Pharmacotherapy for Adults With Alcohol Use Disorders in Outpatient Settings: A Systematic Review and Meta-analysis

Importance  Alcohol use disorders cause substantial morbidity and early mortality yet remain greatly undertreated. Medications are considerably underused.

Objective  To conduct a systematic review and meta-analysis of the benefits and harms of medications (US FDA-approved and others) for adults with alcohol use disorders.

Data Sources  PubMed, Cochrane Library, PsycINFO, CINAHL, EMBASE, FDA website, and clinical trials registries (January 1, 1970, to March 1, 2014).

Study Selection  Two reviewers selected randomized clinical trials (RCTs) with at least 12 weeks’ duration that reported eligible outcomes and head-to-head prospective cohort studies reporting health outcomes or harms.

Data Extraction and Synthesis  We conducted meta-analyses using random-effects models and calculated numbers needed to treat for benefit (NNTs) or harm (NNHs).

Main Outcomes and Measures  Alcohol consumption, motor vehicle crashes, injuries, quality of life, function, mortality, and harms.

Results  We included 122 RCTs and 1 cohort study (total 22 803 participants). Most assessed acamprosate (27 studies, n = 7519), naltrexone (53 studies, n = 9140), or both. The NNT to prevent return to any drinking for acamprosate was 12 (95% CI, 8 to 26; risk difference [RD], −0.09; 95% CI, −0.14 to −0.04) and was 20 (95% CI, 11 to 500; RD, −0.05; 95% CI, −0.10 to −0.002) for oral naltrexone (50 mg/d). The NNT to prevent return to heavy drinking was 12 (95% CI, 8 to 26; RD −0.09; 95% CI, −0.13 to −0.04) for oral naltrexone (50 mg/d). Meta-analyses of trials comparing acamprosate to naltrexone found no statistically significant difference between them for return to any drinking (RD, 0.02; 95% CI, −0.03 to 0.08) or heavy drinking (RD, 0.01; 95% CI, −0.05 to 0.06). For injectable naltrexone, meta-analyses found no association with return to any drinking (RD, −0.04; 95% CI, −0.10 to 0.03) or heavy drinking (RD, −0.01; 95% CI, −0.14 to 0.13) but found an association with reduction in heavy drinking days (weighted mean difference [WMD], −4.6%; 95% CI, −8.5% to −0.56%). Among medications used off-label, moderate evidence supports an association with improvement in some consumption outcomes for nalmefene (heavy drinking days per month: WMD, −2.0; 95% CI, −3.0 to −1.0; drinks per drinking day: WMD, −1.02; 95% CI, −1.77 to −0.28) and topiramate (% heavy drinking days: WMD, −9.0%; 95% CI, −15.3% to −2.7%; drinks per drinking day: WMD, −1.0; 95% CI, −1.6 to −0.48). For naltrexone and nalmefene, NNHs for withdrawal from trials due to adverse events were 48 (95% CI, 30 to 112) and 12 (95% CI, 7 to 50), respectively; risk was not significantly increased for acamprosate or topiramate.

Conclusions and Relevance  Both acamprosate and oral naltrexone were associated with reduction in return to drinking. When directly compared with one another, no significant differences were found between acamprosate and naltrexone for controlling alcohol consumption. Factors such as dosing frequency, potential adverse events, and availability of treatments may guide medication choice.

Alcohol use disorders (AUDs) are common, cause substantial morbidity, and result in 3-fold increased rates of early mortality (eTable 1 in the Supplement).^1-8 Treating AUDs is difficult but may be aided by using medications. Pharmacotherapy for AUDs was initiated in the 1950s and consisted only of disulfiram (Antabuse). In the 1990s, naltrexone (oral and intramuscular formulations) and acamprosate were approved by the US Food and Drug Administration (FDA) (eTable 2 in the Supplement).

Fewer than one-third of patients with AUDs receive treatment,⁶ and only a small percentage (<10%) receive medications to assist in reducing alcohol consumption. To evaluate the benefits and harms of medications for the treatment of adults with AUDs, we conducted a systematic review. A larger, more comprehensive technical report for the Agency for Healthcare Research and Quality was prepared (eTable 3 in the Supplement).⁹ This article summarizes findings from the larger report on the efficacy of various medications used for the treatment of AUDs in reducing alcohol intake or improving health outcomes and on the adverse effects of these medications.

We developed and followed a standard protocol. A technical report that details methods, search strategies, and additional information is available online.⁹

We searched PubMed, the Cochrane Library, PsycINFO, CINAHL, and EMBASE from January 1, 1970, to October 11, 2013, for the technical report; we updated searches through March 1, 2014, for this article. An experienced Evidence-based Practice Center (EPC) librarian ran all searches; another EPC librarian peer-reviewed them. We manually searched reference lists of pertinent reviews and trials for relevant citations that our searches missed.

We searched for unpublished studies using ClinicalTrials.gov, the World Health Organization International Clinical Trials Registry Platform, and the FDA website. In addition, the Scientific Resource Center of the Agency for Healthcare Research and Quality requested unpublished studies and data from manufacturers.

We included studies enrolling adults with AUDs that evaluated an FDA-approved medication or any of 23 off-label medications for at least 12 weeks in an outpatient setting. Studies were required to assess one of the following outcomes: (1) consumption—return to any drinking, return to heavy drinking, drinking days, heavy drinking days (≥4 drinks per day for women; ≥5 for men), drinks per drinking day; (2) health outcomes—accidents (ie, motor vehicle crashes), injuries, quality of life, function, and mortality; or (3) adverse effects.

Double-blind randomized clinical trials (RCTs) comparing one of the medications with placebo or another medication were eligible. Prospective cohort studies that compared 2 medications were eligible if they reported a health outcome. For adverse effects, the following designs were eligible if they compared 2 drugs of interest: nonrandomized or open-label trials, subgroup analyses from trials, prospective cohort studies, and case-control studies.

Two investigators independently reviewed each title and abstract. Studies marked for possible inclusion by either reviewer underwent dual, independent full-text review. If reviewers disagreed, we resolved conflicts by consensus.

We used structured data extraction forms to gather relevant data from each article. All data extractions were reviewed for completeness and accuracy by at least 2 investigators.

To assess the risk of bias of studies, we used predefined criteria based on established guidance.^10,11 We included questions about adequacy of randomization, allocation concealment, similarity of groups at baseline, blinding, attrition, validity and reliability of measures, whether intention-to-treat (ITT) analysis was used, methods of handling missing data, and fidelity. We rated the studies as low, medium, high, or unclear risk of bias.^10,11 Two independent reviewers assessed risk of bias for each study. Disagreements were resolved by consensus.

We conducted meta-analyses of RCTs using random-effects models.¹² For continuous outcomes, we used weighted mean differences (WMDs) and 95% CIs. For binary outcomes, we calculated risk differences (RDs) between groups and 95% CIs. We did not include studies rated as high or unclear risk of bias in our main analyses but included them in sensitivity analyses. When possible, we conducted post hoc subgroup analyses to assess whether pooled results differed for studies rated as low risk of bias. We calculated the I² statistic to assess statistical heterogeneity.^13,14 We examined potential sources of heterogeneity by analysis of subgroups defined by patient population (eg, US vs non-US studies). Analyses were conducted using the metan, metafunnel, and metabias commands in Stata version 11.1 (StataCorp). Statistical significance was assumed when 95% CIs of pooled results did not cross 0. All testing was 2-sided. We calculated numbers needed to treat (NNTs) and numbers needed to harm (NNHs) when pooled RDs found a statistically significant result. When appropriate^15,16 (eg, ≥10 studies in a meta-analysis), we assessed for publication bias by visually examining funnel plots and using the Begg-Mazumdar¹⁷ test. None of the funnel plots or statistical tests indicated concern for publication bias. When quantitative synthesis was not appropriate (eg, insufficient numbers of similar studies), we synthesized the data qualitatively.

We graded the strength of evidence as high, moderate, low, or insufficient based on established guidance.¹⁸ The approach incorporates 4 key domains: risk of bias, consistency, directness, and precision. Two reviewers assessed each domain for each outcome and determined an overall grade. Differences were resolved by consensus.

Quiz Ref IDWe did not combine medications with similar mechanisms or in the same drug class in our analyses because we aimed to determine which medications (not classes) have evidence supporting associations with improved outcomes. For example, nalmefene is an opioid receptor antagonist like naltrexone, but we analyzed them separately.

We included 151 articles reporting on 123 studies (Figure 1). Of these, one was a prospective cohort study²⁹; the rest were RCTs; the total number of participants was 22 803. Characteristics of included studies are shown in eTable 4 in the Supplement. Most studies assessed acamprosate (27 studies, n = 7519), naltrexone (53 studies, n = 9140), or both. Sample sizes ranged from 21 to 1383. Treatment duration ranged from 12 to 52 weeks. Mean age was usually in the 40s. All participants met criteria for alcohol dependence in the vast majority of trials. Most studies enrolled patients after detoxification or required a period of sobriety (at least 3 days). Studies typically included psychosocial co-interventions; thus, effect sizes reflect the added benefits of medications beyond those of psychosocial interventions and placebo. The largest trial, COMBINE,³⁰ randomized 1383 treatment-seeking patients to receive medical management with naltrexone (plus 1 placebo), acamprosate (plus 1 placebo), both, or 2 placebos, with or without a combined behavioral intervention or to receive a combined behavioral intervention only (no pills).

We included 22 placebo-controlled trials of acamprosate, 4 of disulfiram, and 44 of naltrexone. For medications used off-label, we included 1 placebo-controlled trial for each of the following: aripiprazole, atomoxetine, desipramine, fluvoxamine, gabapentin, imipramine, olanzapine, ondansetron, and paroxetine. We included multiple placebo-controlled trials for baclofen, buspirone, citalopram, fluoxetine, nalmefene, quetiapine, sertraline, topiramate, valproic acid, and varenicline. We included 4 trials directly comparing acamprosate with naltrexone, 1 comparing disulfiram with naltrexone, and 4 comparing naltrexone with the off-label medications (aripiprazole, desipramine, paroxetine, sertraline, and topiramate).

Acamprosate and naltrexone were associated with improvement in consumption outcomes (Table 1, Figure 2, and Figure 3). To prevent 1 person from returning to any drinking, the NNTs were 12 (95% CI, 8 to 26; 16 trials, n = 4847) and 20 (95% CI, 11 to 500; 16 trials, n = 2347) for acamprosate and oral naltrexone (50 mg/d), respectively. For return to heavy drinking, acamprosate was not associated with improvement, whereas oral naltrexone (50 mg/d) was associated with improvement with an NNT of 12 (95% CI, 8 to 26; 19 trials, n = 2875). For injectable naltrexone, our meta-analyses found no statistically significant association with return to any drinking or return to heavy drinking but found an association with reduction in heavy drinking days (WMD −4.6%; 95% CI, −8.5% to −0.56%; 2 trials, n = 926). Quiz Ref IDEvidence from well-controlled trials of disulfiram does not adequately support an association with preventing return to any drinking or improvement in other alcohol consumption outcomes (Table 1). The largest disulfiram trial (n = 605) reported fewer drinking days for participants who returned to drinking and had a complete set of assessments.³² Results of sensitivity analyses that included studies rated as high or unclear risk of bias were similar to the results of our main analyses (eFigures 1 and 2 in the Supplement).

Post hoc subgroup analyses by risk of bias (separating studies rated as low risk of bias) did not reveal any notable differences or were underpowered to find differences for most outcomes and medications (eFigures 3 through 10 in the Supplement). However, the subgroup analysis for return to any drinking for acamprosate compared with placebo showed a decreasing effect size from high/unclear (RD, −0.13; 95% CI, −0.20 to −0.06; 3 trials, n = 757) or medium (RD, −0.11; 95% CI, −0.16 to −0.06, 12 trials, n = 3438) to low (RD, −0.02; 95% CI, −0.09 to 0.05, 4 trials, n = 1409) risk of bias (eFigure 3 in the Supplement). Although the confidence intervals for pooled estimates of all subgroups overlapped, the pooled estimate for the low risk of bias subgroup was not statistically significant, and the 2 studies^30,37 rated as low risk of bias that contributed the largest number of events found lack of efficacy for acamprosate.

Our meta-analyses of head-to-head RCTs comparing acamprosate with naltrexone^30,38-40 found no statistically significant difference between the 2 medications (Table 2). COMBINE was one of the RCTs.³⁰ It found that patients receiving medical management with naltrexone, a combined behavioral intervention, or both had better drinking outcomes than those who received placebo, but acamprosate showed no evidence of efficacy.

For the vast majority of medications used off-label, evidence was either insufficient to determine whether they are associated with reduced consumption or evidence suggested that they are not (eTable 5 in the Supplement). We found some exceptions (eTable 5, Figure 3). For topiramate, evidence supports an association with fewer drinking days (WMD, −6.5%; 95% CI, −12.0% to −1.0%; 2 trials,^77,78 n = 541), heavy drinking days (WMD, −9.0%; 95% CI, −15.3% to −2.7%; 3 trials,^77-79 n = 691), and drinks per drinking day (WMD, −1.0; 95% CI, −1.6 to −0.48; 3 trials,^77-79 n = 691). For nalmefene, evidence supports an association with fewer heavy drinking days per month (WMD, −2.0; 95% CI, −3.0 to −1.0; 2 trials,^80,81 n = 806) and drinks per drinking day (WMD, −1.02; 95% CI, −1.77 to −0.28; 3 trials,^41,82,83 n = 608). Finally, limited evidence from 2 small RCTs^42,43 (total n = 88), one enrolling people with bipolar disorder, supports an association between valproic acid and improvement in some consumption outcomes.

We found insufficient direct evidence from RCTs to determine whether or not treatment with medications leads to improvement in health outcomes (Table 1 and eTable 5 in the Supplement). Quiz Ref IDVery few trials reported health outcomes, and the included trials were not designed or powered to assess health outcomes: they typically focused on consumption outcomes. COMBINE reported some evidence of improvement in quality of life with naltrexone plus behavioral intervention (on the physical health scale from the 12-item Short Form health survey, version 2), but the difference between groups did not reach a clinically meaningful threshold.³³

There was insufficient evidence regarding many potential adverse events precluding determination of risks associated with these medications. In most cases, inadequate precision (ie, wide confidence intervals that contained clinically distinct conclusions) resulted in our inability to arrive at conclusions about medication risk. For most of the specific adverse events, point estimates favored placebo (ie, more adverse events with medications), but differences were not statistically significant. In head-to-head studies, the risk of withdrawal due to adverse events was not significantly different between acamprosate and naltrexone, but the risks of headache and vomiting were slightly higher for those treated with naltrexone (eTable 6 in the Supplement).

Compared with placebo, patients treated with naltrexone or nalmefene had a higher risk of withdrawal from trials due to adverse events (NNH, 48; 95% CI, 30 to 112; 17 trials, n = 2743; and NNH, 12; 95% CI, 7 to 50; 5 trials, n = 2054, respectively); we found no significant difference for acamprosate or topiramate. Compared with placebo, patients treated with acamprosate had a higher risk of anxiety (NNH, 7; 95% CI, 5 to 11; 2 trials, n = 624), diarrhea (NNH, 11; 95% CI, 6 to 34; 12 trials, n = 2978), and vomiting (NNH, 42; 95% CI, 24 to 143; 4 trials, n = 1817); those treated with naltrexone had a higher risk of dizziness (NNH, 16; 95% CI, 12 to 28; 13 trials, n = 2675), nausea (NNH, 9; 95% CI, 7 to 14; 24 trials, n = 4655), and vomiting (NNH, 24; 95% CI, 17 to 44; 9 trials, n = 2438); those treated with nalmefene had a higher risk of dizziness (NNH, 7; 95% CI, 5 to 10; 4 trials, n = 1944), headache (NNH, 26; 95% CI, 15 to 143; 3 trials, n = 1401), insomnia (NNH, 10; 95% CI, 8 to 17; 5 trials, n = 2049), nausea (NNH, 7; 95% CI, 5 to 11; 5 trials, n = 2049), and vomiting (NNH, 17; 95% CI, 11 to 48; 3 trials, n = 1679); and those treated with topiramate had a higher risk of cognitive dysfunction (NNH, 12; 95% CI, 7 to 84; 2 trials, n = 521), paresthesias (NNH, 4; 95% CI, 3 to 7; 3 trials, n = 691), and taste abnormalities (NNH, 7; 95% CI, 5 to 15; 2 trials, n = 477) (eTable 7 in the Supplement).

When used in conjunction with psychosocial co-interventions, addition of several medications resulted in better alcohol consumption outcomes. Acamprosate and oral naltrexone (50 mg/d) have the best evidence supporting their benefits. Trials comparing these medications have not established a difference in outcomes between them.

When clinicians decide to use one of the medications, a number of factors may help with choosing which medication to prescribe, including the medication’s efficacy, administration frequency, cost, adverse events, and availability. In some health systems, these medications may not be on the formulary. Quiz Ref IDAcamprosate is given 3 times daily and is somewhat less convenient to use than oral naltrexone that only requires 1 daily tablet. Acamprosate is contraindicated with severe renal impairment and oral naltrexone is contraindicated with acute hepatitis, liver failure, concurrent opioid use, or an anticipated need for opioids.⁸⁴

Because of its long-standing availability, clinicians may be more familiar with disulfiram than naltrexone or acamprosate. However, well-controlled trials of disulfiram did not show overall reductions in alcohol consumption. In a subgroup analysis of the largest disulfiram trial,³² there were fewer drinking days for patients who returned to drinking and had a complete set of assessments. This suggests that disulfiram may benefit some AUD patients. However, none of the disulfiram trials evaluated supervised medication delivery, potentially underestimating the benefits of the drug when used in supervised treatment programs.

The evidence from trials was insufficient to make any conclusions about improved health outcomes attributable to pharmacotherapy of AUDs. Epidemiologic studies consistently relate high average alcohol consumption and heavy per-occasion use to increased risks for health problems. These include cancers (eg, mouth, esophagus, colon, liver, and breast); cognitive impairment; liver cirrhosis; chronic pancreatitis; stroke; depression; suicide; and injuries and violence.^5,85-91 Given the epidemiologic evidence for adverse health consequences of heavy alcohol use, improved health outcomes should occur with AUD treatment. A recent modeling study estimated that increasing treatment coverage to 40% of all people with alcohol dependence in the European Union would reduce alcohol-attributable mortality by up to 13%.⁹² Several AUD treatment combinations including pharmacotherapy, when compared with placebo plus medical management, reduced costs from health care, arrests, and motor vehicle accidents in a cost analysis of the COMBINE trial.⁹³

All participants met criteria for alcohol dependence in most of the studies we reviewed. Based on the studies’ time period, they used Diagnostic and Statistical Manual of Mental Disorders (DSM) Third Edition or Fourth Edition criteria for alcohol dependence. The Fifth Edition, DSM-5, was released in 2013 and describes a single AUD category measured on a continuum from mild to severe (eTable 1 in the Supplement). DSM-5 no longer has separate categories for alcohol abuse and dependence.^94,95 Using DSM-5 terminology, most participants in the studies we reviewed likely had moderate to severe AUDs. As a consequence, applicability of our findings regarding pharmacological treatment for AUDs to patients with mild disorders is uncertain. The mean age of participants was generally in the 40s. We did not find evidence to confirm or refute whether treatments are likely to be more or less beneficial for older or younger subgroups, different sex groups, racial or ethnic minorities, smokers or nonsmokers, and those with certain coexisting conditions.⁹

The majority of placebo-controlled trials assessing acamprosate were conducted in Europe (16/22) and a minority were conducted in the United States (4/22). In contrast, the opposite occurred for naltrexone: 27 of 44 trials were conducted in the United States and 8 of 44 were carried out in Europe. The few US-based acamprosate studies did not find it to be efficacious. The European trials of acamprosate typically identified patients from inpatient settings or treatment programs, whereas the US-based trials recruited patients using advertisements and referrals. Differences in how patients were recruited into the trials might have resulted in populations with differing AUD severity and differing potential for benefit.

Most studies required patients to abstain for at least a few days prior to initiating medication. Medications for AUDs are generally recommended for maintenance of abstinence. Acamprosate and injectable naltrexone are only approved for use in patients who have established abstinence. However, some studies enrolled patients who were not yet abstinent and reported reduced heavy drinking with naltrexone^44,96 or acamprosate.⁴⁵

The US Preventive Services Task Force recommends screening adults for alcohol misuse.⁹⁷ Screening will inevitably identify some individuals with AUDs. Clinicians must then decide whether to refer to specialized treatment or intervene within their practice. Like primary care–based behavioral counseling interventions for risky drinking, implementing pharmacotherapy and psychosocial interventions for AUDs may require formal protocols, staffing (eg, multidisciplinary team-based care), support systems, and additional provider and staff training.^86,98 Some experts advocate chronic care management, a systematic approach to treatment and follow-up similar to how the health care system approaches heart failure, diabetes, and other chronic diseases.⁹⁹

Barriers to prescribing medications for AUDs in primary care may include lack of familiarity with the medications, lack of confidence in their effectiveness, or inability to provide suitable psychosocial co-interventions—eg, because of competing demands or insufficient practice resources, personnel, or training. Historically, primary care providers have referred patients with AUDs for specialized treatment. However, these medications are underutilized,^100,101 and many patients may not be willing to pursue or may not have access to specialized treatment. Thus, offering treatment through primary care has the potential to reduce morbidity for many patients with AUDs.

We found scant evidence from primary care settings. One trial (n = 100) that recruited participants primarily by advertisement in 2 family medicine settings found no significant treatment effect for acamprosate.⁴⁶ The only other trial meeting our inclusion criteria conducted in primary care settings compared nalmefene with placebo in 15 sites (about half were primary care) in Finland.⁸³ It found no significant difference in percentage of drinking days but reported a lower percentage of heavy drinking days (18.1% vs 29.7%, P = .02) and fewer drinks per drinking day (WMD, −1.0; 95% CI, −2.0 to −0.02) for patients treated with nalmefene than for those who received placebo.

Some included studies conducted in non–primary care settings used interventions that could be adapted for delivery in primary care. For example, in the COMBINE study,³⁰ providers delivered a medical management intervention comprised of up to 9 manual-guided counseling visits (at weeks 0, 1, 2, 4, 6, 8, 10, 12, and 16). The first visit was approximately 45 minutes, and follow-up visits were about 20 minutes each. Medical management included advice for reducing drinking, inquiries about medication adverse effects, and emphasis on the importance of adherence. Participants were encouraged to attend support groups available in the community (eg, Alcoholics Anonymous). The Medical Management Treatment Manual provides direction for clinicians to provide medical management, a combined behavioral intervention, and medical treatment with naltrexone or acamprosate as provided in the COMBINE trial.¹⁰²

Regarding implementation of treatment programs for AUDs in primary care, we identified 4 other publications that did not meet our inclusion criteria (because of study design or comparators) that have important implications.^103-106 Although these studies found conflicting results, they demonstrate approaches to managing AUDs in primary care. Further details of these studies are available in the eDiscussion in the Supplement. In general, the interventions involve formal clinic structure, staffing, and protocols. They used variations of chronic care management, multidisciplinary team-based care, and care coordination between primary care and mental health providers.

We only considered trials with at least 12 weeks of treatment. Longitudinal studies have found that shorter treatment periods may yield misleading conclusions about benefits, due to fluctuations in drinking typical of the course of AUDs.^107,108 Next, we did not assess how medications and psychosocial interventions compare with each other. Our review focused on studies assessing benefits and harms of medications and how they compare with other medications, and our findings reflect the added benefits of medications beyond those of psychosocial co-interventions. Studies used a variety of different psychosocial co-interventions. This heterogeneity limits our certainty about the benefits of medications when used alone (with no co-intervention) or when added to a particular psychosocial intervention. Further, we did not specifically assess benefits for patients without a goal of abstinence.

We combined studies that included populations with a dual diagnosis (eg, alcohol dependence and depression) and those that did not in our meta-analyses. To determine whether this potential population heterogeneity had a significant influence on our conclusions, we conducted sensitivity analyses for acamprosate and naltrexone, stratifying by whether or not studies reported enrolling a dual diagnosis population (data in full report⁹). Effect sizes did not change significantly.

Most studies were rated as medium risk of bias. We rated few studies as low risk of bias (8/123 included studies; 4/27 studies assessing acamprosate; and 4/53 studies assessing naltrexone). Most studies rated as medium, rather than low, risk of bias lacked complete reporting of information about several of the following: randomization sequence generation, allocation concealment, fidelity, adherence, or outcome assessor masking. For most outcomes and medications, our post hoc subgroup analyses separating studies rated as low risk of bias did not suggest notable differences or were underpowered to find differences. But a subgroup analysis for return to any drinking for acamprosate showed that the pooled effect of the studies rated as low risk of bias found no significant difference between acamprosate and placebo. Possible explanations include population differences (eg, severity, country), other heterogeneity, no true association between acamprosate and return to drinking (ie, the effect found in overall pooled analyses represents bias), random error, or a combination of these factors. The 2 studies (out of 4) rated as low risk of bias that contributed by far the largest number of events were both conducted in the United States and relied on advertisements and referrals to identify participants. In contrast, the vast majority of the 15 studies rated as medium, high, or unclear risk of bias were conducted in European countries (1 was in the United States and 1 in Brazil) and typically identified patients from inpatient settings or treatment programs. It is possible that this resulted in populations with differing AUD severity and differing potential for benefit or that having gone through a program may increase adherence to treatments and improve potential for benefit.

In addition, publication bias and selective reporting are potential limitations. However, funnel plots did not raise concern for publication bias, and we searched for unpublished studies and unpublished outcomes and did not find direct evidence of either of these biases.

Quiz Ref IDBoth acamprosate and oral naltrexone (50 mg/d) were associated with reduction in return to drinking. They have the best evidence for improving alcohol consumption outcomes for patients with AUDs. Head-to-head trials have not established superiority of either medication. Among medications used off-label, moderate evidence supports an association with improvement in some consumption outcomes for nalmefene and topiramate.

Corresponding Author: Daniel E. Jonas, MD, MPH, University of North Carolina at Chapel Hill, Department of Medicine, 5034 Old Clinic Bldg, CB No. 7110, Chapel Hill, NC 27599 (daniel_jonas@med.unc.edu).

Author Contributions: Dr Jonas had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Study concept and design: Jonas, Amick, Feltner, Kim, Gass, Garbutt.

Acquisition, analysis, or interpretation of data: Jonas, Amick, Feltner, Bobashev, Thomas, Wines, Kim, Shanahan, Gass, Rowe, Garbutt.

Drafting of the manuscript: Jonas, Amick, Feltner, Bobashev, Wines, Kim, Shanahan, Rowe, Garbutt.

Critical revision of the manuscript for important intellectual content: Jonas, Feltner, Bobashev, Thomas, Kim, Gass, Garbutt.

Statistical analysis: Jonas, Amick, Feltner, Bobashev, Wines, Kim.

Obtained funding: Jonas.

Administrative, technical, or material support: Jonas, Amick, Thomas, Wines, Kim, Shanahan, Gass, Rowe.

Study supervision: Jonas, Feltner, Garbutt.

Conflict of Interest Disclosures: All authors have completed and submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest and none were reported.

Funding/Support: This project was funded by the Agency for Healthcare Research and Quality (AHRQ), US Dept of Health and Human Services (contract HHSA290201200008I_HHSA29032002T).

Role of the Sponsors: AHRQ staff participated in development of the scope of work and reviewed drafts of the manuscript. Approval from AHRQ was required before the manuscript could be submitted for publication, but the authors are solely responsible for its content and the decision to submit it for publication. AHRQ staff did not participate in the conduct of the review, data collection, data management, data analysis, interpretation of the data, or preparation of the manuscript.

Disclaimer: Statements in the report should not be construed as endorsement by the Agency for Healthcare Research and Quality or the US Dept of Health and Human Services.

Additional Contributions: We express appreciation to the other members of the RTI International–University of North Carolina Evidence-based Practice Center who contributed to this effort: Christiane Voisin, MSLS, for help in literature searching; Carol Woodell, BSPH, for her project management expertise; Claire Baker for help in article retrieval; and Loraine Monroe and Jennifer Drolet, MA, for editing and formatting the technical report. These individuals received salary support from a contract from the AHRQ. We also gratefully acknowledge the continuing support of our AHRQ Task Order Officers, Carmen Kelly, PharmD, RPh; Christine Chang MD, MPH; and Aysegul Gozu, MD, MPH; they were not compensated for their contributions besides their salaries.