Journal of Substance Abuse Treatment 28 (2005) S73–S82

Article

Predictors of alcohol and crack cocaine use outcomes over a 3-year follow-up in treatment seekers James R. McKay, (Ph.D.)a,b,*, Carol Foltz, (Ph.D.)b, Richard C. Stephens, (Ph.D.)c, Peter J. Leahy, (Ph.D.)c, Evelyn M. Crowley, (Ph.D.)b, Wendy Kissin, (Ph.D.)d a

Department of Psychiatry, University of Pennsylvania, Philadelphia, PA 19104, USA b DeltaMetrics, Philadelphia, PA 19106, USA c University of Akron, Institute for Health and Social Policy, Akron, OH 44325, USA d WESTAT, Rockville, MD 20850, USA

Received 18 February 2004; received in revised form 24 September 2004; accepted 28 October 2004

Abstract This study identified predictors of long-term alcohol and crack cocaine use outcomes in individuals participating in the Persistent Effects of Treatment Study. The domains that were assessed included motivation, self-efficacy, social support, psychiatric severity, employment, housing status, and self-help group attendance at baseline and 6, 12, 24, and 30 month follow-ups. In alcohol users, higher perceived seriousness of substance use problems, self-efficacy, and self-help group attendance, as well as lower social support for substance use, consistently predicted better alcohol use outcomes in the subsequent assessment period. In crack cocaine users, only self-efficacy consistently predicted cocaine use outcomes. Higher self-efficacy during follow-up was predicted by lower perceived seriousness of substance use and lower alcohol use frequency in the prior assessment period, whereas greater self-help group attendance was predicted by greater perceived seriousness of substance use, and lower substance use frequency. D 2005 Elsevier Inc. All rights reserved. Keywords: PETS; Long-term treatment outcome; Alcohol; Crack cocaine; Post-treatment predictors

1. Introduction At this point, there is considerable evidence that alcohol and drug use disorders are chronic problems—at least for many individuals who enter the treatment system. The continued vulnerability to relapse exhibited by many treatment seeking substance abusers appears to be a function of poor compliance with treatment and other behavioral change regimens, as well as a number of biological, psychological, behavioral, interpersonal, and environmental factors (Lehman & Simpson, 1990; Moos, Finney, & Cronkite, 1990). Poor performance while in treatment, as indicated by lack of

* Corresponding author. University of Pennsylvania, Treatment Research Center, 3900 Chestnut St., Philadelphia, PA 19104, USA. Tel.: +1 215 746 7704; fax: +1 215 746 7733. E-mail address: mckay_ [email protected] (J.R. McKay). 0740-5472/05/$ – see front matter D 2005 Elsevier Inc. All rights reserved. doi:10.1016/j.jsat.2004.10.010

compliance, weak therapeutic alliance, and failure to complete treatment, is associated with worse drinking and drug use outcomes (Connors, Carroll, DiClemente, Longabaugh, & Donovan, 1997; McKay & Weiss, 2001), as is lack of participation in self-help group programs posttreatment (Crits-Christoph et al., 2003; McKay, Merikle, Mulvaney, Weiss, & Koppenhaver, 2001; Tonigan, Toscova, & Miller, 1996; Vaillant, 1996). Co-morbid psychiatric disorders, such as major depression and antisocial personality disorder, have been linked to poorer outcomes in some, but not all, studies (Kranzler, Del Boca, & Rounsaville, 1996; McLellan, Luborsky, Woody, Druley, & O’Brien, 1983; Project MATCH Research Group, 1997). Negative mood states have been associated with poor drinking outcomes (Meyers, 2001), and low positive mood has predicted poorer cocaine outcomes (Hall, Havassy, & Wasserman, 1991; McKay, Merikle, et al., 2001). With regard to other psychological and behavioral factors, low

S74

J.R. McKay et al. / Journal of Substance Abuse Treatment 28 (2005) S73–S82

readiness to change, lack of commitment to abstinence, low self-efficacy, poor coping behaviors, and greater craving have consistently been associated with poor drinking and drug use outcomes (Connors, Maisto, & Zywiak, 1996; Hall, Havassy, & Wasserman, 1990; Hall et al., 1991; McKay, Merikle, et al., 2001; Miller, Westerberg, Harris, & Tonigan, 1996; Moos et al., 1990; Morgenstern, Frey, McCrady, Labouvie, & Neighbors, 1996; Morgenstern, Labouvie, McCrady, Kahler, & Frey, 1997; Project MATCH Research Group, 1998). Interpersonal problems and social support for continued substance use have also been associated with worse drinking and drug use outcomes (Havassy, Wasserman, & Hall, 1995; Longabaugh, Wirtz, Beattie, Noel, & Stout, 1995; Moos et al., 1990), as has life stress (McKay & Weiss, 2001; Miller et al., 1996; Moos et al., 1990). Not surprisingly, post-treatment scores on these measures are often more strongly associated with subsequent substance use than pretreatment/ baseline values. Although relationships between various during- and post-treatment factors and subsequent substance use outcomes have been examined in numerous studies, relatively few studies have looked at a full range of post-treatment predictors and substance use outcomes over repeated follow-ups. In a prior attempt to provide a comprehensive picture of the factors that predict outcomes over time, we followed a group of 132 cocaine dependent male VA patients who had completed a 1-month intensive outpatient program for 2 years, with assessments at 6, 12, 18, and 24 months (McKay, Merikle, et al., 2001). Univariate analyses indicated that commitment to abstinence, self-efficacy, support from family, positive mood, employment, self-help group participation, and attendance in continuing care were all associated with less cocaine use during the subsequent 6-month follow-up period. Many of these relationships held over most or all of the 2-year follow-up. When all predictors were included in multivariate analyses, however, only selfhelp group participation consistently made a unique and significant contribution to the prediction of cocaine use in the subsequent 6-month follow-up period. The present study seeks to extend our prior work (McKay, Merikle, et al., 2001) by making use of data collected as part of the Persistent Effects of Treatment Studies (PETS). One of the performance sites for this Center for Substance Abuse Treatment (CSAT) funded initiative was Cuyahoga County in Ohio, which includes the city of Cleveland. The sample consisted of treatment seeking substance abusers who completed a comprehensive baseline assessment at study intake and were re-assessed at 6, 12, 24, 30, and 36 months post baseline. The sample included both men and women, and the majority of participants were treated in either inpatient, intensive outpatient, or standard outpatient care following intake into the study. The participants reported significant levels of use of a range of substances at baseline, including alcohol, crack-cocaine, and other drugs. Therefore, the present study included a larger and more diverse sample of substance

abusers, followed for a longer period of time, compared to our prior study. Using scales and items from the assessment battery in the study, we were able to measure most of the constructs that had been examined in our prior study (McKay, Merikle, et al., 2001) and in other studies cited here (Connors et al., 1996; Hall et al., 1991; Miller et al., 1996; Moos et al., 1990), including motivation, self-efficacy, social support, psychiatric severity, employment, and self-help group group attendance. Data were also available on the quality of the participantsT living situations. Multivariate analyses were performed to determine the independent contribution of each variable to predicting alcohol and crackcocaine outcomes in the subsequent follow-up period, after baseline demographic and substance use variables were controlled. Additional analyses were also planned to explore the predictors and correlates of any factors that emerged as consistent predictors of either alcohol or crack-cocaine use.

2. Materials and methods 2.1. Participants Participants were individuals seeking treatment for substance abuse from adult treatment sites in Cuyahoga County, Ohio, with data initially collected as part of the Target Cities Program funded by CSAT. 2.2. Measures Data were collected at intake and follow-up by means of the Computer Assisted Central Intake Assessment Instrument (CIAI-C, Version 5.1). This comprehensive, multidimensional structured assessment interview was developed by multiple contributors including the Office of Treatment Improvement within the Alcohol, Drug Abuse, and Mental Health Administration, CSAT, the Department of Sociology of the University of Akron, and the Alcohol and Drug Addiction Services Board of Cuyahoga County, Ohio. The CIAI-C assesses lifetime and current problem severity levels in the same domains that are included in the Addiction Severity Index (ASI; McLellan, Luborsky, Woody, & O’Brien, 1980; McLellan et al., 1985), and most of the ASI items needed to generate the ASI composite scores were embedded in the instrument. The CIAI-C also assesses in detail several areas that are not or only lightly covered by the ASI, including treatment services received, self-help group attendance, cognitions and beliefs concerning addiction (e.g., self-efficacy, beliefs about treatment, etc.), housing situation, and substance use patterns of family members and friends. Although comprehensive psychometric analyses of the CIAI-C have not yet been undertaken, there is reason to believe that it possesses adequate reliability and validity, given that the basic structure and many of the key items are similar to those in the

J.R. McKay et al. / Journal of Substance Abuse Treatment 28 (2005) S73–S82

ASI. The following indicators and scales were derived from the CIAI-C: Quality of Living Situation. This indicator assesses whether the participant is in a high risk living situation (i.e., either in a homeless shelter/on the street/jail or with a number of substance abusers, vs. adequate living situation and few or no substance abusers). The indicator is scored such that 1 = good/adequate situation, 0 = high risk situation. General Social Support. This continuous measure is a count of the number of relatives and friends whom the participant enjoys getting together with and who provide support when problems arise. Higher scores indicate more support (coefficient alpha = .79). Substance Abuse Specific Social Support. This continuous measure is a count of the number of relatives and friends whom the participant enjoys getting together with and who would provide support when problems arise, but who also have alcohol or drug problems. The measure is reverse scored so that higher scores indicate less social support for substance use (coefficient alpha = .76). Self-help Group Attendance. This continuous measure assesses the number of self-help group meetings attended in the prior 6 months. Perceived Seriousness of Substance Use Problem. This continuous measure assesses the participantTs perceptions of how serious his/her substance abuse problem is and openness to making changes to address the problem. Higher scores indicate greater perceived seriousness and desire to change (coefficient alpha = .84). Self-efficacy. This continuous measure assesses the participantTs degree of confidence that he/she can cope with a particular stressor without using alcohol or drugs. Higher scores indicate greater confidence (coefficient alpha = .92). Psychiatric Severity. This continuous measure assesses perceived severity of psychiatric symptoms, and is a proxy for the ASI psychiatric composite score. Higher scores indicate greater current psychiatric symptom severity (coefficient alpha = .71). Employment Status. A dichotomous measure was used to indicate current employment status (1 = currently employed full or regular part time; 0 = not currently employed). Alcohol and Cocaine Use. The CIAI-C includes several continuous measures of alcohol and drug use, including average frequency of use over the prior 6 months, days of use in the prior 30 days, and average number of drinks or drug uses per day of use. We elected to use a frequency of use measure because it is the most common outcome measure in addictions treatment outcome research (Babor et al., 1994; Carroll, 1995). The frequency of use measure with the best psychometric properties was average frequency of use over the prior 6 months (coded 1 = never/ none; 2 = one time; 3 = less than once per week; 4 = about once per week; 5 = 2 to 6 times per week; 6 = about once per day; 7 = 2 to 3 times almost every day; 8 = 4 or more times a day almost every day).

S75

Corroboration of Self-Reports of Cocaine use. Urine samples were obtained from approximately one third of study participants at the 24, 30, and 36 month follow-ups and tested for the cocaine metabolite benzoylecgonine. The percentage of participants who reported no cocaine use but had a positive urine test was 26% at 12 months (48 of 188), 23% at 24 months (49 of 216), and 23% at 36 months (44 of 194). Therefore, among participants from whom a urine sample was obtained, approximately three quarters of the self-reports of no cocaine use at each follow-up were supported by a negative urine test. 2.3. Procedures Study enrollment occurred at two Central Intake Units (CIU) and 17 Service Delivery Units (SDU) between October 1996 and August 1998. Participants entering the Cleveland system through the CIU were enrolled by CIU staff, who also conducted the baseline interviews. Clients presenting directly to any one of 17 SDUs were eligible to participate in the study if they had not visited that SDU during the previous 2 years, a restriction added to increase comparability with CIU-recruited participants. The 17 participating SDUs were the largest ones in Cuyahoga County system, treating the vast majority of clients in the system and representing all major treatment modalities. Reports from the SDUs suggest that failure to contact clients was the major reason for nonparticipation in the study. Informed consent was obtained for all participants. Most of the participants in the study began treatment within 2 weeks of their baseline assessment, although some entered later or failed to enter. Extensive efforts were made by research technicians, who were independent of the CIU staff, to gather data from the full sample at each follow-up wave. Follow up rates for completed interviews were 72.1% at 6 months, 73.5% at 12 months, and 75.0% at 24 and 30 months, and 80.1% at 36 months. 2.4. Statistical analyses Two parallel sets of analyses were conducted. The first set included all participants who reported alcohol use at baseline; in these analyses, the dependent variable was frequency of alcohol use over the past 6 months at each follow-up point. The second set included all participants who reported crack cocaine use at baseline; in these analyses, the dependent variable was frequency of crack cocaine use over the past 6 months. Multiple regressions were used to evaluate the unique contribution of each independent variable to the prediction of average alcohol or cocaine use frequency in subsequent periods. Separate regressions were done with the baseline, 6, 12, 24, and 30-month variables to predict alcohol and cocaine use in the subsequent follow-up period. The Model 1 regressions included six baseline control variables (e.g., baseline value of the outcome measure, years of alcohol or cocaine use,

S76

J.R. McKay et al. / Journal of Substance Abuse Treatment 28 (2005) S73–S82

approach is that one analysis can incorporate data from each point in the follow-up, thereby integrating information contained in the separate regressions, and increasing the sensitivity of the analyses. However, the limitation of this approach is that only a few variables can be examined in any one analysis. Therefore, these analyses complement the separate, lagged regressions described above. Second, lagged correlations between possible explanatory variables assessed prior to the predictor and the predictor variable itself were performed to identify correlates of the significant predictors. For example, if self-help group attendance proved to consistently predict alcohol or cocaine outcomes, then the analyses would focus on identifying other variables that predicted self-help group attendance. For all regressions, analyses were conducted to determine if multicollinearity problems were present. According to accepted conventions, the criteria for multicollinearity are a conditioning index N 30 and at least two variance proportions N .50 for each root number (Belsley, Kuh, & Welsch, 1980; Tabachnick & Fidell, 2001). These criteria were not met in any of the regression analyses. The only variable that came at all close to meeting these criteria was perceived seriousness of substance use problems, but dropping this variable had essentially no effect on the models. We also examined the correlations between all predictors at baseline and then again at 24 months. At both time points, fewer than 10% of the variables were correlated z .20, and 1% were correlated N .50.

years education, gender, marital status, and race), and the eight main predictor variables (e.g., general social support, substance use specific social support, quality of living situation, perceived seriousness of alcohol or drug problem, self-efficacy, self-help group group attendance, psychiatric severity, and employment status). The purpose of this first set of models was to determine the unique contribution of the main predictor variables, after baseline factors were controlled. The Model 2 regressions included all the variables from the Model 1 regressions, along with current alcohol or cocaine use (i.e., days of alcohol or cocaine use in the prior 30 days) at the same point at which the eight main predictors were assessed. In describing the results, we use the month at which the predictor variables were assessed to indicate the part of the follow-up under consideration. For example, analyses that linked predictors at 12 months to substance use at 24 months are referred to as b12 month analyses.Q The magnitude and significance level of each predictor was indicated by standardized beta coefficients and p values. Additional analyses were also planned for predictor variables that demonstrated consistent significant relations to subsequent alcohol or cocaine use. First, true longitudinal (i.e., mixed effect, repeated measures) analyses, in which a limited number of predictors were entered as time-varying covariates, were conducted to better understand the dynamic relation between the predictors and substance use over time at the level of individual participants. The strength of this

Table 1 Prediction of alcohol use outcomes in baseline drinkers Average frequency of alcohol use in months 1–6 Control Variables Gender (1 = male) Education (1 z HS) Race (1 = minority) Marital status (1 = married) Years using alcohol Baseline value of outcome Predictors General social support Social support for use (reversed) Quality of living situation (1 = good) Perceived seriousness of AOD Self-efficacy Self-help group attendance Psychiatric severity Employment status (1 = working) N F R2

.04 .02 .02 .02 .09 .19*** .07 .10* .11* .06 .11* .09 .02 .03 510 3.60*** .09

7–12 .07 .04 .02 .01 .05 .10* .05 .03 .06 .12** .26*** .22*** .05 .02 549 11.03*** .22

19–24

25–30

31–36

.04 .07 .01 .05 .06 .15***

.02 .03 .00 .03 .05 .14***

.04 .02 .01 .00 .08* .16***

.04 .09* .07 .15*** .28*** .15*** .01 .05 536 10.23*** .22

.04 .12** .04 .13** .36*** .16*** .04 .04 590 15.38*** .27

.06 .12** .02 .19*** .33*** .21*** .02 .11* 547 16.01*** .30

Note: Lagged analyses: baseline predictors/mo 1–6 alcohol use, 6 month predictors/mo. 7–12 alcohol use, 12 month predictors/mo 19–24 alcohol use, 24 month predictors/mo 25–30 alcohol use, 30 month predictors/mo 31–36 alcohol use. * p b .05. ** p b .01. *** p b .001.

J.R. McKay et al. / Journal of Substance Abuse Treatment 28 (2005) S73–S82

3. Results 3.1. Characteristics of study participants Descriptive data on the characteristics of study participants were based on the 810 individuals who were included in any of the analyses to predict alcohol use outcomes, and the 610 individuals who were included in any of the analyses to predict crack cocaine outcomes. The sample size varied, as only those with alcohol use at baseline were included in the alcohol analyses, and only those with cocaine use were included in the cocaine analyses. Moreover, participants had to have provided complete data at two adjacent follow-up points to be included in an analysis. The maximum numbers of participants in any of the lagged regressions to predict alcohol and cocaine use were 590 and 450, respectively. Of the participants in the alcohol analyses, 59% were male, 21% were white, 79% were minorities (primarily African American), 50% were high school graduates, 17% were currently married, and 24% were currently employed. They were, on average, 35.5 years old (SD = 8.4). They averaged 13.7 (SD = 11.1) days of alcohol use out of the prior 30, 20.6 (SD = 8.8) years of alcohol use, and 2.0 (SD = 3.7) prior treatments for substance abuse. Of the participants in the cocaine analyses, 53% were male, 13% were white, 87% were minorities (primarily African American), 49% were high school graduates, 15% were currently married, and 19% were currently employed. They were, on average, 35.5 years old (SD = 7.6). They averaged 12.9 (SD = 11.3) days of crack cocaine use out of the prior 30, 8.5 (SD = 5.0) years of crack use, and 2.5 (SD = 4.30) prior treatments for substance abuse. Participation in methadone maintenance at some point in the prior 3 years was reported at baseline by 9.3% of those in the alcohol analyses, and 6.5% of those in the cocaine analyses. 3.2. Prediction of alcohol use outcomes in alcohol users The results of the five lagged regressions to predict alcohol use are presented in Table 1. Of the six demographic/baseline control variables, only average frequency of alcohol use over the prior 6 months consistently predicted alcohol use outcomes. Four of the eight main predictor variables were consistently related to alcohol use in subsequent periods, as indicated by significant findings in the majority of the assessment points. These variables were social support for substance use, perceived severity of substance use problems, self-efficacy, and self-help group attendance. In each cases, higher scores on these measures predicted better alcohol use outcomes (with social support for substance use reverse scored). The standardized beta coefficients in Table 1 indicate that the magnitude of these relationships within each variable was generally consistent over time, at least from the 6 month assessment onward. For example, the coefficients for self-efficacy

S77

assessed at 6, 12, 24, and 30 months ranged from .26 to .36 (all ps b .001), and the coefficients for self-help group attendance ranged from .15 to .22 (all ps b .001). With the exception of the analysis with all baseline predictors, these regressions accounted for 22–30% of the variance in alcohol use during the follow-up. The second set of regression models to predict alcohol use included one additional variable, days of alcohol use in the prior 30 days at the point at which the other predictors were assessed. The purpose of these analyses, which are presented in Table 2, was to determine whether the other predictor variables would still account for alcohol use outcomes when current alcohol use was controlled. At each assessment point, current alcohol use predicted subsequent alcohol use, with coefficients ranging from .31 to .52. Perceived seriousness of substance use problems and selfefficacy still predicted subsequent alcohol use at each time point. Self-help group attendance at 6 and 30 months also predicted alcohol use in the next period. None of the other variables was still consistently predictive of alcohol use when current alcohol use was included in the model. These regressions accounted for 26–45% of the variance in alcohol use outcomes. Longitudinal mixed effect regressions were also done that included self-help group and self-efficacy at each assessment point as time varying covariates to predict alcohol use in the subsequent assessment point. Higher scores

Table 2 Prediction of alcohol use outcomes in baseline drinkers, controlling for current use Average frequency of alcohol use in months 7–12

19–24

25–30

31–36

Significant Control Variables Education .08* Years using alcohol .07* Current alcohol use .42*** .31*** .48*** .52*** Significant Predictors Quality of living .11** situation Perceived seriousness .10** .11* .08* .09* of AOD Self efficacy .14*** .17*** .18*** .09* Self help group .14*** .14*** attendance Psychiatric severity .07* .07* N 547 535 588 546 F 20.16*** 12.86*** 29.37*** 31.55*** R2 .35 .26 .42 .45 Note: Summary statistics at bottom of table are based on regression with all 15 predictor variables included. However, only significant predictors ( p b .05) are presented in the table. Lagged analyses: 6 month predictors/mo. 7–12 alcohol use, 12 month predictors/mo 19–24 alcohol use, 24 month predictors/mo 25–30 alcohol use, 30 month predictors/mo 31–36 alcohol use. * p b .05. ** p b .01. *** p b .001.

S78

J.R. McKay et al. / Journal of Substance Abuse Treatment 28 (2005) S73–S82

on both variables were significant predictors of alcohol use outcomes across the 3-year follow-up: self-help group [ F(1,1544) = 31.98, p b .0001], self-efficacy [ F(1,1544) = 47.28, p b .0001]. 3.3. Prediction of crack cocaine use outcomes in cocaine users The results of the five lagged regressions to predict crack cocaine use are presented in Table 3. None of the six demographic/baseline control variables were related to cocaine use outcomes in more than one of the five analyses. Of the eight primary predictor variables, only self-efficacy consistently predicted crack cocaine use in the subsequent follow-up period. The coefficients for self-efficacy at each assessment point ranged from .20 to .29 (all p b .001). Greater general social support and self-help group attendance, and being employed, were also generally associated with less cocaine use at subsequent follow-ups; however, only a few of these relations were significant. These regressions accounted for 10–18% of the variance in crack cocaine use outcomes. The results of the second set of regressions also included current cocaine use at each follow-up point, and are presented in Table 4. At each point, current cocaine use predicted subsequent cocaine use, with coefficients ranging from .28 to .36. Self-efficacy was still a significant predictor of subsequent cocaine use at three of the four follow-ups, with coefficients ranging from .14 to .19. These

regressions accounted for 17–26% of the variance in crack cocaine use outcomes. In longitudinal mixed effect regressions with timevarying covariates, higher self-efficacy scores predicted less crack cocaine use across the follow-up [ F(1,1222) = 58.46, p b .0001]. Higher self-help group scores also predicted less crack use, but at the level of a trend [ F(1, 1222) = 3.64, p b .06]. 3.4. Predictors of self-help group attendance and self-efficacy A series of lagged regressions were done to identify the predictors of self-help group attendance and self-efficacy. The predictors were the same 13 variables that were included in the Model 1 analyses (i.e., presented in Tables 1 and 3). These regressions were done separately for those participants with alcohol use at baseline and those with cocaine use at baseline. In the alcohol participants, greater self-help group attendance was consistently predicted by greater perceived seriousness of substance use problems (standardized beta coefficients ranged between .17 to .30, all p b .001) and lower alcohol use frequency (coefficients ranged from .17 to .35, all p b .001). Higher self-efficacy predicted more self-help group attendance at two followup points (coefficent = .12, p b .01). None of the other variables was a significant predictor at more than one time point. Very similar results were obtained in the cocaine participants, although in these individuals, a better living

Table 3 Prediction of crack cocaine use outcomes in baseline users Average frequency of crack cocaine use in months 1–6 Control Variables Gender Education Race Marital status (1 = married) Years of use Baseline value of the outcome Predictors General social support Social support for use (reversed) Quality of living situation (1 = good) Perceived seriousness of AOD SelfQefficacy SelfQhelp group attendance Psychiatric severity Employment status (1 = working) N F R2

7–12

.06 .01 .03 .15** .07 .02

.08 .00 .10* .02 .06 .12*

.04 .06 .07 .00 .20*** .09 .05 .04 382 2.77*** .10

.12* .05 .04 .04 .23*** .16** .02 .01 417 6.44*** .18

19–24

25–30

31–36

.05 .02 .02 .00 .02 .01

.08 .03 .03 .09 .09 .02

.09 .02 .05 .01 .02 .04

.09 .05 .03 .04 .21*** .02 .04 .08 424 3.45*** .11

.12* .06 .01 .05 .26*** .05 .05 .12* 450 6.15*** .17

.07 .07 .01 .01 .29*** .08 .06 .00 438 5.07*** .14

Note: Lagged analyses: baseline predictors/mo 1–6 crack cocaine use, 6 month predictors/mo. 7–12 crack cocaine use, 12 month predictors/mo 19–24 crack cocaine use, 24 month predictors/mo 25–30 crack cocaine use, 30 month predictors/mo 31–36 crack cocaine use. * p b .05. ** p b .01. *** p b .001.

J.R. McKay et al. / Journal of Substance Abuse Treatment 28 (2005) S73–S82 Table 4 Prediction of crack cocaine use outcomes in baseline users, controlling for current use Average frequency of crack cocaine use in months 7–12

19–24

25–30

31–36

Significant Control Variables Race .09 Current cocaine use .28*** .32*** .36*** .36*** Significant Predictors General social support SelfQefficacy .16** .14** .19*** Current employment .09* N 417 424 450 438 F 8.71*** 6.09*** 10.91*** 10.04*** R2 .23 .17 .26 .25 Note: Summary statistics at bottom of table are based on regression with all 15 predictor variables included. However, only significant predictors ( p b .05) are presented in the table. Lagged analyses: 6 month predictors/ mo. 7–12 crack cocaine use, 12 month predictors/mo 19–24 crack cocaine use, 24 month predictors/mo 25–30 crack cocaine use, 30 month predictors/ mo 31–36 crack cocaine use. * p b .05. ** p b .01. *** p b .001.

situation predicted more self-help group attendance at three follow-up points (coefficient = .12, p b .01). With regard to self-efficacy, higher scores in participants in the alcohol analyses were consistently predicted by lower perceived seriousness of substance abuse problems (coefficients ranged between .12, p b .01, and .17, p b .001), and lower alcohol use frequency (coefficients ranged between .38 and .46, p b .001). Greater self-help group attendance predicted higher self-efficacy at three follow-up points (coefficients = .09 to .11, p b .03), and general social support predicted higher self-efficacy at two follow-ups (coefficients = .10, p b .02). The only other variable that predicted self-efficacy at more than one follow-up was race, with white participants having lower self-efficacy at three follow-ups (coefficients ranged from .08, p b .05, to .15, p b .001). In the analyses with cocaine users, higher selfefficacy was consistently predicted by lower crack cocaine use frequency (coefficients ranged from .31 to .45, all p b .001), and greater self-help group attendance (coefficients = .13 to .16, p b .01). Once again, white participants had lower self-efficacy at two follow-ups (coefficients = .09, p b .05).

4. Discussion 4.1. Summary and review of findings The primary goal of this study was to identify factors that predict long-term outcomes after substance use treatment, with particular emphasis on post-treatment factors. A secondary goal was to identify the predictors of any post

S79

treatment factors that emerged as consistent predictors of substance use outcomes. Among participants with alcohol use at baseline, four measures consistently predicted better alcohol use outcomes in the subsequent assessment period across most or all of the follow-up: lower social support for substance use, greater perceived seriousness of substance use problems, higher self-efficacy, and more self-help group attendance. Even after current alcohol use at the follow-up was included, self-efficacy, perceived seriousness of substance use, and self-help group attendance still predicted alcohol use in the subsequent follow-up period. Among participants reporting crack cocaine use at baseline, only self-efficacy emerged as a consistent predictor of crack cocaine use in subsequent follow-up periods. Because self-efficacy and self-help group attendance were such robust predictors across time points and for both alcohol and cocaine outcomes, we asked what predicted higher scores on these variables. Results showed that higher self-efficacy during follow-up was predicted by lower perceived seriousness of substance use problems and lower substance use frequency in the prior assessment period, whereas greater self-help group attendance was predicted by greater perceived seriousness of substance use problems and lower substance use frequency in the prior period. The finding in the alcohol sample that self-efficacy and self-help group attendance were consistent predictors of subsequent alcohol use is in agreement with the findings of a number of prior studies (Connors et al., 1996; CritsChristoph et al., 2003; Miller et al., 1996; Moos et al., 1990; Morgenstern et al., 1997). In the alcohol-using participants, lower social support for substance use also consistently predicted less alcohol use in the Model 1 multivariate regressions. General social support, on the other hand, was not a significant predictor of subsequent alcohol use at any point in the follow-up. Prior studies of alcoholic treatment samples have also found that less social support for substance use predicts better alcohol use outcomes (Longabaugh et al., 1995; Longabaugh, Wirtz, Zweben, & Stout, 1998; Project Match Research Group, 1997). Abstinence specific social support is usually defined as the number of persons that the substance abuser spends time with regularly (i.e., family or friends) who are not substance abusers themselves (Longabaugh et al., 1995). In this study, however, the measure was a count of the number of relatives and friends whom the substance abuser enjoys getting together with and who would provide support, but who also have alcohol or drug problems. To be consistent with prior literature, this measure was reverse scored, so that higher scores indicated less social support for substance use. Indications from the results suggest that having fewer relatives and family members who are substance abusers predicts better substance use outcomes. This finding also is consistent with common sense and prior literature, although the magnitude of the effect was modest, and it was no longer significant when level of current alcohol or cocaine use was included in the regressions.

S80

J.R. McKay et al. / Journal of Substance Abuse Treatment 28 (2005) S73–S82

There have been relatively few studies of factors that predict long-term outcomes in cocaine patients, compared to the number of such studies done with alcohol and opiate dependent patients. This study afforded the opportunity to compare predictors of alcohol and crack cocaine outcomes. The findings suggest that there are both similarities and differences in patterns of predictors. Self-efficacy was a common and consistent predictor of both alcohol and cocaine use outcomes. However, three other factors—social support for substance use, perceived seriousness of substance use problems, and self-help group attendance— predicted alcohol outcomes but not cocaine outcomes. These findings are at odds with those from a prior study (McKay, Merikle, et al., 2001), in which self-help group involvement and support from family did predict subsequent cocaine use outcomes. The participants is this prior study were all graduates of an intensive outpatient treatment program, and the self-help group measure focused on total involvement rather than attendance only, which may account for the different findings. One other notable difference in the results from the alcohol and crack cocaine regressions was that the former consistently accounted for a higher percentage of variance in outcome than the latter. A good part of this was apparently due to larger correlations between alcohol use at each time point than between crack-cocaine use at each point, which is in keeping with basic differences in alcohol and cocaine use patterns—i.e., alcohol use tends to be fairly stable, whereas crack-cocaine use is more episodic. However, coefficients for the other predictors also tended to be somewhat higher in the alcohol regressions than in the cocaine regressions. It is possible that crack cocaine relapse episodes have a more impulsive quality than alcohol relapses and are determined to a greater extend by proximal factors, which would make them harder to predict with measures that are gathered at relatively widely spaced intervals (e.g., 6 to 12 months; Marlatt & Gordon, 1985; Shiffman, 1989). Finally, other variables, not included in the study, may have been stronger predictors of crack-cocaine use. Although it is useful to identify factors that predict longer-term outcome in substance abusers, this process provides little information on how patients achieve favorable scores on these factors. For example, if higher selfefficacy at 12 months predicts better long-term alcohol and cocaine use outcomes—as it does in this study—how do patients achieve this favorable marker? Most studies of posttreatment predictors of substance use outcomes have focused on identifying predictors, not on identifying the processes that account for the predictors themselves. Here, we tried to understand the factors that predicted self-efficacy and self-help group participation over the course of the follow-up, as these factors were significant predictors of outcomes in some of the analyses reported here, as well as in a number of prior studies. According to self-efficacy theory (Bandura, 1977), beliefs about oneTs ability to execute a behavior successfully

have a strong influence on future behavior. Higher selfefficacy is developed through the accomplishment of specific behavioral tasks, and therefore repeated success with a particular task leads to increased self-efficacy beliefs concerning that task (Marlatt & Gordon, 1985). Our findings support this theoretical position, in that the best predictor of high self-efficacy for being able to cope with problems without using alcohol or drugs was less alcohol and drug use in the prior follow-up period. This was particularly the case for crack cocaine users. Unfortunately, because self-efficacy is strongly related to recent substance use, it has limited usefulness as an independent goal for treatment interventions. Programs may find it more useful to focus directly on substance use, unless there is reason to believe that patients will not provide accurate reports. However, these findings are consistent with the idea that mastery of new and improved coping responses may produce improve self-efficacy if it also leads to better within-treatment and early post-treatment substance use outcomes (Annis & Davis, 1989; Marlatt & Gordon, 1985). Self-help group participation was also predicted by prior substance use frequency, but an even stronger predictor in the longitudinal analyses was perceived seriousness of substance use problems and desire to change. The participants with frequent attendance at self-help group meetings tended to be those who had low levels of alcohol or drug use in the prior period but who still believed that their substance abuse represented a serious problem that must be addressed. Prior studies of the predictors of self-help group attendance have usually focused on pretreatment or baseline factors, and one common finding from these studies is that greater severity of substance use problems tends to predict more self-help group attendance (Emrick, Tonigan, Montgomery, & Little, 1993; Weiss et al., 2000). It is possible that individuals with worse substance use histories are more likely to continue to perceive their problems as more severe, even when they are doing relatively well during follow-up. Self-help group programs place particular emphasis on the idea that substance abusers in recovery should always remember how severe their addiction was, and remain strongly committed to the process of change. The philosophy is reflected in program saying such as bkeep it freshQ and bdonTt ever forget where you came from.Q Moreover, one of the benefits to becoming a sponsor in such programs is the frequent reminder of how destructive active substance abuse can be. Although these findings regarding the correlates of factors like self-efficacy and self-help group attendance that predict subsequent substance use are of some interest, further work of this sort is clearly needed. The lack of more substantive findings suggests that future studies should examine a wider range of potential variables or processes, and employ more sophisticated mediational models (Longabaugh & Wirtz, 2001; MacKinnon, Taborga, & Morgan-Lopez, 2002). In particular, it would be useful to identify the bactive ingredientsQ of treatment that produce better short-term results on markers

J.R. McKay et al. / Journal of Substance Abuse Treatment 28 (2005) S73–S82

of progress, including self-efficacy and self-help group attendance (McLellan, McKay, Forman, Cacciola, & Kemp, in press). We were not able to assess treatment process measures while participants were in treatment, which precluded an examination of whether progress toward the goals of treatment was related to factors like self-efficacy and selfhelp group attendance at post-treatment follow-up points. 4.2. Strengths and limitations The study had a number of significant strengths. A large sample of participants were recruited from a number of treatment facilities, including residential, intensive outpatient, and standard outpatient programs. The study sample included men and women, and both alcohol and cocaine abusers were represented. A comprehensive, multidimensional assessment instrument was used to gather predictor variable and outcome data. Finally, participants were followed for 3 years to examine the stability of findings over a relatively long period. With regard to limitations, the study involved a naturalistic follow-up. Because none of the study variables was manipulated, it is not clear that any of the relations that were observed between predictors and outcomes were actually causal. The study might have yielded different results had the follow-up assessments been done more frequently, such as every 2 months. For example, high levels of depression or anxiety might predict proximal relapses, but not those 6 to 12 months later. The scales that were derived from the CIAI-C had good internal consistency, but formal psychometric evaluations of the test-retest reliability and validity of this measure have not yet been undertaken. Finally, because urine toxicology data were gathered from a minority of participants, it was not possible to run separate analyses with urine results as a measure of cocaine use. Moreover, agreement between self-reports of cocaine use and urine toxicology results were only marginal; the percentage of participants who reported no cocaine use but had a positive urine test was over 20% at each follow-up. Therefore, the difficulty in predicting cocaine use outcomes might be due in part to inaccurate self-reports of cocaine use. 4.3. Implications for a continuing care model of addiction treatment The findings reported here suggest that the relationships between most predictor variables and subsequent alcohol and crack cocaine outcomes remain relatively stable from the 6-month point onward, out to 36 months. This does not support a model of recovery in which different tasks or goals are important at different points. Rather, the goals that are important at 6 months are still important at 30 months. However, the degree to which treatment is necessary to achieve these goals, or the optimal approach when treatment is needed may change over time. Under our current treatment system, patients receive formal treatment for relatively

S81

short periods of time, and most addiction management is done through self-help group programs. Patients are not monitored by treatment providers; in fact, contact with formal treatment is usually resumed only after a period of severe relapses and further negative consequences to the substance abuser and loved ones (McLellan, Lewis, OTBrien, & Kleber, 2000). When the patient is back in treatment, counselors often note that the bwarning signsQ of an impending relapse have usually been evident for some time. Another approach to treatment involves shifting from a reactive to a proactive system in which patients are monitored for substantial periods of time following an episode of treatment to improve the management of their disorder (Dennis, Scott, & Funk, 2003; McKay, 2001; Stout, Rubin, Zwick, Zywiak, & Bellino, 1999). This sort of low intensity, low cost monitoring is designed to catch warning signs and initiate stepped-up services before severe deterioration occurs, thereby improving the long-term outcomes of substance abuses. The data presented here suggest that for alcoholics, ongoing monitoring of self-efficacy, self-help group attendance, social support for substance use, and perceived seriousness of substance use problems could be useful in identifying individuals who may be in need of stepup services, along with assessment of current alcohol use. The data from this study are less informative with regard to monitoring the status of patients with cocaine use disorders, although assessment of self-efficacy may be useful when accurate reports of substance use are not available.

Acknowledgments Preparation of this manuscript was supported by funding from the Center for Substance Abuse Treatment (CSAT), Substance Abuse and Mental Health Services Administration, Department of Health and Human Services Contract No. 270-97-7011 (Persistent Effects of Treatment Studies), and a National Institute on Drug Abuse Independent Scientist (K02) award to the first author. The content of this publication does not necessarily reflect the views or policies of the Department of Health and Human Services, nor does mention of trade names, commercial products, or organizations imply endorsement by the U.S. government. Parts of this paper were presented at the 2001 annual conference of the American Evaluation Association, St. Louis, Missouri, in a paper by McKay, Foltz, Leahy, & Stephens (2001), entitled Relation of Post-Treatment Factors to Alcohol and Cocaine Outcomes Over 36 Months.

References Annis, H. M., & Davis, C. S. (1989). Relapse prevention. In R. K. Hester & W. R. Miller (Ed.), Handbook of alcoholism treatment approaches (pp. 170 – 182). New York7 Pergamon Press. Babor, T. F., Longabaugh, R., Zweben, A., Fuller, R. K., Stout, R. L.,

S82

J.R. McKay et al. / Journal of Substance Abuse Treatment 28 (2005) S73–S82

Anton, R. F., & Randall, C. L. (1994). Issues in the definition and measurement of drinking outcomes in alcoholism treatment research. Journal of Studies on Alcohol. Supplement 12, 101 – 111. Bandura, A. (1977). Self efficacy: Toward a unifying theory of behavioral change. Psychological Review, 84, 191 – 215. Belsley, D. A., Kuh, E., & Welsch, R. E. (1980). Regression diagnostics. New York7 Wiley. Carroll, K. M. (1995). Methodological issues and problems in assessment of substance use. Psychological Assessment, 7, 349 – 358. Connors, G. J., Carroll, K. M., DiClemente, C. C., Longabaugh, R., & Donovan, D. M. (1997). The therapeutic alliance and its relationship to alcoholism treatment participation and outcome. Journal of Consulting and Clinical Psychology, 65, 588 – 598. Connors, G. J., Maisto, S. A., & Zywiak, W. H. (1996). Understanding relapse in the broader context of post-treatment functioning. Addiction, 91, S173 – S190. Crits-Christoph, P., Connolly Gibbons, M. B., Barber, J. P., Gallop, R., Beck A. T., Mercer, D., Tu, X., Thase, M. E., Weiss, R. D., & Frank, A. (2003). Mediators of outcome of psychosocial treatments for cocaine dependence. Journal of Consulting and Clinical Psychology, 71, 918 – 925. Dennis, M. L., Scott, C. K., & Funk, R. (2003). An experimental evaluation of recovery management checkups (RMC) for people with chronic substance use disorders. Evaluation and Program Planning, 26, 339 – 352. Emrick, C. D., Tonigan, J. S., Montgomery, H., & Little, L. (1993). Alcoholics Anonymous: What is currently known? In B. S. McCrady & W. R. Miller (Eds.), Research on Alcoholics Anonymous: Opportunities and Alternatives (pp. 41 – 76). New Brunswick, NJ7 Rutgers Center of Alcohol Studies. Hall, S. M., Havassy, B. E., & Wasserman, D. A. (1990). Commitment to abstinence and acute stress in relapse to alcohol, opiates, and nicotine. Journal of Consulting and Clinical Psychology, 58, 175 – 181. Hall, S. M., Havassy, B. E., & Wasserman, D. A. (1991). Effects of commitment to abstinence, positive moods, stress, and coping on relapse to cocaine use. Journal of Consulting and Clinical Psychology, 59, 526 – 532. Havassy, B. E., Wasserman, D. A., & Hall, S. M. (1995). Social relationships and abstinence from cocaine in an American treatment sample. Addiction, 90, 699 – 710. Kranzler, H. R., Del Boca, F. K., & Rounsaville, B. J. (1996). Comorbid psychiatric diagnosis predicts three-year outcomes in alcoholics: A posttreatment natural history study. Journal of Studies on Alcohol, 57, 619 – 626. Lehman, W. E. K., & Simpson, D. D. (1990). Prediction of Year 12 Outcomes. In D. D. Simpson & S. B. Sells (Eds.), Opioid addiction and treatment: A 12 year follow-up (pp. 203 – 220). Malibar, FL.7 Krieger. Longabaugh, R., & Wirtz, P. W. (Eds.). (2001). Project MATCH hypotheses: Results and causal chain analyses. Bethesda, MD7 U.S. Department of Health and Human Services. Longabaugh, R., Wirtz, P. W., Beattie, M. C., Noel, N., & Stout, R. L. (1995). Matching treatment focus to patient social investment and support: 18 month follow-up results. Journal of Consulting and Clinical Psychology, 63, 296 – 307. Longabaugh, R., Wirtz, P. W., Zweben, A., & Stout, R. L. (1998). Network support for drinking, alcoholics Anonymous, and long-term matching effects. Addiction, 93, 1313 – 1333. MacKinnon, D. P., Taborga, M. P., & Morgan-Lopez, A. A. (2002). Mediation designs for tobacco prevention research. Drug and Alcohol Dependence, 68, S69 – S83. Marlatt, G. A., & Gordon, J. R. (Eds.). (1985). Relapse prevention: Maintenance strategies in the treatment of addictive behaviors. New York7 Guilford. McKay, J. R. (2001). The role of continuing care in outpatient alcohol treatment programs. In M. Galanter (Ed.), Recent developments in alcoholism, Vol. 15: Services research in the era of managed care (pp. 357 – 372). New York7 Kluwer Academic/Plenum. McKay, J. R., Foltz, C., Leahy, P. J., & Stephens, R. C. (2001). Relation

of post-treatment factors to alcohol and cocaine outcomes over 36 months. Paper presented at the annual conference of the American Evaluation Association, St. Louis, MO. McKay, J. R., Merikle, E., Mulvaney, F. D., Weiss, R. V., & Koppenhaver, J. M. (2001). Factors accounting for cocaine two years following initiation of continuing care. Addiction, 96, 213 – 225. McKay, J. R., & Weiss, R. V. (2001). A review of temporal effects and outcome predictors in substance abuse treatment studies with long-term follow-ups: Preliminary results and methodological issues. Evaluation Review, 25, 113 – 161. McLellan, A. T., Lewis, D. C., OTBrien, C. P., & Kleber, H. D. (2000). Drug dependence, a chronic medical illness: Implications for treatment, insurance, and outcomes evaluation. Journal of the American Medical Association, 284, 1689 – 1695. McLellan, A. T., Luborsky, L., Cacciola, J., Griffith, J., Evans, F., Barr, H., & O’Brien, C. P. (1985). New data from the Addiction Severity Index: Reliability and validity in three centers. Journal of Nervous and Mental Disease, 173, 412 – 423. McLellan, A. T., Luborsky, L., Woody, G. E., Druley, K. A., & O’Brien, C. P. (1983). Predicting response to alcohol and drug abuse treatments: Role of psychiatric severity. Archives of General Psychiatry, 40, 620 – 625. McLellan, A. T., Luborsky, L., Woody, G. E., & O’Brien, C. P. (1980). An improved diagnostic evaluation instrument for substance abuse patients: The Addiction Severity Index. Journal of Nervous and Mental Disease, 168, 26 – 33. McLellan, A. T., McKay, J. R., Forman, R., Cacciola, J., & Kemp, J. (in press). Reconsidering the evaluation of addiction treatment: From retrospective follow-up to performance monitoring. Addiction. Meyer, R. E. (2001). Finding paradigms for the future of alcoholism research: An interdisciplinary perspective. Alcoholism: Clinical and Experimental Research, 25, 1393 – 1406. Miller, W. R., Westerberg, V. S., Harris, R. J., & Tonigan, J. S. (1996). What predicts relapse? Prospective testing of antecedent models. Addiction, 91, S155 – S172. Moos, R. H., Finney, J. W., & Cronkite, R. C. (1990). Alcoholism treatment: Context, process, and outcome. New York7 Oxford Press. Morgenstern, J., Frey, R., McCrady, B., Labouvie, E., & Neighbors, C. (1996). Mediators of change in traditional chemical dependency treatment. Journal of Studies on Alcohol, 57, 53 – 64. Morgenstern, J., Labouvie, E., McCrady, B., Kahler, C., & Frey, R. (1997). Affiliation with Alcoholics Anonymous following treatment: A study of its therapeutic effects and mechanisms of action. Journal of Consulting and Clinical Psychology, 65, 768 – 778. Project Match Research Group. (1997). Matching alcoholism treatments to client heterogeneity: Project MATCH post-treatment drinking outcomes. Journal of Studies of Studies on Alcoholism, 58, 7 – 29. Project MATCH Research Group. (1998). Matching alcoholism treatments to client heterogeneity: Project MATCH three-year drinking outcomes. Alcoholism: Clinical and Experimental Research, 22, 1300 – 1311. Shiffman, S. (1989). Conceptual issues in the study of relapse. In M. Gossop (Ed.), Relapse and addictive behavior (pp. 149 – 179). Kent, England7 Croom Helm Ltd. Stout, R. L., Rubin, A., Zwick, W., Zywiak, W., & Bellino, L. (1999). Optimizing the cost-effectiveness of alcohol treatment: A rationale for extended case monitoring. Addictive Behaviors, 24, 17 – 35. Tabachnick, B. G., & Fidell, L. S. (2001). Using multivariate statistics, (4th ed.). New York7 Harper Collins. Tonigan, J. S., Toscova, R., & Miller, W. R. (1996). Meta-analysis of the literature on Alcoholics Anonymous: Sample and study characteristics moderate findings. Journal of Studies on Alcohol, 57, 65 – 72. Vaillant, G. E. (1996). A long-term follow-up of male alcohol abuse. Archives of General Psychiatry, 53, 243 – 249. Weiss, R. D., Griffin, M. L., Gallop, R., Luborsky, L., Siqueland, L., Frank, A., Onken, L. S., Daley, D. C., & Gastfriend, D. R. (2000). Predictors of self-help group attendance in cocaine dependent patients. Journal of Studies on Alcohol, 61, 714 – 719.