Efficacy of the alcohol use disorders identification test as a screening tool for hazardous alcohol intake and related disorders in primary care: a validity studyBMJ 1997; 314 doi: https://doi.org/10.1136/bmj.314.7078.420 (Published 08 February 1997) Cite this as: BMJ 1997;314:420
- Marco Piccinellia, researcher (, )
- Elisabetta Tessari, clinical psychologistb,
- Marco Bortolomasi, resident in psychiatryb,
- Orazio Piasere, resident in psychiatryb,
- Massimo Semenzin, resident in psychiatryb,
- Nicola Garzotto, associate professor of psychiatryb,
- Michele Tansella, professor of psychiatrya
- a Servizio di Psicologia Medica, Istituto di Psichiatria, Università di Verona, Verona, Italy
- b Istituto di Psichiatria, Università di Verona, Verona, Italy
- Correspondence to: Dr Marco Piccinelli Servizio di Psicologia Medica, Istituto di Psichiatria, Ospedale Policlinico, 37134 Verona, Italy.
- Accepted 6 December 1996
Objective: To determine the properties of the alcohol use disorders identification test in screening primary care attenders for alcohol problems.
Design: A validity study among consecutive primary care attenders aged 18-65 years. Every third subject completed the alcohol use disorders identification test (a 10 item self report questionnaire on alcohol intake and related problems) and was interviewed by an investigator with the composite international diagnostic interview alcohol use module (a standardised interview for the independent assessment of alcohol intake and related disorders).
Setting: 10 primary care clinics in Verona, north eastern Italy.
Patients: 500 subjects were approached and 482 (96.4%) completed evaluation.
Results: When the alcohol use disorders identification test was used to detect subjects with alcohol problems the area under the receiver operating characteristic curve was 0.95. The cut off score of 5 was associated with a sensitivity of 0.84, a specificity of 0.90, and a positive predictive value of 0.60. The screening ability of the total score derived from summing the responses to the five items minimising the probability of misclassification between subjects with and without alcohol problems provided an area under the receiver operating characteristic curve of 0.93. A score of 5 or more on the five items was associated with a sensitivity of 0.79, a specificity of 0.95, and a positive predictive value of 0.73.
Conclusions: The alcohol use disorders identification test performs well in detecting subjects with formal alcohol disorders and those with hazardous alcohol intake. Using five of the 10 items on the questionnaire gives reasonable accuracy, and these are recommended as questions of choice to screen patients for alcohol problems.
Hazardous alcohol intake and related disorders are a major public health issue. Data from the World Health Organisation's collaborative project on psychological problems in general health care have shown that alcohol dependence or harmful use of alcohol as defined by the 10th revision of the International Classification of Diseases (ICD-10) is present in about 6% of primary care attenders, ranking third in frequency after major depression and generalised anxiety.1
In addition to formal alcohol disorders such as dependence or harmful use, increasing attention has been paid to hazardous alcohol intake, defined as a level of consumption or pattern of drinking which, if it persists, is likely to result in harm. Hazardous alcohol intake is directly or indirectly implicated in many physical, psychological, and social problems, imposing a substantial financial burden on the drinkers and on society.2 3 4 Moreover, drinking at levels causing detectable biochemical abnormalities is associated with a mortality that is twice that of the normal population.5
Primary prevention often requires national strategies promoting an overall decrease of alcohol consumption in the population. By contrast, secondary prevention can effectively be undertaken at the primary care level by means of early detection of people with hazardous alcohol intake and time limited interventions aimed at decreasing alcohol consumption and thus the likelihood of subsequent harm and dependence. Though several screening instruments have been developed that are fairly short and easy to administer, they tend to detect severe alcohol disorders such as dependence and overlook hazardous drinking. The WHO therefore devised a 10 item questionnaire–the alcohol use disorders identification test6–whose distinct advantage is the ability to detect both formal alcohol disorders and hazardous alcohol intake.
We investigated the screening properties of the alcohol use disorders identification test in the detection of primary care attenders with formal alcohol disorders or hazardous alcohol intake.
Subjects and method
Ten primary care physicians in Verona, north eastern Italy, allowed investigators to visit their clinics twice a week, once in the morning and once in the afternoon. Among patients aged 18-65 attending other than for a prescription, every third patient was approached up to a total of 50 patients at each clinic. Subjects were informed about the project and told that responses would be kept confidential. Those agreeing to participate had the size of a standard drink7 explained to them (see box) and then completed the alcohol use disorders identification test in the waiting room. In addition, the alcohol use module of the composite international diagnostic interview8 9 was administered by an investigator at the clinic on the same day or at the patient's home within a week. Investigators included three doctors and a final year student in psychology; they received group training in administering the composite international diagnostic interview and practised individually in role play sessions before the fieldwork. Finally, for each eligible subject the primary care physician rated on a form a list of clinical signs often related to alcohol consumption (for example, abnormal skin vascularisation, jaundice, hand tremor, liver characteristics); noted drinking behaviour over the previous 12 months (no alcohol abuse, occasional alcohol abuse, regular alcohol abuse); and noted the intake of psychotropic drugs during the two weeks before examination.
The alcohol use disorders identification test is a self administered questionnaire including three items on the amount and frequency of drinking, three on alcohol dependence, and four on common problems caused by alcohol (see 3). Each item is scored 0-4, giving a total score of 40.
The composite international diagnostic interview is a standardised diagnostic interview for assessing mental disorders according to criteria of the ICD-1010 and the Diagnostic and Statistical Manual of Mental Disorders, Third Edition, Revised (DSM-III-R).11
English versions of both instruments were translated into Italian, and the Italian versions were independently translated back into English; changes were made where necessary in order to ensure close correspondence between the original and Italian versions.
The screening properties of the alcohol use disorders identification test were tested against the standard criteria listed in the box. Criteria were fulfilled during the 12 months before examination and based on responses to the alcohol use module of the composite international diagnostic interview, which was the standard for the study.
Alcohol dependence and harmful use were diagnosed according to ICD-10 criteria. Defining hazardous alcohol intake was difficult, as the risk associated with alcohol consumption lies along a continuum. Recommendations on levels of safe drinking published in the United Kingdom by the Health Education Authority and supported by the Royal College of Psychiatrists, the Royal College of General Practitioners, and the Royal College of Physicians12 13 suggest that 30 g pure ethanol daily in men and 20 g daily in women constitute hazardous alcohol intake. The definitions of hazardous alcohol intake in this study (see box), based on categories of quantity and frequency of alcohol consumption from the alcohol use module of the composite international diagnostic interview, closely corresponded to the recommendations reported above.
Standard diagnostic criteria used in validating alcohol use disorders identification test
Alcohol dependence (at least three items required)(ICD-10)
Strong desire or sense of compulsion to take the substance
Impaired capacity to control substance taking behaviour in terms of onset, termination, or levels of use
Physiological withdrawal state when substance use is reduced or stopped or use of the substance to relieve or avoid withdrawal symptoms
Evidence of tolerance to the effects of the substance
Other pleasures or interests being given up or reduced because of the substance use
Persistent substance use despite clear evidence of harmful consequences
Harmful alcohol use (ICD-10)
Clear evidence that the substance use is responsible for (or is substantially contributing to) physical or psychological harm
The nature of the harm is clearly identifiable and specified
The pattern of use has persisted for at least one month or has occurred repeatedly within the 12 month period
The subject does not fulfil criteria for alcohol dependence
Hazardous alcohol intake
Men: Three to seven drinks almost every day or seven or more drinks at least three times a week
Women: Two to five drinks almost every day or five or more drinks at least three times a week
A standard drink was defined as equivalent volumes containing an average of 13.5 g ethanol. Definitions of a standard drink were based on local alcoholic beverages and included one glass of wine (125 ml), one bottle of beer (500 ml), and one measure of spirits (40 ml)7
The screening properties of the alcohol use disorders identification test were investigated by receiver operating characteristic analysis. This technique summarises the validity coefficients of a test and provides an overall index of diagnostic accuracy (that is, the area under the receiver operating characteristic curve) by plotting sensitivity against the false positive rate for all possible cut off scores. An area under the receiver operating characteristic curve of 0.5 is obtained when the discriminatory ability of a test is no better than chance; a value of 1.0 represents perfect discriminatory ability.14 A computer program for receiver operating characteristic analysis similar to that developed by Dorfman and Alf15 and modified by Metz et al16 was used in this study.
Logistic regression analysis was performed to identify linear combinations of items in the alcohol use disorders identification test that minimised the probability of misclassification between subjects with and without alcohol dependence, harmful use, or hazardous intake. A stepwise selection of predictor variables was adopted by using the likelihood ratio statistic as a test for removal and a probability level of 0.10 to remove a variable.
Five hundred subjects were approached at the primary care clinics, of whom 489 (97.8%) agreed to participate and 482 (96.4%) completed the evaluation. Most were women (n=306; 63.5%), married (290; 60.2%), and employed (274; 56.8%) and had low educational attainment (320 (66.4%) educated to secondary school level only). Mean age was 42.2 (SD 14.4) years. Seven subjects (1.5%) fulfilled ICD-10 criteria for alcohol dependence; all were men, with a median age of 43 years (range 21-61 years). Fifteen subjects (3.1%) fulfilled ICD-10 criteria for harmful alcohol use; 13 (86.7%) were men, with a median age of 50 years (range 24-65 years). Lastly, 62 subjects (12.9%) satisfied criteria for hazardous alcohol intake; 51 (82.3%) were men, with a median age of 48 years (range 21-65 years).
The screening characteristics of the alcohol use disorders identification test were initially tested separately against the diagnostic criteria listed in the box. The questionnaire performed well in detecting subjects with alcohol dependence (area under receiver operating characteristic curve 0.91; 95% confidence interval 0.88 to 0.94), harmful alcohol use (0.90; 0.88 to 0.92), and hazardous alcohol intake (0.92; 0.90 to 0.93). However, though sensitivity and specificity were above 0.8 irrespective of the criterion used, positive predictive values (that is, the probability of having the disorder among patients with positive test results) were low, indicating a high proportion of false positive results.
As the alcohol use disorders identification test is expected to be more suitable for initial screening of people with probable alcohol problems of any type rather than for accurate detection of people with formal alcohol disorders, the screening characteristics of the questionnaire were tested against all three drinking categories considered together. Table 1 shows that the performance of the questionnaire was high, with an area under the receiver operating characteristic curve of about 0.95. The cut off score of 5 provided a good trade off between sensitivity (0.84) and specificity (0.90); however, the positive predictive value was comparatively low, indicating that 40% of subjects scoring 5 or higher were false positive cases. Higher positive predictive values were found at higher cut off scores, though at the expense of decreased sensitivity; higher positive predictive values might be expected at lower cut off scores in populations with a higher prevalence of alcohol problems.
As low positive predictive values might result from the 10 items of the questionnaire being given the same weight in computing a total score, logistic regression analysis was performed to identify the items minimising the probability of misclassification between subjects with and without alcohol dependence, harmful use, or hazardous intake considered together. Estimated coefficients and related statistics from logistic regression analysis are not reported here but are available on request. Five items were retained in the model (goodness of fit 556.5; df=463, P=0.002): item 1 (frequency of drinking), item 2 (number of drinks on a typical day), item 4 (unable to stop drinking), item 5 (failing to do what was normally expected), and item 10 (another person concerned about subject's drinking or suggesting that it should be cut down). The discriminatory ability of the total score resulting from summing the responses to the five items is shown in table 2. Overall performance was high, with an area under the receiver operating characteristic curve of 0.93. A total score of 5 or more on the five selected items was associated with a sensitivity of 0.79, a specificity of 0.95, and a positive predictive value of 0.73; moreover, the probability of a subject scoring less than 5 having alcohol problems was less than 4%.
These findings can be compared with the low ability of doctors to detect patients with hazardous alcohol intake or formal alcohol disorders, only 39% of these patients being rated as abusers of alcohol either occasionally or regularly.
This study shows that the alcohol use disorders identification test is a simple questionnaire that takes only a few minutes to complete and performs well in detecting both people with formal alcohol disorders and those with hazardous alcohol intake. As five of the 10 items on the questionnaire are reasonably accurate for screening, physicians or other primary care professionals are recommended to use them as questions of choice to screen patients for alcohol problems of any type. Subsequent detailed evaluation can then be offered to those with positive test results in order to reach firm diagnostic conclusions. Our findings are similar to those from the exploratory WHO multicentre study,17 in which the 10 item alcohol use disorders identification test had a mean sensitivity of 0.80 and a mean specificity of 0.89 across participating centres.
Several screening instruments for alcohol disorders have been tested, including the Michigan alcoholism screening test18 and its shorter versions19 20 21, the CAGE questionnaire,22 the Veterans alcoholism screening test,23 and the primary care evaluation of mental disorders.24 In general the ability of these instruments to detect formal alcohol disorders is comparable to that of the alcohol use disorders identification test.24 25 However, most of the instruments have not been tested in the detection of hazardous alcohol intake; when this was done sensitivity failed at unacceptable levels.26 Other instruments, such as the Munich alcoholism test,27 require clinical examination to elicit physical signs related to excessive alcohol consumption, which makes them less likely to be used by busy physicians or prevents their use by non-medical professionals. Hence the alcohol use disorders identification test has definite advantages over existing screening instruments, as it can screen both for hazardous alcohol intake (possibly in patients before symptoms begin or in those with mild symptoms) and for formal alcohol disorders and can be used by health workers with no formal medical training.
We acknowledge that our study has possible limitations. Firstly, as data on alcohol consumption in the area were not available we did not perform a power calculation for required sample size and selecting comparatively few patients with alcohol problems might have affected the findings. Secondly, a proportion of subjects with alcohol problems might be expected to underreport them both on the alcohol use disorders identification test and at the diagnostic interview, with validity coefficients of the questionnaire being artificially raised. Independent data provided by primary care physicians suggest that this bias was limited, as three quarters of subjects with physical signs possibly due to excessive drinking reported alcohol problems at interview. No other sources of information (for example, spouse or other key informants, hospital records, biological markers, etc) were available to examine this issue further. Finally, some items included in the alcohol use disorders identification test were embodied within standard validating criteria, which might also have resulted in inflated estimates of test accuracy. Other validity studies using different sources of information and standard criteria may be useful to clarify this issue.
We thank the following primary care physicians for collaborating: M Bagnani, A Battagia, F Boninsegna, G Dal Cortivo, G Insom, R Montolli, G Piubello, G V Romanelli, P Sandri, and L Serra. We are also grateful to Dr Giulia Bisoffi for commenting on the study design.
Conflict of interest: None.