Efficacy of Relapse Prevention: A Meta-Analytic Review

Abstract

Although relapse prevention (RP) has become a widely adopted cognitive-behavioral treatment intervention for alcohol, smoking, and other substance use, outcome studies have yielded an inconsistent picture of the efficacy of this approach or conditions for maximal effectiveness. A meta-analysis was performed to evaluate the overall effectiveness of RP and the extent to which certain variables may relate to treatment outcome. Twenty-six published and unpublished studies with 70 hypothesis tests representing a sample of 9,504 participants were included in the analysis. Results indicated that RP was generally effective, particularly for alcohol problems. Additionally, outcome was moderated by several variables. Specifically, RP was most effective when applied to alcohol or polysubstance use disorders, combined with the adjunctive use of medication, and when evaluated immediately following treatment using uncontrolled pre-post tests.

Full text

Page 1 of 12
http://spider.apa.org/ftdocs/ccp/1999/august/ccp674563.html 8/30/2000
Efficacy of Relapse Prevention
A Meta-Analytic Review
Jennifer E. Irvin
Department of Psychology University of Central Florida
Clint A. Bowers
Department of Psychology University of Central Florida
Michael E. Dunn
Department of Psychology University of Central Florida
Morgan C. Wang
Department of Statistics University of Central Florida
ABSTRACT
Although relapse prevention (RP) has become a widely adopted cognitive—behavioral
treatment intervention for alcohol, smoking, and other substance use, outcome studies have
yielded an inconsistent picture of the efficacy of this approach or conditions for maximal
effectiveness. A meta-analysis was performed to evaluate the overall effectiveness of RP and
the extent to which certain variables may relate to treatment outcome. Twenty-six published
and unpublished studies with 70 hypothesis tests representing a sample of 9,504 participants
were included in the analysis. Results indicated that RP was generally effective, particularly for
alcohol problems. Additionally, outcome was moderated by several variables. Specifically, RP
was most effective when applied to alcohol or polysubstance use disorders, combined with the
adjunctive use of medication, and when evaluated immediately following treatment using
uncontrolled pre—post tests.
Jennifer E. Irvin is now at the Department of Psychology, University of South Florida.
Portions of this article were presented at the 105th Annual Convention of the American Psychological
Association, Chicago, Illinois, August 1997.
Correspondence may be addressed to Clint A. Bowers, Department of Psychology, University of Central
Florida, Orlando, Florida, 32816-1390.
Electronic mail may be sent to bowers@pegasus.cc.ucf.edu
Received: May 1, 1997
Revised: December 15, 1998
Accepted: January 6, 1999
Despite the increasing appeal of relapse-prevention (RP) interventions based on the work of Marlatt and
Gordon (1985) , studies have failed to yield a consistent picture of the effectiveness of this cognitive—
behavioral approach ( Rawson, Obert, McCann, & Marinelli-Casey, 1993 ). In a recent narrative review of
24 studies, Carroll (1996 ) concluded (a) that RP appeared to be effective relative to no-treatment control
groups, (b) that RP appeared to be equally effective as other active treatments, and (c) that comparison of
Journal of Consulting and Clinical Psychology ©
1999 by the American Psychological Association
August 1999 Vol. 67, No. 4, 563-570 For personal use only--
not for distribution.

Page 2 of 12
http://spider.apa.org/ftdocs/ccp/1999/august/ccp674563.html 8/30/2000
RP with attention and discussion control groups yielded inconclusive results. Carroll also concluded that the
effectiveness of RP did not vary with class of substance use disorder. Given the broad-based use of RP and
the uncertainty regarding its overall effectiveness, there is a need to empirically evaluate RP interventions and
the extent to which moderator variables relate to treatment outcomes. Toward that end, we conducted a
meta-analysis to quantify the effect of RP on substance use behavior and overall psychosocial adjustment
and to identify moderator variables that may relate to heightened or attenuated effectiveness of RP.
The RP model hypothesizes that there are common cognitive, behavioral, and affective mechanisms that
underlie the process of relapse, regardless of the particular problem behavior considered ( Marlatt &
Gordon, 1985 ). Thus, it seems that RP should be efficacious in preventing relapse across various classes of
addictive behaviors. In addition, the modality in which RP is delivered should be associated with relative
costs and benefits. For example, individual treatment may provide the best format for targeting clients'
individual needs, whereas group treatment may provide increased social support and create important
opportunities for practicing new skills ( Schmitz et al., 1997 ). Furthermore, working with others in a
collective attempt to maintain behavior changes may foster support and encouragement ( Marlatt & Gordon,
1985 ).
Although there has been considerable controversy regarding the relative effectiveness of inpatient versus
outpatient treatment for substance use, evaluations of this issue have indicated that treatment setting may
have little to do with outcome (e.g., Miller & Hester, 1986 ). The theoretical rationale on which RP is based,
however, suggests that treatment delivery in an inpatient setting may not provide patients with real-life
opportunities to practice newly acquired skills or give them the opportunity to encounter problems while they
still have the close support of treatment professionals. Therefore, in investigating moderators, it was
important to examine the extent to which the effectiveness of RP was impacted by treatment setting.
The RP model conceptualizes addiction as stemming from a collection of maladaptive habit patterns rather
than from purely physiological responses to substance use ( Marlatt & Gordon, 1985 ). Therefore, specific
predictions regarding the potential efficacy of the adjunctive use of medication with RP are not addressed
theoretically. Similarly, type of substance use outcome measure used to evaluate RP is not theoretically
addressed by the model but is of considerable importance because of the controversy surrounding this issue
(see L. C. Sobell & Sobell, 1990 ). Studies included in the present analysis typically used verified self-
report, unverified self-report, or biochemical measures of substance use, creating an opportunity to compare
these methods.
One goal of the present analysis was to evaluate the overall efficacy of RP. Therefore, it was important to
examine the effects of RP when compared with other active interventions, no-additional-treatment controls,
discussion controls, physician advice, and psychoeducation. Finally, the theoretical rationale underlying RP
suggests that a critical stage of the habit—change process begins after initial cessation of substance use
( Marlatt & Gordon, 1985 ) and relapse is conceptualized as a transitional process in which individuals may
or may not return to habitual substance use following a lapse. In support of this concept, Moos and Finney
(1983) reported that a considerable portion of the variability associated with the long-term effectiveness of
alcohol treatment is related to environmental factors that emerge after completion of formal treatment. In
addition, RP seeks to provide individuals with coping strategies to prevent a "slip" from becoming a full-
blown relapse because the model asserts that beliefs and expectations regarding a lapse may influence
continued use. Lapses are viewed as a normal part of acquiring abstinence and constitute opportunities for
new learning. Therefore, it was theoretically important to evaluate the efficacy of RP at different
posttreatment intervals.

Page 3 of 12
http://spider.apa.org/ftdocs/ccp/1999/august/ccp674563.html 8/30/2000
Method
Potential studies were identified through computerized databases, reference lists, and the contacting of
prominent researchers to request unpublished data. To be included in the analysis, a study had to identify the
treatment approach as RP or be clearly consistent with Marlatt and Gordon's (1985) approach.
Additionally, studies were required to report test statistics associated with hypothesis tests of the
effectiveness of RP reflecting outcomes when compared with no-additional-treatment controls, other active
interventions, discussion controls, physician advice, or uncontrolled pre—post tests. Treatment approaches
that used general cognitive—behavioral techniques or evaluated an RP program couched in a larger, broader
based approach were not included. Twenty-two published studies and 4 unpublished studies conducted
between 1978 and 1995 met the inclusion criteria. The independent variable was RP, and the two
dependent measures of RP that we examined were substance use and overall psychosocial adjustment (e.g.,
clients' ratings of the severity of their problem, marital adjustment, acquisition of cognitive and behavioral
coping and problem-solving skills to avoid relapse, self-efficacy, locus of control, and depression). Data
were coded for the seven moderators described previously: modality of treatment, treatment setting, type of
outcome measure used to evaluate RP, adjunctive use of medication, class of substance use disorder,
comparative effectiveness of RP, and effectiveness at various lengths of follow-up.
Twenty-six studies with 70 hypothesis tests representing a sample of 9,504 participants were included in the
analysis. All main effect sizes are reported in terms of the weighted averaged correlation coefficient r . We
calculated effect size estimates using r rather than d or g because it is a readily interpretable statistic from a
practical standpoint. More detailed information regarding the use of r as the index of effect size and the
general statistical procedures used in meta-analysis is provided by Rosenthal (1984) . A discussion regarding
the use of r as an unbiased estimator of effect size can be found in the work of Hedges and Olkin (1985) .
We report 95% confidence intervals (CIs) for all effect size estimates. CIs that do not contain zero can be
considered statistically significant from zero at the p < .05 level. The n s reported in this article represent the
number of studies included in each analysis.
The first step in conducting the analysis was to aggregate findings across studies. We converted all test
statistics to the effect size index r with the following formulas:
The n used to convert z to r reflected the overall sample size in each study. The n used to convert χ 2 to r
reflected the sum of sample sizes for the two groups compared (i.e., n 1 + n 2 ).
Second, we computed the overall CI of r . Several studies contributed more than one hypothesis test to the
meta-analysis, violating the assumption of independence. Therefore, we conducted a mini-meta-analysis

Page 4 of 12
http://spider.apa.org/ftdocs/ccp/1999/august/ccp674563.html 8/30/2000
within each study to find the combined effect size estimate for each study. For purposes of the mini-meta-
analyses, r s were transformed to Fisher's z s (i.e., z r ; Rosenthal, 1984 ). Next, the weighted averaged
formula for z r was used to compute the combined mean estimate of Fisher's z s (i.e., z combined ) and the
standard error. In computing the combined sample size for each study, we used the smallest sample size
associated with tests within each study because this represented the most conservative approach. A 95% CI
of [ z Ι , z µ ] was computed. Subsequently, z combined was converted back to r combined and [ z Ι , z µ ] to
[ r Ι , r µ ] (see Table 1 ). The method of the weighted averaged r was used to cumulate the effect sizes
across studies. This method was used to account for different variances across studies. The weight used was
the reciprocal of the corresponding variance estimate. Thus, studies with more precise effect size estimates
(i.e., smaller variance) were given larger weights (see Cooper & Hedges, 1994 ).
Finally, we constructed a dot plot to provide a readily interpretable graphical depiction of the results of the
meta-analysis (see Figure 1 ; see Wang & Bushman, 1999 ). A dot plot provided an effective method of
displaying correlation coefficient estimates and CIs for each study in relation to every other study and to the
average of the combined r s represented by the column of r s down the center of the dot plot. CIs were
included for each effect size estimate as an index of the reliability of each estimate. The studies are rank
ordered by effect size. That the display depicts a gradual progression of effect sizes from small to large with
no gaps between effect sizes indicates that the present data are especially appropriate for meta-analysis
( Wang & Bushman, 1999 ). Positive correlations indicated that RP fared better than the treatment(s) with
which it was compared. Negative correlations indicated that RP fared worse than the treatment(s) with
which it was compared.
Results
General Effects
The overall treatment effect of RP interventions for alcohol use disorders, substance use disorders, and
smoking was r = .14 ( n = 26). The obtained effect was highly reliable. As can be seen in Figure 1 , the
lower bound limit of the 95% CI was r = .10, which is well above zero. Additionally, the CI was relatively
small (95% CI = .10 to .17). The effect of RP on reducing substance use was r = .14 and was reliable (95%
CI = .10 to .17, n = 22). This analysis differed from the overall treatment effect analysis in that it did not
include the 4 studies that only measured psychosocial outcomes. The effect of RP on improving overall
psychosocial adjustment was of dramatically larger magnitude ( r = .48) and reliable (95% CI = .42 to .53,
n = 10).
Because meta-analytic results may be vulnerable to biases related to the "file drawer" problem, it is important
to provide an estimation of the number of studies averaging null results that would be required to reduce the
obtained effect size to the .05 level ( Rosenthal, 1984 ). The fail-safe n indicates the number of undiscovered
studies averaging no effect of RP interventions that would be required to reduce the obtained relationship to
zero. The fail-safe n obtained for the overall effect of RP was 388 ( p = .05), which indicates that this finding
is quite tolerant of future undiscovered null results.
Class of Substance Use Disorder
The effect sizes for different substance use disorders were r = .37 (95% CI = .28 to .45, n = 10) for alcohol
use, r = .27 (95% CI = .17 to .37, n = 5) for polysubstance use, r =
−
.03 (95% CI =
−
.17 to .11, n = 3)

Page 5 of 12
http://spider.apa.org/ftdocs/ccp/1999/august/ccp674563.html 8/30/2000
for cocaine use, and r = .09 (95% CI = .04 to .13, n = 8) for smoking. The finding for cocaine use should
be interpreted with caution, as only 3 studies contributed to the analysis. Contrast analyses indicated that the
difference between alcohol and polysubstance use was not significant (contrast = .10, 95% CI = − .04
to .25, z = 1.40, p = .1612). However, the effect sizes obtained for RP in treating both alcohol use and
polysubstance use were substantially greater than the effect size obtained for smoking (contrast comparing
alcohol with smoking = .29, 95% CI = .19 to .38, z = 5.66, p = .0000; contrast comparing polysubstance
use with smoking = .19, 95% CI = .08 to .31, z = 3.26, p = .0011). Therefore, RP was significantly more
effective in treating alcohol and polysubstance use than in treating smoking or cocaine use. This certainly
attenuated the overall effect size obtained for RP and complicated further analyses.
Treatment Modality
The effect size for studies that used RP in an individual format was r = .10 (95% CI = .05 to .15, n = 5). Of
these 5 studies, however, 3 were based on alcohol use and the other 2 were based on cocaine and smoking.
An examination of the dot plot (see Figure 1 ) indicates that the effect sizes for the alcohol studies were r
= .33 and r = .35. The effect size for studies that used RP in a group format was r = .16 (95% CI = .11
to .20, n = 19). Again, an examination of the dot plot indicates that effect sizes for alcohol studies tended to
be substantially larger than the overall effect size for group therapy (i.e., r = .37 vs. r = .16). The difference
between individual and group modalities was not statistically significant (contrast = − .06, 95% CI = − .13
to .01, z = − 1.58, p = .1140). The effect size for studies that used RP in a conjoint marital therapy format
was r = .32 (95% CI = .10 to .50, n = 2). The latter finding is based on only 2 studies and should be
interpreted with caution. Furthermore, both of these studies were alcohol studies, which were associated
with generally larger effect sizes than studies based on other classes of substance use disorders, as noted
above. Unfortunately, there were too few studies that used RP in a conjoint marital therapy format to
perform contrast analyses with the other modalities.
Setting of Treatment
Studies conducted in an inpatient setting yielded a treatment effect of r = .11 (95% CI = .06 to .16, n = 8).
The treatment effect for studies conducted in an outpatient setting was of slightly greater magnitude at r = .16
(95% CI = .11 to .20, n = 18). Although this difference was not statistically significant (contrast = − .04,
95% CI = − .11 to .03, z = − 1.24, p = .2160), 7 of the 8 inpatient studies addressed alcohol use or
polysubstance use and were, therefore, associated with generally larger effect sizes. Therefore, findings may
have supported the superiority of outpatient treatment if there had been enough studies to conduct analyses
separately for each type of substance.
Adjunctive Use of Medication
The treatment effect without the adjunctive use of medication was r = .09 (95% CI = .05 to .13, n = 19),
whereas the effect size for RP with adjunctive use of medication was r =.48 (95% CI = .38 to .56, n = 4).
This difference was significant (contrast = − .40, 95% CI = − .50 to − .30, z = − 6.78, p = .0000). Three of
the 4 studies that evaluated RP in combination with medication were alcohol studies and the 4th was a
cocaine study, which may partially account for the much larger effect size obtained for the adjunctive use of
medication. All 4 studies used different medications. The treatment effect that emerged for the adjunctive use
of nicotine gum in smoking studies was r = .18 (95% CI = .07 to .29, n = 3). This finding was based on only
3 studies and should be interpreted with caution.
Outcome Measures

Page 6 of 12
http://spider.apa.org/ftdocs/ccp/1999/august/ccp674563.html 8/30/2000
The treatment effect that emerged for role-play and problem-solving tests was r = .64 (95% CI = .57
to .70, n = 2 [2 polysubstance use studies]). This analysis was based on only 2 studies and should be
interpreted with caution. Self-reported use yielded an effect size of r = .17 (95% CI = .11 to .22, n = 15
[11 alcohol studies, 2 smoking studies, 1 polysubstance use study, and 1 cocaine study]). The effect size
based on biochemical measures of use (e.g., urinalysis and blood alcohol tests) was of smaller magnitude ( r
= .12, 95% CI = .06 to .19, n = 5 [3 polysubstance use studies, 1 cocaine study, and 1 smoking study]).
The difference between these two effect sizes was not statistically significant (contrast = .05, 95% CI = −
.04 to .13, z = 1.08, p = .2805). The effect size for self-report verified by biochemical tests was even
smaller in magnitude at r = .09 (95% CI = .04 to .13, n = 6 [5 smoking studies and 1 cocaine study]). The
difference between self-report and biochemically validated self-report was statistically significant (contrast
= .09, 95% CI = .01 to .16, z = 2.53, p = .0243). All studies that used validated self-report addressed
either cocaine use or smoking, which probably contributed to the smaller effect size.
Comparative Efficacy of RP
The strongest treatment effect emerged when RP was evaluated using uncontrolled pre- and posttests ( r
= .59, 95% CI = .50 to .67, n = 5 [3 alcohol studies and 2 polysubstance use studies]). The effect size for
RP when compared with wait-list or no-additional-treatment controls was r =.11 (95% CI = .06 to .15, n =
7 [2 polysubstance use studies, 4 alcohol studies, and 1 smoking study]). When compared with other active
interventions, the effect size was r = − .19 (95% CI = − .34 to − .03, n = 4 [2 cocaine studies, 1 alcohol
study, and 1 smoking study]). This negative effect size may be attributable to the number of smoking and
cocaine studies that contributed to this analysis, as RP seems to be less effective for these substances.
Compared with discussion controls, the treatment effect for RP was r = .17 (95% CI = .08 to .26, n = 6 [2
alcohol studies, 3 smoking studies, and 1 cocaine study]). Two studies found discussion and no-additional-
treatment controls to be equivalent, and we combined them for analyses. The effect size for these studies
was r = .09 (95% CI = .01 to .16, n = 2 [1 alcohol study and 1 smoking study]). Compared with physician
advice, however, RP appears to be more effective ( r = .33, 95% CI = .02 to .58, n = 1 [1 smoking
study]). This finding was based on only 1 study, and the 95% CI associated with this finding was very large.
Compared with educational groups, the effect size for RP was r = .20 (95% CI = .06 to .33, n = 3 [1
polysubstance use study and 2 smoking studies]). This finding was based on only 3 studies and should be
interpreted with caution.
Length of Posttreatment Follow-Up
Treatment effects were largest when outcomes were assessed immediately following treatment ( r = .27,
95% CI = .23 to .32, n = 10) but tended to be smaller at longer lengths of follow-up (see Figure 2 for effect
sizes with CIs). Specifically, at 1-month follow-up, r = .20 (95% CI = .04 to .34, n = 2); at 3-month
follow-up, r = .19 (95% CI = .02 to .35, n = 3); at 6-month follow-up, r = .19 (95% CI = .11 to .26, n =
13); and 1-year following treatment, r = .09 (95% CI = .05 to .13, n = 11). The classes of substance use
disorders that were evaluated did not appear to be overrepresented during any particular length of follow-
up. Effect sizes at the various lengths of follow-up were not compared statistically because they are
correlated with one another, and the use of contrast analysis requires that effect sizes be independent.
Discussion
There has been a trend in the psychotherapy literature toward investigating factors that maximize the

Page 7 of 12
http://spider.apa.org/ftdocs/ccp/1999/august/ccp674563.html 8/30/2000
effectiveness of specific treatment strategies in addition to establishing general efficacy. The results of our
meta-analytic review indicate that RP was effective across levels of the moderators and appeared to have
more impact on improving psychosocial functioning than on reducing substance use. Efficacy of RP varied as
a function of five of the seven potential moderators that were investigated. Treatment effects were strong and
reliable for alcohol use and for polysubstance use but dramatically weaker for smoking. This finding was in
line with narrative reviews of the smoking literature (e.g., Lichtenstein & Glasgow, 1992 ) and with primary
research investigating the efficacy of RP with smoking ( Curry, Marlatt, Gordon, & Baer, 1988 ). Although
the results suggest that RP may not be as effective as other approaches in treating cocaine use, further
research is needed to draw a more definitive conclusion. These findings were surprising in light of Carroll's
(1996) review, which concluded that little support existed for the notion that RP was differentially effective
across classes of substance use disorders. Thus, it seems of considerable merit to investigate whether
specific components of RP treatment impact its efficacy across different classes of substances.
The modality in which RP was delivered was not shown to moderate its effectiveness. Findings were
complicated, however, by the differential effectiveness of RP with different types of substances. The average
effect sizes for alcohol studies were very similar, regardless of modality. Although the effect size for conjoint
marital therapy was of stronger magnitude than for the other modalities, it should be interpreted with
considerable caution as this analysis was based on only 2 studies, both of which were alcohol studies. The
effect size for individual treatment was not significantly lower than the effect size for group treatment but was
based on only 2 cocaine or smoking studies and 3 alcohol studies. Although conclusions should be drawn
with considerable caution, it appears that individual, group, and marital modalities are equally effective,
although they all seem to be more effective in treating alcohol problems than problems with other substances.
Given the current trend toward more cost-efficient methods of delivery of mental health services, the
potential benefits of matching clients to modalities should be investigated to evaluate the extent to which
different types of individuals benefit from different treatment formats.
Analyses of treatment setting revealed no significant differences between outpatient and inpatient modalities
of treatment. This finding was consistent with several primary studies and reviews of the literature that found
no real advantages of inpatient versus outpatient treatment for alcoholism (e.g., Miller & Hester, 1986 ). It is
important to note, however, that 7 of the 8 inpatient studies addressed alcohol use or polysubstance use,
which were associated with generally larger effect sizes. Therefore, the superiority of outpatient over
inpatient treatment may become apparent when class of substance use disorder is considered. It is also
possible, however, that problem severity was greater among individuals receiving inpatient care.
Medication may contribute substantially to enhancing treatment effectiveness, particularly for alcohol
problems. Unfortunately, this analysis was based on only 4 studies (most of which were alcohol studies), and
all 4 studies used different medications (fluoxetine, desiprimine, Naltrexone, and citrated calcium carbimide).
Therefore, additional research is needed to verify the contribution of these medications. The analysis of
smoking studies revealed a weak treatment effect for the adjunctive use of nicotine gum with RP. Although
consistent with the smoking literature, this analysis was based on only 3 studies and must be interpreted
cautiously.
Studies that used unverified self-report yielded a significantly larger overall effect size than those that used
biochemically verified self-report. The unverified self-report effect size was also larger than that of studies
that used biochemical measures, although the difference was not statistically significant. Unfortunately,
differential effectiveness of RP across type of substance use complicates the interpretation of these findings.
The majority of the unverified self-report studies addressed alcohol use, whereas all of the biochemical or
biochemically validated self-report studies addressed smoking or cocaine use. Therefore, it is impossible to

Page 8 of 12
http://spider.apa.org/ftdocs/ccp/1999/august/ccp674563.html 8/30/2000
draw any conclusions regarding the veracity of unverified self-report on the basis of this analysis because it is
likely that effect sizes would change substantially if they were based on the same substance.
RP demonstrated a high level of efficacy when evaluated using pretest—posttest designs and when
compared with physician advice and more moderate efficacy when compared with psychoeducational
groups or discussion controls. The comparisons with physician advice and psychoeducational groups were
based on only 1 and 3 studies, respectively, and should be interpreted cautiously. RP yielded a negative
effect size when compared with other active interventions, but differences in type of substance addressed
make this finding difficult to interpret. Two of the 4 studies comparing RP with other active interventions
evaluated its efficacy in treating cocaine use, and 1 evaluated its efficacy in the treatment of smoking. These
two classes of substance use disorders represented the smallest effect sizes for RP. The 4th study evaluated
the effectiveness of RP in preventing a return to alcohol use and found few differences overall in comparing
RP with another active intervention. Therefore, it may be that other active interventions are more effective
than RP for some substances. Unfortunately, too few studies compared RP with active interventions to
address this question meta-analytically.
Somewhat perplexing was the finding of a weak treatment effect for RP compared with no-additional-
treatment controls. This finding may be explained by the fact that, in the majority of hypothesis tests, RP was
administered immediately following another primary intervention or functioned as a supplemental intervention
to another treatment. Only 1 study included in the present analysis compared RP with true no-treatment
controls ( Stevens, Glasgow, Hollis, Lichtenstein, & Vogt, 1993 ). In light of the fact that RP demonstrated
considerable efficacy when evaluated through pre- and posttests and when compared with discussion
controls, it may be that the effects of the preceding treatment obscured RP effects. Although additional
research comparing RP with no-treatment controls would be necessary to elucidate these differences, a
trend in outcome studies in support of the efficacy of psychotherapy compared with wait-list controls may
render this comparison unnecessary.
Effect sizes for RP were strongest when outcomes were evaluated immediately following the completion of
treatment and tended to be weaker (albeit significant) when measured at increased intervals after treatment.
It is to be expected that the proportion of people who relapse will increase with increased amounts of time
after treatment because additional time offers a greater number of opportunities to return to substance use.
Additionally, other factors may change or become more salient over time (e.g., social support, changes in
self-efficacy, and exposure to high-risk situations). In interpreting these findings, however, it is important to
consider that effect sizes representing 1- and 3-month follow-ups were computed based on substantially
fewer studies than effect sizes representing immediate, 6-, and 12-month follow-ups.
In sum, the present analysis supported the overall efficacy of RP in reducing substance use and improving
psychosocial adjustment. Treatment outcome varied with levels of five of the seven moderators that were
investigated (class of substance use disorder, adjunctive use of medication, type of outcome measure, type
of comparison group, and length of posttreatment follow-up). RP was most effective when applied to alcohol
or polysubstance use disorders with the adjunctive use of medication. These moderators of treatment
describe RP at its highest level of efficacy. Overall results, however, suggested that RP was effective across
the board, and outcome did not appear to vary with treatment modality or setting. In interpreting these
findings, it is important to consider their potential clinical significance ( Jacobson & Truax, 1991 ). Although
RP accounted for only a relatively small proportion of variance in actual substance use outcomes, the
tremendously negative consequences of relapse may serve to magnify its clinical importance. Furthermore,
this analysis was based on a very current body of literature, suggesting that the findings are highly applicable
to current practice.

Page 9 of 12
http://spider.apa.org/ftdocs/ccp/1999/august/ccp674563.html 8/30/2000
Although the present meta-analysis provided a precise and statistically powerful picture of the overall
efficacy of RP, certain methodological limitations should be addressed. The differential effectiveness of RP
across class of substance use disorders complicated the findings. Unfortunately, there was an insufficient
number of studies to adequately assess the impact of each moderator within each class of substance use
disorder and other potentially interesting moderators could not be examined. For example, it seems
important to investigate whether specific components of RP are more influential than others in preventing
relapse (e.g., cognitive restructuring vs. behavioral skills training). Further, do certain components lead to
differential effectiveness across classes of substance use disorders in particular treatment settings or
modalities of treatment? Another important question is whether RP is an effective treatment for maintaining
long-term changes over a period of years. Very few studies analyzed in this review collected 1-year follow-
up data, and no studies collected follow-up data after 1 year. Thus, additional research is needed to
investigate these questions.
References
* Peachey, J. E. (1992). The use of calcium carbimide in relapse prevention counselling: Results of a
randomized controlled trial. British Journal of Addiction, 87, 63-72.
* (1990). Outcome evaluation of a relapse prevention and a drug education program with federal
inmates. (Unpublished doctoral dissertation, Florida StateUniversity, Tallahassee.)
* (1988). Relapse prevention and interpersonal psychotherapy: An outcome study of psychotherapies
for cocaine abusers. (Unpublished doctoral dissertation, University ofMinnesota, Twin Cities Campus,
Minneapolis.)
Carroll, K. M. (1996). Relapse prevention as a psychosocial treatment: A review of controlled clinical trials.
Experimental and Clinical Psychopharmacology, 4, 46-54.
* Rounsaville, B. J., Nich, C. & Gordon, L. T. (1994). One-year follow-up of psychotherapy and
pharmacotherapy for cocaine dependence: Delayed emergence of psychotherapy effects. Archives of
General Psychiatry, 51, 989-997.
* O'Leary, M. R. & Marlatt, A. G. (1978). Skill training with alcoholics. Journal ofConsulting and
Clinical Psychology, 46, 1092-1104.
Cooper, H. & Hedges, L. V. (Eds.) (1994). The handbook of research synthesis. (New York: Russell
Sage Foundation)
Curry, S. J., Marlatt, G. A., Gordon, J. & Baer, J. S. (1988). A comparison of alternative theoretical
approaches to smoking cessation and relapse. Health Psychology, 7, 545-556.
* Niaura, R., Follick, M. J. & Abrams, D. B. (1989). Effects of behavioral skills training on schedule of
nicotine gum administration on smoking cessation. American Journal of Psychiatry, 146, 56-60.
* (1984). The efficacy of cognitive—behavioral treatment aimed at relapse prevention in smokers.
(Unpublished doctoral dissertation, Iowa State University,Ames.)
* Rugg, D., Tunstall, C. & Jones, R. T. (1984). Preventing relapse to cigarette smoking by behavioral skill
training. Journal of Consulting and Clinical Psychology, 52, 372-382.
* Catalano, R. F., Gillmore, M. R. & Wells, E. A. (1989). Skills training for drug abusers: Generalization,
maintenance, and effects on drug use. Journal of Consulting and Clinical Psychology, 57, 559-563.
* Catalano, R. F. & Wells, E. A. (1986). Measuring effects of a skills training intervention for drug abusers.
Journal of Consulting and ClinicalPsychology, 54, 661-664.
Hedges, L. V. & Olkin, I. (1985). Statistical methods for meta-analysis. (San Diego, CA: Academic
Press)

Page 10 of 12
http://spider.apa.org/ftdocs/ccp/1999/august/ccp674563.html 8/30/2000
* Rigdon, M. & Johnson, D. (1993). Behavioral smoking cessation treatment for older chronic smokers.
BehaviorTherapy, 24, 321-329.
* Donovan, D. M. & Hall, J. J. (1988). Relapse prevention in alcohol aftercare: Effects on drinking
outcome, change process, and aftercare attendance. BritishJournal of Addiction, 83, 171-181.
Jacobson, N. S. & Truax, P. (1991). Clinical significance: A statistical approach to defining meaningful
change in psychotherapy research. Journal of Consulting and ClinicalPsychology, 59, 12-19.
* Simpson, D. D. & Dansereau, D. F. (1994). Knowledge mapping: A psychoeducational tool in drug
abuse relapse prevention training. Journal of Offender Rehabilitation, 20, (3—4) 187-205.
* Burleston, J. A., Korner, P. K., DelBoca, F. K., Bohn, M. J., Brown, J. & Liebowitz, N. (1995).
Placebo-controlled trial of fluoxetine as an adjunct to relapse prevention in alcoholics. American Journal
ofPsychiatry, 152, 391-397.
Lichtenstein, E. & Glasgow, R. E. (1992). Smoking cessation: What have we learned over the past decade?
Journal ofConsulting and Clinical Psychology, 60, 518-527.
* McKay, J. R. & O'Farrell, T. J. (1995). Relapse precipitants and behavioral marital therapy. Addictive
Behaviors, 20, 383-393.
Marlatt, G. A. & Gordon, J. R. (Eds.) (1985). Relapse prevention: Maintenance strategies in the
treatment of addictive behaviors. (New York: Guilford Press)
Miller, W. R. & Hester, R. K. (1986). Inpatient alcoholism treatment: Who benefits? American
Psychologist, 41, 794-805.
Moos, R. H. & Finney, J. W. (1983). The expanding scope of alcoholism treatment evaluation.
AmericanPsychologist, 38, 1036-1044.
* (1987). Effectiveness of an alcohol relapse prevention program. (Unpublished doctoraldissertation,
Fordham University, Bronx, New York.)
* Choquette, K. A., Cutter, H. S. & Brown, E. D. (1993). Behavioral marital therapy with and without
additional couples relapse prevention sessions for alcoholics and their wives. Journal of Studies on
Alcohol, 54, 652-666.
* Jaffe, A. J., Chang, G., Schottenfeld, R. S., Meyer, R. E. & Rounsaville, B. (1992). Naltrexone and
coping skills therapy for alcohol dependence: A controlled study. Archives of General Psychiatry, 49,
881-887.
* Kearns, W. D., Murrin, M. R., Dolente, A. S. & May, R. L. (1993). Examining the effectiveness of in-jail
substance abuse treatment. Journal ofOffender Rehabilitation, 19, (3—4) 1-39.
Rawson, R. A., Obert, J. L., McCann, M. J. & Marinelli-Casey, P. (1993). Relapse prevention models for
substance abuse treatment. Psychotherapy, 30, 284-298.
Rosenthal, R. (1984). Meta-analytic procedures for social research. (Beverly Hills, CA: Sage)
* Ronnberg, S. (1990). Brief group psychotherapy in relapse prevention for alcohol dependent patients.
International Journal of Group Psychotherapy, 40, 453-476.
Schmitz, J. M., Oswald, L. M., Jacks, S. D., Rustin, T., Rhoades, H. M. & Grabowski, J. (1997). Relapse
prevention treatment for cocaine dependence: Group vs. individual format. Addictive Behaviors, 22, 405-
418.
Sobell, L. C. & Sobell, M. B. (1990). Self-report issues in alcohol abuse: State of the art and future
directions. Behavioral Assessment, 12, 77-90.
* Sobell, L. C. & Gavin, D. R. (1995). Portraying alcohol treatment outcomes: Different yardsticks of
success. Behavior Therapy, 26, 643-669.
* Glasgow, R. E., Hollis, J. F., Lichtenstein, E. & Vogt, T. M. (1993). A smoking-cessation intervention for
hospital patients. Medical Care, 31, 65-72.
* Hollis, J. F. (1989). Preventing smoking relapse, using an individually tailored skills-training technique.
Journal of Consulting and ClinicalPsychology, 57, 420-424.
* Colletti, G. (1984). Relapse coping and problem solving training following treatment for smoking.

Page 11 of 12
http://spider.apa.org/ftdocs/ccp/1999/august/ccp674563.html 8/30/2000
Addictive Behaviors, 9, 401-404.
* Horn, J. L. & Fairchild, D. (1974). Hospital versus community treatment of alcoholism problems.
International Journal of Mental Health, 3, (2—3) 160-176.
Wang, M. C. & Bushman, B. J. (1999). Integrating results through meta-analytical review using SAS
software. (Cary, NC: SAS Institute)
* Peterson, P. L., Gainey, R. R., Hawkins, J. D. & Catalano, R. F. (1994). Outpatient treatment for cocaine
abuse: A controlled comparison of relapse prevention and twelve-step approaches. American Journal of
Drugand Alcohol Abuse, 20, 1-17.
* Brandon, T. H., Jorenby, D. E. & Baker, T. B. (1992). Measures of affect and nicotine dependence
predict differential response to smoking cessation treatments. Journal of Consulting and Clinical
Psychology, 60, 943-952.
Table 1. Studies Included in Meta-Analysis and Effect Sizes of Individual Studies
Figure 1. Dot plot representing effect sizes of individual studies included in the meta-analysis. The asterisk
represents the r value. CI = confidence interval.
Figure 2. Combined effect sizes with upper and lower bound limits of 95% confidence intervals immediately
following treatment and at 1-, 3-, 6-, and 12-month follow-ups.

Page 12 of 12
http://spider.apa.org/ftdocs/ccp/1999/august/ccp674563.html 8/30/2000