Abstract
Background
This study used survival analysis to examine risk factors for substance abuse relapse among residents in Oxford Houses (OH) a national network of self-run, self-financed aftercare homes for individuals recovering from substance use disorders.
Methods
Participants who entered OH within 60 days of a one-year longitudinal study (N=268) were selected from of a nationally representative U.S. sample. Discrete-time survival analysis compared baseline risk of relapse to four hypothesized survival models that included time-invariant and time-varying factors across three subsequent time periods.
Results
The model predicting higher risk for more severe substance use disorders and psychiatric problems was supported. The hypothesized model that predicted time-varying increases in alcohol (but not drug) abstinence self-efficacy significantly affected risk of relapse. Hypothesized demographic and employment variables did not significantly predict relapse risk.
Conclusions
Results suggested that OH recovery homes may reduce relapse by providing closer monitoring and referring additional services to new residents with more severe prior addiction severity. Risk for relapse may also be reduced by enhancing abstinence self-efficacy for alcohol regardless of drug of choice.
Keywords: Oxford House, substance abuse treatment, aftercare settings, relapse, survival analysis, abstinence self-efficacy
INTRODUCTION
Individuals with substance use disorders may improve their chances to maintain sobriety after treatment by living in aftercare settings that are substance-free and provide social support to prevent relapse 1,2. The current study is the first to use discreet-time survival analysis statistical techniques to examine the risk for alcohol and other drugs (AOD) relapse among a U.S. national sample of residents of Oxford Houses (OHs). OHs are an international network of over 1,700 aftercare settings sharing similar traits: all OHs are non-professional, self-run, self-financed, and require residents to remain abstinent from AOD with no maximum length of stay 3.
Factors affecting risk for AOD relapse after treatment include addiction severity, co-occurring psychiatric disorders, employment, and changes in abstinence self-efficacy. Higher Addiction Severity Index (ASI) scores for drug use and employment problems predicted relapse in 224 parolees in weekly drug counseling 4. Pedersen & Hesse 5 adapted the ASI to produce a Risk of Alcoholic Relapse Scale and found it to predict relapse at 6 months and three years after treatment based on addiction severity. Langdon et al.6 reviewed the characteristics of 128 clients in repeated AOD treatments in the U.K. and found that comorbid psychological disorders to be among the strongest predictors of relapse and re-admission. Xie et al 7 performed a ten-year prospective study of 223 individuals with co-occurring psychiatric disorders and substance use disorders. Xie et al. found men, participants with less than a high school education, and those who lived independently versus in a group setting had higher risk of relapse. In a study of 53 substance abuse treatment participants, Kushner et al. 8 found that clients with comorbid AOD problems and anxiety (55% of their sample) relapsed more frequently in a 4-month post-treatment follow-up than those individuals assessed with AOD problems alone. Low rates of employment have also been associated with AOD relapse 9,10. Gaining employment can increase self-esteem and self-efficacy, autonomy, and better functioning and increased skills in daily living 11, and stable employment can reduce relapse risk 12. Finally, Brown 13 found significantly higher levels of self-efficacy at three months for a group of abstainers compared to a relapse group, and abstainers had increased their self-efficacy scores from baseline to three months, while the relapse group’s self-efficacy scores decreased. Brown concluded that effective coping resources – self-efficacy and good social support – moderated the effects of significant life adversity that lead to lower relapse rates. Davis and Jason 14 found that longer length of stays in an OH increased abstinence self-efficacy for 87 men and women OH residents.
The current study proposed four hypotheses affecting hazard ratios compared to baseline. 1) Participants who were younger, un-married, and with fewer years of education, would have significantly higher hazard ratios than older, married, and more educated participants. 2) Study participants who had shorter lengths of sobriety, more previous treatments, and higher ASI composite scores for alcohol problems, drug problems, and psychiatric problems, would have higher hazard ratios. 3) Participants with ASI scores indicating more severe problems with employment, lower initial income upon entry into OH, and fewer days employed, would predict significantly higher hazard ratios. 4) Participants whose self-efficacy scores increased over time will have lower hazard ratios. The present study the first study to use survival analysis techniques to examine the relationship between demographic and psychological variables to predict relapse among OH residents.
METHODS
Research Participants
Data were taken from an archival data set of a study conducted by DePaul University researchers in 2000-2003 of 897 men and women participants already living in 170 OHs in a U.S. national sample. Data was gathered at baseline and at three subsequent periods at day 120, 240, and at the end of one year. The current study selected new residents who had entered OH within ±60 days of the beginning of the original study, replicating a similar selection technique performed on other OH studies 15,16. This resulted in 268 cases at baseline consisting of 162 men and 106 women.
Measures
The Addiction Severity Index-Lite ASI-Lite: 17 assesses problem areas commonly related to AOD problems. This study used ASI data taken at baseline for age, marital status, education; alcohol problem severity composite scores and number of alcohol treatments; drug problem severity composite score and number of drug treatments; employment composite scores, days employed in the last 30 days, and income from employment in the last 30 days; and psychiatric problem severity composite scores. ASI composite scores ranged from 0 to 1.
Alcohol and Drug Abstinence Self-efficacy
At baseline and at day 120, 240, and at the end of one year, participants were administered the 20-item Alcohol Abstinence Self-Efficacy scale (AASE) 18 and the Drug Abstinence Self-Efficacy scale (DASE). The AASE/DASE asked respondents to indicate how confident (from 0% to 100%) participants were in resisting alcohol/drug use in 20 high-risk situations using a 5-point Likert scale. The DASE was modified such that the words “drink alcohol” were replaced by the words “use drugs.”
Relapse Measure
Miller and Del Boca’s 19 Form 90 Timeline Follow Back (modified to record 120 days rather than 90 days) to measure past alcohol and drug use was administered at baseline and at three subsequent periods at day 120, 240, and at the end of one year. A triggering event for relapse occurred if the participant answered positively to any use of drugs or alcohol.
Procedure
The research project was conducted under the supervision of DePaul University’s LRB and IRB Review and received approval as a secondary data analysis project. The DePaul IRB classified this project as non-reviewable on November 14, 2013. DePaul University IRB granted permission for this study as a non-renewable exempt study on the condition that all data could not be traceable to the original participants.
We constructed baseline survival curves and hazard ratios using the lifetables technique using the statistical package SPSS version 21. We then entered hypothesized variables into the Cox proportional hazard regression models that compared the resulting hazard ratios against the baseline hazard ratio. If the hazard ratio with covariates differed significantly from the baseline ratio, the null hypothesis was rejected. We also used Cox proportional hazard regression models entering the time-varying AASE and DASE values as covariates at baseline and across each discreet data-gathering periods at day 120, 240, and at the end of one year. If the model using time-varying covariates affected the hazard ratio compared to the baseline ratio for that discreet time period, the null hypothesis was rejected.
RESULTS
Table 1 shows the demographic information of the 268 participants, means of variables, and comparisons with noteworthy p-values based on gender. By the end of the study, 87 participants reported no use, 75 had relapsed, and 102 had dropped out of the study. The mean survival time was 308 days (95% CI 296.4-322.6). Comparisons of all the demographic variables between the no use, relapsed, and dropped groups were not significant.
Table 1. Participant Socio-demographic and ASI Variable Comparisons at Baseline.
| Sample (N=268) |
Male (n=162; 60.1%) |
Female (n=106; 39.9%) |
|||
|---|---|---|---|---|---|
| M (SD) | M (SD) | M (SD) | p | η 2 | |
| Age | 35.17 (8.94) | 36.01 (9.00) | 33.91 (8.75) | .06 | |
| Education (months) | 149.44 (26.57) | 150.96 (24.54) | 147.14 (29.34) | ||
| Employment | |||||
| Days paid work last 30 days |
11.67 (11.40) | 12.11 (11.12) | 10.96 (11.86) | ||
| Employment income last 30 days Length of sobriety (days) |
$479.45 ($754.43) |
$608.86 ($874.73) |
$280.26 ($453.85) |
** | .0 5 |
| Alcohol | 298.03 (661.67) | 345.90 (741.36) | 225.99 (514.24) |
||
| Drugs | 345.31 (951.22) |
406.42 (1105.57) |
253.35 (648.04) |
||
| Number of previous treatments (lifetime) |
|||||
| Alcohol | 2.98 (4.59) | 3.59 (5.49) | 2.03 (2.36) | ** | .0 3 |
| Drugs | 2.91 (3.37) | 3.02 (3.89) | 2.74 (2.42) | ||
| Prior criminal charges | 4.90 (7.20) | 4.99 (7.43) | 4.75 (6.86) | ||
| ASI Composite Scores (.00- 1.00; lower is better) |
|||||
| Medical | .25 (.26) | .24 (.28) | .27 (.24) | ||
| Alcohol | .16 (.11) | .16 (.11) | .15 (.11) | ||
| Drug | .08 (.05) | .07 (.04) | .08 (.05) | ||
| Legal | .14 (19) | .13 (.18) | .16 (.19) | ||
| Family/Social | .30 (.16) | .27 (.16) | .34 (.16) | ** | .0 6 |
| Psychiatric | .19 (.23) | .18 (.22) | .22 (.23) | ||
| Employment (reverse scored) |
.61 (.30) | .62 (.30) | .61 (.31) |
| % (n) | % (n) | % (n) | P | ||
|---|---|---|---|---|---|
| Race/Ethnicity | |||||
| White/Caucasian | 66.4 (178) | 68.3 (110) | 63.6 (68) | ||
| Black/African American | 26.1 (70) | 23.0 (37) | 30.8 (33) | ||
| Hispanic/Latino | 4.9 (13) | 6.2 (10) | 2.8 (3) | ||
| Other | 2.6 (7) | 2.5 (4) | 2.8 (3) | ||
| Marital status | |||||
| Never married | 49.6 (133) | 49.7 (80) | 49.5 (53) | ||
| Divorced//widowed | 29.4 (78) | 31.7 (51) | 25.2 (27) | ||
| Separated | 16.0 (43) | 13.0 (21) | 20.6 (22) | ||
| Married | 4.1 (11) | 3.7 (6) | 4.7 (5) |
p≤ 0.05
p≤ 0.01
Table 2 displays the hazard ratios, confidence intervals, and p-values for each of the hypothesized variables entered into the Cox proportional hazards regression models. Cox proportional hazards regression analyses indicated that demographic (age, level of education, and marital status), and employment factors (ASI employment composites, days of paid work in the last 30 days, and income from work in the last 30 days) were not associated with significant effects on hazard ratios compared to baseline ratios.
Table 2. Covariate Analyses of Individual Variables using Cox Proportional Hazards Regression.
| Covariate | B | SE | Wald χ2 | p | Hazard Ratio (95% CI) |
|---|---|---|---|---|---|
| Age | −.001 | .013 | .011 | .92 | .99 (.97-1.03) |
| Marital Status | |||||
| Never Married (n=133) | − | − | 4.18 | .243 | − |
| Married (n=11) | −.477 | .258 | 3.41 | .065 | .621 (.37-1.03) |
| Separated (n=43) | .250 | .609 | .168 | .682 | 1.28 (.39-4.24) |
| Divorced/Widowed (n=78) |
−.283 | .358 | .626 | .43 | .75 (.37-1.52) |
| Level of Education | .002 | .004 | .263 | .61 | 1.00 (.99-1.01) |
| # Previous Treatments for Alcohol |
.065 | .031 | 4.50* | .034 | 1.07 (1.00-1.13) |
| # Previous Treatments for Drugs |
.081 | .028 | 8.40** | .004 | 1.08 (1.03-1.15) |
| ASI Alcohol Composite | 2.11 | .99 | 4.58* | .033 | 8.23 (1.18-57.24) |
| ASI Drug Composite | 3.31 | 2.91 | 1.29 | .256 | 27.42 (.091- 8306.06) |
| ASI Psychiatric Problems Composite |
1.03 | .50 | 4.28* | .039 | 2.80 (1.06-7.42) |
| Days since last Used | −.001 | .001 | 3.08 | .079 | .99 (.998-1.00) |
| ASI Employment Composite |
−.385 | .375 | 1.05 | .305 | .68 (.326-1.42) |
| Days employed last 30 days | .010 | .010 | .877 | .349 | 1.01 (.99-1.03) |
| Income from employment last 30 days |
.000 | .000 | .002 | .960 | 1.00 (1.00-1.00) |
| Alc. Abstinence Self- Efficacy |
−.012 | .004 | 9.37** | .002 | .988 (.980-.996) |
| Drug Abstinence Self- Efficacy |
.001 | .004 | .120 | .729 | 1.001 (.994- 1.009) |
p≤ 0.05
p≤ 0.01
Cox proportional hazards regression indicated the number of previous treatments for alcohol was significant, such that one additional alcohol treatment episode was associated with a 6.7% increase in the hazard ratio. The number of previous treatments for drugs was also significant, such that one additional drug treatment episode was associated with an 8.4% increase in the hazard ratio. ASI alcohol composite scores were also significant; for each .10 increase in the ASI alcohol composite was associated with an increase in the hazard ratio by 82%. ASI psychiatric severity was also significant, such that each .10 increase in the ASI psychiatric composite was associated with an increase in the hazard ratio by 28%. However, days since last substance used and the ASI drug severity composite scores were not significantly associated with hazard ratios.
Cox proportional hazards regression using time-varying AASE and DASE self-efficacy composites at each 120-day time point indicated significant differences compared to the baseline hazard ratios, Wald χ2 (2) = 27.75, p<.001. Individually, the effect of alcohol abstinence self-efficacy (AASE) was significant, such that each additional unit in alcohol self-efficacy at each 120-day time point was associated with a decrease the hazard ratios by 1.2%. However, drug abstinence self-efficacy (DASE) was not significantly associated with hazard ratios, B=.001, Wald χ2 (1) = .120, p=.729, Exp(B)=1.001.
DISCUSSION
The goal of this study was to determine what factors were associated with risk for relapse among residents living in OH over a period of one year. Results indicated that higher ASI scores for alcohol problems, psychiatric severity, and the greater number of prior alcohol or drug treatment episodes prior to enter OH predicted higher risks for relapse. The period of greatest hazard is the first 90 days of the study, with 18% (31 participants) relapsed during this interval. Hazard reduced slightly between days 180 and 270 to 16%. This supports previous OH research that found stays of six months or more promoted long-term recovery 20,21.
It was surprising only increases in alcohol abstinence self-efficacy significantly affected hazard. These findings suggested a possible lack of instrument sensitivity in measuring drug versus alcohol self-efficacy, and that conflating situational triggers for using drugs and alcohol might be misguided. El-Sheikh and Bashir found differences in abstinence self-efficacy based on substances and situations between former alcohol and heroin groups using two different measures 22. Similarly, Sklar and colleagues 23 compared self-efficacy scores of 344 alcohol and 253 cocaine clients in treatment and found that the alcohol group had lower self-efficacy scores in interpersonal conflict situations, whereas the cocaine group had lower self-efficacy scores in temptation-related situations.
This study has potential limitations. It is unknown what effects the ±60-day selection window had on relapse; the rationale for this selection allowed for a substantial increase in data while avoiding exposing or depriving participants 1/3 or less of the six-month dosage found to be effective in prior OH studies 20,24. Another limitation of this study is the paucity of data gathered at each time period that were expected to vary across time, such as employment and income, which could have had significant effects on relapse. Finally, there were no distinctions among this sample to designate primary AOD of choice. As only increases in alcohol abstinence self-efficacy significantly reduced hazard ratios in this study, future research should distinguish primary drug of choice to explore potential alcohol and drug self-efficacy interactions.
Conclusion
There are compelling benefits for examining the phenomenon of relapse that occur over time in specific settings 25. Kushner et al. 8 concluded that pre-screening for comorbid disorders such as anxiety and depression should be used as a marker for potential relapse relative to singly diagnosed clients. Knowledge of relapse factors of previous participants could inform intake and screening procedures and could help determine whether a setting is appropriate for the individual 26,27. In addition to matching clients to settings, analysis of the patterns and timing of relapse can alert staff or co-residents to emphasize certain aspects of treatment to individuals, or offer additional resources for vulnerable populations at specific phases of treatment 28,29. In short, awareness of client factors related to relapse in OH could advance treatment matching and person-environment fit to improve outcomes 30. Future research on OH using survival analysis techniques and relapse could reveal such relationships and should be encouraged.
Acknowledgments
FUNDING
The authors appreciate the financial support from the National Institutes of Health/National Institute on Drug Abuse grants DA13231 and DA19935 (PI: Dr. Jason). Grant DA13231 funded the collection, management, and analysis of the longitudinal study that produced these data for this paper. Grant DA19935 funded the salary of a graduate student/project director (Ronald Harvey) on an unrelated research project during which this data was analyzed.
Footnotes
AUTHOR CONTRIBUTIONS
Dr. Harvey had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. Dr. Jason is the first author’s advisor, dissertation committee chair, and principal investigator (PI) of the grant which generated this research data. Dr. Ferrari was on the first author’s dissertation committee and co-PI of the grant which generated this research data.
The authors declare that they have no conflicts of interest.
REFERENCES
- 1.Jason LA, Olson BD, Ferrari JR, Davis MI. Substance abuse: The need for second-order change. Int J Self Help Self Care. 2003;2:91–109. [Google Scholar]
- 2.Moos RH, Nichol AC, Moss BS. Risk factors for symptom exacerbation among treated patients with substance use disorders. Addiction. 2002;97(1):75–85. doi: 10.1046/j.1360-0443.2002.00063.x. [DOI] [PubMed] [Google Scholar]
- 3.Oxford House Inc. Oxford House Manual: An Idea Based On a Sound System For Recovering Alcoholics and Drug Addicts to Help Themselves. Oxford House World Services, Inc; Silver Spring, MA: 2011. [Google Scholar]
- 4.Hanlon TE, O’Grady KE, Bateman RW. Using the Addiction Severity Index to Predict Treatment Outcome Among Substance Abusing Parolees. J Offender Rehabil. 2000;31:67–79. [Google Scholar]
- 5.Pedersen MU, Hesse M. A simple risk scoring system for prediction of relapse after inpatient alcohol treatment. Am J Addict. 2009;18(6):488–493. doi: 10.3109/10550490903205983. [DOI] [PubMed] [Google Scholar]
- 6.Langdon PE, Yágüez L, Brown J, Hope A. Who walks through the ‘revolving-door’ of a British psychiatric hospital? J Ment Health. 2001;10(5):525–533. [Google Scholar]
- 7.Xie H, McHugo GJ, Fox MB, Drake RE. Substance abuse relapse in a ten-year prospective follow-up of clients with mental and substance use disorders. Psychiatr Serv. 2005;56:1282–1287. doi: 10.1176/appi.ps.56.10.1282. [DOI] [PubMed] [Google Scholar]
- 8.Kushner MG, Abrams K, Thuras P, Hanson KL, Brekke M, Sletten S. Follow-up Study of Anxiety Disorder and Alcohol Dependence in Comorbid Alcoholism Treatment Patients. Alcohol Clin Exp Res. 2005;29(8):1432–1443. doi: 10.1097/01.alc.0000175072.17623.f8. [DOI] [PubMed] [Google Scholar]
- 9.Baldwin ML, Marcus SC, De Simone J. Job loss discrimination and former substance use disorders. Drug Alcohol Depend. 2010;110:1–7. doi: 10.1016/j.drugalcdep.2010.01.018. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Slaymaker VJ, Owen PL. Employed men and women substance abusers: job troubles and treatment outcomes. J Subst Abuse Treat. 2006;31:347–354. doi: 10.1016/j.jsat.2006.05.008. [DOI] [PubMed] [Google Scholar]
- 11.Reif S, Horgan CM, Ritter GA, Tompkins CP. The Impact of Employment Counseling on Substance User Treatment Participation and Outcomes. Subst Use Misuse. 2004;39(13-14):2391–2424. doi: 10.1081/ja-200034661. [DOI] [PubMed] [Google Scholar]
- 12.Rabinowitz J, Mark M, Popper M, Slyuzberg M. Predicting revolving-door patients in a 9-year national sample. Soc Psychiatry Psychiatr Epidemiol. 1995;30(2):65–72. doi: 10.1007/BF00794944. [DOI] [PubMed] [Google Scholar]
- 13.Brown SA, Vik PW, Patterson TL, Grant I, Schuckit Ma. Stress, vulnerability and adult alcohol relapse. J Stud Alcohol. 1995;56:538–545. doi: 10.15288/jsa.1995.56.538. [DOI] [PubMed] [Google Scholar]
- 14.Davis MI, Jason LA. Sex Differences in Social Support and Self-Efficacy Within a Recovery Community. Am J Community Psychol. 2005;36(3):259–274. doi: 10.1007/s10464-005-8625-z. [DOI] [PubMed] [Google Scholar]
- 15.Aase DM, Jason LA, Ferrari JR, et al. Anxiety symptoms and alcohol use: a longitudinal analysis of length of time in mutual help recovery homes. Int J Self Help Self Care. 2005;4(1):21–35. [Google Scholar]
- 16.Aase DM, Jason LA, Olson BD, et al. A Longitudinal Analysis of Criminal and Aggressive Behaviors Among a National Sample of Adults in Mutual-Help Recovery Homes. J Groups Addict Recover. 2009;4(1/2):82. doi: 10.1080/15560350802712454. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.McLellan AT, Kushner H, Metzger D, Peters R. The fifth edition of the Addiction Severity Index. J Subst Abuse Treat. 1992;9(3):199–213. doi: 10.1016/0740-5472(92)90062-s. [DOI] [PubMed] [Google Scholar]
- 18.DiClemente CC, Carbonari JP, Montgomery RPG, Hughes SO. The Alcohol Abstinence Self-Efficacy scale. J Stud Alcohol. 1994;55(2):141–148. doi: 10.15288/jsa.1994.55.141. [DOI] [PubMed] [Google Scholar]
- 19.Miller WR, Del Boca FK. Measurement of drinking behavior using the Form 90 family of instruments. J Stud Alcohol. 1994;(12):112–118. doi: 10.15288/jsas.1994.s12.112. [DOI] [PubMed] [Google Scholar]
- 20.Jason LA, Davis MI, Ferrari JR. The need for substance abuse after-care: A longitudinal analysis of Oxford House. Addict Behav. 2007;32:803–818. doi: 10.1016/j.addbeh.2006.06.014. [DOI] [PubMed] [Google Scholar]
- 21.McKay JR. Is there a case for extended interventions for alcohol and drug use disorders? Addiction. 2005;100:1594–1610. doi: 10.1111/j.1360-0443.2005.01208.x. [DOI] [PubMed] [Google Scholar]
- 22.El-Sheikh SE-G, Bashir TZ. High-risk relapse situations and self-efficacy: comparison between alcoholics and heroin addicts. Addict Behav. 2004;29:753–758. doi: 10.1016/j.addbeh.2004.02.003. [DOI] [PubMed] [Google Scholar]
- 23.Sklar SM, Annis HM, Turner NE. Group comparisons of coping self-efficacy between alcohol and cocaine abusers seeking treatment. Psychol Addict Behav. 1999;13:123–133. [Google Scholar]
- 24.Groh DR, Jason LA, Davis MI, Olson BD, Ferrari JR. Friends, Family, and Alcohol Abuse: An Examination of General and Alcohol-Specific Social Support. Am J Addict. 2007;16(1):49–55. doi: 10.1080/10550490601080084. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Moos RH. Understanding environments: The key to improving social processes and program outcomes. Am J Community Psychol. 1996;24(1):193–201. doi: 10.1007/BF02511887. [DOI] [PubMed] [Google Scholar]
- 26.Melnick G, De Leon G, Thomas G, Kressel D. A client-treatment matching protocol for therapeutic communities: First report. J Subst Abuse Treat. 2001;21(3):119–128. doi: 10.1016/s0740-5472(01)00191-x. [DOI] [PubMed] [Google Scholar]
- 27.Walters GD. Lessons Learned From Project MATCH. Addict Disord Their Treat. 2002;1:135–139. [Google Scholar]
- 28.Ilgen M, Tiet Q, Finney J, Moos RH. Self-efficacy, therapeutic alliance, and alcohol-use disorder treatment outcomes. J Stud Alcohol. 2006;67(3):465–472. doi: 10.15288/jsa.2006.67.465. [DOI] [PubMed] [Google Scholar]
- 29.Moos RH. Addictive disorders in context: Principles and puzzles of effective treatment and recovery. Psychol Addict Behav. 2003;17(1):3–12. doi: 10.1037/0893-164x.17.1.3. [DOI] [PubMed] [Google Scholar]
- 30.Babor TF, Del Boca FK, editors. Treatment matching in alcoholism. Cambridge University Press; New York, NY US: 2003. Treatment matching in alcoholism. [Google Scholar]