validity of drug use reporting in a high-risk community sample- a comparison of cocaine and heroin...

8/3/2019 Validity of Drug Use Reporting in a High-Risk Community Sample- A Comparison of Cocaine and Heroin Survey Re

1/8

American Journal of EpidemiologyCopyright 1999 by The Johns Hopkins University School of Hygiene and Public HealthAll rights reserved

Vol. 149, No. 10Printed in U.S.A.

Validity of D rug Use Reporting in a High-Risk Com mun ity Sam ple: ACom parison of Co caine and H eroin Survey Reports with Hair Tests

Michael Fendrich,1 Timothy P. Johnson,2 Seymour Sudman,3 Joseph S. Wislar,1 and Vina Spiehler4

Hair specimens were collected from 322 subjects and analyzed as part of an experimental studyadministering household surveys during 1997 to a high-risk community sample of adults from Chicago, Illinois.Toxicologic results were compared w ith survey responses abo ut recent and lifetime drug use. Abou t 35% of thesample tested positive for cocaine, and 4% tested positive for heroin. Sample prevalence estimates of cocaineuse based on toxicologic results were nearly five times the survey-based estimates of past month use an d nearlyfour times the survey-based estimates of past year u se. With the hair test results as the standard , cocaine andheroin use were considerably underreported in the survey. Underreporting was more of a problem for cocainethan for heroin. Among those who tested positive, survey disclosure of cocaine use was associated with higherlevels of cocaine detected in hair. In general, when recent drug use w as rep orted, it was usually detected in hair.Whe n a drug was detected in hair, use was usually not reported in the survey. When heroin was detected in hair,cocaine was almost always detected as well. Am J Epidemiol 1999;149:955-62.cocaine; drug testing; heroin; substance abuse; survey research

Increasing concerns have been raised about thevalidity of drug use prevalence estimates based on sur-vey reports (1-3). Toxicologic advancements, particu-larly those that facilitate the detection of cocaine andopiates in hair (46), promise to enhance our under-standing of the limits of drug use surveys. Studiesincorporating both hair testing and survey researchmetho ds have used the standard of the hair test to eval-uate the accuracy of reports of recent drug use (7-10).To date, most studies that use these comparative pro-cedures have focused on specialized samples in limitedcontexts, such as inmates in jail or prison (11-14),addicts in drug treatment programs (7), obstetricpatients (9), or employees on the job (10). Sincenational drug use prevalence estimates are most com-monly based on results from household surveys con-ducted in the community, insights concerning the lim-

Received for publication March 31 , 1998, and accepted for publi-cat ion September 21 ,199 8.1 University of Illinois at Chicago, Department of Psychiatry,Institute for Juvenile Research, Chicago, IL.2 University of Illinois at Chicago, College of Urban Planning andPublic Affairs, Survey Research Laboratory, Chicago Office,Chicago, IL.3 University of Il l inois at Urbana Champaign, Survey ResearchLaboratory, Urbana Office, Urbana, IL.4 Spielher and Associates, Newport Beach, CA.Reprint requests to Dr. Michael Fendrich, University of Illinois atChicago, Institute for Juvenile Research, 907 South Wolcott Avenue,M/C 747, Chicago, IL 60612.

itations of drug use reporting in this context are of crit-ical importance (15). This report summarizes theresults of one of the first studies to directly comparehair test results with survey responses from householdinterviews obtained from a community sample.MATERIALS AND METHODSSample and data collection

The main study from which this sample is derivedwas based on a community sample of residents aged18-35 years who resided within the city of Chicago,Illinois, in 1997. The pool of neighborhoods selected forthe study was restricted to those with above-averageadmissions to state-supported drug and alcohol treat-ment programs. In addition, neighborhoods were strat-ified according to 1990 Census data to assure a bal-anced distribution across race/ethnicity groups (i.e.,one third of the neighborhoods were predominantlyAfrican American, one third were predominantlywhite, and one third were predominantly Hispanic).Randomly selected blocks from eligible neighbor-hoods were randomly assigned to one of two interviewconditionsa "control" condition, closely followingthe procedures and instruments used in the 1995National Household Survey on Drug Abuse (16) andan experimental" condition that incorporated cognitiveinterview procedures (17). The latter condition included955

y

y

,

j

j

g
http://aje.oxfordjournals.org/http://aje.oxfordjournals.org/http://aje.oxfordjournals.org/http://aje.oxfordjournals.org/http://aje.oxfordjournals.org/http://aje.oxfordjournals.org/http://aje.oxfordjournals.org/http://aje.oxfordjournals.org/http://aje.oxfordjournals.org/http://aje.oxfordjournals.org/http://aje.oxfordjournals.org/http://aje.oxfordjournals.org/http://aje.oxfordjournals.org/http://aje.oxfordjournals.org/


2/8

956 Fendrich et al.the use of a computer-assisted, self-administered inter-view, clearer definitions for the drugs inquired about,and extensive cognitive probes designed to stimulateand facilitate recall of recent behavior. Within selectedhouseholds, a variant of the Troldahl-Carter-Bryantprocedure (18, 19) was used to select randomly fromamong eligible respondents aged 18-35 years. Duringthe screening phase of the interview, informants fromselected households were asked to participate in aninterview "for our study on alcohol, tobacco, and othersubstances." Forty-seven percent of eligible infor-mants agreed to participate.Trained interviewers oriented each respondent to theassigned protocol. Each household interview protocolincluded detailed ques tions about past drug use of a totalof 12 different substances. Questions were also askedabout impairment resulting from drug use and recentinvolvemen t in drug treatment. Respondents were notoffered a financial incentive for participation in either ofthe interview conditions. Upon the conclusion of theinterview, however, interviewers asked respondents toprovide a specimen of their hair for a $10 incentive. Aspart of the consent process, respondents were told thathair samples would be tested for drugs. If a respondentconsented to participate, interviewers cut approximately60 strands of hair from the head. For three cases, sam-ples were taken from facial hair. Respondents were nottold about the hair test until after they had completed thedrug use survey. Hair segments w ere measured, wrappedin foil with the root ends marked, sealed in a paper enve-lope, and enclosed in a plastic bag. Specimens were sentto United States Drug Testing Laboratories in DesPlaines, Illinois, for toxicologic analysis.Variables

Drug use reporting. Three sets of parallel drug usereporting measures were derived from the two proto-cols. Respondents were asked about their lifetime andmost recent use of cocaine and/or crack (labeled in theremainder of this report as "cocaine") and heroin. Onthe basis of responses to these questions, subjects wereclassified as lifetime, past year, past month, andnonusers of each of these substances.Passive exposure. Researchers have identifiedthe potential for passive exposure to drugs, especiallycocaine powder, to result in positive hair tests results(20). To adjust for the potential influence of passiveexposure on hair test results, all respondents wereasked whether they had recently been in the sameroom with someone ingesting cocaine. Subjects whoreported that they had been in the same room withsom eone ingesting this substance on at least 1 day dur-ing the past 30 days were counted as having experi-enced passive exposure.

Hair cocaine and opiate analysis. Hair segmentswere washed and then screened using a fluorescencepolarization assay process as previously described (21 ).Analyses w ere performed on hair segmen ts cut to 3 cmin length from the root ends, corresponding to about 3months hair growth (22). Samples shorter than 3 cmwere analyzed in toto. All specimens exceeding 0.05ng/mg of cocaine or its metabolic equivalents (ben-zoylecgonine or cocaethylene) were confirmed forcocaine using a gas chromatography/mass spectrome-try procedure. Similarly, all specimens exceeding 0.5ng/mg of opiate equivalents (codeine, morphine, ormonoacetyl morphine) were confirmed for opiatesusing gas chromatography/mass spectrometry. All con-firmed cocaine samples that identified cocaine or itsmetabolites at a level exceeding 0.01 ng/mg were con-sidered to be cocaine-positive specimens. All con-firmed opiate samples that identified opiates or theirmetabolites at a level exceeding 0.05 ng/mg were con-sidered to be opiate-positive specime ns. Toxicologicanalyses of hair suggest that monoacetyl morphine ormorphine are most commonly detected in heroin users(6). Accordingly, only confirmed specimens that iden-tified either of these metabolites (either alone or incombination with codeine) were considered positivetests.

Background variables. Since prior research(23-25) suggests that black informants may be lesswilling to disclose drug use than are white informants,race/ethnicity comparisons are of particular interest.Other background variables examined as potential cor-relates of drug use disclosure were gender, highestlevel of education completed, employment status, age,and whether a subject was assigned to the control orexperimental condition of the interview.Statistical analyses

Unweighted sample prevalence estimates were com-pared across measures within each substance. Paireddifferences between the proportion indicating use onthe survey and that testing positive were evaluatedusing the McN emar test (26). Agreement betweendrug tests and interviewer reports was evaluated byexamining sample estimates for sensitivity (using thehair test as the standard for a positive result), coeffi-cient kappa (26), and conditional kappa (27).Conditional kappa is an appropriate statistic when oneof the indicators (or ratings) in a comparison is con-sidered to be a "standard." Since biologic measures areoften considered the standard in drug use studies, con-ditional kappa has been used to evaluate agreementbetween biologic measures and survey reports (7, 8,24). Thus, in our study, the conditional kappa evaluatedchance-corrected agreement only for those subjects

Am J Epidemiol Vol. 149, No. 10, 1999

atUniversitiSainsM

alaysiaonMay4,2011

aje.oxfordjournals.org

Downloadedfrom
http://aje.oxfordjournals.org/http://aje.oxfordjournals.org/http://aje.oxfordjournals.org/http://aje.oxfordjournals.org/http://aje.oxfordjournals.org/http://aje.oxfordjournals.org/http://aje.oxfordjournals.org/http://aje.oxfordjournals.org/http://aje.oxfordjournals.org/http://aje.oxfordjournals.org/http://aje.oxfordjournals.org/http://aje.oxfordjournals.org/http://aje.oxfordjournals.org/http://aje.oxfordjournals.org/http://aje.oxfordjournals.org/http://aje.oxfordjournals.org/http://aje.oxfordjournals.org/http://aje.oxfordjournals.org/http://aje.oxfordjournals.org/http://aje.oxfordjournals.org/http://aje.oxfordjournals.org/http://aje.oxfordjournals.org/http://aje.oxfordjournals.org/http://aje.oxfordjournals.org/http://aje.oxfordjournals.org/


3/8

Validity of Drug Use Reporting 95 7

who had been classified by the hair test as "positive"for a particular drug (i.e., agreement level conditionalon a positive drug test). Since prior research suggestedthat drug users were more comfortable disclosing usein more remote periods of time (24), agreementbetween drug tests and lifetime, past year, and pastmonth reports were evaluated. Analyses also evaluatedthe sensitivity and specificity of interview reports foreach of these time periods. Bivariate cross-tabularanalyses evaluated the association between interviewcondition and background variables and drug use dis-closure among those with positive cocaine hair tests.RESULTS

Hair specimens were obtained from 322 respon-dents, or about 56 percent of the sample of 568 respon-dents. Compared with the rest of the sample, thosewith usable specimens were significantly more likelyto be female, white or other, and unemployed (table 1).As a verification of the sensitivity of the laboratoryprocedures, we split three samples known to be positivefor opiates and cocaine by prior toxicologic analysesand submitted them as blind specimens to the labora-tory on two occasions. Two of the specimens were

TABLE 1. Demographic differences between those with andthose without sp ecimens, Chicago, Illinois, 1997

Variable Hair test providedYes No Total(n = 322) (n = 246) (n = 568)Age groupt (years)18-2 5 38.5 44.9 41.326- 30 28.0 25.7 27.0>31 33.5 29.4 31.7Sex*Male 30.4 50.8 39.3Female 69.6 49.2 60.7Race/ethnicity*African American 50.9 54.9 52.6Hispanic 16.5 25.2 20.2White/other 32.6 19.9 27.1Education^Less than high school 23.9 21.3 22.8High school graduale/GED 34.2 30.3 32.5High school graduate or more 41.9 48.4 44.7Employment*,tFull-time 38.1 55.1 45.5Part-time 16.6 13.1 15.0Unemployed 45.3 31.8 39.5*p


4/8

95 8 Fend rich et al.tion of the M cNem ar chi-square test statistics indicatesthat hair testing yields a higher sam ple prevalence ratethan do any of the survey-based measures of recentuse. The sample prevalence estimate based on hairtests is more than 4.8 times the estimates generatedfrom survey reports of past month use and 3.6 timesthe prevalence estimate generated from survey reportsof past year use. The test statistics suggest that subjectswere significantly less likely to disclose past month,past year, and lifetime use on the survey than theywere to test positive for cocaine on the hair test.Similar, albeit less marked, trends are observed forcomparisons between the prevalence rate generatedfrom heroin tests and rates derived from surveyresponses about heroin use. The test-based estimate is2.2 times the rate generated from survey reports of pastmonth day use and 1.6 times the rate generated fromsurvey reports of past year use. McNemar tests indi-cate that differences between hair test results and sur-vey reports of heroin use were not statistically signifi-cant. Of the three survey versus drug test heroincomparisons, only the comparison involving pastmonth reports approached statistical significance.Survey report and drug test overlap w ere explored ingreater detail in table 3. Agreement between surveyreport and drug test results was particularly poor forcocaine, with none of the kappa or conditional kappameas ures exceeding 0.20. Few respondents whoreported recent (past month or past year) cocaine usehad negative hair tests. Although survey report sensi-tivity clearly increased with more rem ote time periodsfor reporting, sensitivity of cocaine reporting was gen-erally low, irrespective of survey measure. Recentcocaine (past month or past year) use w as disclosed byfewer than one in five subjects who tested positive forthis substance. Just over one in four subjects who test-ed positive for cocaine reported that they had ever usedthis substance.

There was a somewhat higher level of overlapbetween survey reports of heroin and hair test results.Kappa and conditional kappa coefficients for surveymeasures of heroin ranged from 0.29 to 0.43.Although sensitivities for heroin reports were some-what higher than those for cocaine reports, overallrates of disclosure among those who tested positivewere still low. Fewer than half of those w ho tested pos-itive for heroin reported that they had ever used thissubstance.Considering the high cocaine prevalence rate fromhair tests, we evaluated the im pact of raising the cutofflevel for positive test results on our comparative analy-sis. We note that Kline et al. (9), following the sug-gestions of Baumgartner et al. (28), reported use of acutoff corresponding to 0.2 ng/mg. Increasing the cut-off from 0.05 ng/mg to this value results in a net lossof five cocaine-positive cases and had minimal impacton the overall sensitivity of survey reports. Sensitivitywas 19, 20, and 27 percent for past month, past year,and lifetime reports, respectively. A shift up to the 0.5ng/mg cutoff used by Farabee and Fredlund (14)resulted in a net loss of 14 positive cases. Again, thisupward shift had only a marginal impact on validitystatistics, with sensitivity increasing to 21 , 22, and 30percent for past month, past year, and lifetime reports,respectively.Toxicologists have suggested that the presence ofmetabolites at confirmation may be a better indicatorof cocaine ingestion than the presence of parentcocaine alone (29). Accordingly, we reanalyzed com-parisons after counting the 20 confirmed positive caseswithout metabolites as negatives. This change, as withthe other changes noted above, had a marginal impacton sensitivity, with this statistic increasing to 22 per-cent for both past month and past year use and 31 per-cent for lifetime use. We did note, however, that kappaand conditional kappa values showed more than a mar-

TABLE 3. Validity statistic s: survey report versus hair test, Chicag o, Illinois, 1997Surveymeasure

CocainePast monthPast yearLifetimeHeroinPast monthPast yearLifetime

Specificity

98.695.381.5

99.498.496.1

SR-/H-*

208/211201/211172/211

307/309304/309297/309

Validity measureSensitivity

18.018.927.0

30.830.846.2

SR+/H+t20/11121/11130/111

4/134/136/13

K

0.200.170.09

0.410.340.36

K t

0.120.100.07

0.300.290.43* Survey report "noThair test negative,t Survey report "yes"/hair test positive.X Conditional kappa, w ith hair test as the standard.


atUniversitiSainsM

alaysiaonMay4,2011


Downloadedfrom


5/8

Validity of Drug Use Reporting 959

ginal increase when the metabolite criteria were used.The kappas increased to 0.27, 0.22, and 0.14 (from0.20, 0.17, and 0.09), and the conditional kappasincreased to 0.16, 0.14, and 0.12 (from 0.12, 0.10, and0.07) for past month, past year, and lifetime cocaineuse comparisons, respectively. We noted, however,that these values were still well below the cutoff con-sidered to be "fair" agreement (26).Two potential limitations emerged from our surveyreport comparisons. Our recent use survey compar-isons were limited to the periods past month and pastyear. The 3-cm segment that was typically analyzedcorresponded to a period of 3 months of growth. Thislack of overlap in use reference period potentiallyaffected the overall levels of agreement between thetwo sources of use information. To investigate thisissue, we examined the extent to which drug usereports for different use time frames were actuallydetected by the hair test (table 4). Although 87 percentof the self-reported past month cocaine users weredetected by the hair test, only 12.5 percent of thosewhose reported useoccurred between the past monthand the past year were detected by the hair test.Similarly, 67percent of self-reported heroin users forthe past month were detected by the hair test, whilenone of those whose reported use occurred betweenthe past month and past year were detected by the hairtest.In making these observations, we note that self-reported recent last use beyond the past month periodwas relatively rare for either cocaine orheroin. Therewere only eight reports of cocaine use and three reportsof heroin use within this time frame. Most recent usersreported that their last cocaine orheroin use occurredwithin the past month. We also note that while thepotential lack of overlap in the detection period mayaffect the overall agreement statistics on table 3 (i.e.,kappa and conditional kappa), it should not haveaffected the sensitivity rates. Use in the past 3 monthsis encompassed by past year use. Thus, if the drug testis accurate for the detection of use within a 3-month

TABLE 4. Hair test detection by survey reporting status bytime frame, Chicago, Illinois, 1997

Drug use status(survey report)

Ever userMost recently reported useUsed inpast monthUsed inpast yearUsed >1 year agoNever userTotal

SubstanceCocaine

positive43.587.012.523.732.034.5

No .tested692383825 3

32 2

Heroinpositive

33.366.70. 022.22. 3

4. 0

No .tested18

63930 432 2

period, honest informants whose usewas detected bythe hair test should have disclosed use in a time frameencompassed by the past year.Another potential limitation in our comparisons con-cerns the finding that nearly 10 percent of our speci-mens were below the length typically required foranalysis. The analysis of shorter strands potentiallylimits the length of the detection period. To whatextent did the analysis of short strands (i.e., strandsshorter than 3 cm) adversely affect our comparativeresults? The analysis of short strands wo uld potentiallyhave resulted in a lower prevalence of positive testresults. Inparticular, we might expect the use of shortstrands to have resulted in a proliferation of thosereporting recent use that was not detected by the hairtest. Data from table 4 can be used to show that only10 subjects reported recent (past month or past year)cocaine use that was not also detected by the hair test.Only five subjects reported recent heroin use that wasnot also detected by the hair test. Drug test insensitiv-ity to self-reports of recent use was not a prevalentproblem in this study. Nevertheless, we evaluated thelength of analyzed hair segments for those whose hairtests were negative but whose self-reports suggestedrecent cocaine and heroin use. Of the 10 self-reportedrecent cocaine users with negative hair tests, only twohad a specimen with a length smaller than 3 cm. Offive self-reported recent heroin users with negativehair tests, none had a specimen with a length smallerthan 3 cm. Thus, hair test insensitivity in this samplewas probably not a consequence of insufficient hairlength. In addition, we underscore that the major con-sequence of short specimens is on the ov erall lev els ofagreement (kappa and conditional kappa) and not onthe sensitivity rate.We investigated potential correlates of recent (pastyear) use disclosure among the subgroup of respon-dents who tested positive for cocaine (n = 111). Dueto the small number of positive heroin tests, no similarcomparisons w ere made for this substance. There wereno associations between past year use disclosure andthe age, education, employment status, race/ethnicity,or sex of the respondents. Disclosure was also notassociated with interview condition. We do note, how-ever, that respondent disclosure was associated withthe amount of cocaine detected in the hair.Specifically, 34 percent of the respondents with levelsof cocaine at or above the sample median value (3.8ng/mg) disclosed past year use, compared with just 4percent of respondents with levels below the median( 3 ^ = 1 = 16.6; p


6/8

96 0 Fendrich et al.

reported never having used this substance, only fourreported passive exposure.Additional bivariate comparisons evaluating associ-ations between demographic characteristics and life-time drug use disclosure yielded two significant vari-ables: race/ethnicity and education. Compared withthose with more education, those with less than a highschool education had higher rates of lifetime disclo-sure (X2df=2 = 8.64; p < 0.05). Compared with H ispanicand black subjects, white/other subjects had higherrates of lifetime disclosure (X2d{=2 = 6.07; p < 0.05). Inthe context of multiple logistic regression simultane-ously controlling for race/ethnicity, gender, educationstatus, age, and study condition, only race/ethnicitymaintained significance. Adjusted odds ratios suggestedthat, compared with black respondents who tested pos-itive for cocaine, white/other respondents who testedpositive had 1.9 times the odds of disclosing lifetimedrug use (95 percent confidence interval 1.3-33.2); thecontrast between Hispanic and white/other respondentswas not significant.DISCUSSION

The findings presented in this report suggest thatsurvey reports of cocaine and heroin use do not pro-vide an accurate picture of the prevalence of drug usein high-risk communities. Among those who consentedto the hair test, drug use was underreported.Underreporting did not vary according to subject back-ground characteristics or interview conditions.Disclosure did, however, vary by drug. It appears thatheroin use was more likely to be disclosed than wascocaine use. In addition, those with higher concentra-tions of cocaine in their hair were more likely to dis-close use of this s ubstance. Finally, subjects w ho testedpositive for cocaine were more willing to discuss life-time use than recent use. These findings suggest thatepidemiologic researchers should have more confi-dence in lifetime than in recent drug use prevalencerates. When recent drug use was reported, it was usu-ally detected by the hair test. On the other hand, whenrecent drug use was detected by the test, it was usual-ly not reported.Findings with respect to the more accurate reportingof heroin use are consistent with at least two otherstudies of reporting behavior among drug users seek-ing treatmen t. Wish et al. (13) reported that clients w hovolunteered for drug treatment and who tested positiveby hair for heroin nearly always reported recent heroinuse at an initial intake screening; in contrast, clientswho tested positive for cocaine reported recent useonly about 79 percent of the time. While these authorssuggest that treatment system incentives may be thecause of differences in disclosure rates, our findings

derived in the context of a community sample suggestthat reporting of heroin may be less sensitive thanreporting of cocaine. Although the precise reasons forthis need further exploration, we believe that thesecomparative findings support the observations ofMagura and Kang (8) regarding the highly stigmatizednature of cocaine use, especially crack cocaine use,among young adults (30, 31).Differential rates of disclosure with level of cocainedetected raise the possibility that our findings maypotentially be an artifact of false positives amongnonusers (i.e., passive exposure) among a significantportion of our sample. On the one hand, we note thatfew subjects in our sample who tested positive reportedpassive exposure. On the other hand, there is the possi-bility that subjects may be just as reluctant to disclosetheir associations with drug users as they are to disclosetheir drug use; passive exposure may have b een under-reported here. Among the five cocaine-positive subjectswho admitted to passive exposure but not to use, themedian cocaine level was 0.91, a level well above the"high cutoff'used by Farabee and Fredlund (14). Wealso note that some toxicologists (20) point out that pas-sive exposure canno t be differentiated according to lev-els of cocaine detected in the hair. We believe that themost accurate interpretation of our findings is that thosewho use cocaine heavily are more comfortable dis-cussing their involvement with this substance than aremore marginal users. This interpretation is consistentwith findings from a previous analysis (23) indicatingthat respondents who reported lower levels of cocaineuse in an initial interview were more likely to fail toreport lifetime use at follow-up.When heroin was detected by hair tests in our sam-ple, cocaine was also almost always detected. Thisfinding that heroin use in the community usually coin-cides with cocaine use is of considerable epidemiolog-ic importance and provides empirical support to ethno-graphic accounts of the emerging phenomenon of"speedballing" (32). We have no way of knowingwhether the actual use of these substances for the 11subjects with co-occurring positive tests was simulta-neous. Our study, like many epidemiologic surveys,considers heroin and cocaine as separate drug cate-gories. These findings underscore the importance ofincorporating survey measures specifically aimed atascertaining whether heroin and cocaine were actuallyused in combination or separately.The lack of association between demographic char-acteristics and recent drug use disclosure was some-what surprising. The 111 subjects who tested positivefor cocaine were demographically more homogeneousthan was the rest of the samp le, especially w ith respectto race/ethnicity. The subgroup that tested positive for


atUniversitiSainsM

alaysiaonMay4,2011


Downloadedfrom


7/8

Validity of Drug Use Reporting 961

cocaine was 80 percent black, 11 percent Hispanic, andonly 9 percent white/other. Sample homogeneity, alongwith the reduced sample size, diminished the power todetect differences in disclosure rates across demo-graphic subgroups. The finding suggesting that blackinformants w ere less likely todisclose lifetime usethanwere white/other informants, a finding that held up in amultivariate context, is consistent with recent literaturediscussing social differences with respect to drug dis-closure (2 3- 25). Thecontrast between recent and life-time analyses reflects a general reluctance to discloserecent drug use among all subgroups testing positivefor cocaine. Again, these findings support the notionthat prevalence rates constructed from lifetime usereports have more validity than do rates derived fromresponses to questions about recent use.Although the utility of hair testing has been demon-strated here and in other contexts, we underscore animportant limitation of this procedure in community-based epidemiologic research. Just over half of thesubjects in our study provided useable specimens. Thisparticipation rate is consistent with other work in sam-ples not from corrections departments or treatmentfacilities (9). Nevertheless, the key demographic dif-ferences observed between participants and nonpartic-ipants, especially with respect torace and gender, sug-gest that prevalence results based solely on thisprocedure may be subject to considerable bias.Within the context of this study, there is the possi-bility that the $10 incentive motivated higher levels ofdrug test participation among drug users, especiallynondisclosing drug users. Such differential selectioninto the test would yield comparative analyses thatoverestimate the extent of underreporting and druguse in the general population. Supplementary analysessuggested that there were nodifferences in the rates ofdisclosure of any recent and lifetime drug use betweendrug test participants and nonparticipants (33).Further, we note that even if every hair test nonpartic-ipant were to test negative for cocaine, the rate ofcocaine use detected by thehair test for the entire sam-ple would still greatly exceed rates of recent usederived from the survey. Thehypothetical rate testingpositive under this unlikely scenario would be 19.5percent (111 of 568); this compares with 5.3 percent(3 0 of 568) reporting use in the past month, and 7.9percent (45 of 568) reporting use in the past year. Theaccuracy of our estimates hinges both on the overalllevel of recruitment into the drug test and on respon-dent behavior (especially recent drug use) associatedwith test participation. Future studies need to morecarefully assess the precise reasons and mechanismsassociated with participation in hair and other drugtesting procedures in community research.

ACKNOWLEDGMENTSSupported in part by National Institute on Drug Abusegrants R01DA 09285 andR01DA09286-02.

REFERENCES1. Reuter P. Prevalence estimation and policy formation. J DrugIssues 1993;23:167-84.2. Leen J. A shot in the dark on drug use? The GAO and othersfind the government's numbers aren't very precise. TheWashington Post National Weekly Edition, Vol. 15, no. 11.January 12,1998:32-33.3. General Accounting Office (GAO). Drug use measurement:strengths, limitations and recommendations for improvement.Report no. GAO/PEMD-93-18. Washington, DC: US GPO,1993.4. Cone EJ. New developments in biological measures of drugprevalence. In: Harrison L, Hughes A, ed s. The validity of self-reported drug use: improving the accuracy of survey estimates.NIDA research monograph no. 167. Rockville, MD: US

Department of Health and Human Services, 1997:108-29.5. Baumgartner WA, Hill V. Hair analysis for organic analytes:methodology, reliability issues, and field studies. In: Kintz P,ed. Drug testing in hair. Boca Raton, FL: CRC Press,1996:223-66.6. Garside D, G oldberger B A. Determination of cocaine and opi-oids in hair. In: Kintz P, ed. Drug testing in hair. Boca Raton,FL : CRCPress, 1996:151-80.7. Magura S, Freeman R, Siddiqi Q, et al. The validity of hairanalysis for detecting cocaine and heroin use among addicts.Int J Addictions 1992;27:54-69.8. Magura S, Kang S. The validity of self-reported drug use inhigh risk populations: a metaanalytic review. Subst UseMisuse 1996;31:1131-53.9. Kline J, Ng SK, Schittini M, et al. Cocaine use during preg-nancy: sensitive detection by hair assay. Am J Public Health1997;87:352-8.10. Cook RF, Bernstein AD, Andrews CM. Assessing drug use inthe workplace: a comparison of self-report, urinalysis, andhairanalysis. In: Harrison L, Hughes A, eds. The validity of self-reported drug use: improving the accuracy of survey estimates.NIDA research monograph no. 167. Rockville, MD: USDepartment of Health and Human Services, 1997:247-72.11. Mieczkow ski, T, Barzelay, D, Gropper, B, et al . Concordance ofthree measures of cocaine use in an arrestee population: hair,urine, and self-report. J Psychoactive Drugs 1991;23:241-9.12. Feucht TE, Stephens RC, Walker ML. Drug use among juve-nile arrestees: a comparison of self-report, urinalysis and hairassay. J Drug Issues 1994;24:99-116.13. Wish ED,Hoffman JA, Nemes S. The validity of self-reportsof drug use at treatment admission and follow up: comparisonswith urinalysis and hair assays. In: Harrison L, Hughes A, eds.The validity of self-reported drug use: improving theaccuracyof survey estimates. NIDA research monograph no. 167.Rockville, MD: US Department of Health and HumanServices, 1997:200-26.14. Farabee D,Fredlund E. Self-reported drug use among recentlyadmitted jail inmates: estimating prevalence and treatmentneeds. Substance UseMisuse 1996;31:423-35.15. Miller PV. Review: survey on drug abuse. Public Opinion Q1997 ; 61 : 627^1 .16. Office of Applied Studies. National household survey ondrugabuse advance report no. 18. Washington, DC: SubstanceAbuse and Mental Health Services Administration, 1996.17 . Schwarz N, Sudman S, eds. Answering questions.Methodology for determining cognitive and communicativeprocesses in survey research. San Francisco, CA: Jossey Bass,1996.


atUniversitiSainsM

alaysiaonMay4,2011


Downloadedfrom


8/8

96 2 Fendrich et al.18. Troldahl V, Carter RE . Random selection of respondents with-in households in phone surveys. J Marketing Res 1964;l:71-6.19. Bryant BE. Respondent selection in a time of changing house-hold composition. J Marketing Res 1975;12:129-35.20. Kidwell DA, Blank DL. Environmental exposurethe stum-bling block of hair testing. In: Kintz P, ed. Drug testing in hair.Boca Raton, FL: CRC Press, 1996:17-68.21 . Lewis D, Moore C, Morrissey P, et al. Determination of drugexposure using hair: application to child protective cases.

Forensic Sci Int 1997;84:123-8.22 . Sachs H. Forensic applications of hair analysis. In: Kintz P, ed.Drug testing in hair. Boca Raton, FL: CRC Press, 1996:17-68.23 . Fendrich M, Vaughn CM. Diminished lifetime substance useover time: an inquiry into differential underreporting. PublicOpinion Q 1994;58:96-123.24. Fendrich M, Xu Y. The validity of drug use reports from ju ve-nile arrestees. Int J Addictions 1994;29:971-85.25 . Johnston LD, O'Malley PM. The recanting of earlier drug useby young adults. In: Harrison L, Hughes A, eds. The validityof self-reported drug use: improving the accuracy of surveyestimates. NIDA research monograph no. 167. Rockville, MD:US Department of Health and Human Services, 1997:59-80.

26. Fleiss JL. Statistical methods for rates and proportions. NewYork, NY: John Wiley & Sons, 198 1.27 . Bishop YMM, Fienberg SE, Holland PW. Discrete multivari-ate analysis: theory and practice. Cambridge, MA: MIT Press,1975.28 . Baumgartner WA, Hill VA, Blahd WH . Hair analysis for drugsof abuse. J Forensic Sci 1989;34:1433-53.29 . Society of Hair Testing. Criteria for obtaining a positive result.Forensic Sci Int 1997;84:3-6.30. Dunlap E, Johnson BD. The setting for the crack era: macroforces, micro consequences 1960-1992. J Psychoactive Drugs1992;24:307-22.31. Hamid A. The development cycle of a drug epidemic. Thecocaine smoking epidemic of 1981-1991. J PsychoactiveDrugs 1992;24:337-48.32 . Office of National Drug Control Policy. Pulse check. Nationaltrends in drug abuse, Summer 1997. Washington, DC: Officeof National Drug Control Policy, 1997.33 . Fendrich M, Johnson T, Wislar JS, et al. The feasibility of hairtesting in a household survey on drug abuse. In: M ieczkowskiT, ed. Drug testing methods: assessm ent and evaluation. BocaRaton, FL: CRC P ress (In press).


atUniversitiSainsM

alaysiaonMay4,2011


Downloadedfrom

validity of drug use reporting in a high-risk community sample- a comparison of cocaine and heroin...

Documents