Addiction (1993) 88, 791-804

RESEARCH REPORT

Development of the Alcohol Use DisordersIdentification Test (AUDIT):WHO Collaborative Project on EarlyDetection of Persons with Harmful AlcoholConsumption—II

JOHN B SAUNDERS/ OLAF G. AASLAND,' THOMAS F. BABOR,'JUAN R. DE LA FUENTE' & MARCUS GRANT'

'Centre for Drug and Alcohol Studies, Royal Prince Alfred Hospital, and Departments ofMedicine and Psychiatry, University of Sydney, Sydney, Australia, ^Ministry of Health andSocial Services, Oslo, Norway, ^Alcohol Research Center, Department of Psychiatry,University of Connecticut Health Center, Earmington, Connecticut, USA, *Instituto Mexicanode Psiquiatria, Mexico City, Mexico & ^Programme on Substance Abuse and Mental HealthDivision, World Health Organization, Geneva, Switzerland

AbstractThe Alcohol Use Disorders Identification Test (AUDIT) has been developed from a six-country WHOcollaborative project as a screening instrument for hazardous and harmful alcohol consumption. It is a10-item questionnaire which covers the domains of alcohol consumption, drinking behaviour, and alcohol-related problems. Questions were selected from a 150-item assessment schedule (which was administered to1888 persons attending representative primary health care facilities) on the basis of their representativenessfor these conceptual domains and their perceived usefulness for intervention. Responses to each question arescored from 0 to 4, giving a maximum possible score of 40. Among those diagnosed as having hazardous orharmful alcohol use, 92% had an AUDIT score of 8 or more, and 94% of those with non-hazardousconsumption had a score of less than 8. A UDIT provides a simple method of early detection of hazardousand harmful alcohol use in primary health care settings and is the first instrument of its type to be derivedon the basis of a cross-national study.

Introduction problems. It is a pro-active technique which aimsEarly intervention is a new approach to the pre- to identify persons with hazardous or harmfulvention and management of alcohol-related alcohol consumption* before dependence and

*In WHO terminology hazardous consumption is alcohol ^«"°"S l̂ "̂™. ^^ve occurred, and tO provide briefconsumption which confers the risk of physical and/or psycho- therapy, typically at the point of first Contactlogical harm (Edwards, Arif& Hodgson, 1981); harmful alcohol (World Health Organization, 1980; Babor, Rit-use (an ICD-10 diagnosis) is defined by the/iresencc of physical O T T J ,^C^ C^ J , . ^ ^ . , T .or psychological complications (World Health Organization, S°" '^ HodgSOn, 1986; Saunders, 1987; Institute'992), of Medicine, 1990; Saunders & Foulds, 1992).

791

792 John B. Saunders et al.

The primary goal of intervention is to facilitatereduction in alcohol intake to non-hazardouslevels, and thereby lessen the risk of harmfulconsequences of drinking. There is increas-ing evidence for its efficacy in this regard(Kristenson et al, 1983; Wallace, Cutler &Haines, 1988; Institute of Medicine, 1990;Saunders & Foulds, 1992). Through earlyidentification, treatment is provided well beforethe stage alcohol-affected patients usuallypresent for treatment, which typically is whenthey have suffered major health problems, psy-chological impairment and social decline. Earlyintervention is particularly appropriate for pri-mary health care settings, where the highthroughput of patients offers great opportunitiesfor systematic screening and therapy.

An essential component of any early interven-tion procedure is a simple and valid screeninginstrument to detect persons with hazardous orharmful alcohol consumption before dependenceand permanent harm have developed. Thepresent paper describes the derivation of a 10-item screening questionnaire which is basedupon experience in a World Health Organizationcollaborative study based in six countries (seeAppendix). It follows a companion paper (Saun-ders et al., 1993) which describes the prevalenceof alcohol use disorders and the inter-relation-ships of consumption and harm in a primaryhealth care population. The instrument, the Al-cohol Use Disorders Identification Test(AUDIT) is a development of the provisional'core' screening instrument described in the orig-inal report of the collaborative project (Saunders& Aasland, 1987).

Screening instruments for alcohol use disor-ders are not new. Nearly all existing instrumentshave, however, been developed to detect alco-holism, and not to screen for problem drinkingwhich is at an earlier or milder stage. The mostwidely known alcoholism questionnaire is proba-bly the Michigan Alcoholism Screening Test(MAST) (Selzer, 1971), of which several ver-sions now exist (Pokomy, Miller & Kaplan,1972; Selzer, Vinokur & Van Rooijen, 1975;Swenson & Morse, 1975). Questions wereselected for the MAST on the basis of theircapacity to discriminate between inpatient alco-holics and psychiatric patients who had nodiagnosis of alcoholism. The MAST correctlyidentified 98% of alcoholics and only 5% ofnon-alcoholic patients had a positive score. Sev-

eral items in the MAST reflect advancedphysical dependence or severe harm, or whetherthe subject identifies as an alcoholic. Examplesinclude "Have you ever had delirium tremens(DT's)?" and "Have you ever attended a meet-ing of Alcoholics Anonymous?". Among thebroader spectrum of problem drinkers, theMAST is less sensitive: 59% of persons con-victed of drunken and disorderly behaviour andonly 11% of persons whose vehicle licences wereunder review because of an alcohol-related of-fence were identified as 'cases' (Selzer, 1971). Ina community study, Saunders & Kershaw foundthat the short version of the MAST and a relatedinstrument, the CAGE, identified only a minor-ity of problem drinkers (Saunders & Kershaw,1980). One may conclude that these instrumentsare useful and very sensitive in screening foradvanced problems such as alcoholism, but areless suitable for detecting those with less severedrinking problems, who actually form a largerproportion of the general population (Mayfield,McLeod & Hall, 1974; Saunders & Kershaw,1980; Bemadt et al, 1982; Jacobson, 1983;Ewing, 1984; Hedlund & Vieweg, 1984; Bush etal, 1987; Waterson & Murray-Lyon, 1989).Similar comments may be applied to severalother screening instruments such as the MunichAlcoholism Test and the MacAndrew Alco-holism Scale (MacAndrew, 1965; Jacobson,1983; Feuerlein et al, 1986; Gottesmann &Prescott, 1989).

In Europe, screening techniques for alco-holism have been developed which are based onthe presence of abnormal physical findings. Theprincipal exponent of this approach has been theFrench physician Le Go. The 'Le Go Grid' (LeGo, 1976) is a brief clinical screening procedurewhich involves the detection of, inter alia, con-junctival injection, abnormal vascularization ofthe facial skin, coating of the tongue and hep-atomegaly. It has been taken up widely in Francefor the detection of alcoholism (Babor et al,1988). To date there has been little work on theuse of the Le Go Grid in detecting less severedrinking problems.

Biological markers of heavy alcohol consump-tion have also been used to screen for problemdrinking. They include serum gammaglutamyl-transferase (GGT) activity, serum aspartate andalanine aminotransferase activities (AST andALT respectively), HDL-cholesterol, uric acidand erythrocyte mean cell volume (MCV)

Development of AUDIT 793

(Rosalki & Rau, 1972; Wu, Chanarin & Levi,1974; Whitfield et al, 1978; Chick, Kreitman &Plant, 1981; Bemadt et al, 1982; Skinner et al,1984). The sensitivity of these markers is vari-able: up to 80% of alcoholics admitted to generalmedical or gastroenterology units have a raisedGGT or MCV (Rosalki & Rau, 1972; Wu et al,1974). However, in psychiatric hospitals and ingeneral population samples the sensitivity ofthese tests for the diagnosis of alcoholism ismuch lower, and they identify only 10-20% ofpersons with less severe drinking problems.(Chick et al, 1981; Bemadt et al, 1982). Allthese markers are also non-specific for the detec-tion of problem drinking, being abnormal inmany disease states. Recently discovered mark-ers, such as desialotransferrin (Stibler, Borg &Joustra, 1986; Storey et al, 1987) and antibodiesto acetaldehyde-altered plasma proteins (Israel etal, 1986), may prove to be more sensitive thanexisting tests. Another approach is to assay alco-hol in body fluids, such as saliva and sweat. Thisis a highly specific method, and an 'alcohol dip-stick' has been developed for the purposes ofmass screening (Kapur & Israel, 1984). Thedipstick provides a rapid, semi-quantitativeestimate of alcohol concentration in saliva.However, all tests based on detecting alcohol arelimited to identifying consumption within theprevious 24 hours and cannot in themselves esti-mate the severity of an underlying clinicalsyndrome.

At the time the present collaborative study wasbeing devised (in 1983) there seemed to beseveral major limitations to the screening instru-ment currently available. Screening question-naires, clinical examination procedures andlaboratory tests had been developed mainly todistinguish between hospitalised alcoholics andnormal drinkers. Their suitability for identifyingthe broad spectrum of problem drinkers waslimited or had not been tested. Several appearedto be highly culturally specific, and as Room hascommented, 'off the shelf screening tests origi-nating in North America, the United Kingdomand Europe may not be relevant for other cul-tures (Room, 1988). Thirdly, few had beendeveloped for primary health care settings. Theaim of the study was therefore to develop aninstrument that would screen effectively for abroad spectrum of problem drinking andspecifically for hazardous and harmful alcoholconsumption. Thus, it should identify subjects

before dependence, physical disease or major lifeproblems had developed. It should be valid indifferent cultures. It should be suitable for pri-mary health care, and therefore would need to bebrief and easily understood to encourage itswidespread use.

MethodSampling frameAs described in the preceding paper (Saunders etal, 1993), the collaborative project involved re-search centres in Australia, Bulgaria, Kenya,Mexico, Norway and the USA. Subjects wererecruited from representative primary health carefacilities and classified on the basis of their re-sponses in a structured interview as either 'nondrinkers' (total abstainers or those who drankalcohol on three or fewer occasions per year andhad never been treated for an alcohol problem),'drinking patients' (who drank on at least fouroccasions per year and had never received treat-ment for a drinking problem) and 'alcoholics'(who had been diagnosed as alcoholic, had re-ceived treatment for an alcohol-related disorderor were seeking treatment). In all, 1905 subjectswere interviewed in the six countries. After 17subjects were excluded because of incompletedata on alcohol consumption, 1888 remained toform the study sample. Of these 678 (36%) wereclassified as non-drinkers, 913 (48%) wereclassified as drinking patients, and 297 (16%)were termed alcoholics. Only data obtained firomthe drinking patients were used to select itemsfor AUDIT. The non-drinkers and alcoholicsformed reference groups for instrument valida-tion.

AssessmentEach subject was interviewed by a trained inter-viewer who administered a 150-item schedulewhich encompassed socio-demographic vari-ables, presenting conditions, current sympto-matology, past medical history, alcoholconsumption, other substance use, diet, drinkingbehaviour, psychological reactions to alcohol,alcohol-related problems, family history of alco-holism, and self-perception of an alcoholproblem (Saunders & Aasland, 1987). The inter-view was supplemented by a clinical exami-nation, and blood samples were taken for

794 John B. Saunders et al.

biological markers of alcohol consumption.Scores for several scales within the questionnairewere calculated according to a frequency rating(from 0 = never to 4 = daily) or a bimodalresponse (0 = no, 1 = yes) for individual items.

Data analysisThe analyses proceeded in a stepwise fashion.Firstly, the firequency distribution of responsesby the drinking patients to all questions in theinterview schedule was examined. Questions towhich fewer than 2% of respondents answered inthe affirmative were excluded from considerationfor the screening instrument. Following this theintrascale reliability of each conceptual domainwas determined for each national sample bycomputing Cronbach's alpha coefficient. A valueof 0.7 or more was taken as indicating satisfac-tory reliability. The score for each scale wascorrelated with various measures of alcohol in-take, including mean daily alcohol intake andfrequency of having six drinks or more per ses-sion, both of which were transformed beforecorrelation analysis by taking the natural loga-rithm. A correlation coefficient of 0.40 or morewas taken as indicating a sufficiently close rela-tionship with alcohol consumption to warrantfurther examination. For scales which fulfilledthese criteria in most national samples, the itemto total-minus-item correlation coefficients werecalculated to identify the most representativeitems. A correlation matrix was then constructedof all scales and individual variables, and factoranalysis was performed to examine their inter-relationships and identify any domains stronglyassociated with alcohol intake which had notbeen covered by candidate questions. Theseanalyses were carried out for each national sam-ple and then for the aggregated data. Thetest-retest reliability of potential questionnaireitems was then assessed, together with theirvalidity in comparison with information fromcollateral informants.

Questions were not selected for AUDIT solelyon the basis of statistical parameters. We made ajudgement on the suitability of questions forscreening and on their usefulness as a focus fortherapy. Their face validity was an importantconsideration. Would they be understood readilyby patients and provide a basis for discussion ofthe person's drinking problem? In addition, thevalue of including the major conceptual domains

(alcohol consumption, drinking behaviour (de-pendence), and alcohol-related problems) andthe adequacy of coverage of conceptual elementswithin a domain were taken into consideration.

Reference standards and instrument validationTo calculate the interim sensitivity andspecificity of AUDIT, several reference stan-dards were devised. They include hazardousalcohol consumption, which was defined for thisanalysis as an average daily alcohol intake ex-ceeding 60 g per day for men and 40 g per dayfor women. Recurrent intoxication was definedas consumption of 60 g in a single session dailyor almost daily, or 120 g per session at leastweekly. 'Abnormal drinking behaviour' was diag-nosed when subjects fulfilled at least onecriterion of the alcohol dependence syndrome (ata minimum frequency of monthly) on the drink-ing behaviour scale. An alcohol dependencesyndrome was diagnosed when at least threecriteria were fulfilled (World Health Organiza-tion, 1992). 'Alcohol-related problems' encom-passed domestic, legal and occupationaldifficulties related to alcohol, traumatic injurycaused by drinking, and concern expressed byfamily, friends or health professionals. Finally, acomposite index of hazardous or harmful alcoholuse was constructed. A positive 'case' wasdefined as having any one of the following: ahazardous daily level of consumption; recurrentintoxication; abnormal drinking behaviour; atleast one alcohol-related problem in the last year;an alcohol-related disease; or a perceived drink-ing problem. The instmment finally selected wasscored from 0 to 40. Cut-off values were exam-ined by receiver operating characteristic (ROC)analysis to identify the point(s) of maximal sensi-tivity and specificity with respect to hazardousand harmful alcohol use across the six samples.The sensitivity of the final instrument for detect-ing known alcoholics (an external referencegroup) was also determined.

ResultsIntrascale reliability of conceptual domainsExamination of the alpha coefficients showedthat among the drinking patients the drinkingbehaviour and adverse psychological reactionsdomains had high intrascale reliability, withmean values of 0.93 and 0.81 respectively

Development of AUDIT 795

Table 1. Item correlations and intrascale reliability coefficients for the drinking behaviour (alcohol dependence) scale

1. Not able to stopdrinking once started

2. Difficult to stop beforebecame completely intoxicated

3. Fail to do what was expectedbecause of drinking

4. Drunk for days5. Moming drinking6. Difficult to get the thought

of alcohol out of your mind7. Skipped meals because

of drinking8. Drinking at unusual times9. Moming shakes

10. Tried to reduce consumptionbut failed

11. Needed more alcohol thanbefore to get the desired effect

12. Drank more than friends13. Gulping drinks to

speed the effect

Cronbach's alpha

Australia

0.49

0.38

0.560.440.49

0.40

0.550.380.57

0.26

0.210.27

0.28

0.80

Bulgaria

0.71

0.71

0.640.800.78

0.58

0.560.790.62

0.48

0.770.61

0.39

0.90

Kenya

0.92

0.89

0.850.850.64

0.82

0.810.830.76

0.81

0.880.88

0.74

0.97

Mexico

0.91

0.94

0.930.960.92

0.83

0.910.900.86

0.94

0.870.83

0.56

0.98

Norway

0.87

0.88

0.890.830.79

0.84

0.690.640.70

0.69

0.490.58

0.66

0.94

USA

0.76

0.81

0.530.370.47

0.65

0.550.450.59

0.51

0.570.61

0.54

0.88

Weighted meancorrelation

0.81

0.81

0.780.780.73

0.71

0.700.700.69

0.67

0.660.64

0.53

0.93

(Tables 1 and 2). There was relatively littlevariatioti from country to country. The values forthe two domains 'alcohol problems in the lastyear' and 'alcohol problems ever' were lower at0.69 and 0.65 respectively (Table 3), and variedsignificantly among the six samples (fromapproximately 0.35 to 0.83). There was a mod-erately strong correlation between the drinkingbehaviour (dependence) and mean daily alcoholconsumption (r=0.53) and between adversepsychological reactions, alcohol problems in theprevious year and alcohol problems ever scales,and intake (r= 0.50, 0.50 and 0.51 respectively).The non-alcohol-specific domains such as cur-rent symptoms, past medical history and clinicalexamination findings had much lower values forintrascale reliability, and weaker and often non-significant correlations with mean daily alcoholintake (Saunders & Aasland, 1987). Factor anal-ysis reported in more detail in the previous paper(Saunders et al, 1993) showed that the alcohol-specific scales and measures of consumptionloaded on a dominant first factor. Current symp-toms was the only non-alcohol specific domainto have a factor coefficient exceeding 0.40 (actu-ally 0.46), and was also located in the thirdfactor extracted, together with the clinical exam-ination scale. On the basis of these analyses andto ensure adequate coverage of the most relevant

domains, it was decided that the screening in-strument should include questions on thefollowing: (1) alcohol consumption; (2) drinkingbehaviour (dependence); (3) adverse psychologi-cal reactions; and (4) alcohol-related problems.It was decided not to include any questions fromthe non-specific domains.

Selection of itemsFrom the drinking behaviour scale, three ques-tions were selected: (1) How often during thelast year have you found that you were not ableto stop drinking once you had started? (2) Howoften . . . have you failed to do what was nor-mally expected from you because of drinking?and (3) How often . . . have you needed a firstdrink in the moming to get yourself going after aheavy drinking session? TTie first was selectedbecause it had the highest weighted mean item-to-total correlation coefficient (0.81), andshowed the most uniform pattem of correlationacross the centres (Table 1). It thus appeared tobe the most representative item for the scale. Itrefiects impaired control over consumption ofalcohol. The second question had the thirdhighest overall correlation coefficient (0.78), anda narrow range of values fi-om centre to centre. Itrepresents salience of drinking, i.e. neglect of

796 John B. Saunders et al.

Table 2. Item correlations and intrascale reliability coefficients for the scale 'adverse psychological reactions'

1. Had a feeling of remorseafter drinking

2. Become depressedafter drinking

3. Been unable to rememberwhat happened the night beforebecause you had been drinking

4. Become angry after drinking

Cronbach's alpha

Australia

0.38

0.49

0.480.310.64

Bulgaria

0.31

0.15

0.200.260.73

Kenya

0.73

0.70

0.590.630.76

Mexico

0.69

0.46

0.510.460.90

Norway

0.56

0.21

0.470.370.79

USA

0.55

0.31

0.290.590.75

Weighted meancorrelation

0.53

0.44

0.440.430.81

alternative interests and obligations in favour ofdrinking. The third question had the fifth highestcorrelation (0.73) and refers to relief of malaiseafter a drinking session by further consumptionof alcohol. Although this may be a relativelyitinocuous experience at first, refiecting relief ofhangover symptoms, when it occurs on a dailybasis it indicates physical dependence. It wasconsidered advantageous for both behaviouraland physiological aspects of drinking behaviourto be represented. The response categories,'never', 'less than monthly', 'monthly', 'weekly'and 'daily' or almost daily', are identical to thosein the original assessment instrument. Responsesare scored from 0 to 4 respectively.

Two questions were selected to represent ad-verse psychological reactions: (1) How oftenduring the last year have you had a feeling ofguilt or remorse after drinking? and (2) Howoften during the last year have you been unableto remember what happened the night beforebecause you had been drinking? The first had thehighest mean item-to-total coefficient in thisscale (Table 2). The second showed the mostuniform pattem of correlations of the three re-maining questions. Refiecting as it does theexperience of alcohol-induced amnesic episodes('blackouts'), it was considered to be a particu-larly useful cue for further enquiry about adverseeffects of alcohol. Responses are scored as for thequestions on drinking behaviour.

The two sections on alcohol-related problems('in the last year' and 'ever') were more het-erogenous than the other alcohol domains, withmean alpha coefficients of only 0.69 and 0.65(Table 3). The overall item-to-total correlationsof the five questions were very similar (Table 3)but there was appreciable variation from country

to country. Pragmatic considerations enteredinto the selection of items to a greater extentthan elsewhere. It was decided to include a ques-tion on concem shown by family or healthworkers and one on alcohol related-injuries, asfollows: (1) Has a relative or friend, or a doctoror other health worker, been concerned aboutyour drinking or suggested you cut down? and(2) Have you or someone else been injured as aresult of your drinking? Response categorieswere selected so that two different time frameswere covered. An answer which indicates thatthe event occurred in the last year scores 4,whereas if it had happened before that, the scoreis 2.

The remaining questions measure alcohol con-sumption: (1) How often do you have a drinkcontaining alcohol? (2) How many drinks con-taining alcohol do you have on a typical daywhen you are drinking? and (3) How often doyou have six or more drinks on one occasion?They were selected as being the simplest way ofcapturing the usual fi-equency of drinking, thequantity consumed and the frequency of episodicheavy drinking. Episodic heavy drinking of thisorder would typically result in blood alcoholconcentrations that would cause impairment offunction. One 'drink' is assumed to contain ap-proximately 10 g of alcohol; if in a particularculture, a typical drink contains substantiallymore or less than 10 g alcohol, these questionsshould be rephrased or the response categoriesaltered.

Performance of the instrumentThe 10 items described above form the AlcoholUse Disorders Identification Test (AUDIT).

Development of AUDIT 797

Table 3. Item correlations and intrascale reliability coefficients for 'alcohol problems ever'

1. Family or friend concemedabout your drinking

2. Work difficulties becauseof drinking

3. Doctor or other health workerconcemed about your drinkingor suggested that you cut down

4. Legal trouble in connectionwith drinking

5. You or someone else injuredas a result of your drinking

Cronbach's alpha

Australia

0.38

0.49

0.48

0.31

0.42

0.65

Bulgaria

0.31

0.15

0.20

0.26

0.22

0.40

Kenya

0.73

0.70

0.59

0.63

0.48

0.83

Mexico

0.69

0.46

0.51

0.46

0.61

0.77

Norway

0.56

0.21

0.47

0.37

0.26

0.60

USA

0.55

0.31

0.29

0.59

0.42

0.67

Weighted meancorrelation

0.53

0.44

0.44

0.43

0.40

0.65

Questions 1-3 measure alcohol consumption,4-6 drinking behaviour, 7-8 adverse reactionsand 9-10 alcohol-related problems. Each ques-tion is scored fi-om 0 to 4, and the range ofpossible scores is fi-om 0 to 40. The performanceof AUDIT was examined in two ways, firstly bycomparing scores against the diagnoses of haz-ardous and harmful alcohol use in the samples ofdrinking patients, and secondly by calculating itssensitivity among the extemal reference group ofknown alcoholics. Cut-off points for AUDITwere determined by receiver operating character-istic (ROC) analysis. The sensitivity andspecificity of scores from 0 to 40 were calculatedfor the diagnoses of (1) hazardous daily alcoholconsumption and/or recurrent intoxication, (2)abnormal drinking behaviour, (3) alcohol relatedproblems in the last year, and (4) the combinedindex of hazardous and harmful alcohol con-sumption. Two cut-off points, 8 and 10 wereidentified which resulted in maximal sensitivityand specificity.

The results which are presented in Tables 4and 5 should be regarded as interim estimates ofthe sensitivity and specificity of AUDIT. Theyrepresent a comparison between the AUDITscores and diagnoses based on multiple items ofinformation obtained during a comprehensive,structured interview, supplemented by physicalexamination and laboratory findings. Using thelower cut-off point of 8, the sensitivity ofAUDIT for hazardous alcohol consumptionranged fi-om 95% to 100%, for abnormal drink-ing behaviour fi-om 93% to 100%, for thealcohol dependence syndrome it was 100%, and

for "alcohol problems in the last year" it rangedfrom 91% to 100% (Table 4). The overall sensi-tivity for hazardous and harmful alcohol use was87% to 96%, with an overall value of 92%(Table 4). The corresponding specificity was81% to 98%, with an overall value of 94%.When the cut-off point of 10 was taken, sensitiv-ity was lower (Table 5), with an overall value of80% for the combined index. The specificity wascorrespondingly higher: for the combined indexvalues ranged from 95 to 100%, and the overallvalue was 98%.

The validity of AUDIT was then determinedamong the extemal reference groups of knownalcoholics and non-drinkers. Of the alcoholics,99% had a score of 8 or more, 98% had a scoreof 10 or more and when those who were cur-rently abstinent were excluded, all scored 10 ormore. Only three of 678 non-drinkers (0.5%)had a score of 8 or more. As a final check on theaccuracy of the process for selecting items, theAUDIT questions were compared vnxh 20 othercombinations of questions for their ability todiscriminate between patients with hazardous orharmful alcohol use and those with non-hazardous consumption. The AUDIT questionsclassified subjects as accurately as any othercombination.

DiscussionThe development of screening instruments isnecessarily linked to prevailing concepts of alco-hol problems. The first screening tests wereaimed at identifying alcoholics and were success-

798 John B. Saunders et al.

Table 4. Sensitivity and specificity of the 'AUDIT' questionnaire at cut-off point of 8

AustraliaBulgariaKenyaMexicoNorwayUSAAll countries

Hazardousconsumption/

recurrentintoxication

100/7495/7697/66

100/77100/8996/7497/78

Abnormal drinking behaviour(at least one element ofdependence at specified

minimum frequency)

.97/75100/7497/6993/8994/9197/7996/81

Alcohol-relatedproblems in the

last year

95/81100/72100/7194/9891/9492/8395/85

Combined indexof hazardousand harmfulalcohol use

93/8296/8795/8188/9887/9790/9292/94

ful in doing so. Sensitivities exceeding 95% forthe MAST and its progeny attest to this. How-ever, the ability of this family of instruments toidentify patients with hazardous or harmful alco-hol use is much lower (Selzer, 1971; Saunders& Kershaw, 1980; Hedlund & Vieweg, 1984).Another problem with these questiotmaires isthat because questions are phrased in terms oflifetime occurrence, it may be impossible todistinguish current alcoholics or problemdrinkers from those who have ceased drinking orare no longer symptomatic. Indeed, this problemhas led to the development of a revised responseformat for the MAST that provides better dis-crimination between current and lifetimesymptoms. Reworking of the MAST has not,however, resulted in an instrument which em-braces the broad spectrum of problem drinkers.

Clinical examination findings (Le Go, 1976;Skinner et al., 1986; Babor et al., 1988) have alsobeen employed to screen for alcoholism and itsphysical sequelae. Although the medical impactof heavy drinking is widespread, clinical abnor-malities occur relatively late in the evolution of adrinking problem (Babor, Kranzler & Lauerman,1987) and appear to be of limited value for earlyintervention. Other correlates of chronic drink-ing, such as hypertension, are not sufficientlyspecific to be of much value in screening. How-ever, they may help to confirm an impressionthat alcohol consumption is harmful. Establishedbiological markers have also proved relativelyinsensitive in screening for hazardous alcoholconsumption (Whitfield et al., 1978; Chick et al,1981; Bemadt et al, 1982). They are alsoaffected by substances other than alcohol as wellas diseases unrelated to drinking. Serum trans-ferrin and new immunological tests developed tomeasure acetaldehyde bound to plasma protein

show promise as more specific markers of heavydrinking, but further research is needed toconfirm their usefulness in routine screening(Stibler et al, 1986; Israel et al, 1986; Storey etal, 1987). The ideal biological marker would beone that is even more sensitive and able toidentify gradations in alcohol use. Such a markeris not yet available.

Composite instruments have been devisedwhich seek to capture a broad spectrum of phys-ical and psychosocial consequences of alcoholuse. It was hoped this would increase the sensi-tivity and yield of the screening process. Wilkinsdeveloped a two-stage procedure which incorpo-rated a check-list of clinical indicators and adisguised questionnaire, the "Spare Time Activi-ties Questionnaire" (Wilkins, 1974). TheMunich Alcoholism Test (MALT) (Feuerlein etal, 1986) and the Alcohol Clinical Index (Skin-ner et al, 1986) also include clinical findings aswell as questionnaire items. Although they areconceptually appealing and, for example, theAlcohol Clinical Index provides good discrimina-tion between non-dependent problem drinkersand non-problem drinkers, they are relativelycomplex and they have not been taken up widelyin clinical practice. Any method based on clinicalexamination findings would normally require theinvolvement of a medical practitioner. Likewise,most biological tests require sophisticated labo-ratory equipment and computer technologywhich might preclude their use in primary healthcare settings, particularly in developing coun-tries.

Selection of items for AUDIT was guidedboth by the statistical analysis and certain opera-tional requirements. The latter included thedesirability of representing the conceptual do-mains of alcohol consumption, dependence and

Development of AUDIT 799

Table S. Sensitivity and specificity of the AUDIT' questionnaire at cut-off point of 10

AustraliaBulgariaKenyaMexicoNorwayUSAAll countries

Hazardousconsumption/

recurrentintoxication

97/8684/8794/79

100/8390/9486.8192/87

Abnormal drinking behaviour(at least one element ofdependence at specified

minimum frequency)

88/8692/8594/8291/9588/9686/8590/90

Alcohol-relatedproblems in the

last year

85/9284/8297/8484/10074/9887/9186/92

Combined indexof hazardousand harmfulalcohol use

79/9879/9691/9579/10067/9976/9680/98

problems, the need for a rounded number ofquestions and for all questions to be easily un-derstandable, valid across different cultures, andcapable of providing the framework for subse-quent intervention. Analysis revealed that insome scales (for example, drinking behaviour)many items were virtually interchangeable interms of their correlation with total scores, orcapacity to distinguish between hazardous andharmful drinkers and those in the non-hazardous range. The final 10-item instrumentincluded three questions on intake, four on alco-hol-related problems and adverse reactions, andthree on drinking behaviour. Because these do-mains conform approximately to the concepts ofhazardous alcohol use, harmful use and depen-dence in WHO terminology and the ICD-10psychoactive substance use disorders section(World Health Organization, 1992), the test wasnamed the Alcohol Use Disorders IdentificationTest (AUDIT).

The differences between AUDIT and mostexisting questionnaires can be summarised asfollows. Firstly, it seeks to detect problemdrinkers at the less severe end of the spectrum,rather than targetting persons with establisheddependence or alcoholism. Secondly, it placesconsiderable emphasis on hazardous consump-tion and firequency of intoxication comparedwith drinking behaviour and adverse conse-quences. It refers to alcohol experiences in thepast year as well as lifetime experience. Thisimproves its relevance to current drinking status.It does not require the individual to identify as aproblem drinker. The responses are not yes/noones but are based on the frequency of theexperience, and range from 'never' and 'less thanmonthly' through to daily. It is anticipated that

this will reduce under-reporting of adverseeffects. Not surprisingly, there are also somesimilarities with existing instruments. Two of theten items (guilt about drinking, and drinking inthe moming) are similar to those in the CAGE(Ewing, 1984), and five resemble questions inthe MAST (Selzer, 1971). AUDIT is shorterthan both the full AdAST and a more recentlyintroduced instrument wth a similar conceptualbasis, the Canterbury Alcoholism Screening Test(Elvy & Wells, 1984). It can be embedded withingeneral health and lifestyle questionnaires with-out making them unwieldly or upsetting theirbalance. This would provide an element of dis-guise which may be useful in some settings.

The comparative merit for screening purposesof "consumption" questions and "adverse conse-quences" ones has been the subject of debate.Ryder and colleagues reported that questions onalcohol-related problems were more sensitive in-dicators of problem drinking than consumptionones (Ryder et al., 1988). In an antenatal popu-lation, however, quantity-frequency questionsgave a higher yield than the CAGE or BriefMAST (Waterson & Murray-Lyon, 1989). Themain conclusion from these and related studies isthat a screening instrument for general purposesshould combine consumption and conse-quences questions (Barrison et al., 1982; Cutler,Wallace & Haines, 1988; Cyr & Wartman, 1988;Persson & Magnusson, 1988; Babor et ai,1989a).

The unique feature of AUDIT is that it hasbeen derived from a cross-national data set. Onlythose questions which could be translated, liter-ally and idiomatically, into multiple languageswere included in the original assessment sched-ule. Questions were selected on their repre-

800 John B. Saunders et al.

sentativeness in the pooled data set and checkswere made to ensure that none performed poorlyin an individual national sample. The sensitivityand specificity of AUDIT (Tables 4 and 5) aresimilar from country to country, and there is noevidence of dominance by one particular cultureas judged by these parameters. Experience usingexisting screening instruments in other cultureshas often been unsatisfactory. Only five of the 31criteria in the Munich Alcoholism Test, derivedin Germany, were found to be 'relatively firee ofcultural differences' when the test was applied inSpain and Ecuador (Gorenc et al., 1984). Use ofan alcoholism questionnaire in an Alaskan Es-kimo population resulted in 75% of the adultpopulation being classified as 'probable alco-holics', a proportion considered to be far inexcess of actuality (Klausner & Foulks, 1982).The problem is less when there is careful selec-tion of items and adequate checks on translation:the MAST discriminated well between alcoholicand non-alcoholic subjects in an Italian sample(Garzotto et al., 1988). One can be optimisticthat AUDIT will prove useful in countries withsimilar cultural, political and economic charac-teristics to those represented in the presentstudy.

The validity of AUDIT was examined in rela-tion to its ability to discriminate between personswith harmful or hazardous alcohol consumptionand those with non-hazardous consumption.This approach brings certain problems in itswake. Definitions of hazardous alcohol intake,recurrent intoxication, abnormal drinking be-haviour and alcohol-related problems had to beestablished by the collaborating investigatorssince there were no intemationally agreed crite-ria at the time the analyses were undertaken.They were based on the level of consumption orharm above which intervention was judged tobe preferable to no intervention. A compositereference standard was also defined whichencompassed these entities and also included thepresence of an alcohol-related disease and ac-knowledgement by the subject of a drinkingproblem. This reference standard was judged toincorporate the key elements of hazardous andharmful consumption, as defined by WHO andin the ICD-10 system, then in the process ofdevelopment.

Calculations of sensitivity and specificity in thepresent report should be supplemented by re-sults of further field studies. The questions

which form AUDIT were embedded within theassessment schedule and were not administeredseparately. An individual's response to questionswhich are grouped together may differ fromwhen they form part of a larger schedule. Inpractice, however, AUDIT may be used morefi-equently when embedded within a broadly-based lifestyle questionnaire than as a discretealcohol screening instrument. Secondly, a com-posite question was formed to inquire of concemshown by family or health professionals. Thirdly,the response categories for the two questions onalcohol-related problems differ from those usedin the assessment instrument. Finally, the criteriaagainst which AUDIT was judged (e.g. haz-ardous consumption, abnormal drinkingbehaviour, alcohol-related problems) are repre-sented in it. Validating a new screening test byreference to a broader assessment schedulewhich incorporates the same key elements mayresult in classification accuracy and validitycoefficients being infiated (Babor & Kadden,1985). Nevertheless, the present study includedan extemal criterion group (known alcoholics)and AUDIT performed well with this sample.

It is now appropriate to submit the instrumentto extensive field testing. A user's manual andresearch agenda have been developed for thispurpose (Babor et al., 1989b). Among the ques-tions that need to be answered are its accuracyand usefulness in different health care facilities,such as general practitioners' (family physicians')rooms, community health clinics, hospital outpa-tient clinics, and inpatient services. It may alsofind a place in epidemiological surveys of general(i.e. non-clinical) populations. Although thecountries represented in the present project wereculturally diverse, it would be appropriate in thenext phase of work to include additional cul-tures. The issue of whether AUDIT should beadministered as it exists at present or incorpo-rated into a larger questiormaire which enquiresof other lifestyle issues (such as diet, cigarettesmoking and exercise), merits exploration: a de-gree of disguise to the purpose of screening mayfacilitate accurate self-reporting. Ultimately, thepredictive validity of the instmment needs to bedocumented. Reassessment of sub-samples ofsubjects in the present study is presently beingundertaken three years after they were originallyinterviewed.

At the outset of the study it was hoped toidentify questions which did not mention alcohol

Development of AUDIT 801

explicitly which would prove to be useful foridentifying harmful alcohol consumption. It wasconsidered that such questions would be unaf-fected by response bias in self-reports on alcoholuse and consequences. This did not eventuate.Correlations between gastrointestinal and otherphysical symptoms and psychological distur-bances, and measures of alcohol intake weregenerally low or non-significant. There did notappear to be adequate grounds for includingthem in a screening instrument.

The present study is part of a WHO colla-borative programme on early intervention. Theultimate aim is to make available a validscreening procedure, and an effective and robustform of brief therapy which can be applied at thepoint of first contact. This approach capitaliseson the accessibility of primary health care andthe high throughput of patients, and eliminatesthe problem of attrition where referral to aspecialist service is the only option. More fun-damentally, it identifies the primary health careworker as the key person in the strategy toreduce alcohol-related harm throughout thecommunity.

AcknowledgementsThis paper is based on the data and experienceobtained during the participation of the authorsin the WHO Collaborative Projects on Detectionand Treatment of Persons with Harmfiil AlcoholConsumption, sponsored by the World HealthOrganization and the participating field researchcentres. The collaborating investigators in thisstudy are: at Headquarters in WHO, Geneva:Ms K. Canavan, Mr M. Grant, Mr J. Ordingand Dr N. Sartorius; in the field research centresin Australia: Dr F. Harding Bums, Dr R. Reznikand Dr J. B. Saunders; in Bulgaria: Dr M.Boyadjieva; in Kenya: Dr S. W. Acuda; inMexico: Dr J. R. de la Fuente; in Norway:Dr O. Aasland, Mr A. Amundsen and Dr J.Morland; and in the USA: Dr T. Babor, Dr H.Kranzler and Dr R. Lauerman. The viewsexpressed in this paper are the authors' own anddo not necessarily represent the opinions ofthe other investigators participating in theproject nor the views or policies of sponsoringagencies. The advice and comments of DrP. Anderson (Intstat Australia) are much appre-ciated.

ReferencesBABOR, T . F . & KADDEN, R. (1985) Screening for

alcohol problems: conceptual issues and practicalconsiderations, in: CHANG, N . & CHAO, H . M . (Eds)Early Identification of Alcohol Abuse. NIAAA ResearchMonographs No. 17, pp. 1-30 (Washington, DC, USGovemment Printing Office).

BABOR, T . F., RITSON, E. B. & HODGSON, R. J. (1986)Alcohol-related problems in the primary health caresetting. A review of early intervention strategies,British Joumal of Addiction, 81, pp. 23-46.

BABOR, T . F., KRANZLER, H . R. & LAUERMAN, R. J.(1987) Social drinldng as a health and psychosocialrisk factor: Anstie's limit revisted, in: GALANTER, M .(Ed.) Recent Advances in Alcoholism, Vol. 5, pp.373-402 (New York, Plenum Press).

BABOR, T . F., WEILL, J., TREFFARDIER, M . &BERNARD, Y. (1988) Detection and diagnosis ofalcohol dependence using the Le Go method, in:CHANG, N . C. & CHAO, H . M. (Eds) EarlyIdentification of Alcohol Abuse. NIAAA ResearchMonograph No. 17 (Washington, DC, US Govem-ment Printing Office).

BABOR, T . F., KRANZLER, H . R. & LAUERMAN, R. J.(1989a) Early detection of harmful alcohol con-sumption: comparison of clinical, laboratory andself-report screening procedures. Addictive Be-haviours, 14, pp. 139-157.

BABOR, T . F., DE LA FUENTE, J. R., SAUNDERS, J. B. &GRANT, M . (1989b) AUDIT—The Alcohol Use Disor-ders Identification Test: Guidelines for Use in PrimaryHealth Care (Geneva, World Health Organization).

BARRISON, I. G., VIOLA, L., MUMFORD, J., MURRAY, R.M., GORDON, M. & MURRAY-LYON, I. M. (1982)Detecting excessive drinking among admissions to ageneral hospital, Health Trends, 14, pp. 80-83.

BERNADT, W., MUMFORD, J., TAYLOR, C , SMITH, B.& MURRAY, R. M . (1982) Comparison of question-naire and laboratory tests in the detection ofexcessive drinking and alcohoUsm, Lancet, i, pp.325-328.

BUSH, B., SHAW, S., CLEARY, P., DELBANCO, T . L. &ARONSON, M . D . (1987) Screening for alcohol abuseusing the CAGE questionnaire, American Joumal ofMedicine, 82, pp. 231-235.

CHICK, J., KREITMAN, N . & PLANT, M . (1981) Meancell volume and gamma-glutamyltranspeptidase asmarkers of drinking in working men. Lancet, i, pp.1249-1251.

CUTLER, S. F., WALLACE, P. G. & HAINES, A. P.(1988) Assessing alcohol consumption in generalpractice patients: a comparison between question-naire and interview. Alcohol and Alcoholism, 23, pp.441^50.

CYR, M . & WARTMAN, S. A. (1988) The effectivenessof routine screening questions in the detection ofalcoholism, Joumal of the American Medical Associa-tion, 259, pp. 51-54.

EDWARDS, G., ARIF, A. & HODGSON, R. (1981)Nomenclature and classification of drug and alcoholrelated problems: a WHO memorandum. Bulletin ofthe World Health Organization, 59, pp. 225-242.

ELVY, G. A. & WELLS, J. E. (1984) The CanterburyAlcoholism Screening Test (CAST): a detection

802 John B. Saunders et al.

instrument for use with hospitalised patients. NewZealand Medical Joumal, 97, pp. 111-115.

EWING, J. A. (1984) Detecting alcoholism—the CAGEquestionnaire, Joumal of the American Medical Associ-ation, 252, pp. 1905-1907.

FEUERLEIN, W . , RINGER, C, KUFNER, H . & ANTONS,K. (1986) The diagnosis of alcoholism—the MunichAlcoholism Test (MALT), Intemational Joumal ofRehabilitation Research, 2, pp. 533-534.

GARZOTTO, N . , BARATTA, S., PISTOSO, S. & FACCIN-CANI, C. (1988) Validation of a screeningquestionnaire for alcoholism (MAST) in an Italiansample. Comprehensive Psychiatry, 29, pp. 323-329.

GoRENC, K.-D., BRUNER, C . A. , NADELSTICHER, A . ,PACURUCH, S. & FEUERLEIN, W . (1984) A compari-son of German, Spanish and Ecuadorian alcoholicsusing the Munich Alcoholism Test (MALT), Ameri-can Joumal of Drug and Alcohol Abuse, 10, pp.429-446.

GoTTESMAN, I. I. & PRESCOTT, C. A. (1989) Abusesof the MacAndrew MMPI alcoholism scale: a criticalreview. Clinical Psychology Review, 9, pp. 223-242.

HEDLUND, J. L. & VIEWEG, B . W . (1984) The Michi-gan Alcoholism Screening Test (MAST); acomprehensive review, Joumal of Operational Psychi-atry, 15, pp. 55-65.

iNSTlTt/TE OF MEDICINE (1990) Broadening the Base ofTreatment for Alcohol Problems (Washington, DC, Na-tional Academy Press).

ISRAEL, Y . , HURWITZ, R. , NIEMELA, O . & ARNON, R.(1986) Monoclonal and polyclonal antibodiesagainst acetaldehyde-containing epitopes in ac-etaldehyde-protein adducts. Proceedings of theNational Academy of Sciences of the United States ofAmerica, 83, pp. 7923-7967.

JACOBSON, G.R. (1983) Detection, assessment anddiagnosis of alcoholism: current techniques, in:GALANTER, M . (Ed.) Recent Developments in Alco-hoUsm, pp. 377-413 (New York, Plenum Press).

KAPUR, B . M . & ISRAEL, I. (1984) Alcohol Dipstick—ARapid Method for Analysis of Ethanol in Body Fluids(Toronto, Addiction Research Foundation).

KLAUSNER, S. Z . & FOULKS, E. F. (1982) EskimoCapitalists: Oil, Politics and Alcohol (Totowa, NJ,Ailanheld Osmun).

KRISTENSON, H . , OHLIN, H . , HULTEN-NOSSLIN,M-B., TRELL, E . & HOOD, B . (1983) Identificationand intervention of heavy drinking in middle-agedmen: results and follow-up of 24-60 months of longterm study with randomized controls. Alcoholism:Clinical and Experimental Research (NY), 7, pp. 203-209.

L E G O , P . M . (1976) Le Depistage Precoce et Systema-tique du Buveur Excessif (Paris, Departmentd'Alcoologie Therapeutic de Rion Laboratories).

MACANDREW, G . (1965) The differentiation of malealcoholic outpatients from non-alcoholic psychiatricoutpatients by means of the MMPI, Quarterly Jour-nal of Studies on Alcohol, 26, pp. 238-246.

MAYFIELD, D . G . , M C L E O D , G . & HALL, P. (1974)The CAGE questionnaire: validation of a new alco-holism screening instrument, American Joumal ofPsychiatry, 131, pp. 1121-1123.

PERSSON, J. & MAGNUSSON, P. H. (1988) Comparison

between different methods of detecting patients withexcessive consumption of alcohol, Acta Medica Scan-dinavica, 223, pp. 101-109.

PoKORNY, A. D., MILLER, B . A. & KAPLAN, H . B .(1972) The Brief MAST: a shortened version of theMichigan Alcoholism Screening Test, AmericanJoumal of Psychiatry, 129, pp. 342-345.

ROOM, R . (1988) Cross-cultural research in alcoholstudies: research traditions and analytical issues, in:TOWLE, L . H . & HARFORD, T . C . (Eds) Cultural

Infiuences and Drinking Pattems: A Eocus on Hispanicand Japanese Populations. NIAAA Research Mono-graph No. 19 (Washington, DC, US GovemmentPrinting Office).

ROSALKI, S. B . & RAU, D . (1972) Serum GGT activityin alcoholism, Clinica Chimica Acta, 39, pp. 41-47.

RYDER, D . , LENTON, S., HARRISON, S. & DORRICOTT,J. (1988) Alcohol-related problems in a general hos-pital and a general practice: screening and thepreventive paradox. Medical Joumal of Australia, 149,pp. 355-360.

SAUNDERS, J. B. (1987) The WHO projects an earlydetection and treatment of harmful alcohol con-sumption, Australian Drug and Alcohol Review, 6, pp.303-308.

SAUNDERS, J. B. & AASLAND, O . G . (1987) WHO

Collaborative Project on the Identification and Treat-ment of Persons with Harmful Alcohol Consumption.Report on Phase I: Development of a Screening Instru-ment (Geneva, World Health Organization).

SAUNDERS, J. B. & FOULDS, K . (1992) Brief and earlyintervention: experience from studies of harmfuldrinking, Australian and New Zealand Joumal ofMedicine, 22, pp. 224-230.

SAUNDERS, W . M . & KERSHAW, P. W. (1980) Screen-ing tests for alcoholism: findings from a communitystudy, British Joumal of Addiction, 75, pp. "il-AX.

SAUNDERS, J. B., AASLAND, O . G . , AMUNDSEN, A. &GRANT, M . (1993) Alcohol consumption and relatedproblems among primary health care patients. WHOCollaborative Project on Early Detection of Personswith Harmful Alcohol Consumption. I, Addiction,88, pp. 349-362.

SELZER, M . L . (1971) The Michigan AlcoholismScreening Test: the quest for a new diagnostic in-strument, American Joumal of Psychiatry, 127, pp.1653-1658.

SELZER, M . L . , VINOKUR, A. & VAN ROOIJEN, L . (1975)A self-administered Shon Michigan AlcoholismScreening Test (SMAST), Joumal of Studies on Alco-hol, 36, pp. 117-126.

SKINNER, H . A., HOLT, S., SCHULLER, R . , ROY, J. &ISRAEL, Y . (1984) Identification of alcohol abuseusing laboratory tests and a history of trauma. Annalsof Intemal Medicine, 101, pp. 847-851.

SKINNER, H . A., RINGER, S., SHEU, W . J. & ISRAEL, Y .(1986) Clinical versus laboratory detection of alco-hol abuse: the Alcohol Clinical Index, British MedicalJoumal, 292, pp. 1703-1708.

STIBLER, H . , BORG, S. & JOUSTRA, M . (1986) Microanion exchange chromatography of carbohydrate-deficient transferrin in serum in relation to alcoholconsumption. Alcoholism: Clinical and ExperimentalResearch, 10, pp. 535-544.

Development of AUDIT 803

STOREY, E. L., ANDERSON, G. J., MACK, U., POWELL, WHITFIELD, J. B., HENSLEY, W. J., BRYDEN, D . &L. W. & HALUDAY, J. W. (1987) Desialated trans- GALLAGHER, H . (1978) Some laboratory correlatesferrin as a seroiogical marker of chronic excessive oi drin\^nghdh\K, Annals of Clinical Biochemistry, 15,alcohol consumption. Lancet, i, pp. 1292-1294. pp. 297-303.

SWENSON, W. M. & MORSE, R. M . (1975) The use WILKINS, R. (1974) The Hidden Alcoholic in Generalof a self-administered alcoholism screenng test Practice (London, Elek Science).(SAAST) in a medical center. Mayo Clinical Proceed- WORLD HEALTH ORGANIZATION (1980) Problems Re-ings, 50, pp. 204-208. lated to Alcohol Consumption. Report ofa WHO Expert

WALLACE, P., CUTLER, S. & HAINES, A. (1988) Ran- Committee. Technical Report Series No. 650 (Geneva,domised controlled trial of general practitioner Worid Health Organization),intervention in patients with excessive alcohol con- WORLD HEALTH ORGANIZATION (1992) Intemationalsumption, British Medical Joumal, 297, pp. 663-668. Classification of Diseases, 10th revision. Chap. 5.

WATERSON, E. J. & MURRAY-LYON, I. M. (1989) Mental and Behavioural Disorders due to PsychoactiveScreening for alcohol related problems in the ante- Substance Use (Geneva, World Health Organization),natal clinic: an assessment of different methods, Wu, A., CHANARIN, I. & LEVI, A. J. (1974) Macrocyto-Alcohol and Alcoholism, 24, pp. 21-30. sis of chronic alcoholism. Lancet, i, pp. 829-831.

Appendix

AUDIT

Please circle the answer that is correct for you

1. How often do you have a drink containing alcohol?Never Monthly Two to four Two to three Four or more

or less times a month times a week times a week2. How many drinks containing alcohol do you have on a typical day when you are drinking?

1 or 2 3 or 4 5 or 6 7 to 9 10 or more

3. How often do you have six or more drinks on one occasion?Never Less than Monthly Weekly Daily or

monthly almost daily

4. How often during the last year have you found that you were not able to stop drinking once you had started?Never Less than Monthly Weekly Daily or

monthly almost daily

5. How often during the last year have you failed to do what was normally expected from you because of drinking?Never Less than Monthly Weekly Daily or

monthly almost daily

6. How often during the last year have you needed a first drink in the moming to get yourself going after a heavydrinking session?Never Less than Monthly Weekly Daily or

monthly almost daily

7. How often during the last year have you had a feeling of guilt or remorse after drinking?Never Less than Monthly Weekly Daily or

monthly almost daily

8. How often during the last year have you been unable to remember what happened the night before because youhad been drinking?Never Less than Monthly Weekly Daily or

monthly almost daily

9. Have you or someone else been injured as a result of your drinking?No Yes, but not in Yes, during

the last year the last year

10. Has a relative or friend, or a doctor or other health worker been concerned about your drinking or suggestedyou cut down?No Yes, but not in Yes, during

the last year the last year

804 John B. Saunders et al.

Procedure for Scoring AUDIT

Questions 1-8 are scored 0, 1, 2, 3 or 4. Questions 9 and 10 are scored 0, 2 or 4 only. The response coding is asfollows:

Question 1

Question 2

Questions 3-8

Questions 9-10

0

Never

1 or 2

Never

No

1

Monthlyor less

3 or 4

Less thanmonthly

2

Two tofour timesper month

5 or 6

Monthly

Yes, butnot in thelast year

3

Two tothree timesper week

7 to 9

Weekly

4

Four or moretimes per week

10 or more

Daily oralmost daily

Yes, duringthe last year

The minimum score (for non-drinkers) is 0 and the maximum possible score is 40.A score of 8 or more indicates a strong likelihood of hazardous or harmful alcohol consumption.