Factors affecting the discriminant function analysis of blood chemistry profiles

The discriminant function analysis (DFA) of a battery of blood chemistry tests is generally more sensitive and specific than single biochemical test in detecting alcoholism or hazardous drinking. This study examined factors affecting the DFA of blood chemistry profiles (BCP) from alcoholics (ALC) in treatment, primary care outpatients (PC), and the general population (GP). Our data indicate that of the 32 variables in the BCP, 14 are not correlated with alcohol intake or consequences and can be deleted from the set considered for use in the DFA. Confounding medical conditions in the data set used in the DFA did not adversely affect the correct classification of ALC or heavy drinkers. It was not necessary to use only extreme groups (e.g., alcoholics vs. light drinkers/abstainers) to develop the discriminant function. Transformation of some blood variables greatly improved the selection of key variables in the DFA. Another finding was that there was little benefit from fitting the DFA to the demographics of subjects. When the BCP was combined with the results of a brief questionnaire, the TWEAK, it only improved slightly the classification power of the DFA based only on the TWEAK questions.