Warner LA, Alegría M, Canino G. Remission From Drug Dependence Symptoms and Drug Use Cessation AmongWomen Drug Users in Puerto Rico. Arch Gen Psychiatry. 2004;61(10):1034-1041. doi:10.1001/archpsyc.61.10.1034
Studies of remission from drug dependence have most often been based
on treatment samples, with limited generalizability to persons who may benefit
from but never seek substance abuse treatment. Little is known about remission
patterns among drug users in the community.
To identify patterns and predictors of remission in a community sample
of drug users followed up prospectively.
Three waves of data on a range of individual and interpersonal correlates
of drug abuse and health care service use were collected between April 1997
and October 2000.
Areas of metropolitan San Juan where drug sales were known to occur.
Two hundred seventy-five women aged 18 to 35 who were crack cocaine
or injecting drug users.
Main Outcome Measures
Self-reported drug use validated with urine screens and drug use dependence
criteria based on the DSM-IV.
Most (86.9%) of the women were drug dependent at baseline. By wave 3,
fewer than half (42.6%) of the women were dependent, 13.8% had subthreshold
disorder, and 17.8% used substances but did not endorse any dependence criteria.
Cessation of use and decreases in the number of dependence criteria endorsed
were significantly less likely for women with depressive symptoms (odds ratio
[OR], 0.92; 95% confidence interval [CI], 0.88-0.96; and OR, 0.88; 95% CI,
0.86-0.90; respectively), with a partner who engaged in criminal activities
(OR, 0.30; 95% CI, 0.16-0.58; and OR, 0.63; 95% CI, 0.46-0.85; respectively),
and who traded sex for money or drugs (OR, 0.12; 95% CI, 0.05-0.29; and OR,
0.26; 95% CI, 0.19-0.35; respectively).
Drug use patterns and rates of dependence fluctuated substantially over
time among drug users recruited from the community. Findings regarding the
characteristics that impede remission suggest that mental health practitioners
have an important role to play in community-based outreach and interventions
designed to support women’s efforts to stop using drugs.
Compared with the amount of research on the development of substanceuse disorders, there has been less emphasis on patterns of remission oncedrug use disorders are established. Available data on remission are primarilyfrom treatment-based samples or from community-based samples that representpopulations who are dependent on alcohol rather than illicit substances. Thegeneralizability of these studies to community samples of persons who abuseillicit substances is not known; however, their lower degree of “recoverycapital”1 is likely to yield differentremission patterns. Moreover, respondents in longitudinal community studiesthat speak to remission from alcoholism have been almost exclusively male2- 4 or predominately EuropeanAmerican,5- 9 andlongitudinal studies of remission from heroin tend to be based on convenience10 or treatment samples, also predominately male.11,12 Consequently, our understanding ofremission from drug dependence among ethnic minority women is limited.
According to data from longitudinal community-based studies,2,3,5 most persons with alcoholismdo not receive treatment for their condition. Rather, people with alcoholproblems often “age out” of substance use behaviors8,9,13- 15 orremit from alcoholism without substance abuse treatment (ie, “spontaneously”),6,7,16- 18 especiallyif the drinking problems are less serious.19 Modestevidence about remission from other substances also suggests that high ratesof cessation occur without treatment,20 althoughdata on respondents’ drug abuse or dependence were not reported. Theearliest empirical evidence that “hard” drug users can achieveremission, despite never receiving treatment, was based on a sample of malesoldiers who had been opiate users in Vietnam and then decreased or stoppedtheir use on return to the United States.10 Morerecent research supports the argument that remission from heroin use mostlikely occurs without treatment,21 althoughtreatment-related remission may be more likely than spontaneous remissionfor users who are dependent on heroin.22
Insofar as posttreatment factors that support sustained remission areindependent of the treatment experience, studies of relapse following treatmentwere used to identify relevant predictors of remission in this community sample,along with cross-sectional studies of the correlates of substance abuse amongethnic minority women. Three main categories were identified: individual (sexualabuse history, psychiatric symptoms, and levels of social status attainment),interpersonal (relationship with a partner and social support from familyand friends), and institutional (contact with social, health, and criminaljustice systems). Low socioeconomic status, comorbid psychiatric conditions,and lack of family and social supports are consistent markers of noncompliancewith treatment and relapse following treatment completion.23 Forwomen in particular, symptoms of depression are believed to play an importantrole in persistence of substance abuse.24- 26 Relationshipswith a partner are associated with treatment retention,27 posttreatmentdrug use continuation,28,29 anddrug use persistence.30 Childhood experienceof sexual abuse is a significant predictor of drug use disorder, as well asa risk factor for unsuccessful treatment.31 Healthservices research suggests that remission among the most serious substanceusers tends to be associated with the use of health services rather than addictionor mental health treatment.32- 35
This study aims to advance our understanding of remission using datafrom the Inner-City Latina Drug Use Study (ICLDUS),36 acommunity-based longitudinal study of women who live in the extended metropolitanarea of San Juan. First, we describe patterns of remission in terms of reductionsin the number of DSM-IV drug dependence criteriaand total cessation of drug use. Both types of remission are considered becausea sole focus on cessation of all drug use would overlook the chronic relapsingnature of drug and alcohol addictions, which is underscored by a literaturereview that reported that 40% to 60% of samples return to substance use withina year of treatment and that a smaller proportion (15% to 30%) returns todependent use.37
Second, individual, interpersonal, and institutional predictors wereanalyzed separately and together to identify the characteristics that remainsignificant in the context of risk factors from other domains. By identifyingkey predictors, and the type of remission for which they are most salient,the results may be useful for program planners and clinicians who seek tosupport women’s efforts to stop using drugs.
The ICLDUS is a 3-wave longitudinal research project funded by the NationalInstitute on Drug Abuse, fielded between April 1997 and October 2000, in SanJuan, to study drug use patterns and correlates among low-income, inner-citywomen, aged 18 to 35. Two groups of women were recruited: crack cocaine andinjecting drug users from drug sale areas and non–drug users who livedwithin a 1.6-km radius of these areas. Because the research question focuseson remission from substance use and dependence, only respondents from thefirst sample were included in the analysis.
A 2-stage sampling procedure was used. First, research staff visitedlocations that had constituted the sampling frame of a prior study38 and updated the list, including new areas and excludingareas that had closed down, only operated at night, or did not service femaleclients. Sixty areas were identified and ordered by random assignment forthe purpose of recruitment.
During the first year of the study, 316 women were approached at thedrug sale areas. Three women refused to participate, and 38 were ineligiblebecause of the age criteria, yielding 275 female drug users in the baselinesample (time 1 response rate, 98.8%). On average, 16.5 months elapsed betweentimes 1 and 2, and 10.1 months between times 2 and 3. Ten percent could notbe located at times 2 and 3 for interviews, 4.4% were not reinterviewed attime 2, and 6.6% were not reinterviewed at time 3, but they provided informationfor the other data collection periods. Substance use patterns did not differsignificantly between the women with complete and incomplete data accordingto analyses of variance adjusted for multiple comparisons, with a mean of4.7 drug dependence symptoms experienced at time 1 by women with completeinformation, 4.5 among women missing at time 2 only, 5.3 among women missingat time 3 only, and 5.1 among women missing at times 2 and 3.
Urine and hair samples for drug assays, as well as serum for human immunodeficiencyvirus (HIV) testing, were obtained from consenting participants. Data fromurine tests are particularly important for this sample, who may underreportdrug use because of a strong stigma associated with illicit drug use amongLatino groups39 and because women may facegreater consequences for reporting drug use than their male counterparts.40
The women were offered HIV and sexually transmitted disease precounselingand were given a directory of sexually transmitted disease and drug treatmentfacilities. All recruitment and follow-up procedures were consistent withthe study protocol approved by the Institutional Review Board of the Universityof Puerto Rico Medical Sciences Campus.
Respondents reported on their use of 8 illicit substances (crack cocaine,cannabis, opiates, sedatives, amphetamines, hallucinogens, inhalants, andphencyclidine [phenylcyclohexyl piperidine]), as well as alcohol. Resultsof urine tests for crack cocaine and heroin, the most prevalent drugs in thissample, were included to improve the accuracy of drug use estimates. The urinetests were conducted with a rapid on-site protocol that has shown a sensitivityof 100%, specificity of 90%, and efficiency of 95.1% for detection of cocaineand heroin use.41
Substance dependence was measured with the Spanish version of the substanceabuse and dependence module of the Composite International Diagnostic Interview(CIDI).42 Past-year dependence was definedby DSM-IV criteria,43 andwomen who endorsed 3 or more of the criteria for a given drug within the yearbefore interview were counted as substance dependent.
Categorical and continuous measures were used to describe substanceuse remission patterns. The categorical measure of substance use had 4 levels:(1) no substance use, (2) substance use but no dependence criteria or diagnosis,(3) substance dependence criteria but no dependence diagnosis, and (4) fulfillssubstance dependence diagnosis. Women who self-reported no substance use buttested positive for drug use on the basis of a urine test were included inthe level 2 category. The number of DSM-IV criteria(range, 0-7) relevant to a specific drug at each time point was used as thecontinuous measure of substance use. For the multivariate analyses, a shiftfrom any use (level 2, 3, or 4) at time 1 to no use (level 1) at time 2 wascounted as a yes answer for the dichotomous substance use cessation variable,otherwise as a no. The same assessment was made for changes in substance usebetween times 2 and 3. A decrease in the number of substance dependence criteriafrom one time point to the next was considered symptom reduction in multivariateanalyses based on the continuous measure.
The model underlying the ICLDUS study design assumes that substanceabuse is a multidimensional phenomenon, and information was collected on awide range of known drug abuse correlates and risk factors, many of whichhad been used in previous studies36,44,45 ofsimilar populations. Guided by results of these studies, specific variablesfrom the data set were selected to represent individual, interpersonal, andinstitutional domains. The distribution of these variables at baseline ispresented in Table 1. To minimize multicollinearityin the predictive models, correlational analyses were performed using thetime 1 variables, before other data reduction or recoding occurred; variablesthat were highly correlated with one another (P<.001)were combined into aggregated measures or prioritized for inclusion basedon the extent to which they could be targets for intervention.
The individual domain included 3 distal and 5 current predictors. Educationwas a continuous variable based on number of years of education completed,the continuous “substance use problem onset age” was based onthe youngest age the respondent reported for any CIDI DSM-IV substance use disorder criteria, and sexual abuse before the age of18 was indicated if the respondent endorsed any of 4 items regarding sexualrelations with someone at least 5 years older than the respondent or witha family member.46 Current predictors fromthis domain included trading sex for money or drugs in the past year (basedon a yes or no question), number of depressive symptoms in the past year fromthe CIDI depression module,42 physical functioningin the past month based on summed responses to Medical Outcomes Study 36-ItemShort-Form Health Survey items (with a high score indicating better physicalfunctioning),47 HIV status (counted as positiveif the respondent reported an age at onset of HIV or had a positive HIV bloodtest result, otherwise not), and religiosity (respondents who answered thatreligion is “somewhat” or “very important” to themwere counted as religious, otherwise not). Interpersonal domain variablesaddressed the partner’s criminal activity (dichotomized into an indicatorvariable, with 1 assigned if the total number of 14 possible activities inthe past year summed to 1 SD above the mean), social support (a compositemeasure of family social support and friend social support, each of whichwas based on 4 items assessing the frequency of behaviors such as expressinginterest in the respondent’s well-being, Cronbach α for the itemsis.87), and drug use among family and friends (coded yes  if a family memberwas identified as a drug user or the number of friends using drugs was >1).The institutional domain included dichotomous measures reflecting past-yearcontact in the following 5 settings: substance abuse treatment (ie, outpatient,detoxification, methadone, residential, and drug treatment in jail), criminaljustice (arrested for any of 12 reasons, including robbery and drug possessionor sales), social services, medical services (ie, a physician’s office,health care center, or group practice), and hospitals (ie, admitted for ≥1day for a physical health problem).
To control for repeated observations on the same individual, multivariateanalyses were based on general estimating equations methods. These methodsare based on the assumption that dropout cases are missing completely at random.48 To evaluate if this assumption was satisfied, dropoutstatus was regressed on the variables identified for inclusion in the analyses.The results (data not shown) indicated that dropout status was associatedwith depressive symptoms (Wald χ2 = 3.98, P = .046; means for women with complete and incompletedata are 8.2 and 10.2, respectively) and positive HIV status (Wald χ2 = 7.24, P = .007; percentagespositive among women with complete and incomplete data are 22.4% and 38.2%,respectively). Given the violation of the assumption about the nature of themissing data, we ran the analyses in 2 ways.49 First,we deleted all women who did not provide data at time 2 or time 3 and ranthe analysis on the women who responded at all time points (ie, the completecases). We also imputed data, guided by the results of the attrition analysis.Specifically, for women who were missing at time 2 or time 3, the number ofdepressive symptoms was set to 10.2, and HIV status was set to 1, indicatingyes. Mean values for the other predictor variables were estimated based onthe women who provided data at time 2 and replaced the missing values forwomen who were not observed at time 2. The same procedure was used to imputemissing values at time 3. The multivariate results presented are based onimputed data.
The person-level data were transformed so that each woman was representedin the data set by 3 data points for each variable (825 observations withimputed data and 767 observations for complete cases). The PROC GENMOD procedurein SAS50 software was used to estimate themultivariate models. Logistic models were run to predict substance use cessationat time t, and multivariate proportional odds modelswere estimated to predict the probability of reduction in symptoms by time t. For comparability across models, the parameter estimatesand corresponding 95% confidence intervals were exponentiated, and odds ratios(ORs) are reported. Each domain was first analyzed separately to identifythe variables that were associated with the outcomes, and then the domainswere entered together to identify the predictors that remained significantwhen controlling for other domains.
Table 2 shows the longitudinalaggregate patterns of substance use. All of the women reported some levelof drug use at time 1, and 86.9% qualified for past-year substance dependence,with a mean number of 4.9 criteria. Most of the women were dependent on heroinor cocaine, and about one quarter of the women had co-occurring alcohol dependence.One tenth (10.6%) were not dependent but endorsed 1 or 2 DSM-IV criteria as assessed by the CIDI, and 2.6% reported substanceuse but did not endorse any diagnostic criteria. Over time, the rate of dependencedeclined significantly (F = 17.65, P<.001),as did the mean number of symptoms experienced (F = 7.97, P<.001), while the rates of substance use increased(F = 4.65, P = .01). At time2, no substance abuse was reported by 9.1%, but about 40% of these women testedpositive for cocaine or heroin. Incorporating this information into the prevalenceestimate results in only 5.1% of the women with no substance use and in 9.5%of the women with substance use but no report of dependence symptoms. At time3, fewer than half (42.6%) of the women were dependent, and 13.8% reportedsymptoms only. Based on adjustments because of positive urine screens forwomen who had self-reported no substance use, 17.8% of the women used substanceswithout reporting symptoms at time 3.
Individual-level patterns of substance use among women who were interviewedat each time point show that 86.2% of women were persistent users, 3.2% didnot use substances at either of the follow-ups, and the remainder fluctuatedfrom use to no use or the reverse (data not shown). The patterns of symptomreduction are more complicated. About one third (33.7%) of the women followedpathways that suggest continued improvement (eg, no symptoms at time 2 ortime 3, fewer symptoms at time 2 than time 1, and no symptoms at time 3).
Results from the models predicting cessation from substance use arepresented separately for each domain in the “Partial Models” columnsof Table 3. In the first model, 2 elementsof the individual domain (ie, engaging in sex work and physical functioning)are significantly associated with cessation of substance use. Women who tradesex for money or drugs are substantially less likely (OR, 0.07) to stop usingsubstances compared with women who do not, and women with positive physicalfunctioning are more likely to stop drug use. In the interpersonal domain,partner criminality reduces the probability of cessation by close to 75%;on the other hand, women with high levels of social support are close to twiceas likely to stop substance use compared with women with low levels of socialsupport. Only substance abuse treatment in the institutional domain predictscessation; women who had contact with any substance abuse service in the pastyear were less likely to stop using substances (OR, 0.42).
When all the domains are analyzed together (Table 3, “Full Model” columns), the relationships betweenseveral of the variables in the individual domain and cessation shift modestly.For example, depressive symptoms, which had been significant at the margin(P = .059) in the partial model, significantlyreduce the odds of cessation in the full model (OR, 0.92), whereas positivefeelings about religion increase the odds of cessation (OR, 1.64). A partner’scriminality persists as an interpersonal correlate of cessation in the fullmodel, and none of the institutional variables are significant. There waslittle difference between these results and the results of the models thatused data on complete cases only; the same parameters were significant, butat P<.05 rather than more conservative levelsof significance.
Overall, more of the estimates of remission from symptoms are significantthan estimates of cessation of use (Table 4).Significant predictors in the partial models include education, sex work,depressive symptoms, and physical functioning (individual domain). Women whotrade sex for money or drugs are about one fourth less likely (OR, 0.25) toimprove (ie, experience fewer symptoms) as women who do not, and with eachadditional CIDI depressive symptom, the probability of symptom reduction decreasesby approximately 10% (OR, 0.87). On the other hand, women with positive physicalfunctioning are more likely to report a reduction in drug symptoms. All ofthe interpersonal predictors in the partial model are significantly associatedwith symptom remission. Partner criminality and drug use among family or friendsare associated with lower odds of remission (OR, 0.57 and 0.54, respectively),while social support significantly increases the odds of remission (OR, 1.22).
All of the institutional predictors, with the exception of social servicecontact, significantly predict remission. However, the association is positiveonly for women who received medical care in the prior year; the probabilityof remission increases by about 70% when women receive medical care comparedwith when they do not. Hospitalization, arrest, and substance abuse treatmentare predictive of lower odds of remission.
Results of the full model for symptom remission are consistent withthose of the partial models. Seven of the 11 significant predictors from thepartial models are also significant in the final model, but drug use amongfamily or friends, social support, being arrested, and hospitalization arenot significant in the full model. When the models were rerun for cases withcomplete data, the estimates for education and physical functioning were nolonger significant.
Most women in this community sample were dependent on substances orhad problems associated with substance use when they were recruited. Althoughthere was a significant reduction in substance dependence and number of symptomsover time, most women continued to use substances during the subsequent 2years, often with some accompanying substance dependence symptoms. Althoughcaution should be exercised when comparing the ICLDUS and other longitudinalcommunity-based studies because of substantial differences in sample characteristicsor the type of substances studied, it is notable that the rates of persistenceand numbers of dependence symptoms tend to be substantially higher in theICLDUS. An exception comes from a prospective study51 ofolder, predominately white male problem drinkers, in which only approximately10% resolved their drinking problems. On the other hand, studies such as a1-year follow-up of long-term cannabis users in Australia found that halfof the persons who remitted from dependence reported no drug use whatsoever,52 and in a rural, predominately white sample, alcoholdependence decreased from 41% to 16% in 18 months.53 Self-reportsin these studies were not supplemented with urine tests, however. Other studies54,55 of persons with severe alcohol dependencereport that abstinence is more likely than controlled drinking, with the implicationthat controlled drinking without relapse is difficult to sustain for formerlydependent persons. Additional data are needed to determine the extent to whichthe persistence of substance use in the absence of meeting diagnostic criteriais a predictor of recurrent disorder or an expected stage in the achievementof complete abstinence.
The trading of sex for money or drugs is the most consistent predictorfrom the individual domain, significantly decreasing the odds of symptom reductionand use cessation in partial and full models. Depressive symptoms also appearto be an important individual factor in both remission outcomes, as does physicalfunctioning. Religiosity is the sole characteristic to be significant in onlyone model (ie, the full model predicting drug use cessation). A partner’scriminality is a robust negative predictor in the partial and full modelsfor symptom remission and substance use cessation, whereas other interpersonalvariables are not significant in the full models. When all domains are consideredjointly, none of the institutional domain variables are significantly associatedwith drug use cessation, but medical care and substance abuse treatment aresignificantly associated with symptom remission.
Sexual abuse history was not a significant predictor of either outcome,possibly because the measure did not assess severity or duration. Moreover,the measure was designed to capture the distal experience of sexual abuseduring childhood, and as such its effect on substance use may be mediatedby recent experiences of physical attack or rape56 orongoing psychosocial responses to the abuse.57,58 Togetherwith other research that shows more effective outcomes when substance abusetreatment providers address issues related to sexual abuse,59,60 theseresults suggest that a history of abuse may be important to address once womenare involved in treatment. Substance abuse interventions that seek to engagewomen in treatment may need to attend to current partner issues, especiallythose related to the criminal activity of the partner, and to the economicand psychosocial mechanisms that support trading sex for money or drugs.
It is also possible that the effect of sexual trauma during childhoodmight be indirectly captured through current depressive symptoms among thewomen in the ICLDUS. This interpretation is supported by a longitudinal study61 of posttreatment drinking outcomes in which the significanceof sexual abuse history disappeared when depression and other psychiatricdisorders were added to the models. The strong association between depressivesymptoms and lower probabilities of remission and cessation in the ICLDUSis also consistent with studies4,62- 64 thatshow associations between depression and relapse to drug use after treatment.
The effect of the variables in the institutional domain is importantto examine further with data that provide more refined measures of temporalordering within the year before interview. In particular, the present studyindicates that substance abuse treatment is associated with lower odds ofremission from drug dependence symptoms and has no significant relationshipwith cessation of drug use. Although conclusions regarding a causal connectionare not justified, women with the most severe substance abuse at any pointin the past year may be the most likely to receive treatment, and detectionof a substantial change in their drug use behavior would require a longerfollow-up. It is also possible that the drug abuse services available to thesewomen do not include sex-specific programs, which have been shown to yieldpositive outcomes for women.59,65 Analternative explanation is that drug treatment programs as implemented areintermittently helpful in decreasing the negative consequences that accompanywomen’s drug use and in improving functioning, but they do not solvethe interpersonal and individual problems that influence continued drug use.
In the ICLDUS, rates of contact with drug treatment providers and healthcare providers are similar, but health care use resulted in significantlypositive outcomes in the multivariate models (ie, symptom remission). Althoughthese data do not allow examination of the actual services received in healthcare settings, the results suggest that engaging women in the recovery processmay be achieved through an emphasis on health rather than drug use behavioror on collaborative service delivery between health care and substance abuseproviders.51,63,66,67 Suchan emphasis may minimize the stigma associated with drug abuse treatment.Moreover, it may offer opportunities to develop effective ways to manage symptomsof depression and to develop positive relationship skills.
The complicated missing data patterns, which are likely a reflectionof the disorganized lives of the women in the study, presented some analyticchallenges. Data were incomplete for about one fifth of the women in the sample,who were more depressed and more likely to be HIV positive than women withcomplete data. However, analyses based on complete case data and imputed datayield similar results, suggesting that the predictors identified are robustregardless of follow-up pattern.
The results should be interpreted with some caution because of the samplesize. Substance use cessation is rare, and a larger sample could have resultedin a greater number of significant predictors. Similarly, the levels of significanceassociated with the predictors decreased when all the domains were analyzedtogether for substance use cessation and symptom remission models. Replicationof the study with a larger sample size is needed to discern the effect ofspecific predictors when multiple variables are controlled for.
The urinalysis measure in addition to self-report measures of drug useprovided more accurate information about drug use than would otherwise havebeen available, because the women who self-reported no drug use but had positiveurine screens could be counted as users. However, it was not possible to determineif these women were drug dependent or experienced any drug dependence symptoms,possibly biasing downward the estimates of symptom reduction and substancedependence.
With these limitations in mind, the ICLDUS provides a unique opportunityto study longitudinal changes in substance use behavior in a high-risk communitysample. The results indicate that outreach and treatment efforts may be mosteffective when they are coordinated with programs that address women’ssocioeconomic and mental health needs, or if multimodal programs are developed.In particular, the salience of trading sex for money or drugs argues for targetedoutreach to these drug users. Once engaged, positive outcomes are more likelyto be achieved when program objectives include identifying and treating women’sdepression, increasing women’s capacity for developing healthy relationships,and developing viable work strategies as alternatives to sex work.
Correspondence: Lynn A. Warner, PhD, Schoolof Social Work, Rutgers University, 536 George St, New Brunswick, NJ 08901(firstname.lastname@example.org).
Submitted for Publication: November 7, 2003;final revision received April 15, 2004; accepted April 21, 2004.
Funding/Support: This study was funded by grantDA A09438-05 from the National Institute on Drug Abuse, Bethesda, Md.
Previous Presentation: This study was presentedpreviously at the Latino Research Program Project Young Investigator’sConference; July 3, 2001; San Juan, Puerto Rico.
Acknowledgment: We thank Patrick Shrout, PhD,for assistance with analysis in an early version of the manuscript.