Skip navigation

Identifying heavy drinkers


A simple test to identify heavy drinkers, the CAGE test, was given in Bandolier 35 . Four simple questions are asked:



Scores of 3 or 4 were highly suggestive of heavy drinking in a large teaching hospital medical outpatient population with a 36% prevalence [1]. So how do laboratory tests compare?

Modelling laboratory tests


A study in 426 heavy drinkers and 188 light drinkers evaluated the performance of 40 laboratory tests [2], with the patient population split into a training data set of 411 subjects and a validation set of 171. As well as the 40 common laboratory tests, three new tests (carbohydrate-deficient transferrin, haemoglobin-associated acetaldehyde and ß-hexosaminidase) designed to tell heavy from non-drinkers were tested. A number of diagnostic models found from systematic search of the literature were also examined.

Method


Strengths of association between a number of laboratory tests and true result were found and a model constructed called the heavy drinking index (HDI). A score of 0 on the HDI indicated a non-drinker, and 0.26 and above (up to 1) a heavy drinker.

Results


Ten laboratory tests were included in the final model. Not even the most significantly associated test (chloride) would be helpful alone though. Even though it had the highest odds ratio, the mean value of 104.1 in heavy drinkers was not so very different from the value of 102.5 found in light drinkers.

The HDI successfully separated heavy from light drinkers, with trivial overlap and a sensitivity of 98% and specificity of 95% (likelihood ratio 20). No other model of laboratory test came close to the efficiency of this, with one exception, though all were significantly associated with heavy drinking. The three new single tests had sensitivities of 40% or less at a specificity of 95% (likelihood ratio 8 at maximum).

Comment


This model was designed and tested in an unusual population, in which 69% of subjects were heavy drinkers. Would its efficiency be the same in a primary care population with a much lower prevalence of heavy drinking? And who, in primary care, would be able to run a complicated model like this?

The extraordinary cleverness of a model that does well in one particular setting is lost if it is unusable. Bandolier often wonders why a morning blood sample is simply not sent for an ethanol estimation. Simple, cheap, and with a CAGE score might well provide the ability to detect heavy drinking even when prevalence is low.

References:

  1. DG Buchsbaum, RG Buchanan, RM Centor, SH Schnoll, MJ Lawton. Screening for alcohol abuse using CAGE scores and likelihood ratios. Annals of Internal Medicine 1991 115: 774-7.
  2. AJ Hartz, C Guse, A Kajdacsy-Balla. Identification of heavy drinkers using a combination of laboratory tests. Journal of Clinical Epidemiology 1997 50: 1357-68.




previous or next story in this issue