This was a meta-analysis featured on Drugs and Alcohol Findings in April and looking at the questions used to assess and detect unhealthy alcohol use.
Both AUDIT and AUDIT-C are known to accurately detect unhealthy drinking, but is one more accurate than the other? This paper looks for answers in 14 studies from across Europe and in the United States.
Summary The Alcohol Use Disorders Identification Test (AUDIT) was developed in 1993 by the World Health Organization, and is one of the most frequently recommended and researched diagnostic tests for detecting unhealthy drinking, alongside the Cut-Down, Annoyed, Guilty, and Eye-opener (CAGE) questionnaire and the Michigan Alcoholism Screening Test (MAST).
An abbreviated version of AUDIT, the Alcohol Use Disorders Identification Test-Consumption (AUDIT-C), was introduced a few years later. AUDIT-C is made up of three questions about alcohol intake from the full AUDIT, which also asks another seven questions about alcohol-related problems and symptoms indicative of dependence.
Evidence shows that AUDIT-C can be a useful and valid screening test for unhealthy drinking, as demonstrated in this study with patients in a Brazilian emergency department, and this study in three Veterans Affairs general medical clinics in the United States. Some differences have been found in the performance of AUDIT compared with AUDIT-C, but the authors of the present paper argue that the accuracy of one versus the other has not (until this study) “undergone systematic examination”.
This paper is based on a meta-analysis, amalgamating results from 14 studies which directly compared the accuracy of AUDIT-C with AUDIT for the detection of unhealthy drinking. The authors selected these studies as directly relevant and meeting pre-defined inclusion and quality criteria from a systematic search of six online databases, within a date range of 1998 and 2008.
The reported findings focus on primary care, as this was the most intensively researched setting. Eight of 14 studies were based in primary care, compared with four in general population samples, two in inpatient samples, and zero studies in an emergency department. The ‘target conditions’ of the 14 studies varied, but included risky drinking, harmful drinking, alcohol abuse, alcohol dependence, alcohol use disorder and unhealthy drinking (the latter covering the full spectrum from risky drinking to any alcohol use disorder).
Accuracy of the two tests was measured in a range of ways. For the purposes of this account, those easiest to interpret and most meaningful were “sensitivity” and “specificity”. In the current context, sensitivity can be understood as the proportion of respondents identified as risky drinkers by AUDIT or AUDIT-C who really are (normally as judged by a more comprehensive assessment) risky drinkers, while specificity is the proportion found not to be risky drinkers by these questionnaires who really are abstinent or drinking in a non-risky manner. Together these measures tell us how well the questionnaires pick up on risky drinking without also drawing into the net large numbers of non-risky drinkers.
The authors found considerable variation in the results of the tests, and variation in the way the tests were implemented. For example, the proportions of primary care patients identified as risky drinkers ranged from 11% to 35%, and the choice of threshold scores for risky drinking in primary care settings varied between 4 and 8 for the AUDIT test, and between 3 and 5 for AUDIT-C.
Overall, the accuracy of AUDIT and the AUDIT-C did not differ to a significant degree for screening for risky drinking, alcohol use disorders, or unhealthy drinking in primary care settings. However, one of the ways of comparing the tests did indicate a significant if small advantage for AUDIT over AUDIT-C in the identification of risky drinking (ie, above recommended maximums) in primary care settings. This measure was the ratio between the proportion of patients correctly identified as risky drinkers versus those incorrectly identified – the so-called ‘positive likelihood ratio’. The higher this ratio, the better the test is at correctly distinguishing who really is a risky drinker. For AUDIT the ratio was over twice as high (6.6 versus 3.0), a statistically significant difference.
With any such test there is trade-off between setting the threshold high so that nearly all those identified as risky drinkers really are, versus the increased chance that many risky drinkers will score below this threshold and fail to be spotted. This trade-off worked differently for the two tests. AUDIT-C’s ability to spot risky drinkers suffered less when such a high threshold was set that at least 85% of those who scored as risky drinkers really were. An alternative strategy is to set a low threshold, effectively casting a wide net so that nearly all drinkers in need of intervention are identified, while accepting that at the same time more people will be falsely identified as risky drinkers. In this scenario AUDIT was the more robust test, its ability to correctly exclude non-risky drinkers suffering less.
The authors’ conclusions
Although this study found no significant difference between the accuracy of AUDIT and AUDIT-C in primary care settings, this does not mean that it provides evidence of equivalent levels of accuracy of the two tests. It could be the case that with so few studies and with so many variations between the studies, the tests really do differ in accuracy, but research has yet to establish this. Indeed, some results (such as the difference between positive likelihood ratios when screening for risky drinking in primary care) indicated that AUDIT may be superior to AUDIT-C.
There are also some features beyond the degree of accuracy that may influence a practitioner’s choice of a diagnostic test. For example, the authors argue that the full AUDIT may “serve as a starting point for the exploration of the alcohol problem in a general practice situation because of its questions about the consequences of alcohol use. The MAST can provide a detailed description of a potential alcohol problem in settings where time constraints are not crucial. Finally, the CAGE test, with its 4 easily memorisable yes-or-no questions, may be preferable to both the AUDIT and the AUDIT-C, which have several response categories”.
Kriston L., Hölzel L., Weiser A.K., et al. Annals of Internal Medicine: 2008, 149(12), p. 879–888.
So the study is 8 years old but is a meta-analysis which makes it robust.
I would add that I always used to read the CAGE questions and justify that I didn’t meet the criteria because no one had criticised my drinking and I didn’t meet the eye-opener one because I didn’t drink in the morning. That said there had been occasions when I’d been up until the small hours drinking (4-5am) and we’d gone to the pub for lunch the next day where a bloody mary had been used to manage a hangover – you know hair of the dog and all that. I would argue that the time between stopping drinking the night before and a lunch time livener would have met that definition after all ………