1/18/23, 2:55 AM Explaining abstinence rates following treatment for alcohol abuse: A quanti…: GCU Library Resources – All Subjects
NLM Title Abbreviation:
Other Journal Titles:
Explaining abstinence rates following treatment for alcohol abuse: A quantitative synthesis of patient, research design and treatment effects.
Monahan, Susanne C.. Montana State U, Dept of Sociology, Bozeman, US Finney, John W.
Addiction, Vol 91(6), Jun, 1996. pp. 787-805.
United Kingdom : Blackwell Publishing
British Journal of Addiction
United Kingdom : Wiley-Blackwell Publishing Ltd.
0965-2140 (Print) 1360-0443 (Electronic)
treatment & patient & research design characteristics, abstinence rates following alcohol abuse treatment, alcohol abusers
Examined the relationships of treatment, patient, and research design characteristics to treatment outcome (i.e. abstinence rates) in a sample of 150 treatment conditions drawn from 100 alcohol treatment outcome studies published between 1980 and 1992. Treatment characteristics were related to abstinence rates: more intensive treatments had higher abstinence rates than less intensive treatments, whereas treatments with an expressed goal other than abstinence had lower abstinence rates than treatments with an abstinence goal. When the public vs private ownership status of the treatment facility was taken into account, the presence of behavioral elements in the treatment condition also was related to higher abstinence rates. Research design characteristics were also related to abstinence rates. Treatment conditions with shorter follow-ups and treatments drawn from studies that did not use criteria to exclude more impaired subjects had better outcomes. (PsycINFO Database Record (c) 2019 APA, all rights reserved)
*Alcohol Abuse; *Sobriety; *Treatment Outcomes; *Substance Use Treatment
1/18/23, 2:55 AM Explaining abstinence rates following treatment for alcohol abuse: A quanti…: GCU Library Resources – All Subjects
Medical Subject Headings (MeSH):
Digital Object Identifier:
Adult; Alcoholism; Combined Modality Therapy; Female; Follow-Up Studies; Humans; Male; Outcome and Process Assessment (Health Care); Patient Care Team; Patient Selection; Research Design; Social Adjustment; Substance Abuse Treatment Centers; Temperance; United States
Drug & Alcohol Rehabilitation (3383)
Adulthood (18 yrs & older)
Journal; Peer Reviewed Journal
EXPLAINING ABSTINENCE RATES FOLLOWING TREATMENT FOR ALCOHOL ABUSE: A QUANTITATIVE SYNTHESIS OF PATIENT, RESEARCH DESIGN AND TREATMENT EFFECTS
This content may contain URLs/links that would redirect you to a non-EBSCO site. EBSCO does not endorse the accuracy or accessibility of these sites, nor of the content therein. ✖
Section: RESEARCH REPORT
1/18/23, 2:55 AM Explaining abstinence rates following treatment for alcohol abuse: A quanti…: GCU Library Resources – All Subjects
Abstract We examined the relationships of treatment, patient and research design characteristics to treatment outcome (i.e. abstinence rates) in a sample of 150 treatment conditions drawn from 100 alcohol treatment outcome studies published between 1980 and 1992. Treatment characteristics were related to abstinence rates: more intensive treatments had higher abstinence rates than less intensive treatments, whereas treatments with an expressed goal other than abstinence had lower abstinence rates than treatments with an abstinence goal. When the public vs. private ownership status of the treatment facility was taken into account, the presence of behavioral elements in the treatment condition also was related to higher abstinence rates. Because of inconsistent reporting in primary studies, we assessed the effects of only one patient pre-treatment characteristic; treatment conditions with a higher proportion of socially stable patients had better outcomes. Research design characteristics were also related to abstinence rates. Treatment conditions with shorter follow-ups and treatments drawn from studies that did not use criteria to exclude more impaired subjects had better outcomes. We discuss possible reasons why our findings regarding the effects of treatment intensity and the use of exclusionary criteria differ from those in previous reviews.
Introduction The body of research literature on the effectiveness of alcoholism treatment is vast. Emrick (1974) surveyed 271 studies; a later paper from the same project included an additional 126 studies (Emrick, 1975). For a review of pre-1980 studies, Miller & Hester (1980) reported that they spent 6 months reading 600 research reports. In 1986 they noted that an additional 300 treatment reports had been published (Miller & Hester, 1986a). To make sense out of the hundreds of studies on the outcome of alcoholism treatment, reviewers have used a number of strategies to synthesize research findings: narrative or qualitative reviews; semi-quantitative, non-statistical reviews; quantitative analyses of effect sizes; and statistical analyses modeling outcomes.
Until recently, most reviews were qualitative or narrative in nature (see Miller & Hester, 1980; Annis, 1986; Miller & Hester, 1986a; Elkins, 1991; Mackay, Donovan & Marlart, 1991). Reviewers drew conclusions about treatment effectiveness based on both statistical and non-statistical evidence presented in research reports: pairwise comparisons, overall treatment effects in the absence of pairwise comparisons, improvements in treatment groups over time, process analyses, trends in the data and subjective appraisals of study findings.
In an early effort to quantify the findings of primary studies, Emrick (1974) combined data on treatment outcome across 113 studies and found that 4591 of a total of 13 570 patients (33.8%) were reported to be abstinent. Costello and his colleagues (1975a, 1975b; Costello, Biever & Baillargeon, 1977) also employed a semi-quantitative approach to examine factors associated with treatment outcome. They used success rates (e.g. percentage of patients who were abstinent, improved, drinking in a controlled fashion) to cluster studies by outcome and explored whether certain treatment modalities and study characteristics (e.g. exclusion criteria, mandatory treatment) were associated with outcome groups. More recently, Holder et al. (1991) assigned a Weighted Effectiveness Index (WEIn) score to each of 33
treatment modalities, usually based on whether statistical tests of treatment effects in randomized or matched group studies indicated that a particular modality was superior to some alternative. Based on the resulting WEIn scores, they concluded that some treatments are more effective than others (for an adjusted “box-score” approach, see Finney & Monahan, in press).
Analysis of effect sizes is the current method of choice for quantitative research reviews (Cooper, 1989). When certain conditions hold, such analyses are an effective way to synthesize research on treatment effects across studies. In particular, when various treatment modalities are compared with the same treatment or control condition across studies, effect sizes can be calculated that can be meaningfully compared to assess the relative effectiveness of different treatment modalities. For example, in a focused meta-analysis, Bien, Miller & Tonigan (1993) compared the effectiveness of brief interventions for alcohol abuse with that of no treatment and extended interventions.
In the alcoholism treatment outcome literature, however, different treatments are often compared with different alternatives. For example, in an analysis of effect sizes by Mattick & Jarvis (1993), the effectiveness of social skills training was compared with that of group discussions, cognitive restructuring, relaxation training and providing a list of treatment agencies. In contrast, the effectiveness of marital and family therapy was compared with that of individual therapy, cognitive therapy, social skills training, disulfiram and individual counseling and brief advice. The average effect size for social skills training at 12-23 months was 0.78, whereas the average effect size for marital and family therapy at the same follow-up interval was 0.29. However, because the two treatment modalities were compared with different alternatives, it is not possible to interpret the difference in effect sizes. Is the difference due to the relative effectiveness of the two treatments or to differences in the strength of the competition against which they were pitted? Thus, although an analysis of effect sizes is appropriate when treatments are compared with a standard treatment or control condition across studies, that state of affairs does not hold for the body of research on alcoholism treatment outcomes.
Modeling outcomes across studies is an alternative synthesis approach when a research literature is not well suited to the calculation of effect sizes. Both Emrick (1975) and Costello (1980) employed quantitative methods to assess the relationship between treatment for alcohol abuse and its outcome. Using t-tests, Emrick (1975) concluded that treated alcoholics had significantly higher improvement rates than un-treated alcoholics, but not higher abstinence rates.
In a more elaborate analysis, Costello (1980) used path analysis to predict treatment “success rates”. Treatment success was defined as the proportion of patients in a treatment condition whose drinking was non-problematic at follow-up. Costello found that success rates at a 1- to 3-year follow-up were directly affected by patient social stability and directly and indirectly affected by treatment “extensity”. In particular, more comprehensive treatments and treatments provided to more socially stable (i.e. married
and employed) patients had higher success rates, and more comprehensive treatments were more likely to be followed by active aftercare, which in turn was positively associated with success rates.
Researchers frequently report the proportion of patients who were abstinent at follow-up, making the quantitative analysis of such rates across studies a plausible alternative to an effect size analysis. With a quantitative synthesis of abstinence rates, the effects of treatment variation can be examined. In addition, the effects of variation in study and patient characteristics can be determined across studies and controlled when estimating the effects of treatment variables. Finally, because this approach does not depend on within-study comparisons, data from both multiple and single group treatment studies can be included in the synthesis.
In this paper, we follow Costello’s (1980) lead by conducting a quantitative synthesis of treatment outcome findings from studies of alcoholism treatment published between 1980 and 1992. Included in the review are studies that (1) reported treatment group outcome in terms of abstinence, (2) had a minimum follow-up point of at least 3 months and (3) reported data on patients’ marital and/or employment status. Using multivariate methods, we examine patient social stability, research design characteristics and treatment elements that may explain variation in abstinence rates across studies.
Factors that may affect treatment outcome Treatment characteristics
We focused on five characteristics of treatment for alcohol abuse that various researchers (e.g. Costello, 1980; Miller & Hester, 1986a; Holder et al., 1991; Mackay et al., 1991) have identified as important, effective, or commonplace: (1) treatment goal; (2) treatment intensity; (3) presence of behavioral treatment techniques; (4) availability of disulfiram and other antidipsottopic medications; and (5) involvement of family members or significant others in treatment.
Treatment goal. Because abstinence is only one of several possible goals of alcoholism treatment (Sobell, 1978), the treatment goal should be taken into account when attempting to explain abstinence rates across treatment groups. Treatments having a goal of abstinence may result in higher abstinence rates than treatments with alternative goals (e.g. non-problem drinking, reduction in DUI recidivism, self- selected goals). Lower abstinence rates for treatments with non-abstinence goals do not necessarily reflect treatment ineffectiveness; instead, abstinence rates may not be an appropriate index of treatment effectiveness for studies with alternative goals.
Treatment intensity. In their review of brief interventions, Bien et al. (1993) concluded that “brief intervention yields outcomes . . . often comparable to those of more extensive treatment” (p. 332), including inpatient treatment. Furthermore, recent reviews of inpatient treatment have concluded that inpatient treatment is no more effective than outpatient treatment for most alcoholics (Annis, 1986; Miller & Hester, 1986b). However, Finney, Hahn & Moos (in press) noted that although setting effects did not
emerge in studies comparing inpatient treatment with high-intensity outpatient treatment (i.e. day hospital) or with outpatient treatment following inpatient detoxification, inpatient treatment sometimes had superior effectiveness when compared with low-or middle-intensity outpatient treatments not preceded by inpatient detoxification. Thus, setting effects may be confounded with those of treatment intensity. We examine the effects of treatment intensity here by comparing treatments situated in high- intensity settings–where patients spent a large part of the day in treatment–to treatments offered in lower-intensity outpatient settings.
Antidipsotropics. When combined with alcohol, disulfiram and other antidipsotropics, such as Metronidazole and calcium carbimide, cause an adverse physical reaction whose threat is thought to deter alcohol consumption (Miller & Hester, 1986a). Miller and Hester’s (1986a) review identified antidipsotropic medication as a common element of treatment for alcohol abuse, although evidence for its effectiveness is mixed.
Behavioral treatment. Behavioral approaches to the treatment of alcohol abuse focus on changing maladaptive behaviors that contribute to problematic drinking. Therapeutic techniques include social skills training, self-control training, aversion therapies, relapse prevention and community reinforcement treatment (Mackay et al., 1991). Miller & Hester (1986a) noted that although research supports the effectiveness of behavioral approaches (see also Holder et al., 1991), such techniques were not among those commonly used in the treatment of alcoholism.
Family involvement. Because there is often a reciprocal relation between alcoholism and family relationships, several approaches to the treatment of alcoholism incorporate family involvement: family and marital therapy sessions, family support groups, family member participation in outpatient treatment and co-admission of family members to inpatient treatment. Studies by McCrady et al. (1979), Hedberg & Campbell (1974) and Corder, Corder & Laidlaw (1972) found that treatment that included family involvement had better outcomes than treatment without family involvement.
In trying to isolate treatment effects in a quantitative synthesis of the type reported here, it is necessary to control for patient characteristics. Some research indicates that post-treatment functioning is better predicted by patient characteristics than treatment characteristics (see Costello, 1980; Polich, Armor & Braiker, 1981; Moos, Finney & Cronkite, 1990). Some types of patients have a better prognosis than others; in particular, patients who are more socially stable (i.e. more closely tied to social support networks via marriage and/or employment) tend to function better following treatment (Gibbs & Flanagan, 1977; Costello, 1980; Neubuerger et al., 1981).
In research using experimental designs, random assignment of groups should ensure group equivalence with respect to patient characteristics (e.g. age, race, marital status, employment status); given group
equivalence, patient characteristics should not affect within-study comparisons. In comparisons across studies, however, variation in patient pre-treatment characteristics may play an important role in accounting for variation in post-treatment outcome. Thus, our synthesis across studies assesses the effects of patient social stability on the outcome of treatment for alcoholism.
Research design characteristics
We also focus on several features of research design that may affect abstinence rates: (1) use of criteria to exclude more impaired subjects; (2) varying definitions of abstinence; (3) length of the follow-up window; and (4) follow-up rates.
Exclusion criteria. The use of criteria to exclude more impaired patients from studies may affect treatment outcome. Studies that exclude more impaired or harder-to-treat patients may have higher success rates (Costello, 1975a, 1975b).
Definition of abstinence. In some studies, the definition of abstinence allows “slips”. Studies with strict no-drinking definitions of abstinence may have lower abstinence rates than studies that define abstinence to include slips.
Follow-up interval There is considerable debate regarding how long patients should be followed after the termination of treatment. Some researchers argue that 3 months is appropriate because the risk of relapse is greatest during that period (Hunt, Barnett & Branch, 1971; Mackay et al., 1991). Others, however, argue for follow-up intervals of at least 1 year in order to assess the longer-term impact of treatment (see Nathan & Lansky, 1978; Sobell et al., 1987). Most researchers assume, however, that the relative influence of treatment decreases over time as does its rate of success, at least for several years following treatment (Emrick, 1982; but see Mc-Crady et al., 1991).
Follow-up rate. The proportion of patients successfully followed may affect data on treatment outcome. Studies that follow a higher proportion of patients may report lower abstinence rates because they locate worse-off patients who may be missed by studies with lower follow-up rates (Moos & Bliss, 1978; Polich et al., 1981; Sobell, Sobell & Maisto, 1984; cf LaPorte et al., 1981).
Moderators of treatment effects
Theories of patient-treatment matching specify that patient characteristics moderate the effects of treatment on outcome (Project MATCH Research Group, 1993). That is, certain types of patients may do better in certain types of treatment whereas other types of patients may do better in other types of treatment. We examine whether patient social stability moderates the effects of treatment for alcoholism: that is, whether some treatments are more effective for socially stable patients whereas others have greater effectiveness for less socially stable patients. For example, Welte et al. (1981) found that
treatments of longer duration were more effective for less socially stable subjects, whereas treatment length of stay was unrelated to outcome for more socially stable subjects. Although individual primary studies have examined the matching of individual patients with treatments (e.g. Annis & Chan, 1979; Dorus et al., 1989; Rohensow et al., 1991), matching has not been addressed at the treatment group level across a broad set of primary studies.
In this paper, we use quantitative methods to synthesize findings from alcoholism treatment groups included in outcome studies published between 1980 and 1992. In particular, we focus on the relationship of treatment, patient and research design characteristics to treatment outcome (i.e. abstinence rates). After controlling for patient and research design characteristics in hierarchical regression analyses, we examine the effects of different treatment elements on patient post-treatment abstinence rates.
The sample of studies used in the present analysis was drawn from a larger review (339 studies) of alcoholism treatment outcome studies that were published in English between 1980-92 (for a description of that sample, see Floyd et al., in press). Studies were identified using Medline, ETOH, PsychInfo, PsychLit and Dissertation Abstracts, as well as by examining the reference lists of review articles and primary studies published after 1980. Studies included multiple-group, comparative studies; multiple- group, non-comparative studies (i.e. studies that did not use randomization, matching or statistical adjustments to adjust for patient pre-treatment differences); and single-group studies. For the present analyses we selected all studies that: (1) reported, for at least one treatment condition or group, the proportion of patients who were abstinent; (2) at a minimum follow-up of 3 months; and (3) also reported the proportion of study participants who either were married and/or employed. Thirty-one per cent of the research projects (n = 108) met the inclusion criteria.
The treatment condition, rather than the study, was the unit of analysis. The sample of 108 projects reported data on 172 treatment conditions: 65 projects reported one treatment condition, 26 reported two treatment conditions, 13 reported three treatment conditions, and four reported four treatment conditions. We excluded 12 of the treatment conditions because they comprised no- or minimal- treatment control groups. We also excluded 10 of the remaining treatment conditions because inpatient treatment was provided to some patients and outpatient treatment to others but data for each group were not reported separately. Because eight of those 10 conditions were from single-group studies, we were left with 100 studies reporting data on 150 treatment conditions. A list of studies included in this review is available from the authors upon request.
We computed chi statistics to determine if our study selection criteria (i.e. a minimum 3-month follow-up interval, data on marital and/or employment status, outcome data in the form of abstinence rates) systematically selected certain types of studies. The analyses indicated that we selected a disproportionate number of studies of high-intensity treatment compared with lower-intensity treatments (chi =45.51, df= 1, p<0.0001), a disproportionate number of studies conducted by non-university- affiliated researchers compared with university-affiliated researchers (chi =9.14, df= 1, p<0.01), a disproportionate number of studies that included family involvement (chi = 4.11), and disproportionate number of studies that used criteria to exclude more impaired subjects (chi =4.00). Studies of high- intensity treatments were more likely to report abstinence rates and to have a follow-up point of 3 months or longer than were studies of lower-intensity treatments. Studies conducted by non-university- affiliated investigators were more likely to report abstinence rates than were studies by university- affiliated investigators. Studies with family involvement were more likely to have a follow-up point of 3 months or longer than were studies with no family involvement. Finally, studies that used criteria to exclude more impaired patients were more likely to report marital and/or employment status (from which we computed the social stability index) than were studies that did not use exclusion criteria. The chi analyses indicated that our sample was not biased with respect to the provision of disulfiram or behavioral treatment, or the use of an alternative treatment goal, nor was the sample biased in terms of the educational attainment of the principal investigator.
Thus, the findings from our sample of studies should not be generalized to all alcoholism treatment outcome studies. Our sample consists of a more limited group of studies that report abstinence rates at a minimum 3-month follow-up and basic patient pre-treatment characteristics. Although our selection criteria limit the representativeness of the sample, our selection criteria were as liberal as possible, while still allowing a reasonable synthesis to be conducted. If more studies reported a comparable outcome at a minimally informative follow-up point and also reported patient pretreatment characteristics, quantitative syntheses such as this could include a broader range of studies.
Because we excluded a disproportionate number of studies by university-based investigators we were concerned that the methodological quality might be lower in the studies in our review. However, using an preliminary index developed by Morley et al. (1995) we found that the studies we included in our review were of significantly higher quality than those that were excluded.
The criterion variable was the proportion of abstinent patients, taken at the first follow-up point of 3 months or longer (i.e. if data were reported for a 3-month follow-up and a 6-month follow-up, we used 3- month follow-up data).
We included three sets of predictor variables in the multiple regression models: treatment characteristics, patient characteristics and research design characteristics. Treatment variables included
measures of treatment goal, intensity and content.
Alternative treatment goal. If treatment had an expressed goal other than abstinence (e.g., non-problem drinking, patient-selected goal), a score of ‘1’ was assigned. If the goal of treatment was abstinence, or was not explicitly indicated, a score of ‘0’ was assigned. We assumed that if a treatment condition had a goal other than abstinence, the alternative goal would be reported. Thus, those studies that did not explicitly report treatment goal were assumed to have abstinence-rather than an alternative–as their goal. We analyzed the data excluding those treatment conditions for which the goal was not explicitly stated; the results did not differ substantially from the results of analyses that included treatment conditions for which the goal was not explicitly stated.
Treatment intensity. In constructing a measure of treatment intensity based on treatment setting, we assumed that: (1) inpatient, residential, day hospital and halfway house settings provided high-intensity treatments; (2) treatment that included an inpatient component of 2 weeks or longer constituted high- intensity treatment; and (3) outpatient settings other than day clinics provided lower-intensity treatment. High-intensity treatments received a score of ‘1’, whereas lower-intensity treatments received a score of ‘0’.
Treatment conditions coded as high intensity provided more treatment to patients than treatment conditions coded as low intensity. Using information on the duration of treatment in the primary reports, we calculated the average number of days of inpatient, residential or day-hospital treatment in high- intensity settings and the average number of hours of treatment in outpatient settings. Data on the number of days of treatment was available for 97 of the 111 high-intensity conditions; high-intensity treatments averaged 34.6 days of treatment (SD = 21.2). Assuming 6 hours per day of treatment on weekdays and no treatment at weekends, high-intensity treatments averaged 148 hours of treatment. Data on hours of treatment was available for 24 of the 39 low-intensity conditions; low-intensity treatments averaged 14.1 hours of treatment (SD = 20.0).
We characterized treatment elements using three dichotomous variables indicating the presence or absence of certain components in the treatment description. Treatment descriptions varied widely in the detail provided. For example, 16 treatment conditions provided no information on the treatment provided except for the setting (i.e. inpatient, outpatient) of the treatment. If a component was not reported, we assumed that it was not provided.
Antidipsotropics. Antidipsotropics was scored ‘1’ if disulfiram, calcium carbimide or Metronidazole was prescribed or made available to patients, and ‘0’ if the availability of those medications was not mentioned in the treatment description.
Behavioral therapy. Behavioral therapy was scored ‘1’ if the treatment was described as having a behavioral orientation or if any of the following modalities was included in the treatment description:
aversion therapy, covert sensitization, behavior contracting, behavioral marital therapy, community reinforcement therapy, relapse prevention, cognitive therapy, self-control training, social skills training, stress management training, relaxation training, assertion training, systematic desensitization. In the absence of any of these modalities, a score of ‘0’ was given to the treatment condition.
Family involvement. Family involvement was scored ‘1’ if the treatment was described as involving active family participation, and ‘0’ if no family participation was described.
Because of the incomplete reporting of patient data, we were only able to include one measure of patient characteristics.
Patient social stability. Social stability was computed as the maximum of either the proportion of patients who were married or the proportion who were employed at treatment entry. Thus, if a treatment condition reported marital but not employment status of patients, the social stability index would be the proportion of married patients; if the treatment condition reported employment but not marital status, the social stability index would be the proportion of employed patients. If both were reported, the social stability index was the greater of the two proportions.
We included four variables related to research design.
(1) Abstinence definition. We included a dichotomous variable with ‘1’ indicating that drinking (e.g. slips) was permitted in the definition of abstinence; otherwise, a score of ‘0’ was assigned.
(2) Exclusion criteria. Treatment conditions were assigned a score of ‘l’ if criteria were used to exclude more impaired patients (e.g. those with psychiatric diagnoses, organic brain disorder, impaired cognitive functioning, severe dependence symptoms) from the study, and ‘0’ if more impaired patients were not excluded from the study. We viewed the use of criteria to exclude more impaired patients as a rough measure of the range of patient impairment for each treatment condition: that is, we thought that treatment programs that did not explicitly exclude more impaired patients would be more likely to treat such patients than would more exclusive treatment programs.
(3) Follow-up interval. Follow-up interval was measured in months. For treatments with an inpatient component and other high-intensity treatments, length of follow-up was measured from discharge; for treatments without an inpatient component (i.e. outpatient therapies, drug treatments), follow-up was measured from the initiation of treatment. Length of follow-up for day-hospital treatment was measured from the initiation of treatment because patients were released to their homes only during evenings and at weekends, thus making day-hospital a high-intensity form of outpatient care. We coded the first follow-up point of 3 months or longer for which a study reported abstinence data. Because this variable had a skewed distribution, we used a log transformation of the follow-up interval in the analyses. We
originally planned to include the duration of assessed abstinence in the analyses. It was excluded, however, because it was highly correlated (r = 0.79) with the length of the follow-up interval.
(4) Follow-up rate. Follow-up rate was computed as the number of patients followed, divided by the number of patients who started treatment. Thirty-four treatment conditions did not report the number of patients starting treatment (i.e. they were drawn from retrospective studies); in those cases, we assigned the mean follow-up rate of the other studies (75%). We conducted additional analyses excluding those treatment conditions for which the follow-up rate was unavailable; the results were not substantially different from the analyses that included treatment conditions for which we estimated the follow-up rate.
Reliability of coding. We assessed the reliability of measures by double-coding a randomly drawn set of 26 studies. Correlations were computed for interval-level variables: proportion abstinent, proportion non- problem drinkers, proportion socially stable, follow-up interval and follow-up rate. Correlations between original coding and reliability coding ranged from 0.90 (proportion socially stable) to 0.96 (proportion abstinent). Percentage of agreement was computed for the dichotomous variables: goal, intensity, behavioral treatment, antidipsotropics, family involvement, exclusion criteria and abstinence definition. Goal, intensity and antidipsotropics had more than 95% agreement. Percentage of agreement was somewhat lower for the other variables: 74% for abstinence definition, 81% for family involvement and 85% for exclusion criteria and behavioral treatment.
Data were analyzed using ordinary least squares (OLS) regression. A basic assumption of OLS regression (i.e. independent cases) was violated by our sample because we drew more than one treatment condition from some of the projects. To assess the possible effects of non-independent cases, we compared the results of analyses of the total sample of treatment conditions with those of a smaller set of independent cases that included all single-group studies and one randomly selected treatment condition from each multiple-group study (see Buchanan, Maccoby & Dombusch, 1991 for a discussion of the treatment of non-random cases). The complete sample had 150 treatment conditions; the reduced sample had 100 treatment conditions. Descriptive data on both samples are presented in Table 1; no significant differences were found between the samples. We discuss the results of analyses of the independent cases when they differ in statistical significance or magnitude from the analyses for the sample as a whole.
Results In all, 27 407 patients were treated in the 150 treatment conditions drawn from 100 studies. Treatment groups ranged in size from nine to more than 8000 patients. The average number of patients in each group was 183 patients (SD = 694) and the median number of patients was 53. Abstinence rates ranged
from 0 to 91% across treatment conditions, and averaged 43%. In the average treatment condition, approximately 60% of the patients were employed and/ or married.
The average follow-up point was 10.6 months, although there was wide variation ranging from 3 to 96 months. The variable measuring the follow-up interval was bimodal, with one mode at 3 months (40 treatment conditions) and another mode at 6 months (36 treatment conditions).
Fifty-eight per cent of the treatment conditions used criteria to exclude more impaired patients from the study. Slips were included in the definition of abstinence in only 11% of the treatment conditions.
Most of the treatment conditions (74%) took place in high-intensity settings. In only a few (9%) of the treatment conditions was it indicated that antidipsotropics were available to patients. About a fifth of the treatment conditions had family involvement, and just over a quarter used behavioral treatment elements or a behavioral orientation. Only 5% of the treatment conditions had a treatment goal other than abstinence.
As Table 2 indicates, the zero-order correlations among the predictor variables were generally low (ranging from – 0.15 to 0.15). Although some were statistically significant at p<0.05, none was large enough to present a problem with multi-collinearity in the multivariate analyses. In general, the correlations among the variables in the sample of independent cases were similar to those of the sample as a whole.
By using hierarchical regression analyses to predict abstinence rates across studies we were able to assess the cumulative effects of each set of predictor variables and control for patient and research design characteristics before examining treatment effects. We entered patient characteristics, then research design characteristics and, finally, treatment characteristics. Within each class of predictors, the independent variables were entered simultaneously. Table 3 presents the parameters estimated in the final step of the analyses predicting abstinence rates in the total sample of treatment conditions and the randomly drawn set of independent cases.
Main effects of predictor variables
The social stability index was entered into the model first; it explained 4% of the variance in abstinence rates. When the variables associated with research design were added to the model, explained variance increased to 17% (F = 5.56, p = 0.0003) The addition of the final set of variables, those related to treatment characteristics, increased the explained variance of the model to 34% (F = 7.49, p < 0.0001).
Patient social stability. Patient social stability was positively associated with abstinence rates. Controlling for the other predictors, treatment conditions with a higher proportion of socially stable patients reported
higher abstinence rates at follow-up; an increase of four percentage points in the proportion of socially stable patients was associated with an increase of about one percentage point in abstinence rate.
Research design characteristics. Abstinence rates were predicted by three of the four research design characteristics: follow-up interval, the use of exclusionary criteria and follow-up rate. The follow-up interval was negatively associated with abstinence rates. Because log transformations do not produce linear relationships, the interpretation of this parameter estimate is not straightforward; but, for example, the predicted abstinence rate of a group followed 12 months after treatment was, on average, seven percentage points lower than that of a group followed 3 months after treatment. Initially we predicted abstinence rates using the untransformed follow-up interval; the coefficient estimate for the untransformed follow-up interval was not statistically significant (B = 0.17, SE = 0.12) and, overall, the model did not fit as well as the model predicting abstinence rates with the log transformation of follow-up interval (overall F = 6.54, R = 0.32, adjusted R = 0.27 versus overall F =7.25, R =0.34, adjusted R = 0.30). Thus, it seems reasonable to assume that over time abstinence rates decrease at a decreasing rate.
Surprisingly, studies that used criteria to exclude more impaired patients had lower abstinence rates–by an average of 10 percentage points. The negative relationship persisted even when the definition of exclusionary criteria was expanded to include the exclusion of less-impaired subjects: that is, we found that treatment conditions in studies using any inclusion or exclusion criteria reported lower abstinence rates at follow-up (r = – 0.30, p = 0.003) than conditions in studies that did not use such criteria. We suspected that the finding regarding exclusionary criteria was a spurious relationship driven by research quality: that is, higher quality studies may be more likely to use (or report their use of) inclusion and exclusion criteria, and also may have more accurate (i.e. uninfiated) reports of success rates. Our analyses indicated that higher quality studies reported lower abstinence rates: the correlation between methodological quality and abstinence rates was negative and statistically significant (r = – 0.17, p = 0.03). In addition, there was a borderline relationship between methodological quality and the use of exclusion criteria (r = -0.14, p=0.08), indicating that higher quality studies were more likely to report the use of exclusionary criteria. These relations are weak, however, and do not entirely account for the unexpected association between exclusionary criteria and success rates.
As expected, studies with higher follow-up rates had lower abstinence rates. For each additional five percentage points in follow-up rates, abstinence rates decreased by about one percentage point. The definition of abstinence–that is, whether slips were permitted–was not associated with abstinence rates.
Treatment characteristics. Of the treatment-related variables, intensity and alternative goal were significant predictors of abstinence rates. Treatment conditions in high-intensity settings (e.g. inpatient, residential, halfway house, dayclinic) had higher abstinence rates–15 percentage points higher–than conditions in lower-intensity, outpatient settings, Controlling for the other predictors, abstinence rates for treatments with an abstinence goal averaged 26 percentage points higher than rates for treatments with
raw 2 2
log 2 2
alternative goals. Family involvement, the use of behavioral elements and the use of antidipsotropics were not associated with abstinence rates.
Analysis of independent cases. Some of these relationships were weaker in the sample of independent cases: the parameter estimates for social stability and follow-up rates were consistent in their direction but were not statistically significant. The magnitude of the parameter estimate for social stability decreased by almost one-half in the sample of independent cases, while the standard error increased slightly. The magnitude of the parameter estimate for follow-up rate also decreased in the sample of independent cases–although by only about 15%-and its standard error was also slightly higher.
Weighted regression analysis. In the analyses reported thus far each of the 150 treatment conditions contributed equally to the findings, irrespective of the number of patients in the treatment condition. We also analyzed the data using regression weighted by the number of patients in each treatment condition. The statistical significance and the magnitude of the effects were not substantially different from the un- weighted analyses, except that in the weighted regression, treatment conditions that included family involvement had predicted abstinence rates that were on average 10 percentage points less than treatment conditions that did not include family involvement (B = – 10.5, SE = 4.6, t = – 2.28, p < 0.05).
Additional analyses. Given the results of previous studies that found little difference between high- and low-intensity treatments (see a review by Miller & Hester, 1986b), the positive relation between high- intensity treatment and abstinence rates was unexpected. An anonymous reviewer suggested that the high-intensity treatment conditions in our sample may have been drawn predominantly from private, for- profit, inpatient treatment facilities that de facto exclude individuals with poor prognoses for treatment outcome. This exclusion process would explain the relationship we found between treatment intensity and treatment outcome: high-intensity treatment conditions in private facilities generally treat patients with good prognoses. A concentration of high-intensity treatment conditions in private programs would also explain the perplexing negative relationship between the use of exclusion criteria and abstinence rates: private treatment facilities tend not to be accessible to more impaired patients, thus obviating the need for criteria to exclude such patients from treatment and studies of that treatment.
To explore this interpretation our findings, we coded whether each treatment condition occurred in: (1) a private, for-profit treatment program; (2) a public (e.g. VA Hospital; other military hospital; state, county or city hospital) or non-profit treatment program; or (3) an unidentified treatment program. Sixty-four of the treatment conditions occurred in public or non-profit settings (mean abstinence rate = 36.5%, SD = 21.1), whereas 65 of the treatment conditions were set in private, for-profit treatment facilities (mean abstinence rate=49.7%, SD=22.1). We were unable to determine the facility status of 21 treatment conditions; 17 of those were in treatment facilities outside the United States. The mean abstinence rate for the unidentified treatment conditions was 40.9%. Thus, we concluded that we had a reasonable
distribution of public/non-profit and private, for-profit programs in our sample; in addition, the unidentified treatment programs had a mean abstinence rate more typical of the public than the private programs.
We examined the zero-order correlations between treatment facility type (public facilities were coded ‘0’ and private facilities were coded ‘1’) and the predictor and criterion variables; some of the correlations were statistically significant at p<0.05. Conditions in private treatment facilities had patients who were more socially stable (r=0.20), and–as indicated above–reported higher abstinence rates (r= 0.29) following treatment; in addition, private treatment facilities were: (1) less likely to use behavioral treatment elements (r= -0.21); (2) less likely to include slips in the definition of abstinence (r = – 0.20); and (3) less likely to use criteria to exclude more impaired patients (r = -0.23).
Because facility type (i.e. public or private) might be an important predictor of treatment success rates and confounded with some of the other predictors, we added the facility-type variable to the multiple regression analysis predicting abstinence rates for the 129 treatment conditions for which the facility- type variable was available. Because this sample excludes 21 treatment conditions for which we could not determine the facility type, we present in Table 4 a re-analysis of the original model for the sample of 129 cases next to the augmented model that includes the facility type variable: the magnitudes of the slope estimates and their degree of statistical significance varies little across the full- and the reduced- sample models.
Facility type was a significant predictor of abstinence rates following treatment for alcohol abuse. Controlling for the other predictors, private programs reported abstinence rates that were an average of 10 percentage points higher than rates reported by public treatment pro-grams. The inclusion of the facility-type variable did not, however, alter the findings with respect to (1) the positive relationship between high-in-tensity treatment and abstinence rates; or (2) the negative relation between the use of criteria to exclude more impaired patients and abstinence rates. High-intensity treatment conditions reported higher abstinence rates than low-intensity conditions, even controlling for facility type. Similarly, even controlling for facility type, conditions that excluded more impaired patients reported lower abstinence rates than programs that did not report the use of exclusion criteria.
Behavioral treatment emerged as a significant predictor of abstinence rates in the regression model that included the facility-type variable. Treatment conditions that included behavioral elements had higher abstinence rates–by about nine percentage points–following treatment than did treatments without behavioral elements.
Alternative measures of treatment outcome. We explored the possibility of using a measure of drinking- related outcome other than abstinence rate (non-problem drinking rate). Non-problem drinking rates were reported by only 24 of the eligible studies, and by only six studies that did not also report abstinence rates. Separate analyses of non-problem drinking rates were thus not feasible because of the lack of statistical power to detect significant effects. When we combined abstinence and non-
problem drinking rates in the 156 studies to form a measure of treatment “success”, we found that treatment success rate was highly correlated with abstinence rate (r= 0.91). The results of the multivariate analyses of success rate differed in only one respect from the analyses of abstinence rates: treatment conditions that used behavioral elements reported significantly higher success rates–by about 7 percentage points–at follow-up (B = 7.2, SE = 3.5, t = 2.04, p < 0.05) than did conditions not including behavioral elements. Of the outcome variables widely available in this literature, abstinence and non- problem drinking rates were the most prevalent. We did not analyze other measures of treatment success or improvement because such measures had low cross-study reliability.
Moderators of treatment effects
Finally, we explored whether the social stability of patients moderated the effects of treatment on abstinence rates. Using mean-centered deviation scores (see Jaccard et al., 1990; Aiken & West, 1991), we computed product terms to model potential interactions between social stability and each of the five treatment characteristics. None of the product terms was a significant predictor of abstinence rates.
Discussion Using quantitative methods, we synthesized findings from 100 studies of alcoholism treatment outcome published between 1980 and 1992. We focused on studies that reported the social stability of patients and abstinence rates at a minimum follow-up of 3 months.
Only 31% of the studies published between 1980 and 1992 met the three inclusion criteria for this synthesis. Our focus on abstinence rates allowed us to maximize the number of studies included in the analysis, while minimizing the error inherent in having too broad a definition of treatment “success”. Although we explored the possibility of examining a success measure that combined abstinence and non-problem drinking rates, fewer than 10% of all studies reported the proportion of non-problem drinkers, and most of those studies also reported abstinence rates. Thus, the expansion of the outcome measure to include non-problem drinking added only six studies to our analysis and provided similar results. Had we focused on outcome measures other than abstinence or non-problem drinking rates (e.g. consumption measures), few studies could have been analyzed. Optimally, all alcoholism treatment outcomes studies would provide quantitative information on alcohol consumption. In lieu of that, research syntheses would be facilitated if those researchers who report quantitative measures of alcohol consumption also reported abstinence and/or non-problem drinking rates. Although some researchers may be reluctant to report abstinence rates if abstinence was not the goal of treatment, the present study demonstrates that it is possible for research syntheses to control for the effects of alternative treatment goals on abstinence rates.
We found that variation in abstinence rates was systematically related to patient, research design and treatment characteristics. Our results indicate that treatment is related to treatment outcome, although not necessarily in ways suggested by past research. Most striking was the strong relationship between
the intensity of treatment and abstinence rates. Previous reviews have concluded that brief interventions are as effective as more intensive interventions (Bien et al., 1993) and outpatient treatment is as effective as inpatient treatment (Annis, 1986; Miller & Hester, 1986b). However, we found that high- intensity-mostly inpatient–treatment conditions had higher abstinence rates than less intensive– predominantly outpatient–treatment conditions, after controlling for other variables including patient social stability.
Several factors may account for the discrepancy between our findings and those cited in previous reviews. We examined the plausible explanation that treatment intensity, the use of exclusion criteria, and abstinence rates were all related to facility type. That is, private, for-profit facilities may be less likely to use exclusionary criteria and tend to provide higher intensity treatment. However, they also attract patients with better prognoses. That may be the reason they have higher abstinence rates following treatment. Zero-order correlations indicated that private programs had higher abstinence rates and were less likely to use exclusion criteria; facility type was not related to treatment intensity. When facility type was added to the multiple regression model, it was a significant predictor of abstinence rates (i.e. patients in private facilities had higher abstinence rates), but it did not affect the positive relationship between high-intensity treatment and abstinence rates. Even controlling for facility type, high intensity treatments had higher abstinence rates than lower intensity treatments; nor did the addition of facility type to the multiple regression model affect the negative relation between the use of exclusionary criteria and abstinence rates. Even controlling for facility type, treatment conditions that used exclusionary criteria had lower abstinence rates than those that did not use exclusionary criteria.
Because we did not limit our review to randomized clinical trials, many patients in the studies we reviewed selected a particular treatment rather than being randomly assigned to it. In fact, the majority of the studies in our review were inpatient studies where patients presented for and received inpatient treatment. Such patients may differ along important dimensions from those who seek other types of treatment (Skinner, 1981; Timko et al., 1994). Although undoubtedly many patient characteristics affect treatment selection, we controlled for only three: patient social stability, patient impairment (indirectly, by adjusting for the use of criteria to exclude more-impaired patients from the study) and patient’s ability to pay for treatment (indirectly, by adjusting for the public/private nature of the treatment facility). Patients who select high-intensity treatment may experience better outcomes than those who: (1) present for outpatient treatment (Timko et al., 1993); (2) do not have a preference for what kind of treatment they receive (i.e. are willing to accept random assignment to any treatment–a preselection criterion in most randomized clinical trials); or (3) receive treatment they were not expecting.
Thus, the surprising finding regarding treatment intensity might have resulted from confounding of treatment and patient effects. That is, the finding may reflect a treatment effect, but it may also be a function of patient characteristics: better prognosis patients may be drawn to high-intensity treatment. The same is true of the findings with respect to treatment facility type: private treatment is not necessarily “better” than public treatment, even though its post-treatment abstinence rates are higher. Rather, private treatment may be a prow for patient characteristics, in that patients with a better
prognosis may be drawn to private treatment. Thus, although we can conclude that treatment groups exposed to certain treatment elements have higher post-treatment abstinence rates than treatment groups not exposed to those elements, characteristics of members of the treatment group—rather than the treatment itself–may be the causal factor in the relationship between the treatment received and post-treatment abstinence rates. Our analyses, however, controlled for a number of factors that represent potentially influential patient characteristics (i.e. patient social stability, patient impairment, public/private treatment facility). In spite of this, the positive effects of high-intensity treatment persisted.
The difference in the sample of studies we examined may also partly explain the divergence between our findings and those of earlier reviews. Most narrative and meta-analytic reviews use the study as the unit of analysis and focus on comparisons between treatment conditions within a study. By focusing on treatment condition abstinence rates rather than paired comparisons, we were able to incorporate studies that did not explicitly compare higher-intensity with lower-intensity treatments (e.g. multiple- group studies of variations in inpatient treatment and single-group studies of inpatient treatment).
Finally, the difference in intensity between the high- and low-intensity conditions we examined is greater than that evaluated in some of the relevant randomized controlled trials. For example, Edwards et al. (1977) compared extended treatment with a single session of “advice”. During the first year of the study, however, the advice group received additional treatment whereas the second group was given planned extended treatment on an “as needed” basis. After 1 year, the average member of the extended treatment group had received three times as many hours of treatment as the average member of the advice group (97 hours vs. 30 hours). In contrast, the high-intensity conditions in our sample received on average 10 times as many hours of treatment as low intensity conditions (148 hours vs. 14 hours).
In their review of the research literature on setting effects, Finney et al. (in press) note that inpatient treatment has sometimes been compared with intensive outpatient treatment (e.g. McLachlan & Stein, 1982; Longabaugh et al., 1983; McKay et al., 1995). Even in those cases where inpatient treatment has been compared with less intensive outpatient treatment, the number of hours of treatment generally did not differ by a factor of 10 (but see Eriksen, 1986; Chick, 1988), as it did in the present analysis. Thus, the effects of treatment intensity may have emerged in our analysis because of the striking difference between high-and low-intensity treatments in the number of hours of treatment provided.
Other treatment-related factors were also related to outcome. Although only about 5% of the treatment conditions had an alternative goal (e.g. non-problem drinking, self-selected goal), those conditions had significantly lower abstinence rates than treatment conditions with an abstinence or unspecified goal. Ideally, treatment outcome should be assessed along dimensions that treatment is supposed to affect. For example, one would expect treatment with a non-problem drinking goal to affect non-problem drinking rates but not necessarily abstinence rates. Thus, in our analysis of abstinence rates across studies, we controlled for the goal of treatment. Nonetheless, we were struck by the small number of studies that explicitly examined treatment goals other than abstinence. Our emphasis on abstinence rates did not, however, exclude many studies with non-abstinence goals: only two studies (with six
treatment conditions) that (1) reported non-problem drinking rates but not abstinence rates and (2) met the other inclusion criteria for this study had alternative treatment goals.
Consistent with reviews by Holder et al. (1991) and Finney & Monahan (in press), we found that the type of treatment makes a difference. In the analysis that included public/private treatment facility, treatments that included behavioral elements had higher abstinence rates. This relation is striking because behavioral treatments were positively correlated with the use of an alternative treatment goal, and the use of an alternative treatment goal was negatively correlated with abstinence rates. Thus, although one might expect behavioral treatments to be associated with lower abstinence rates (in conjunction with higher non-problem drinking rates), behavioral treatment was associated with higher abstinence rates. Also, we did not distinguish between behavioral approaches that have more or less evidence of effectiveness (Holder et al., 1991; Finney & Monahan, in press). The other measures of treatment content–antidipsotropics and family involvement–were not associated with treatment outcome, however.
Should we infer that, with the exception of behavioral approaches, type of treatment is unrelated to treatment outcome? Probably not. First, there are treatment elements that we did not examine (e.g. AA, aftercare). More importantly, we suspect that our measures of treatment content may have lacked validity because they were derived from descriptions of treatment included in primary reports. For many studies it was difficult to code treatment content with confidence. As we reported earlier, 16 treatment conditions did not include any information about treatment except for its setting. The descriptions of many other treatment conditions were vague or stressed a single element in a multi-modal treatment program. If a treatment element was not mentioned we assumed it was not used; thus, we may have underestimated the prevalence of various treatment elements. There was no way to know whether a treatment description was complete; nonetheless, the more detailed the description, the more confidence we had coding the treatment elements examined here. In addition, we simply coded whether or not a treatment element was reported; we did not code the amount of that type of treatment or the relative importance of each element in the entire treatment package. Nor did we have any information on the quality of treatment implementation.
The inadequacy of our measures of treatment content may have also affected our analysis of the moderating effects of social stability on the relation between treatment and outcome. Clear and complete descriptions of treatment in primary reports would not only facilitate an understanding of the study’s findings, but also the integration of those findings into the wider body of research on alcoholism treatment outcomes.
Primary reports also did not provide consistent data on patients’ pre-treatment characteristics. Due to the spottiness of such data we were able to assess the effects of only one patient pre-treat-ment variable: patient social stability. Consistent with reviews of within-study findings at the individual patient level (e.g. Gibbs & Flanagan, 1977), we found that treatment conditions with a higher proportion of married and/or employed subjects had higher abstinence rates, although the magnitude of the effect decreased to a non-significant level in the random sample of independent cases. Because of insufficient
reporting of the relevant data, we were unable to include patient pre-treatment severity of alcohol abuse in our models. If researchers were required by journal editors or funding sources to report such data in a standardized manner (e.g. years of problem drinking, number of dependence symptoms, diagnostic status), the effects of patient pre-treat-ment severity could be assessed in future syntheses.
In the absence of a direct measure of patient pre-treatment severity, we included a dichotomous measure indicating if more-impaired patients were excluded from the study. We initially thought that studies excluding more impaired subjects would pre-select people with a better prognosis and would thus have higher abstinence rates. We found, however, that treatment conditions in studies that used exclusion criteria had significantly lower abstinence rates. It is possible that more impaired patients who have “hit bottom” may be more motivated to abstain after treatment for alcohol abuse. In addition, there was some evidence that the use of exclusion criteria and reports of treatment outcome were both related to study quality: that is, higher quality studies were more likely to use exclusion criteria and to have lower abstinence rates.
The length of the follow-up interval was also associated with treatment outcome. Abstinence rates decreased over time following treatment. Although not unexpected, this finding has an alternative explanation: it may be that the length of time over which abstinence was assessed (i.e. the window of observation), and not the length of time after treatment at which follow-up data were collected (i.e. follow-up interval), accounts for this relation. For example, one would expect higher rates if abstinence were based on no drinking over the 3 months immediately preceding a 12-month follow-up than if it were based on no drinking over the entire 12 months of the follow-up interval. There was, however, a high correlation between the length of follow-up interval and the length of observation window for determining abstinence (r = 0.79). It is difficult, if not impossible, to disentangle the independent effects of two such highly correlated variables.
The consolidation of research findings is a vital part of the research process (Cooper, 1989). The results of primary studies represent pieces of a puzzle; when those pieces are assembled through a process of synthesis, a more complete picture of research findings emerges. Qualitative, semi-quantitative and quantitative reviews play different roles in the process of synthesis. A synthesis modeling outcomes, such as the one conducted here, does not depend upon within-study comparisons to draw conclusions about the effectiveness of treatment, and thus can incorporate a wider range of treatment studies (e.g. single-group studies, non-comparative studies) that might otherwise not be integrated into the body of research on alcoholism treatment.
Perhaps, as a result, the present quantitative synthesis revealed some surprising findings, most notably that more intensive treatment had better outcomes than less intensive treatment. Future primary studies can determine whether differences in treatment intensity of the magnitude examined here are linked to differences in abstinence rates and other outcome variables.
Acknowledgements This research was supported by NIAAA Grant No. AA08689, by the Department of Veterans Affairs (VA) Health Services Research and Development Service, and by the VA Mental Health and Behavioral Sciences Service. A number of people made important contributions to the research reported in this paper. Annette Hahn began working on the study at its inception, was involved in the development of initial coding forms and was instrumental in helping to get the project “off the ground”. She, along with Kent Harber, Michelle Pearl, Anthony Floyd, Jeanne Bart Morley and Jennifer Noke, conducted electronic searches of bibliographic databases, coded studies, key-entered data, and/or ran statistical analyses. We are also grateful to Keith Humphreys, Rudolf Moos, Hams Cooper and three anonymous reviewers for their helpful comments on earlier drafts of this paper.
Table 1. Descriptive statistics for the total sample and the
sample of independent cases
All cases independent cases
(n = 150) (n = 100)
Mean proportion abstinent (SD) 42.9 (22.6) 43.9 (22.8)
mean proportion socially stable 60.1 (24.4) 61.2 (23.7)
Research design characteristics
mean follow-up point in months (SD) 10.6 (12.4) 11.4 (13.2)
mean log (follow-up point) (SD) 0.86 (0.35) 0.89 (0.36)
mean follow-up rate 0.75 (0.16) 0.75 (0.16)
% that allowed slips in
the definition of abstinence 11 12
% excluding more impaired patients 58 56
% with a non-abstinence goal 5 6
% high-intensity treatments 74 77
% offering or using antidipsotropics 9 10
% using behavioral treatment 28 23
% using family involvement 21 22
Table 2. Zero-order correlations among criterion and predictor
variables (n = 150 treatment conditions)
Legend For Chart:
A – Variable
B – (1)
C – (2)
D – (3)
E – (4)
F – (5)
G – (6)
H – (7)
I – (8)
J – (9)
K – (10)
B C D E F
G H I J K
(1) % Abstinent
— — — — —
— — — — —
(2) % Socially stable
0.19[a] — — — —
— — — — —
(3) Follow-up interval (log)
-0.13 0.14 — — —
— — — — —
(4) Follow-up rate
-0.12 0.17[a] 0.10 — —
— — — — —
(5) Slips allowed
-0.09 -0.10 -0.02 -0.10 —
— — — — —
(6) Exclusion criteria
-0.30[b] -0.13 -0.07 -0.03 0.09
— — — — —
(7) Alternative goal
-0.29[b] 0.09 0.01 0.07 0.10
0.14 — — — —
0.32[b] -0.02 0.12 0.06 0.07
-0.07 -0.06—- — — —
-0.06 0.10 0.22[b] 0.05 -0.04
0.13 -0.08 -0.07 — —
(10) Behavioral treatment
0.05 -0.13 -0.13 -0.08 0.10
0.08 0.12 0.07 -0.05 —
(11) Family involvement
0.02 0.07 0.05 -0.04 -0.03
0.10 0.10 0.08 0.06 -0.02
a p < 0.05; b p < 0.01.
Table 3. Results of multiple regression predicting abstinence
rates with treatment, patient and research design
Predictor variables (all cases, n = 150) cases, n = 100)
Social stability 0.23[c] 0.13
Research design characteristics
Follow-up interval (log) -12.30[b] -13.02[a]
Follow-up rate -0.20[a] -0.17
Definition of abstinence -4.45 -8.18
Exclusion of more
impaired patients -10.49[b] -12.36[b]
Treatment goal other
than abstinence -25.60[c] -28.95[c]
Treatment intensity 16.39[d] 16.87[c]
Antidipsotropics 0.16 4.17
Behavioral treatment 4.07 3.56
Family involvement 1.83 2.64
Constant 49.31[d] 55.06[d]
Re 0.34 0.38
Adjusted R 0.30 0.31
a p < 0.05; b p < 0.01; c p < 0.001; d p < 0.0001.
Table 4. Results of multiple regression predicting abstinence
rates with treatment, patient, research design
characteristics and treatment type
Predictor variables (n = 129) n = 129)
Social stability 0.25[c] 0.23[b]
Research design characteristics
Follow-up interval (log) -14.00[b] -14.68[b]
Follow-up rate -0.24[a] -0.23[a]
Definition of abstinence -5.18 -2.60
Exclusion of more impaired patients -11.72[b] -9.91[b]
Treatment goal other
than abstinence -26.41[b] 24.82[a]
Treatment intensity 15.06[c] 15.08[d]
Antidipsotropics -0.61 1.74
Behavioral treatment 7.36[e] 8.99[a]
Family involvement 6.11 5.69
Private treatment 9.98 [b]
Constant 51.57[d] 45.79[d]
R 0.36 0.412
Adjusted R 0.31 0.35
1 Excluding those conditions for which public/private treatment type could not be determined. e p< 0.10; a p<O.05; b p<O.01; c p< 0.001; d p< 0.0001.
References AIKEN, L. S. & WEST, S.G. (1991) Multiple Regression: testing and interpreting interactions (Newbury Park, CA, Sage).
ANNIS, H. M. (1986) Is inpatient rehabilitation of the alcoholic cost effective? Con position, Advances in Alcohol and Substance Abuse, 5, 175-190.
ANNIS, H. M. & CHAN, D. (1979) The differential treatment model: empirical evidence from a personality typology of adult offenders, Criminal Justice and Behavior, 10, 159-173.
BIEN, T. H., MILLER, W. R. & TONIGAN, J. S. (1993) Brief interventions for alcohol problems: a review, Addiction, 88, 315-336.
BUCHANAN, C. B., MACCOBY, E. E. & DORNBUSCH, S. M. (1991) Caught between parents: adolescents’ experience in divorced homes, Child Development, 62, 1008-1029.
CHICK, J., RITSON, B., CONNAUGHTON, J., STEWART, A. & CHICK, J. (1988) Advice versus extended treatment for alcoholism: a controlled study, British Journal of Addiction, 83, 159-170.
COOPER, n.M. (1989) Integrating Research: A Guide for Literature Reviews, 2nd erin (Newbury Park, CA, Sage Publications).
CORDER, B. F., COP, DER, R. F. & LAIDAW, N. D. (1972) An intensive treatment program for alcoholics and their wives, Quarterly Journal of Studies on Alcohol, 33, 1144-1146.
COSTELLO, R. M. (1975a) Alcoholism treatment and evaluation: in search of methods, International Journal of the Addictions, 10, 251-275.
COSTELLO, R. M. (1975b) Alcoholism treatment evaluation: in search of methods, II. Collation of two- year follow-up studies, International Journal of the Addictions, 10, 857-867.
COSTELLO, R. M. (1980) Alcoholism treatment effectiveness: slicing the outcome variance pie, in: EDWARDS, G. & GRAnt, M. (Eds) Alcoholism Treatment in Transition, pp. 113-127 (Baltimore, MD,
University Park Press).
COSTELLO, R. M., BIEVER, P. & BAILLARGEON, J. G. (1977) Alcoholism treatment programming: historical trends and modern approaches, Alcoholism: Clinical and Experimental Research, 1, 311-318.
DORUS, W., OSTROW, D.G., ANTON, H. et al. (1989) Lithium treatment of depressed and non- depressed alcoholics, Journal of the American Medical Association, 262, 1646-1652.
EDWARDS, G., ORFORD, J., EGERT, S. et al. (1977) Alcoholism: a controlled trial of ‘treatment’ and ‘advice’, Journal of Studies on Alcohol, 32, 1004-1031.
ELKINS, R. L. (1991) An appraisal of chemical aversion (emetic therapy) approaches to alcoholism treatment, Behavioral Research and Therapy, 29, 387-413.
EMRICK, C. D. (1974) A review of psychologically oriented treatment of alcoholism: I. The use and interrelationships of outcome criteria and drinking behavior following treatment, Quarterly Journal of Studies on Alcohol, 35, 523-549.
EMRICK, C. D. (1975) A review of psychologically oriented treatment of alcoholism: II. The relative effectiveness of different treatment approaches and the effectiveness of treatment versus no treatment, Journal of Studies on Alcohol, 36, 88-108.
EMRICK, C. D. (1979) Perspectives in clinical research: relative effectiveness of alcohol abuse treatment, Family and Community Health, 2 (Part II), 71-88.
EMRICK, C. D. (1982) Evaluation of alcoholism psychotherapy methods, in: PATTISON, E. M. & KAUFMAN, E. (Eds) Encyclopedic Handbook of Alcoholism, pp. 1152-1169 (New York, Gardner Press).
EMRICK, C. D. & HASEN, J. (1983) Assertions regarding effectiveness of treatment for alcoholism: fact or fantasy? American Psychologist, 38, 1078-1088.
ERIKSEN, L. (1986) The effect of waiting for alcoholism treatment after detoxification: an experimental comparison between inpatient treatment and advice only, Addictive Behaviors, 11, 389-397.
FINNEY, J. W. & MONAHAN, S.C. (in press) The cost effectiveness of treatment for alcoholism: a second approximation, Journal of Studies on Alcohol.
FINNEY, J. W., HAHN, A. C. & MOOS, R. H. (in press) The effectiveness of inpatient and outpatient treatment for alcohol abuse: the need to focus on mediators and moderators of setting and moderators
of setting effects, Addiction.
FLOYD, A., MONAHAN, S.C., FINNEY, J. W. & MORLEY, J. B. (in press) The nature of alcoholism treatment outcome research, 1980-1992, Addictive Behaviors.
FULLER, R. K., BRANCHEY, L., BRIGHTWELL, D. R. et al. (1986) Disulfiram treatment of alcoholism: a Veterans Administration cooperative study, Journal of the American Medical Association, 256, 1449- 1455.
GIBBS, L. & FLANAGAN, J. (1977) Prognostic indicators of alcoholism treatment outcome, International Journal of the Addictions, 12, 1097-1141.
HEDBERG, A. G. & CAMPBELL, L. III (1974) A comparison of four behavioral treatments of alcoholism, Journal of Behavioral Therapy and Experimental Psychiatry, 5, 251-256.
HOLDER, H., LONGABAUGH, R., MILLER, W. R. & RUBONIS, A. V. (1991) The cost effectiveness of treatment for alcoholism: a first approximation, Journal of Studies on Alcohol, 52, 517-540.
HUNT, W. A., BARNETT, L. W. & BRACH, L. G. (1971) Relapse rates in addiction programs, Journal of Clinical Psychology, 27, 455-456.
JACCARD, J., WAN, C. K. & TURRISI, R. (1990) The detection and interpretation of interaction effects between continuous variables in multiple regression, Multivariate Behavioral Research, 25, 467-478.
LA PORTE, D. J., McLELLAN, A. T., ERLDEN, F. R. & PARENTE, R. J. (1981) Treatment outcome as a function of follow-up difficulty in substance abusers, Journal of Consulting and Clinical Psychology, 49, 112119.
LONGABAUGH, R., McCRADY, B., FINK, E., STOUT, e., McALULEY, T., DOYLE, C. & McNEILL, D. (1983) Cost effectiveness of alcoholism treatment in partial vs. inpatient settings: six month outcomes, Journal of Studies on Alcohol, 44, 1049-1071.
MACKAY, P. W, DONOVAN, D. M. & MARLATT, G. A. (1991) Cognitive and behavioral approaches to alcohol abuse, in: FRANCES, K. & MILLER, S. (Eds) Clinical Textbook of Addictive Disorders, pp. 452- 481 (New York, Guilford).
MATFICK, R. P. & JARVIS, T. J. (Eds) (1993) An Outline for the Management of Alcohol Problems: quality assurance project (National Drug and Alcohol Research Center, The University of New South
McCRADY, B. S., PAOLINO, T. J. JR., LONGABAUGH, R. & Rossi, J. (1979) Effects of joint hospital admission and couples treatment for hospitalized alcoholics: a pilot study, Addictive Behaviors, 4, 115- 165.
MCCRADY, B. S., STOUT, R., NOEL, N., ABRAMS, D. & NELSON, H. F. (1991) Effectiveness of three types of spouse-involved behavioral alcoholism treatment, British Journal of Addiction, 86, 1415-1424.
McKAY, J. R., ALTERMAN, A. I., MCLELLAN, A. T., SNIDER, E. C., & O’BRIEN, C. P. (1995) Effect of random versus non-random assignment in a comparison of inpatient and day hospital rehabilitation for male alcoholics, Journal of Clinical and Consulting Psychology, 63, 70-78.
McLACHLAN, J. F. C. & STERN, R. L. (1982) Evaluation of a day clinic for alcoholics, Journal of Studies on Alcohol, 43, pp. 261-272.
MILLER, W. R. & HESTER, R. K. (1980) Treating the problem drinkers: modem approaches, in: MILLER, W. R. (Ed.) The Addictive Behaviors: treatment of alcoholism, drug abuse, smoking and obesity, pp. 11 141 (Elmsford, NY, Pergamon Press, Inc.).
MILLER, W. R. & HESTER, R. m. (1986a) The effectiveness of alcoholism treatment: what research reveals, in: MILLER, W. R. & HEATHER, N. (Eds) Treating Addictive Behaviors: processes of change, pp. 121-174 (New York, Plenum Press).
MILLER, W. R. & HESTER, R. K. (1986b) Inpatient alcoholism treatment: who benefits? American Psychologist, 41, 794-805.
MOOS, R. H. & BLISS, F. (1978) Difficulty of follow-up and outcome in alcoholism treatment, Journal of Studies on Alcohol, 39, 473-490.
Moos, R. H., FINNEY, J. W. & CRONKITE, R. C. (1990) Alcoholism Treatment: context, process, and outcome (New York, Oxford University Press).
MORLEY, J. B., FINNEY, J. W., MONAHAN, S.C. & FLOYD, A. (in press) The quality of alcoholism treatment outcome research, 1980-1992, Addictive Behaviors.
NATHAN, P. E. & LANSKY, D. (1978) Common methodological problems in research on the addictions, Journal of Consulting and Clinical Psychology, 46, 713-726.
NEUBUERGER, O., HASHA, N., MATARAZZO, J. D., SCHMITZ, R. E. & PRATT, H. H. (1981) Behavioral-chemical treatment of alcoholism: an outcome replication, Journal of Studies on Alcohol, 42, 806-810.
POLICH, I. M., ARMOR, D. J. & BEAIKER, H. B. (1981) The Course of Alcoholism: four years after treatment (New York, Wiley).
PROJECT MATCH RESEARCH GROUP (1993) Project MATCH: rationale and methods for a multisite clinical trial matching patients to alcoholism treatment, Alcoholism: Clinical and Experimental Research, 17, 1130-1145.
ROHENSOW, D. J., MONTI, P.M., BINKOFF, J. A., LIEPMAN, M. R. & NIRENBERG, T. D. (1991) Patient-treatment matching for alcoholic men in communication skills versus cognitive-behavioral mood management training, Addictive Behaviors, 16, 63-69.
SKINNER, H. A. (1981) Different strokes for different folks: differential treatment for alcohol abuse, in: Evaluation of the Alcoholic: Implications for Research, Theory and Treatment, Research Monograph No 5, pp. 349-367 (Rockville, MD, NIAAA).
SOBELL, M. B. (1978) Goals in the treatment of alcohol problems, American Journal of Drug and Alcohol Abuse, 5, 283-291.
SOBELL, M. B., BROCHU, S., SOBELL, L. C., ROY, J. & STEVENS, J. A. (1987) Alcohol treatment outcome evaluation methodology: state of the art, 1980-1984, Addictive Behaviors, 2, 113-128.
SOBELL, L. C., SOBELL, M. B. & MAISTO, S. a. (1984) Follow-up attrition in alcohol treatment studies: is “no news” bad news, good news, or no news? Drug and Alcohol Dependence, 13, 1-7.
TIMKO, C., FINNEY, J. W., Moos, R. H., Moos, B. S. & STEINEAUM, D. P. (1993) The process of treatment selection among previously untreated help-seeking problem drinkers, Journal of Substance Abuse, 5, 203220.
TIMKO, C., MOOS, R. H., FINNEY, J. W., & MOOS, B. S. (1994) Outcome of treatment for alcohol abuse and involvement in AA among previously untreated problem drinkers, Journal of Mental Health Administration, 21, 145-160.
WELTE, J., HYNES, G., SOKOLOW, L. & LYONS, J.P. (1981) Effect of length of stay in alcoholism treatment on outcome, Journal of Studies on Alcohol, 42, 483-491.
By SUSANNE C. MONAHAN & JOHN W. FINNEY
Center for Health Care Evaluation, VA-Palo Alto Health Care System and Stanford University Medical Center, USA