Reduction of Alcohol Consumption by Brief Alcohol Intervention in Primary Care: Systematic Review and Meta-analysis

Background  Numerous trials of the efficacy of brief alcohol intervention have been conducted in various settings among individuals with a wide range of alcohol disorders. Nevertheless, the efficacy of the intervention is likely to be influenced by the context. We evaluated the evidence of efficacy of brief alcohol interventions aimed at reducing long-term alcohol use and related harm in individuals attending primary care facilities but not seeking help for alcohol-related problems.

Methods  We selected randomized trials reporting at least 1 outcome related to alcohol consumption conducted in outpatients who were actively attending primary care centers or seeing providers. Data sources were the Cochrane Central Register of Controlled Trials, MEDLINE, PsycINFO, ISI Web of Science, ETOH database, and bibliographies of retrieved references and previous reviews. Study selection and data abstraction were performed independently and in duplicate. We assessed the validity of the studies and performed a meta-analysis of studies reporting alcohol consumption at 6 or 12 months of follow-up.

Results  We examined 19 trials that included 5639 individuals. Seventeen trials reported a measure of alcohol consumption, of which 8 reported a significant effect of intervention. The adjusted intention-to-treat analysis showed a mean pooled difference of −38 g of ethanol (approximately 4 drinks) per week (95% confidence interval, −51 to −24g/wk) in favor of the brief alcohol intervention group. Evidence of other outcome measures was inconclusive.

Conclusion  Focusing on patients in primary care, our systematic review and meta-analysis indicated that brief alcohol intervention is effective in reducing alcohol consumption at 6 and 12 months.

Excessive alcohol use is a major public health concern¹ associated with an increased risk of morbidity and mortality.² Numerous trials of the efficacy of brief alcohol intervention (BAI) have been conducted in various settings among individuals with a wide range of alcohol disorders. It has been suggested that BAI has the potential to reduce serious adverse consequences of alcohol consumption.³ Between 1993 and 2002, 8 systematic reviews^4-11 including between 7 and 56 studies were published; 4 of these aggregated data in a meta-analysis.^5,9-11

While BAI applies to various settings, the efficacy of the intervention is likely to be influenced by the context. Admission to an emergency department or hospital differs from an opportunistically delivered intervention among primary care patients. Furthermore, because primary care encounters include a large proportion of the community and because of a low threshold of access, it is important to focus on BAI conducted among primary care patients. Several projects aimed at implementing brief alcohol counseling in primary care are ongoing (eg, the World Health Organization brief intervention project, phase IV¹²). Although conditions for optimal implementation strategies of BAI are yet to be determined,^13,14 a summary of the efficacy of BAI in primary care on non–treatment-seeking patients is needed.

The purpose of this systematic review was to evaluate the evidence of efficacy of BAI aimed at reducing long-term alcohol use and related harm among individuals actively attending primary health care centers but not seeking help for alcohol problems.

Studies included in this review were drawn from different sources: (1) the Cochrane Central Register of Controlled Trials, MEDLINE, and PsycINFO from inception to January 2003, completed with searches in the ISI Web of Science and ETOH databases; (2) bibliographies of retrieved references and previous reviews; and (3) our own bibliographic resources. No restrictions were applied to language or year of publication.

Eligible studies were randomized controlled trials reporting at least 1 outcome related to change in alcohol intake, drinking status, health-related quality of life or functional status, laboratory markers related to alcohol use, utilization of health care resources, or cost data.

Only studies conducted in outpatients who were actively attending a primary health care center or provider were considered. Studies involving alcohol treatment–seeking patients were excluded. We considered individuals who responded to advertisements or who were referred for alcohol treatment to be alcohol treatment seekers. Studies conducted in a hospital ward or in an emergency department were not eligible. Studies that selected patients by means of registers or patient lists or that specifically convened individuals for alcohol screening were also excluded. At least 75% of the study population consisted of primary care patients; for samples with less than 75% primary care patients, subanalyses must have been performed according to population type.

Eligible interventions were as follows: (1) intervention delivered individually that focused on alcohol consumption with a face-to-face component during the initial session, and (2) intervention defined as “brief intervention” or “motivational intervention” or reporting the use of feedback or advice to reduce alcohol consumption. No restrictions were applied to repeated interventions or reinforcement sessions.

Two of us (N.B. and B.B.) reviewed the titles and abstracts independently and in duplicate. Abstracts were selected for full text reading whenever selected by 1 or 2 authors. The same 2 then read the preselected articles independently and in duplicate, and applied inclusion and exclusion criteria. When no consensus was obtained, articles that were not fully rejected or accepted were discussed and submitted to a third author (J.-B.D.).

For each selected article, 2 of us alternatively performed data abstraction independently and in duplicate. Disagreements over the abstraction were discussed and submitted to a third author (N.B., B.B., then J.-B.D., or N.B., J.-B.D., then B.B.) when necessary.

We analyzed several components of treatment and control interventions (contents, provider, duration, and repetition), outcomes, and methods (design and inclusion and exclusion criteria) to understand and evaluate the impact of these elements on our results. Two of us (N.B. and B.B.) assessed trial quality independently and in duplicate and obtained a quality score (by consensus when needed) for each trial by using an instrument adapted from Ferri et al.¹⁵ We added elements derived from the Consolidated Standards of Reporting Trials statement¹⁶ and from Jüni et al.¹⁷ For each trial, a quality score ranging from 0 (low) to 18 (high) was assigned. The instrument evaluated randomization, concealment of allocation, blinding in assessment of outcomes, attrition during follow-up, presence of an intention-to-treat analysis, and additional criteria (clear definition of intervention, selection and performance bias, and presence of a measure of intervention exposure).

In the framework of this systematic review, a meta-analysis was deemed feasible only for the principal outcome of alcohol consumption. Studies reporting alcohol consumption or differences in alcohol consumption at 6 or 12 months of follow-up containing corresponding confidence intervals (CIs), standard deviations, or standard errors at baseline and other follow-up measures were included in the meta-analysis. The magnitude of effect (effect size) was expressed in terms of mean net reduction in alcohol consumption (in grams of ethanol per week) observed in the BAI group compared with the control group. For example, an effect size of −10 g means that, on average, subjects in the BAI group reduced their weekly alcohol consumption by 10 g more than the controls. When data on weekly alcohol consumption were missing (eg, no description of the content of a standard drink in absolute alcohol or description of corresponding equivalences of various alcoholic beverages such as beer, wine, or spirits), we used the definition of the absolute ethanol content in a standard drink used in the country where the study was performed, according to Babor and Higgins-Biddle¹⁸ and Dufour.¹⁹ The variation between countries and studies of a “standard” drink’s ethanol content is also the reason for expressing results in grams of ethanol and not in drinks.

To obtain an intention-to-treat analysis across all studies, we considered all randomized individuals. We chose a conservative option and assigned the mean group baseline alcohol consumption value to randomized individuals excluded from the analyses at follow-up. For studies reporting change in alcohol consumption, missing individuals were counted as showing no change.

Meta-analysis was performed by using the commands “metan” and “meta” of the statistical software Stata 8 (Stata Corp, College Station, Tex). A random effects model was used to examine the heterogeneity between the studies and combined their results. The overall pooled effect was estimated by the weighted average of individual study effects by means of the DerSimonian and Laird weighting method. Between-studies heterogeneity in outcome was tested using the Cochran χ²Q statistic. The I² measure proposed by Higgins et al²⁰ was used to estimate the percentage of total variation in outcome that is due to heterogeneity rather than chance. The extent to which study-level variables explained heterogeneity in the effect size was explored by fitting metaregression models. The following variables were considered: year of study publication, sex of study population, mean alcohol consumption in the BAI group at baseline, length of follow-up, response rate to follow-up, duration of intervention, technique used in intervention, score of methodologic quality, and subscore related to quality of randomization, allocation concealment, and blinding. Publication bias was examined by means of the Begg and Mazumdar adjusted rank correlation test²¹ and the Egger regression asymmetry test.²²

A total of 612 references were obtained after exclusion of redundant references from various sources, from which 135 reports were selected for detailed evaluation. The selection process is summarized in Figure 1. Of the 24 selected articles, all but 1 was retrieved for further evaluation during the abstract selection process by one reviewer (N.B.) and all but 1 by the other (B.B.) (agreement, 92%; κ = 0.65).

The 24 retained reports represent the results of 19 trials^23-46: 9 conducted in North America, 7 in Europe, 2 in Africa, and 1 in Australia. Two reports were published in Spanish, 1 in French, and 21 in English. Four reports were related to Project TrEAT (Trial for Early Alcohol Treatment),^30,32,33,40 1 was an ancillary report connected to the study conducted by Senft et al⁴⁴ that focused on health care utilization,³⁴ and 2 reported results of men and women separately.^23,24 Inclusion and exclusion criteria are summarized in Table 1 and Table 2. All but 4 (26%)^27,44-46 of the 19 selected studies used customary alcohol consumption as an inclusion criterion, and 14 studies (74%) excluded alcohol-dependent individuals.

Study size ranged from 80 to 774 subjects, with a total of 5639. Thirteen trials (74%) included subjects of both sexes, 3 (16%) were males only, and 3 (16%) were initially conducted among male and females individuals but females were not included in the analyses.^25,26,38 One study focused specifically on elderly patients (≥65 years old).³¹

Three trials were cluster randomized,^28,36,42 1 was randomly allocated to groups in weekly blocks,⁴³ and the remaining 15 were individually randomized. Two of the cluster-randomized trials^28,42 used the primary care center as the unit of randomization and the patient as the unit of analysis. The third one³⁶ used the physician as the unit of randomization. The length of follow-up ranged from 6 to 48 months.

Treatment and control interventions are summarized in Table 3. The reported descriptions (eg, content and duration) varied across these studies. All selected studies included some advice being given to the BAI groups, and all except 1³⁶ reported the use of feedback regarding alcohol consumption levels and/or adverse effects of alcohol consumption. Six studies made explicit reference to the Motivational Interview⁴⁷ by Miller and Rollnick,^{23,24,28,39,42-44} 2 studies^37,41 to the cognitive-behavioral techniques developed by Sanchez-Craig,⁴⁸ and 3 studies^30,31,36 to the intervention developed by Wallace et al.⁴⁹ The length of intervention ranged from 5 to 45 minutes. The BAI was repeated or included a booster session in 10 studies, and in 4 additional studies a follow-up visit was offered to participants. The control intervention in 6 studies consisted of up to 5 minutes of advice. The remaining 13 studies had no intervention or usual care as the control group.

Summary results of the systematic review are presented in Table 4. Follow-up rates ranged from 31.5% to 92.4% (mean, 72.4%; median, 77.2%). Nine (47%) of 19 studies had an attrition rate of less than 20%. The methodologic quality scores ranged from 5³⁶ to 14,^30,39 with a mean of 9.6 and a median of 9.0. Ten points was arbitrarily determined to be the cutoff between “high-quality” and “low-quality” studies. A low-quality rating was principally related to inadequate randomization or insufficient reporting of concealment of allocation. Only 2 of the selected studies^39,46 sufficiently reported concealment of allocation, whereas most of the remaining studies either reported it only partially, or documented inadequate concealment of allocation. High-quality studies were more likely than low-quality studies to report statistically significant positive effects of intervention (χ² = 3.9, P = .048).

Among 17 trials reporting a quantified outcome measure of alcohol consumption, 8 reported statistically significant effects of BAI and 7 reported no significant effects. None of the studies reported negative effects of BAI. Two trials reported significant effects on secondary measures (number of drinking days per month³⁸ and usual drinking amount per occasion among women²³) and nonsignificant principal quantity-frequency outcomes. Meta-analysis was restricted to those 10 trials for which the reduction of alcohol consumption could be calculated to assess the efficacy of the intervention.

In studies reporting results at both 6 and 12 months of follow-up, the net reduction in alcohol consumption was similar at 6 and 12 months (intention-to-treat analysis, P = .91). Similarly, only a small difference in effect size (intention-to-treat analysis, P = .75) was found between men and women in studies reporting sex differences. Therefore, meta-analysis was performed on all studies reporting data with either 6 or 12 months of follow-up (12 months if both) independent of sex. For the 3 trials that evaluated 2 different intervention modalities, each intervention group was considered separately to examine its contribution to the effect size. Intention-to-treat analyses were possible for only 12 of 13 intervention groups because the number of subjects randomized in each group at baseline was not reported in 1 study.²⁸ The forest plot of the effect sizes observed in the 12 trials and subtrials included in the meta-analysis is shown in Figure 2. Based on follow-up observations without adjustment for dropout, the overall pooled effect size of the trials consisted of a net change of −50 g of ethanol (about 5 drinks) per week (95% CI, −65 to −34). In relative terms, this net change of −50 g of ethanol corresponds to an additional relative mean reduction of 15% in alcohol consumption when the intervention groups (relative mean reduction, 34%) are compared with the control groups (relative mean reduction, 19%). When only trials for which intention-to-treat analysis was possible are considered, the unadjusted pooled effect size was similar (−47 g/wk; 95% CI, −62 to −31 g/wk) and the adjusted effect size after correction for lost subjects was smaller (−38 g/wk; 95% CI, −51 to −24 g/wk) yet still significantly different from 0. Between-studies heterogeneity in effect size was not significant according to the Cochran χ² test (Q₁₂ = 15.1, P = .24). The measure of heterogeneity recommended by Higgins et al²⁰ indicated a low level of heterogeneity accounting for about one fourth of the total variance in outcome (I² = 25.8%). In the intention-to-treat analysis, the heterogeneity was smaller (Q₁₁ = 6.7, P = .82; I² = 0%).

There was no evidence of publication bias according to either the Begg and Mazumdar adjusted rank correlation test (P for null bias = .58) or the Egger regression asymmetry test (P = .66). Although no correlation was found between the publication year of the studies and effect size by meta-regression (P = .85), a significantly greater effect size was found in trials published after 1996 than in those published earlier (effect, −54.8 vs 6.6; P = .02). Cumulative meta-analysis according to date of publication (Figure 3) clearly displays this time trend in scientific evidence from absence of effects toward statistically significant favorable effects of BAI.

There was no linear association between effect size and quality scores according to meta-regression using quality scores as an explanatory variable (per protocol analysis, P = .31). However, all high-quality trials (score >10) were from the United States, and baseline BAI mean alcohol consumption was less than 300 g/wk. By contrast, all low-quality trials were not from the United States and baseline BAI mean alcohol consumption was greater than 300 g/wk. The heterogeneity was higher among low-quality studies (Q = 12.8; P = .08; I² = 82.9%) than among high-quality studies (Q = 2.1; P = .71; I² = 0%). Meta-regression models showed that the correlation between mean alcohol consumption at baseline and effect size was weak (P = .71) in high-quality trials, but very strong (P<.001) in low-quality studies. When both a dichotomous indicator of low vs high quality and the initial drinking level were included in meta-regression, each variable had an independent and significant contribution to effect size. A high-quality study increased effect size by an additional reduction of −110 g/wk (95% CI, −208 to −11 g/wk; P<.03) in alcohol consumption at follow-up, while each increment of 10 g/wk in mean alcohol consumption at baseline corresponded to an additional reduction of −6.0 g/wk (95% CI, −11.3 to −0.6 g/wk; P = .03) at follow-up. These 2 variables explained about 67% of the between-trials heterogeneity in effect size.

Meta-regression on intervention modalities (type of provider, duration, motivational technique used, written material, and repeated intervention) showed only a minor impact on effect size.

Eight studies used laboratory values as indicators of reduction in alcohol use, including 7 with γ-glutamyltransferase,^{23,24,26,27,35-37,43} 5 with mean corpuscular volume,^{23,24,27,35,36,45} 3 with aspartate aminotransferase and alanine aminotransferase,^23,24,26,27 and 1 with carbohydrate-deficient transferrin.^23,24 Two trials^26,37 reported a statistically significant difference in improvement of γ-glutamyl transferase values between BAI and control groups. One reported an improvement of aspartate aminotransferase²⁷ and one an improvement of alanine aminotransferase.²³ The remaining studies failed to report a significant effect of BAI on laboratory values compared with control groups.

Three studies^32-34,37 reported health care utilization measures to test the impact of BAI on health care costs. In an analysis conducted in a subsample of 29 patients, Israel et al³⁷ found a lower number of medical visits by the BAI group than the control group. In Project TrEAT, self-reports by patients indicated no significant group differences in emergency department visits during 6 and 12 months after BAI. However, there were significantly fewer emergency department visits at 48 months and fewer hospital days in the BAI group at 6, 12, and 48 months.³³ Freeborn et al,³⁴ using an administrative database from a health maintenance organization, found no significant differences in health care utilization during the 2 years after BAI. Fleming et al^32,33 performed a cost-benefit analysis, where the benefits of reduced hospitalizations, reduced emergency department visits, and reduced motor vehicle or criminal events were put in monetary terms and then compared with the overall cost of the screening, assessment, and intervention. Project TrEAT showed a benefit-cost ratio for brief alcohol intervention of 4.3:1 from a medical perspective and 39:1 from a societal perspective. The net benefits per intervention patient are $546 from a medical perspective and $7780 from a societal perspective (US dollars, 1993).

Nine studies^{23-27,35,37,38,43,46} reported data related to mental or physical health perception status, well-being, and alcohol-related problems. There were significant differences among 9 of the 21 measures reported, showing an improved or better quality of life in the BAI group than the control group. Group differences in the other 12 measures were not significant. One study³³ examined mortality and found a significant reduction in the number of deaths at 36 months (P<.05) in the BAI vs the control group. This effect disappeared by 48 months.

We selected 19 trials for this systematic review of the effects of BAI on at-risk drinkers in primary care. The important feature of this review was the selective inclusion of patients who came actively to receive usual primary health care. Except for alcohol consumption, we chose not to pool data. Laboratory results, health care utilization outcomes, and mortality results were reported in only a few studies. The definitions and measures of outcomes such as binge drinking, well-being, and problems related to alcohol drinking were too heterogeneous to allow results to be pooled.

Our systematic review and meta-analysis indicated that BAI is effective for both men and women in reducing alcohol consumption at 6 and 12 months. This is supported by the homogeneity of selected studies and by the clear effect shown in studies of high quality. In the study with the longest follow-up period, this effect was also reported 36 and 48 months after BAI when a cumulative analysis was performed.³³ This is consistent with the results of previously conducted systematic reviews that examined the efficacy of BAI conducted in various settings^5,7-9,11,50 and reported comparable effect sizes. Among the published systematic reviews of BAI, 1 focused partially on primary care but did not conclude that BAI was effective because statistical heterogeneity existed among studies.¹⁰ Several trials conducted in primary care have been published after its completion. Our review focused strictly on primary care, and this restriction might explain why we did not find statistical heterogeneity among studies. A cumulative meta-analysis according to date of publication demonstrated that the BAI effects have shown significance since 1997, which could explain the uncertain results reported in the meta-analysis conducted by Poikolainen.¹⁰ Beich et al¹⁴ recently published a systematic review and meta-analysis of the effectiveness of screening in brief intervention trials conducted in general practice. Most of the focus was on the effectiveness of screening, but they assessed some effects of intervention as well. They reported that only 2 or 3 patients per 1000 screened benefited from BAI, but it seems important to point out that their research was not designed primarily to evaluate the efficacy of BAI but rather to assess the screening process; thus, their conclusions about BAI should be interpreted with caution. All of the studies selected by these authors were also evaluated for suitability in the present work; some were excluded mainly because of the recruitment method.

In April 2004, the US Preventive Services Task Force recommended screening and behavioral counseling interventions to reduce alcohol misuse by adults in primary care settings⁵¹ (grade B recommendation). The present work is consistent with their recommendations. Our meta-analytic approach adds to the extended systematic review presented by the Preventive Services Task Force.

Despite these recommendations, questions remain about the effectiveness of such interventions in daily practice. In addition to difficulties related to motivating and training physicians to perform BAI, another major challenge for BAI implementation is to find an optimal and practical way to screen. Many validated screening tools have been used for research purposes (AUDIT [Alcohol Use Disorder Identification Test], CAGE [see Table 1 for the CAGE questions], and quantity-frequency questions, as well as specific questionnaires for pregnant women), but these tools are difficult to use in a primary care practice.^13,14 The Preventive Services Task Force did not recommend a specific approach for screening.⁵¹ Routine implementation of screening and BAI requires specific strategies (eg, promoting understanding of and adherence to the BAI concept, interactive sessions to improve practice skills, computer-based decisions aids, interventions conducted in the waiting room or by physician assistants). The difficulties of tailoring and disseminating such strategies are being addressed by ongoing initiatives.^52-54

Among the 7 studies that reported significant improvement in BAI groups compared with control groups, all but 1 reported an intervention lasting between 5 and 15 minutes. Similarly, all but 1 reported the use of written material and proposed a repeated intervention, either routinely or after a decision by the patient. Thus, it appears that BAI lasting from 5 to 15 minutes, accompanied by written material and the opportunity for the patient to schedule a follow-up visit, has the potential to significantly reduce alcohol consumption compared with either no intervention, usual care, or less than 5 minutes of intervention.

No evidence of the influence of BAI on laboratory values was observed. The low sensitivity of these tests in a sample of principally non–alcohol-dependent primary care patients and the low response of these tests to moderate reductions in alcohol consumption^55-57 are consistent with our results. Indeed, the use of laboratory values as markers of alcohol consumption was progressively abandoned in more recent trials. The use of BAI seems to reduce health care utilization, but these results were obtained only for health maintenance organization–affiliated patients and thus are not to be generalized to the whole primary care population.^32-34,37 In terms of economic analysis, a favorable benefit-cost ratio for BAI was observed. However, this finding comes from 1 study only, although the largest in our selection. Further research is then needed to evaluate this question more precisely and in various environments. There is scant evidence of the efficacy of brief interventions on morbidity, mortality, and quality-of-life measures. Most studies were not aimed at measuring outcomes by using large sample sizes with longer follow-up intervals; an exception was Project TrEAT, which demonstrated a significant reduction in mortality in its BAI groups.³³ The postulate of BAI reducing morbidity and mortality is suggested by the success of BAI in reducing alcohol consumption coupled with actual knowledge of morbidity and mortality associated with alcohol consumption.²

Three intervention models appeared to have inspired most of the selected interventions. All of them included feedback and advice components, which is consistent with analyses conducted by Bien et al,⁵ but they differed in other components and in the style of relationship between patient and provider. Further research is needed to investigate which components of brief interventions present evidence of efficacy in the primary care setting.⁵⁸

This study had certain limitations. Although our purpose was to focus on BAI applied to patients actively seeking primary health care,we still found heterogeneity between trials. The heterogeneity of BAI and the lack of detailed description reported limited the ability to identify key elements of efficacy. The lack of standardized control groups also limited comparison between studies. Furthermore, the concept of usual care used in many studies as the control group is probably highly variable between and within studies. Only a few authors have described the training of providers or tried to enhance the homogeneity of interventions by practicing role playing and by video recording the interventions. Whether the therapeutic procedures were carried out as intended was not assessed; thus, it is difficult to ascertain whether any of the described interventions were delivered effectively. The greater homogeneity found in the US studies might be explained by more homogeneous drinking habits and greater consensus about risk levels related to alcohol use and similar inclusion criteria. Methodologic quality was a significant determinant of success after correction for baseline alcohol consumption. Most studies reported a 10% to 30% reduction in alcohol consumption in the control group. This could be due to regression to the mean, to a minimal or quasi-intervention given to control groups, to the proper effect of an assessment centered on alcohol consumption, to a Hawthorn effect, or to the natural course of changes in alcohol use. Contamination could have occurred if providers changed their approach to at-risk drinkers and delivered a quasi-intervention to patients included in the control group. Project GOAL (Guiding Older Adult Lifestyles) is the only trial in which there was no reduction in alcohol use in the control group,³¹ but the reason is unclear.

Our results are also restricted by the limitations of the selected studies themselves. In such preventive interventions, double-blind studies are very difficult to implement. The nonvalidated quality scores used to rank the study quality were significantly related to more positive results of intervention. This finding is dissimilar to those of studies in other areas of research that demonstrated that studies of lower quality were more likely to report positive results.²² Our findings might be explained by larger effects of contamination within the studies of lower quality. However, colinearity between several different factors (baseline alcohol consumption, study quality, and studies conducted in the United States) precluded us from verifying this hypothesis.

We tried to reduce selection bias as much as possible by reviewing all articles independently and in duplicate. We did not constrain the literature search to any specific language and thus were less prone to language bias; however, no searches in Grey Literature databases could be performed. We may have overlooked some potentially eligible studies during the selection process.

We conclude that BAI aimed at reducing alcohol consumption is effective in primary care settings on the basis of studies that approximate usual practice and are similar in terms of patient context and statistical homogeneity. The typical effective BAI takes no more than 15 minutes, is accompanied by written material, and offers an opportunity for the patient to schedule a follow-up. Positive effects seem to be sustained beyond a year and can last for as long as 48 months. This finding should encourage further research aimed at determining more precisely the components of efficacy and the relationship of BAI to morbidity, mortality, and quality-of-life–related outcomes. Efforts should be sustained to continue the implementation and evaluation of BAI programs.

Correspondence: Bernard Burnand, MD, MPH, Rue du Bugnon 17, CH-1011 Lausanne, Switzerland (Bernard.Burnand@chuv.ch).

Accepted for Publication: September 22, 2004.

Financial Disclosure: None.

Funding/Support: This study was supported by the Clinical Epidemiology Center and the Alcohol Treatment Center, University Hospital, Lausanne, Switzerland.

Acknowledgment: We thank Muriel Gaulis and Alexis Rohrbach, MD, for their help in data extraction and Spanish translation, and the authors of the examined studies who provided additional information related to their studies.