Screeningfor Alcohol Problemsin the U.S. General Population:A Comparisonof the CAGE and TWEAK by Gender, Ethnicity, and ServicesUtilization* CHERYL J. CHERPITEL, DR.P.H. Western Consortium for PublicHealth,Alcohol Research Group,2000HearstAvenue, Berkeley, California 94709 ABSTRACT:Objective: Thepurpose ofthisstudy wastocompare the performance of twoscreening instruments for alcoholproblems, the CAGE andtheTWEAK, against ICD-10 andDSM-IV criteriafor alcohol dependence by genderandethnicity,andto evaluatewhetherchar- acteristics associated withhealthservices utilizationmayaffectthe performance ofscreening instruments, inarepresentative sample ofthe U.S. adultgeneralpopulation. Method:DataarefromtheAlcoholResearch Group's1995National AlcoholSurvey; thesewereweighted to account for thedesign effectinherent in multistage clustersampling andoversampling of blacksandHispanics. Effectivesample sizewas 2,443:797blacks, 642Hispanics and1,004whites andothers (primarily AsianandNativeAmerican). Sensitivity andspecificity wereexaminedfor bothscreening instruments. Logisticregression wasusedto evaluatethepredictive value,separately, for theCAGE andTWEAK, controlling for gender,ethnicity, regionof thecountryandserviceuse (primarycareandemergency room).Results:The TWEAK wasmore sensitive for menthanfor women,but no differences werefoundby serviceuse,while the CAGE was more sensitivefor men who had used the emergency roomduringthe preceding yearcompared with those whohadnot.TheTWEAK wasmoresensitive thantheCAGEamong whiteandHispanic men,amongmenwhohadnotusedtheemergency room and amongwhiteswho had no serviceuse.Performanceof nei- therthe CAGE nor theTWEAK wasfoundto varyby regionof the country.Conclusions: Data suggestthat while the performance of screening instruments mayvaryacrossdemographic subgroups in the generalpopulation, instruments mayperformequallywellfor identifyingproblemdrinkers in generalpopulations asin clinicalpopulations. (J. Stud.Alcohol60:705-711, 1999) NUMBER OFrelatively shortscreening instrumentsmary care samplesthan in the generalpopulationfrom havebeendeveloped for usein clinicalpopulations to identifyalcoholdependence or alcoholabuse/harmful drink- ing.The validityof theseinstruments, basedon diagnostic criteria(International Classification of Diseases [ICD] orDiagnostic andStatistical Manualof MentalDisorders [DSM]) for problemdrinking,hasbeentestedprimarilyin clinical populations comprised largelyof whitemalepatients. The performance of screening instruments is lesswell establishedamongwomenandethnicminorities, or in general populations. Theprevalence of problem drinking in primary careandemergency room(ER) settings hasrecentlygained considerable attention(Buchsbaum et al., 1991b;Chanet al., 1994a;Cherpitel,1993,1994;FlemingandBarry,1991b; Magruder-Habib et al., 1991;Soderstrom et al., 1992),and theneedforevaluating easilyadministered screening instrumentsto identifythosepatients whomaybenefitfroma brief intervention or referralfor problemdrinkinghasbecome evident(Buchsbaum et al., 1991a;Cherpitel,1997, 1998; 'FlemingandBarry,1991a;Volk et al., 1997). Althoughtheprevalence of alcohol-related problems has beenfoundto be higherin bothemergency roomandpri- Received:June30, 1997. Revision:November10, 1997. *Thisresearch wassupported byNational Alcohol Research Center grant AA 05595 from the National Instituteon Alcohol Abuse and Alcoholism. 705 whichtheycome(Cherpitel,1991,1995a),theprevalence of alcoholdependence in thegeneralpopulation, nevertheless, hasnotbeenfoundto beinconsequential. The 1988National HealthInterviewSurveyfoundthat9% of Americanadults met DSM-III-R criteria for current alcohol abuse and de- pendence(Grant et al., 1991), while the 1990 National Al- cohol Survey found 4% positive for current alcohol dependence alone,accordingto DSM-IV criteria,and5.5% according to ICD-10 criteria,withhighestratesfor Hispanics followedby blacksandthenwhiteson bothdiagnostic schemes(CaetanoandTam, 1995).The 1992NationalLongitudinal AlcoholEpidemiologic Surveyalsofound4% positive for currentalcoholdependence basedon DSM-IV and ICD-10 criteria,and7% and4.6% positiveon DSM-IV and ICD-10 criteria,respectively,for alcoholabuseand/ordependence(Grant, 1996). Relativelyfew studies, however,haveexaminedtheper- formance of screening instruments for identifying problem drinkingin generalpopulations. Onestudyexamined sensitivity of theCAGE,Brief MichiganAlcoholism Screening Test(BMAST)andTWEAK against a past-year diagnosis of alcoholdependence, usingDSM-III-R criteria,in a primary carepopulation compared witha sampleof thegeneralpopulationfromthesamemetropolitan area,andfoundsensitivity for theseinstruments to be higherin the primarycare sample(Chan et al., 1993, 1994a,b).No studyto date, 706 JOURNAL OF STUDIES ON ALCOHOL however,hasreportedthe performanceof screeninginstrumentsin a representative sampleof the U.S. population,nor have manyof thesestudiesanalyzedscreenerperformance acrossgenderand ethnicgroupsin the generalpopulation. While the useof standarddiagnosticinstrumentswould be preferablefor establishing the prevalenceof alcoholdependence in nonclinicalpopulations,such instrumentsare lengthy,andthe useof shorterscreeners would seemadvantageousin this regard. Additionally, such short screeners wouldbe usefulas a first-stagescreeningdevicein general populationsurveysto identify those who would then be givena standarddiagnosticinstrument. Screeninginstruments may notbe expectedto performas well in the generalpopulationasin clinicalpopulations, due to a lower prevalenceof alcoholdependence and harmful drinking/abuse in nonclinicalsamples.The Epidemiological CatchmentArea studyfounda higherproportionof thosein clinical populationshad a symptomcountfor alcoholdependence well abovethediagnostic threshold compared with thosein the generalpopulation,while thosein the general populationwere morelikely to havea symptomcountright at the diagnosticthreshold(Helzer et al., 1985). Differences in characteristics betweenthosein the generalpopulation andmorefrequentusersof clinicalservices (whowouldhave a greaterprobabilityof falling into suchsamplesthan less frequentusersof thesehealthcare services),may alsoexplain differentialperformanceof screeninginstruments in the generalpopulationcomparedto clinical populations. However,a priorcomparison of drinkingpatternsandproblem and demographiccharacteristics betweena representative sample of ER patients in a large HMO and a representative sampleof the generalpopulationof the same countywhoreportedmembership in theHMO foundthemajor differencebetweenthetwo samples wasthefrequencyof ER use(Cherpitel,1992). To fill thisgapin our knowledgeandexaminethe performanceof screeninginstruments in nonclinicalpopulations, the sensitivityandspecificityof theCAGE andTWEAK are comparedacrossgenderand ethnicgroupsfor black,Hispanic and white/otherrespondents from a U.S. national household probabilitysampleof adults.To evaluatewhether characteristics associated with health services utilization may affecttheperformance of screeninginstruments, analysis alsoincludeswhetherthe respondentreportedusingprimary care and/or ER servicesin the year precedingthe interview. Both ICD-10 and DSM-IV criteria for current al- cohol dependenceare used as the gold standardagainst which the performanceof screeninginstrumentsis compared.Logisticregression analysisis usedto evaluatethe predictivevalueof the CAGE and the TWEAK on alcohol dependence, controllingfor demographic characteristics, regionof the countryandserviceuse(ER andprimarycare). Theseanalysesareimportantin determiningthosescreeners that may be most effective for identifyingproblem drinkersin nonclinicalpopulations. / SEPTEMBER 1999 Method Sample The dataarefromtheAlcoholResearchGroup's 1995NationalAlcohol Survey.Fieldworkfor the studywas subcontracted to the Institute for Survey Researchat Temple University.The sampleconsisted of adults18yearsandolder living in householdswithin the 48 contiguousstates.A multistagearea-probabilitysamplewas drawn, using 100 primary samplingunits, with an oversamplingof blacksand Hispanics.Completedinterviewswere obtainedon 1,587 black, 1,598 Hispanic and 1,740 white/other (primarily AsianandNative American)respondents, representing completionratesof 77% for blacksandHispanics,and76% for whites/other.Noninterviewswere dueto refusals(13%), and incapacitation, languagebarhersandfailureto locatetherespondent (10%). Datawereweightedto reflectthenumberof adultsliving in a selectedhouseholdandthe interviewcompletionratein a givenarea.Data were alsoweightedto take into accountthe designeffect inherentin the useof multistageclustersampling,using an approachoriginally suggestedby Kish(1965) for estimatinga designeffectaverage. The averagedesigneffectwas 1.99 for blacks,2.49 for Hispanicsand1.73for whites/others, using48 variablesfromsix domainsof the questionnaire, whichresultedin an effective samplesize (N = 2,443; 52% women) of 797 blacks,642 Hispanicsand 1,004 whites/othersfor analysis."Others"in thislattercategoryincluded29 Asians,22 NativeAmericans and 5 of mixed race. Table 1 showsdemographiccharacteristics of the sample by ethnicity.Whites/others weremorelikely to be older,better educated and to be current drinkers than blacks or His- panics,and lesslikely to have neverbeen married,to be unemployedandto havenohealthinsurance.They werealso morelikelyto haveusedprimarycareservices in thelastyear thanblacksor Hispanics,whileHispanicswerelesslikely to haveusedthe ER duringthe lastyearcomparedwith whites or blacks. No differences were found in the rate of alcohol dependence acrossthe threegroups. Data collection Interviews were conducted with informed consent in the respondents' homesby trainedinterviewers usingstructured interview schedules(questionnaires) of about 1 hour in length.Hispanicrespondents were given a choiceof being interviewedin English,or in Spanishwith a bilingualinterviewer.The Spanishversionof the questionnaire underwent a processof translationandindependent backtranslation. Respondents who self-identifiedas either "white of Hispanic origin"or "blackof Hispanicorigin"(Latino,Mexican,Central or SouthAmerican,or any otherHispanicorigin) were classifiedas Hispanic.Respondentswho self-identifiedas "black, not of Hispanicorigin" were categorizedas black. CHERPITEL T^BI•œ1. Demographic characteristics by ethnicity(in percent) White/other Black Hispanic (n = 1,004) (n = 797) (n = 642) 48 52 45 55 50 50 19 44 37 27 a 46 27 a 34 b 45 22 b 5 10 37 25 23 8• 17• 38 24 13• 27• 18• 29• 17• 9b Marriage/marriage-like relationship Separated/divorced 69 10 49• 14a 68 9 Widowed Never married 7 14 8 29 • 3• 200 Gender Men Women <-8th grade Somehigh school High schoolgraduate Somecollege Collegegraduate Marital status Unemployed 4 10" No health insurance 11 20" 38• UsedER in lastyear Usedprimarycarein lastyear 11 34 12 20" 8• 200 Current drinker 67 54 • 54 b 5 4 5 5 5 6 Alcoholdependence-ICD-10 Alcoholdependence-DSM-IV Region 8• East 21 16 • 14 • South 32 57 a 33 Central West 27 20 19 • 8• 7• 46 • "p < .05, comparisonof proportionsbetweenblacksandwhites. Op< .05, comparison of proportions betweenHispanicsandwhites. All otherrespondents are includedin the "white and other" categoryfor analysesreportedhere. Instruments Among other questions,respondents were askeditems comprisingthe CAGE andTWEAK screening instruments, itemsrelatedto alcoholdependence, useof emergency room or primary care servicesduringthe last year, and demographiccharacteristics includingregionof thecountrywhere the respondentlived (East, South,Central,West). Both the CAGE and the TWEAK are mnemonics. (Beresford et al., 1990: Bernadt et al., 1982; Bush et al., 1987).The CAGE hasbeenusedextensivelyin clinicalpopulations,in bothwrittenandverbalform;however,relatively little dataare availableon its performanceacrossgenderand ethnic minorities. Age 18-29 30-49 50+ Education 707 The CAGE (Ewing, 1984)wasdesignedfor rapidverbalscreeningfor alcoholdependence in clinicalpracticeandis derived from the followingfour items:(1) Have you ever felt you shouldcutdownonyourdrinking?(2) Havepeopleannoyed youaboutyourdrinking?(3) Haveyoueverfelt bador guilty aboutyour drinking?(4) Have you ever had a drink first thingin the morningto steadyyour nervesor get rid of a hangover(eye-opener)?Validity of the CAGE was establishedin relationto a diagnosisof alcoholdependence based on psychiatric evaluationin a VeteransAdministration psychiatricinpatientpopulation(Mayfield andJohnston,1981; Mayfield et al., 1974). Basedon a positiveresponseto two or more items, the CAGE has been found to have a sensitiv- ity rangingfrom 72-91%, and a specificityfrom 77-96% The TWEAK is a recentlydevelopedinstrumentdesigned to identify"atrisk"drinkingin prenatalpopulations (Russell et al., 1994),andasksquestions havingto do with tolerance (holdingmorethanfive drinks),friendsor relativesworried abouta person'sdrinking,takinga drink first thingin the morning(eye-opener),blackouts(amnesia),and felt a need to c(k)utdownondrinking.Two of theTWEAK itemswere taken from the CAGE (eye-openerand cut down) and two (worriedandamnesia),from the full MichiganAlcoholism Screening Test(MAST) (Selzer,1971).Usinga cutpointof 2 (whena weightof 2 is appliedto toleranceandworry,and a weightof 1 to theremainingthreeitems),theTWEAK was foundto havea sensitivityof 79% anda specificityof 83% in prenatalpopulations againsta criterionof drinkingoneor more ouncesof absolutealcoholper day (Russellet al., 1994).Usinga cutpointof 3, asusedin thepresentstudy,the TWEAK hasbeenfoundto havea sensitivityrangingfrom 94-84% (Chan et al., 1993; Cherpitel,1995b) and a specificity from 89-81% (Chan et al., 1993; Cherpitel, 1998) againstdiagnostic criteriafor alcoholdependence in clinical samples.The questionhavingto do with the numberof drinksone can hold (tolerance)presumablymakesthis instrumentmore sensitiveto identifying alcohol problems amongwomensinceit cantakeintoaccounta lowerthresholdfor women(Russell,1994).OnequestionontheTWEAK andall four CAGE questionswereaskedon a lifetime basis. If respondents were positiveon theselifetimebasedquestions,theywereaskedagainregardingthelast 12 months. Alcoholdependence duringthelastyearwasmeasured by 24 items similar to those in the Alcohol Section of the Com- posite InternationalDiagnostic Interview (CIDI) core (Wittchenet al., 1991),whichoperationalized bothICD- 10 (World HealthOrganization,1992)andDSM-IV (American PsychiatricAssociation,1994)criteria(CaetanoandRoom, 1994).Thesesameitemshavebeenusedto operationalizealcoholdependence in prior NationalAlcoholSurveys(Caetano and Tam, 1995; Caetano et al., 1997). Items which operationalized ICD-10 includedquestions relatedto thesix domainsof craving, impaired capacity to control, withdrawal,tolerance,neglectof interests,andcontinuedusedespiteproblems;whileDSM-IV includedquestions relatedto the sevendomainsof tolerance,withdrawal,drinkingmore thanintended,unsuccessful effortsto control,givinguppleasuresor intereststo drink, spendinga greatdeal of time in drinkingactivities,and continueduse despiteproblems.A respondent wasconsidered alcoholdependentif positiveon threeor more domainsin either of the diagnosticschemes considered separately. The standardagainstwhichsensitivity of thescreening instruments wasanalyzedwascomprised of thosepositiveon eitheror bothof the diagnosticschemes. 708 JOURNAL OF STUDIES ON ALCOHOL / SEPTEMBER 1999 T^SL•. 2. Sensitivity(S) and specificity(SP) for the CAGE and TWEAK by genderand ethnicityamongcurrent drinkers Total CAGE TWEAK Men Women S SP S SP (141) (1,297) (1,04) 61 79 a 95 89 59 83 a White Hispanic S SP S SP S SP S SP (686) 37 (611) (52) (620) (47) (376) (42) (301) 94 86 70 66 b 97 93 56 77 a 97 91 70 83 95 91 60 76 93 83 White Black Men CAGE TWEAK Black Women Men Hispanic Women Men Women S SP S SP S SP S SP S SP S SP (39) (316) (13) (304) (33) (194) (14) (183) O1) (175) (11) (124) 97 89 62 58 97 94 70 85 94 89 17 77 95 92 89 76 73 64 98 94 54 83 a 55 81 a Note:WeightedN's are in parentheses; subgroup N's may notsumto totalbecauseof weighting. ap< .05, comparison of proportions of sensitivitybetweenCAGE andTWEAIC bp< .05,comparison of proportions of sensitivity of CAGE andTWEAK betweengenders andbetweenethnicgroups. Amongcurrentdrinkers(thosewho reportedhavingany alcoholicbeverageduringthe past 12 months)20% reported consumingeightor moredrinkson at least1 day duringthe lastyear,and65% of thosepositiveon eitheror bothof the diagnosticschemes reporteddoingso. sensitivityare alsoreportedfor thosewho madean ER visit (Table 3) andthosewho madea primarycarevisit (Table 4) duringtheprevious12 monthscomparedwith thosewho did not, within genderand ethnicsubgroups for the CAGE and TWEAK, and for the CAGE comparedto the TWEAK within serviceuseand demographic subgroups. The datain the tables will be discussed further in the Results section Data analysis which follows. The sensitivity(percentagecorrectlyclassifiedas having thecondition)andspecificity(percentage correctlyclassified as nothavingthe condition)for currentalcohoIdependence (ICD-10 and/orDSM-IV criteria)arereportedat a cutpoint of 2 for theCAGE and3 for theTWEAK for reportingthese experiences as havingoccurredduringthe precedingyear. Two-tailedtestsof significantdifferencesbelweenproportionsof sensitivityacrossgenderand ethnicsubgroups are reportedfor theCAGE andtheTWEAK, andfor theCAGE comparedto the TWEAK within demographicsubgroups (Table 2). Significantdifferencesbetween proportionsof Logisticregression (SPSS,1996),with simultaneous entry of variables,wasthenusedto analyzethepredictivevalueon alcoholdependence,separatelyfrom the CAGE and the TWEAK, controllinggender,ethnicity,regionof thecountry, andserviceuse(ER andprimarycare).CAGE, TWEAK, and genderwere codeddichotomously. Men were comparedto womenas the referencecategory.Both ethnicityandregion were coded as indicator contrasts,with whites/othersas the referencegroupto which blacksand Hispanicswere contrasted,andWestas thereferencegroupto East,Southand Centralwerecontrasted. The interactiontermof screeningin- T^aLE 3. Sensitivity(S) and specificity(SP) for the CAGE andTWEAK by ER useduringthe lastyear amongcurrentdrinkers Total Men Yes CAGE TWEAK No No Yes No S SP S SP S SP S SP S SP S SP (21) (142) (121) (1,154) (14) (78) (91) (607) (7) (64) (30) (547) 76 80 94 89 59 • 79• 95 89 84 81 92 84 55 b 83 a 94 86 63 74 97 95 71 64 97 93 White/other Yes CAGE TWEAK Women Yes Black No Yes Hispanic No Yes No S SP S SP S SP S SP S SP S SP (5) (67) (47) (553) (11) (46) (36) (330) (5) (29) (38) (271) 81 78 98 90 53 70: 97 93 81 81 92 95 66 84 95 90 61 25 90 77 60 76 93 84 Note:WeightedN's are in parentheses; subgroup N's may notsumto totalbecauseof weighting. ap < .05, comparison of proportions of sensitivitybetweenCAGE andTWEAK. •p < .05, comparison of proportions of sensitivity of CAGE andTWEAK betweenthosewhodid anddid not usetheER duringthelastyearwith genderandethnicgroups. CHERPITEL 709 TABLE4. Sensitivity(S) andspecificity(SP) for the CAGE andTWEAK by primarycareuseduringthe last yearamongcurrentdrinkers Total Men Yes CAGE TWEAK No No Yes No S SP S SP S SP S SP S SP S SP (40) (346) (102) (950) (29) (154) (75) (531) (11) (191) (26) (420) 55 78 a 96 91 64 79 a 95 89 51 82 a 95 86 62 83 a 94 85 69 63 97 94 70 68 97 93 White/other Yes CAGE TWEAK Women Yes Black No Yes Hispanic No Yes No S SP S SP S SP S SP S SP S SP (19) (218) (33) (402) (13) (68) (35) (308) (9) (59) (33) (241) 58 74 97 92 54 78 a 97 91 65 90 96 94 71 81 94 90 38 64 92 82 65 79 93 84 Note:WeightedN's arein parentheses; subgroup N's maynotsumto totalbecause of weighting. ap< .05, comparison of proportions of sensitivity betweenCAGE andTWEAK. strument(CAGE or TWEAK) by regionwasalsoenteredinto an additionalregressionfor each instrument,with region codedas a deviationcontrast,comparingeachregionto the overalleffectof theotherregions.Oddsratios(ORs) arereportedfor eachvariablein theequation.ORsgive thelikelihood,for example,thata respondent with a positiveCAGE comparedwith a respondent with a negativeCAGE will be positivefor alcoholdependence, controllingfor all othervariablesin theequation.The 95% confidenceinterval(CI) is reportedfor eachvariablethatreachedstatisticalsignificance. Analysisis cardedout on currentdrinkers,sincethisis the population considered tobeatriskforcurrentproblemdrinking. Results for men regardlessof primary careuse, and was alsobetter amongwhitesfor thosewhodidnotmakea primarycarevisit duringtheprecedingyear. Table 5 showsthe predictive value, separatelyfor the CAGE andtheTWEAK, on alcoholdependence, controlling for gender,ethnicity,regionof thecountry,andER treatment and primarycare treatmentduringthe last year. Both the CAGE andTWEAK werehighly predictiveof alcoholdependence:thosepositiveon the CAGE were 37 timesmore likely than thosenegativeto be alcoholdependent,while thosepositiveon the TWEAK were30 timesmorelikely to be alcoholdependentthan thosenegativeon the TWEAK. Genderwas alsofoundto be a significantpredictorof alcohol dependencewhen CAGE statuswas controlled,but not when TWEAK Table2 showsthesensitivityandspecificityfor theCAGE and TWEAK within genderand ethnicsubgroups. No differenceswerefoundacrosssubgroups for the CAGE, while sensitivityof theTWEAK wasfoundto be significantlybetter amongmenthanamongwomen.Comparedto theCAGE, sensitivityof theTWEAK wasfoundto be bestamongwhite andHispanicmales. Sensitivityand specificityof the CAGE and TWEAK werethencomparedfor thosewho did andthosewho did not reportan ER visit duringthepreceding12 months.As seen in Table 3, sensitivityof the CAGE wasbetteramongmen who had attendedthe ER duringthe last year (84%) than amongthosewhohadnot(55%), butnodifferencewasfound amongwomen,or within ethnicgroups.Sensitivityfor the TWEAK wasnot foundto differ significantlyin relationto whetheranER visit wasmadeduringtheprecedingyear,but was foundto be significantlybetterthanthe CAGE among men andamongwhiteswhodid notmakean ER visit. As seenin Table4, no differencewasfoundin sensitivity of theseinstrumentswithin genderor demographicsubgroupsin relationto primarycareserviceuseduringthelast year. Sensitivityof the TWEAK was betterthanthe CAGE status was controlled, with men almost 3 timesmorelikely thanwomento be alcoholdependent. Regionof the countrywas alsofoundto be significantin both equations, with thoseliving in theEastandCentralregions only a thirdaslikely asthoselivingin theWestto be alcohol TABLE5. Oddsratios(OR) andconfidence intervals(CI) for demographic characteristics, healthservicesuseandscreeneron alcoholdependence CAGE as screener TWEAK (n = 940)a OR Screener(positive) Gender(male) Ethnicity (white/other) Black 37.14* 2.67* CI 20.24-68.17 1.42-4.92 0.80 Hispanic Region(West) 1.11 East South Central ER treatment PC treatment 0.32* 0.62 0.38* 0.76 1.07 as screener (n = 932)b OR CI 30.75* 1.69 16.98-55.71 0.79 1.42 0.13-0.80 0.17-0.90 0.36* 0.69 0.38* 0.92 1.22 0.14-0.89 0.17o0.87 aExcludes threerespondents on whomall datawerenotobtained. bExcludes ninerespondents on whomall datawerenotobtained. *p < .05 (X2 with 1 df); *p < .01 (X2 with 1 df). 710 JOURNAL OF STUDIES ON ALCOHOL dependent.Neitherethnicitynorhealthservicesusewaspredictiveof alcoholdependence in eitherequation.Additional regressions were examinedfor the CAGE andTWEAK includingthe interactionterm of screeninginstrumentby region(not shown),thatwasnotfoundto be significant. Discussion Differenceswere found by gender,ethnicityand service use in sensitivityof both the CAGE and TWEAK. The TWEAK was found to be more sensitive for men than for women, while the CAGE was more sensitive for men and whiteswhohadnotusedtheER duringtheyearprecedingthe interviewcomparedwith thosewho had.Differencesin sensitivitywerealsofoundbetweentheCAGE andtheTWEAK, with the TWEAK performingbetterthanthe CAGE among whites and Hispanicmen (but not black men), amongmen who had not usedthe ER during the last year, and among whiteswho hadnotusedtheER or primarycareservicesduring the precedingyear.It is importantto noteherethatmultiple comparisons were cardedout on thesedata, thereby increasingtheprobabilitythatsomeof thedifferencesfound mayhaveoccurredby chancealone.On theotherhand,small samplesizesin someof the gender/ethnic/service usecategoriesmay haveprecludedthedetectionof additionaldifferencesin instrumentperformance acrosssubgroups. Orderingof thetwo instruments may alsohaveinfluenced their relative performance,to somedegree(the CAGE was administeredfirst, followedby the TWEAK). Both instruments were administered rather late in the interview, after numerousquestions hadbeenaskedregardingquantityand frequencyof drinkingand drinkingpatternsand problems, includingquestions whichobtainedan ICD-10 or DSM-IV diagnosisfor alcoholdependence. This,too,may haveinfluencedinstrument performance--apriorstudyfoundthatsensitivity of the CAGE was reduced when precededby questionsrelated to defining quantity and frequencyof drinking(SteinwegandWorth, 1993). While the prevalenceof alcoholdependencein this general populationsampleis similarto that foundin othernational alcohol surveys(Caetano and Tam, 1995; Grant, 1996), it is substantiallylower than that found in clinical populations, andthismay alsohaveaffectedinstrumentperformance.Usingthe samediagnosticcriteriafor alcoholdependence as that used here, prevalence of alcohol dependence rangedfrom 10% amongHispanicpatientsto 16% amongwhite/otherpatientsin a CaliforniaER sample (Cherpitel,1998) and 9% for black patientsand 12% for whitepatientsin a MississippiER sample(Cherpitel,1997). A comparison of thesensitivityof theCAGE andTWEAK reportedherewith that from thesetwo ER studies,usingthe samegold standard, foundbothinstruments performedbetter within the samegender/ethnicsubgroups in the ER samplesthan in this nationalprobabilitysampleof the general population(Cherpitel,1997,1998).Priorprimarycareclinic / SEPTEMBER 1999 studiesalso report higher sensitivityfor the CAGE and TWEAK than that found here, althoughthesestudieshave not examinedinstrumentperformance within gender/ethnic subgroups (Chan et al., 1993, 1994a:Flemingand Barry, 1991a;Liskow et al., 1995). Few significant differences in theperformance of theseinstruments by ER andprimarycareusewerefound,possibly dueto smallsamplesizesasmentionedabove;andcontrary to what might be expected,with one exceptiondifferences foundsuggestthatthe sensitivityof the CAGE andTWEAK maybe betterfor thosenotusingtheER or primarycareservicesduringthe precedingyear thanfor thosewho did use suchservices.The lack of significantdifferencesfoundby services usemayalsoberelatedto thefactthatthosewhoreporteduse of suchservicesin the precedingyear may not necessarily be heavyusersof theseservices,as are manyof thosesampledin theseclinicalsettings,andwho thushavea greaterlikelihoodof falling into suchsamples.Serviceuse wasbasedhereon reportingevenonevisit to an ER or primary careclinicduringthe previousyear. A comparisonof the performanceof screeninginstrumentsin the two ER studiesdescribedabove(Cherpitel, 1997, 1998)foundthatwhileregionof the country(Southvs West)wasnotanimportantpredictorof alcoholdependence, regionaldifferencesin the performanceof screeninginstrumentsmay exist,even whenethnicityis takeninto account. Logistic regressionanalysisreportedhere, however, suggeststhat regionis an importantpredictorof alcoholdependence,with thoseliving in the East andCentralregionsless likely to be alcoholdependent comparedto thosein theWest, but performanceof neitherthe CAGE nor the TWEAK varied by region. Data reportedheresuggest thatwhile the performance of screeninginstrumentsmay vary acrossdemographicsubgroupsin the generalpopulation,instruments may, nevertheless,performequallywell for identifyingproblemdrinkersin generalpopulationsas in clinical populations.Furtherresearch is needed in order to determine the usefulness of screeninginstrumentsfor problem drinking within gender andethnicsubgroups in nonclinicalpopulations, andwhether regionaldifferences, beyondethnicity,in theperformance of screeninginstruments existin the generalpopulation. References AMERICAN PSYCHIATRIC ASSOCIATION. Diagnosticand StatisticalManual of Mental Disorders(DSM-IV), WashingtonDC, 1994. BERESFORD,T.P., BLOW, F.C., HILL, E., SINGER, K. AND LUCEY, M.R. Comparisonof CAGE questionnaire andcomputer-assisted laboratory profilesin screening for covertalcoholism. Lancet336: 482-485, 1990. BERNADT,M.W., MUMFORD,J., TAYLOR,C., SMITH, B. ANDMURRAY,R.M. Comparison of questionnaire andlaboratory testsin thedetection of excessivedrinkingandalcoholism.Lancet1: 325-328, 1982. BUCHSBAUM,D.G., BUCHANAN,R.G., CENTOR,R.M., SCHNOLL,S.H. ANr• LAWTON,M.J. Screeningfor alcoholabuseusingCAGE scoresand likelihood ratios.Ann. Intern. Med. 115: 774-777, 1991a. CHERPITEL BUCHSBAUM,D.G., BUCHANAN, R.G., LAWTON, M.J. AND SCHNOLL,S.H. Alcohol consumptionpatternsin a primary care population.Alcohol Alcsm 26: 215-220, 1991b. BUSH, B., SHAW, S., CLEARY, P., DELBANCO,T.L. AND IRONSON, M.D. Screeningfor alcoholabuseusingthe CAGE questionnaire. Amer. J. Med. 82: 231-235, 1987. CAETANO, R. ANDROOM,g. Alcoholdependence in the 1990 U.S. NationalAlcoholSurvey:Operationalizing andcomparing twonosological systems.Drug Alcohol Rev. 13: 257-267, 1994. CAETANO,R. AND TAM, T.W. Prevalence and correlatesof DSM-IV and ICD- 10 alcoholdependence: 1990U.S. NationalAlcoholSurvey.Alcohol Alcsm 30: 177-186, 1995. CAETANO,R., TAM, T., GREENFIELD, T.K., CHERPITEL,C.J. ANDMIDANIK, L.T. DSM-IV alcoholdependence anddrinkingin the U.S. population: A risk analysis.Ann. Epidemiol.7: 542-549, 1997. CHIN, A.W., PRISTACH, E.A. ANDWELTE,J.W. Detectionby theCAGE of alcoholismor heavydrinkingin primarycareoutpatients and the general population.J. Subst.Abuse6: 123-135, 1994a. CHIN, A.W., PRISTACH,E.A. ANDWELTE, J.W. Detection of alcoholismin threepopulationsby the Brief-MAST. Alcsm Clin. Exp. Res. 18: 695701, 1994b. CHIN, A.W., PRISTACH, E.A., WELTE, J.W. AND RUSSELL,M. Use of the TWEAK testin screening for alcoholism/heavy drinkingin threepopulations.AlcsmClin. Exp. Res.17:1188-1192, 1993. CHERPITEL, C.J. Drinkingpatternsandproblemsamongprimarycarepatients:A comparison with the generalpopulation.AlcoholAlcsm26: 627-633, 1991. CHERPITEL, C.J. Drinkingpatterns andproblems: A comparison of ER patientsin an HMO andin thegeneralpopulation. AlcsmClin. Exp. Res. 16:1104-1109, 1992. CHERPITEL, C.J. Acoholandinjuries:A reviewof international emergency room studies. Addiction 88: 923-937, 1993. CHERPITEL, C.J. Alcoholuseamongprimarycarepatients:Comparingan HMO withcountyclinicsandthegeneralpopulation. DrugAlcoholDepend.36: 167-173, 1994. CHERPITEL, C.J. Alcoholandcasualties: Comparison of county-wide emergencyroomdatawith thecountygeneralpopulation. Addiction90: 343350, 1995a. CHERPITEL, C.J. Screeningfor alcoholproblemsin the emergencydepartment. Ann. Emer. Med. 26: 158-166, 1995b. CHERPITEL, C.J. Comparisonof screeninginstruments for alcoholproblemsbetweenblackandwhiteemergency roompatientsfrom two regionsof thecountry.AlcsmClin. Exp.Res.21: 1391-1397,1997. CHERPITEL, C.J. Differencesin performance of screening instruments for problemdrinkingamongblacks,whitesandHispanicsin an emergency roompopulation.J. Stud.Alcohol59: 420-426,1998. EWING,J.A. Detectingalcoholism:The CAGE Questionnaire. JAMA 252: 1905-1907, 1984. FLEMING,M.F. ANDBARRY,K.L. A three-sample testof a maskedalcohol screeningquestionnaire. AlcoholAlcsm26:81-91, 1991a. FLEMING,M.F. AND BARRY,K.L. The effectivenessof alcoholism screen- ing in an ambulatorycaresetting.J. Stud.Alcohol52: 33-36, 1991b. 711 GRANT,B.F. Prevalenceand correlatesof drug useandDSM-IV drug dependencein the United Sates:Resultsof the NationalLongitudinalAlcoholEpidemiologicSurvey.J. Subst.Abuse8: 195-210, 1996. GRANT, B.F., HARTFORD,T.C., CHOU, P., PICKERING,R., DAWSON,D.A., STINSON,F.S. AND NOBLE,J. Prevalenceof DSM-III-R alcohol abuse and dependence, United States,1988. Alcohol Hlth Res. World 15: 91-96, 1991. HELZER, J.P., ROBINS,L.N., McEvoY, L.T., SPITZNIGEL, E.L., STOLTZ- MAN,R.K., FARMER, A. ANDBROCKINGTON, I.F. A comparison of clinical and Diagnostic Interview Schedule diagnoses. Physician reexamination of lay-interviewedcasesin the generalpopulation.Arch. Gen. Psychiat.42: 657-666, 1985. gaSH,L. SurveySampling,New York:JohnWiley & Sons,Inc., 1965. LISKOW,B., CAMPBELL, J., NICKEL,E.J. ANDPOWELL,B.J. Validity of the CAGE questionnaire in screeningfor alcoholdependence in a walk-in (triage)clinic. J. Stud.Alcohol 56:277-281, 1995. MAGRUDER-HABIB, K., DURAND,A.M. ANDFREY,K.A. Alcoholabuseandal- coholism in primaryhealthcaresettings. J.Fam.Pract.32: 406-413,1991. MAYFIELD, D.G. ANDJOHNSTON, R.G.M. Screeningtechniques andprevalence estimationin alcoholism.In FINN, W.E. (Ed.) Phenomenology andTreatmentof Alcoholism,New York: SpectrumPublications,Inc., 1981,pp. 33-44. MAYFIELD,D., MCLEOD,G. ANDHALL,P. The CAGE questionnaire: Validationof a new alcoholismscreeninginstrument.Amer. J. Psychiat. 131: 1121-1123, 1974. RUSSELL, M. New assessment toolsfor risk drinkingduringpregnancy: T-ACE, TWEAK, and others. Alcohol Hlth Res. World 18: 55-61, 1994. RUSSELL,M., MARTIER, S.S., SOKOL,R.J., MUDAR, P., BOTTOMS,S., JA- COBSON, S. IND JICOBSON, J. Screeningfor pregnancyrisk-drinking. AlcsmClin. Exp. Res.18:1156-1161, 1994. SELZER,M.L. The Michigan AlcoholismScreeningTest: The questfor a new diagnosticinstrument.Amer. J. Psychiat.127: 1653-1658, 1971. SODERSTROM,C.A., DISCHINGER, P.C., SMITH, G.S., MCDUFF, D.R., HEBEL,J.R. ANDGORELICK, D.A. Psychoactive substance dependence amongtraumacenterpatients.JAMA 267: 2756-2759, 1992. STATISTICAL PACKAGE FOR THE SOCIAL SCIENCES. SPSS for Windows: BaseSystemUser'sGuide.Release6.0, Chicago,IL: SPSS,Inc.,1996. STEINWEG,D.L. ANDWORTH,H. Alcoholism:The keys to the CAGE. Amer. J. Med. 94: 520-523, 1993. VOLK, R.J., STEINBAUER,J.R., CANTOR,S.B. AND HOLZER,C.E., III. The Alcohol Use DisordersIdentificationTest (AUDIT) as a screenfor atrisk drinking in primary care patientsof differentracial/ethnicbackgrounds.Addiction92:197-206, 1997. WITTCUEN, H.-U., ROBINS,L.N., COTTLER,L.B., SARTORIUS, N., BURKE, J.D. ANDRESTOR, D. Cross-cultural feasibility,reliabilityand sources of varianceof theCompositeInternationalDiagnosticInterview(CIDI). Brit. J. Psychiat.159: 645-653, 1991. WORLD HEALTH ORGANIZATION. The ICD- 10 Classification of Mental and BehavioralDisorders:ClinicalDescriptionsandDiagnosticGuidelines, Geneva:World Health Organization,1992.