Aim: To examine methods for combining quantitative results for serum carbohydrate deficient transferrin (CDT), gamma-glutamyltransferase (GGT) and/or aspartate aminotransferase (AST), and refining these by inclusion of patient characteristics.
Methods: Data from 1684 subjects, recruited from the general population, abstainer groups and alcohol treatment centres (participants in the five nations WHO/ISBRA study of biological markers of alcohol use), were used to develop clinical rules for combining results of GGT, AST and CDT. The algorithm derived by Sillanaukee and Olsson was tested, and compared with new algorithms derived by logistic regression and discriminant analysis. Diagnostic accuracy was assessed by area underneath the receiver operator characteristic curve. Effects of adding gender and clinical information to the algorithm were estimated.
Results: The predictive ability of combination rules derived from the two studies and by two different statistical techniques was remarkably consistent. For men, combining lnCDT and lnGGT provided the best accuracy for detecting daily consumption of 60 g ethanol or more in the past 30 days. For women, GGT alone provided the best accuracy for that consumption level. Clinical variables added significantly to the diagnostic accuracy of the models for both men and women, and conversely the test results modified the probability of problem drinking as assessed from clinical data alone. A graphic method was produced to help clinicians estimate probabilities for consumption of 60 g or more per day.
Conclusions: Combining biochemical markers enhances detection of problem drinking in men but not in women. Information on clinical variables increases the ability to correctly detect problem drinking.