, γ-glutamyl-transferase and carbohydrate deficient transferrin, was investigated in 341 blood samples from alcoholics and non-alcoholics. From the history of alcohol consumption, four defined subgroups were formed: non-alcoholics divided into (A) 33 persons with no ethanol consumption during the past year and (B) 60 persons with daily consumption less then 40 g ethanol. Alcoholics were divided into (C) 177 persons with no ethanol at the time of admission/first blood sampling (withdrawal therapy) and (D) 71 persons with positive ethanol levels on admission/first blood sampling. All markers showed different extents of overlap between the collectives of alcoholics and non-alcoholics. By logistic regression, a formula was developed combining these markers with different mathematical weights. Thus an “Alc-Index” could be calculated for each individual. The ROC curve connecting all individual values gives an ideal form with 100% specificity and nearly 93% sensitivity. The threshold between the collectives of alcoholics and non-alcoholics was defined by the Alc-Index value 1.7. This was associated with no false positives among the non-alcoholics while nearly 93% of the alcoholics exceeded this index. The ROC-based calculation of the Alc-Index thus seems to be the most effective method for the diagnosis of alcoholism.