Appendix B. Review of Research Evidence Supporting Guideline Statements

APA recommends (1C) that the initial psychiatric evaluation of a patient with suspected alcohol use disorder include assessment of current and past use of tobacco and alcohol as well as any misuse of other substances, including prescribed or over-the-counter medications or supplements.

APA recommends (1C) that the initial psychiatric evaluation of a patient with suspected alcohol use disorder include a quantitative behavioral measure to detect the presence of alcohol misuse and assess its severity.

APA suggests (2C) that physiological biomarkers be used to identify persistently elevated levels of alcohol consumption as part of the initial evaluation of patients with alcohol use disorder or in the treatment of individuals who have an indication for ongoing monitoring of their alcohol use.

APA recommends (1C) that patients be assessed for co-occurring conditions (including substance use disorders, other psychiatric disorders, and other medical disorders) that may influence the selection of pharmacotherapy for alcohol use disorder.

APA suggests (2C) that the initial goals of treatment of alcohol use disorder (e.g., abstinence from alcohol use, reduction or moderation of alcohol use, other elements of harm reduction) be agreed on between the patient and clinician and that this agreement be documented in the medical record.

APA suggests (2C) that the initial goals of treatment of alcohol use disorder include discussion of the patient’s legal obligations (e.g., abstinence from alcohol use, monitoring of abstinence) and that this discussion be documented in the medical record.

APA suggests (2C) that the initial goals of treatment of alcohol use disorder include discussion of risks to self (e.g., physical health, occupational functioning, legal involvement) and others (e.g., impaired driving) from continued use of alcohol and that this discussion be documented in the medical record.

APA recommends (1C) that patients with alcohol use disorder have a documented comprehensive and person-centered treatment plan that includes evidence-based nonpharmacological and pharmacological treatments.

APA recommends (1B) that naltrexone or acamprosate be offered to patients with moderate to severe alcohol use disorder who

The AHRQ review (Jonas et al. 2014) found that acamprosate treatment at a dose of 666 mg and three times daily (range 1,000 mg to 3,000 mg per day in divided doses) was associated with a decreased likelihood of returning to alcohol use as compared with placebo (moderate strength of evidence; risk difference [RD] –0.09; 95% confidence interval [CI] –0.14 to –0.04; number needed to treat [NNT] = 12) (Table B–1). Number of drinking days was also reduced with acamprosate treatment relative to placebo (moderate strength of evidence; weighted mean difference [WMD] –8.8; 95% CI –12.8 to –4.8; 13 trials). However, for both outcomes, the benefits of acamprosate were seen primarily in studies done outside of the United States. Return to heavy drinking (moderate strength of evidence) and number of heavy drinking days (insufficient strength of evidence) showed no effect of acamprosate. The available evidence also did not permit any conclusions about the effect of acamprosate on outcomes such as quality of life, functioning, accidents, injuries, or mortality. In studies that assessed response rates by sex, men and women did not differ on any measure of efficacy.

In the studies with long-term use of acamprosate (48–52 weeks), there was an 11% absolute reduction in return to any drinking (RD, –0.11; 95% CI, –0.16 to –0.06; 4 trials) and 12.2% fewer drinking days than for those treated with placebo over 48–52 weeks (WMD, –12.2; 95% CI, –16.4 to –8.0; I² 0%).

A number of relevant studies that are not included in the AHRQ meta-analysis or in Table B–1 have shown mixed results for acamprosate. In a pragmatic trial in France, 422 patients treated by 149 practitioners were randomly assigned to standard care (typically outpatient detoxification followed by psychotherapy) or to acamprosate plus standard care (Kiritzé-Topor et al. 2004). The trial reported better outcomes for the acamprosate group on a number of alcohol-related measures, with an NNT of about 7. A 24-week study (total N = 327) with low risk of bias that was conducted in Japan (Higuchi and Japanese Acamprosate Study Group 2015) showed greater rates of abstinence with acamprosate than placebo at 24 weeks (47.2% for acamprosate vs. 36.0% for placebo; p = 0.039), but there was no significant effect of treatment on secondary endpoints (i.e., cumulative days of abstinence during 24 weeks of treatment, time to first relapse, and time to 3 or more days of consecutive drinking). Furthermore, the generalizability of this study to the United States may be limited because patients were enrolled on discharge from 2 months of inpatient detoxification/rehabilitation.

In two additional RCTs, effects of acamprosate did not differ from placebo. The German PREDICT study (Mann et al. 2013), modeled after the COMBINE study, recruited subjects (total N = 426) at time of discharge from medical detoxification (average length of stay 18 days). The time to first heavy drinking (primary outcome) did not differ among the treatment groups. Relapse-free survival at 90 days was 48.3% for acamprosate versus 51.8% for placebo. Another study (total N = 100) with low risk of bias in a primary care setting (Berger et al. 2013) found no effect of acamprosate on percent days abstinent (primary outcome), percent heavy drinking days, or change in gamma-glutamyl transferase (GGT) levels. Nevertheless, both acamprosate and placebo groups showed improvement during the 12-week trial, particularly among individuals with a treatment goal of abstinence.

The AHRQ review (Jonas et al. 2014) found statistically significant increases in diarrhea and vomiting as compared with placebo, although statistical heterogeneity was high, particularly for diarrhea (Table B–2). In studies published since the AHRQ report (Jonas et al. 2014) and not included in Table B–2, diarrhea was also common. In a study by Berger et al. (2013), diarrhea occurred in almost one-third of subjects, but there was no difference between acamprosate and placebo. In a study by Higuchi and Japanese Acamprosate Study Group (2015), diarrhea occurred more frequently with acamprosate than with placebo (12.9% vs. 4.9%, respectively). In a study by Mann et al. (2013), diarrhea was also noted to be greater with acamprosate than with placebo.

Anxiety was also noted to be greater with acamprosate than with placebo in the AHRQ review, but this finding was based on one study, and other studies have noted no difference from placebo (Micromedex 2017a) or less anxiety (Mann et al. 2013) or even somnolence (Forest Pharmaceuticals 2005) with acamprosate. In studies published since the AHRQ report (Jonas et al. 2014), other side effects occurred in less than 10% of individuals treated with acamprosate or placebo (Berger et al. 2013; Higuchi and Japanese Acamprosate Study Group 2015), without differences in overall side effects (Higuchi and Japanese Acamprosate Study Group 2015) or study attrition due to adverse events (Mann et al. 2013) between the two groups.

In the package insert for acamprosate (Forest Pharmaceuticals 2005), adverse events of a suicidal nature were described as somewhat more common with acamprosate as compared with placebo (1.4% vs. 0.5% in studies of 6 months or less; 2.4% vs. 0.8% in year-long studies), with suicide in 3 of 2,272 (0.13%) patients in the pooled acamprosate group and 2 of 1,962 patients (0.10%) in the pooled placebo group. However, the AHRQ report notes that evidence was not sufficient to make a determination about the risk of suicide-related events (Jonas et al. 2014). The package insert also notes that acamprosate is contraindicated with severe renal impairment (creatinine clearance [CrCl] 30 mL/min or less) and requires dose adjustments for moderate renal impairment (CrCl of 30–50 mL/min). Other information on harms of acamprosate comes from nonrandomized trials and drug information databases and is noted in Statement 9, Implementation.

Studies related to acamprosate are listed in Table B–3.

In the AHRQ review (Jonas et al. 2014) (Table B–4), studies of oral naltrexone typically used a dose of 50 mg/day (Table B–5), but a few trials used doses of 100–150 mg/day (Table B–6); trials of long-acting injectable naltrexone used doses of 150–400 mg/month (Table B–7). With naltrexone treatment, 4% fewer subjects returned to any drinking than with placebo (RD, –0.04; 95% CI, –0.07 to –0.01; 21 trials of low or medium bias), and 7% fewer subjects returned to heavy drinking than with placebo (RD, –0.07; 95% CI, –0.11 to –0.03; 23 trials of low or medium bias). For oral naltrexone at a dose of 50 mg/day, the NNT was 20 to prevent one person from returning to any drinking, with a NNT of 12 to prevent one person from returning to heavy drinking. For doses of oral naltrexone of 100 mg/day and for injectable naltrexone, effects were similar to those for oral naltrexone at 50 mg/day but were not statistically significant. As compared with placebo, subjects who received naltrexone also had 4.6% fewer drinking days (WMD, –4.6; 95% CI, –6.6 to –2.5; 19 trials), 3.8% fewer heavy drinking days (WMD, –3.8; 95% CI, –5.8 to –1.8; 11 trials), and 0.5% fewer drinks per drinking day (WMD, –0.54; 95% CI, –1.01 to –0.07; 11 trials). The single study of injectable naltrexone found a large effect size (WMD, –8.6) for fewer drinking days relative to placebo.

Only a limited number of studies assessed factors related to quality of life, and these studies used different measures, making comparison or meta-analysis impossible. In addition, quality of life measures were secondary outcomes, and studies were not adequately powered to assess these effects. One study found better overall mental health, but not physical health, with long-acting injectable naltrexone at 380 mg/month but no benefit on either measure at a dose of 190 mg/month. A placebo-controlled study of 50 mg/day of oral naltrexone found fewer alcohol-related consequences in the naltrexone group (76% vs. 45%, p = 0.02). The other two studies assessing quality of life measures showed no statistical difference with naltrexone as compared with placebo.

One trial did not meet inclusion criteria for the comparative effectiveness review but was described in some detail in the AHRQ report. In this study (O’Malley et al. 2003), individuals all received oral naltrexone with random assignment to 10 weeks of either primary care management (PCM) or cognitive-behavioral therapy (CBT). Responders in each group (84.1% for PCM vs. 86.5% for CBT) continued with their assigned psychosocial treatment and were randomly assigned to continue naltrexone or switch to placebo. In the CBT group, the rates of abstinence decreased in those assigned to placebo but did not reach statistical significance, whereas in the PCM group, the placebo group had a greater reduction in abstinence rates than those who remained on naltrexone (80.8% vs. 51.9%, p = 0.03).

Several studies of oral naltrexone published since the AHRQ review (not included in Tables B–4, B–5, or B–6) have shown minimal benefits. In the German PREDICT study (total N = 426), modeled after the COMBINE study, there was no difference among naltrexone, acamprosate, and placebo groups on the time to first heavy drinking (Mann et al. 2013). In a 12-week, low-risk-of-bias trial, subjects (N = 221) were randomly assigned to 50 mg/day oral naltrexone or placebo in blocks on the basis of their OPRM1 genotype (Oslin et al. 2015). There was no difference in the odds of heavy drinking with naltrexone as compared with placebo for either genotype, although significant reductions in heavy drinking occurred in all treatment groups. A four-arm study (N = 200, medium risk of bias) of men who had sex with men investigated oral naltrexone 100 mg/day versus placebo and brief behavioral compliance treatment with and without modified behavioral self-control therapy (MBSCT) (Morgenstern et al. 2012). MBSCT was associated with a 28% decrease in drinks per week and a 35% decrease in heavy drinking days per week, whereas treatment with naltrexone did not have a statistically significant effect. However, naltrexone did increase the likelihood (odds ratio = 3.3) of achieving nonhazardous levels of drinking, which was the stated goal of study subjects.

In the majority of studies of naltrexone, individuals met criteria for alcohol dependence by DSM-IV criteria; however, one controlled trial of 153 early problem drinkers randomly assigned subjects to naltrexone, targeted naltrexone before high-risk drinking situations, or placebo (Kranzler et al. 2003), and found a reduced likelihood of drinking in the naltrexone or targeted treatment groups over an 8-week period. Although these findings require replication, they are consistent with possible benefit of naltrexone treatment in individuals with less severe AUD.

Although most trials of naltrexone excluded individuals with co-occurring physical or psychiatric illness, one study of naltrexone for smoking cessation conducted a subgroup analysis for individuals who also reported heavy drinking (Fridberg et al. 2014). The total sample included 315 smokers who were randomly assigned to placebo or naltrexone 50 mg/day for 12 weeks. In the subgroup of 69 heavy drinkers (at least two heavy drinking episodes per month), weekly alcohol consumption was reduced with naltrexone treatment (incidence rate ratio [IRR] 0.71, 95% CI = 0.51–1.0, p = 0.049), as was smoking urge. Smoking quit rates with naltrexone as compared with placebo were also significantly better in the heavy drinking subgroup at the end of the study and at 12-month follow-up. Another medium-risk-of-bias study (Foa et al. 2013) was excluded from the AHRQ review because of its study design but is of relevance to clinical practice. Subjects met DSM-IV criteria for posttraumatic stress disorder (PTSD) and for alcohol dependence and were randomly assigned to receive naltrexone 100 mg/day plus prolonged exposure therapy (N = 40), placebo plus prolonged exposure therapy (N = 40), naltrexone 100 mg/day plus supportive therapy (N = 42), or placebo plus supportive therapy (N = 43). Although attrition was relatively high in all groups during the 24-week trial, alcohol craving and the percentage of days drinking alcohol were reduced in all groups, with a greater mean difference in groups that received naltrexone as compared with placebo groups (p = 0.008). PTSD severity was reduced in all groups, with no significant effect of prolonged exposure over supportive therapy; however, those in the prolonged exposure plus naltrexone group were more likely to achieve a low level of PTSD symptoms.

The AHRQ review (Jonas et al. 2014) also examined studies that assessed whether μ opioid receptor gene polymorphism status was associated with a more robust response to naltrexone (Table B–8). The main single-nucleotide polymorphism (SNP) that was tested was an asparagine to aspartate substitution in exon 1 of the μ opioid receptor. (Because of changes in the National Center for Biotechnology Information Human Genome Reference Assembly, this SNP has been referred to by a number of designations, including A118G, Asn40Asp, rs1799971, A355G, and Asn102Asp.) The review found no significant difference between A-allele homozygotes and those with at least one G allele in terms of the outcomes return to any drinking (RD, 0.01; 95% CI, –0.2 to 0.2) and return to heavy drinking (RD, 0.14; 95% CI, –0.03 to 0.3) when all available studies were considered together. However, in its conclusions, the AHRQ report also noted that for return to heavy drinking, “it is possible that patients with at least one G allele of A118G polymorphism of OPRM1 might be more likely to respond to naltrexone” (Jonas et al. 2014, p. 94). The reasons behind this interpretation are severalfold. Of the seven studies, three studies, including the COMBINE study (Anton et al. 2008b), reported positive associations between OPRM1 polymorphisms and naltrexone response. In the COMBINE study, individuals who received medical management (MM) without cognitive-behavioral intervention were more likely to have a good clinical outcome if they had at least one Asp40 allele and received naltrexone (87.1%), as compared with Asn40 homozygotes treated with naltrexone (54.8%). About half of those treated with placebo also had a good outcome, regardless of genotype. This difference in outcomes would be clinically significant. One additional study did not meet a priori inclusion criteria for the systematic review, but it also included information on naltrexone response and OPRM1 genotype (Oslin et al. 2003). This study also found that naltrexone-treated subjects with at least one Asp40 allele as compared with Asn40 homozygotes had significantly lower rates of relapse (p = 0.044) and a longer time to return to heavy drinking (p = 0.04). When the results of this study were added to the meta-analysis in a sensitivity analysis, a positive association between genotype and response emerged (RD, 0.16; 95% CI, 0.02 to 0.29).

Since the AHRQ review, additional studies have not found a relationship between genotype and naltrexone response. As described above, one study randomly assigned subjects (N = 221) to 50 mg/day oral naltrexone or to placebo with stratification on the basis of their OPRM1 genotype (Oslin et al. 2015). In this 12-week trial, there was no difference in the odds of heavy drinking with naltrexone as compared with placebo for either genotype. A secondary analysis of OPRM1 genotype has been conducted in a sample of veterans with alcohol dependence and other psychiatric conditions (Arias et al. 2014). Subjects in this 12-week, medium-risk-of-bias study were randomly assigned to placebo alone (N = 64), naltrexone 50 mg/day (N = 59), disulfiram 250 mg/day plus placebo (N = 66), or naltrexone 50 mg/day and disulfiram 250 mg/day (N = 65). OPRM1 genotyping was conducted for a subset of 107 European American subjects. No significant interactions were found between genotype and the response to naltrexone.

Taken together, the findings on OPRM1 genotype and naltrexone response did not seem to indicate a current role for OPRM1 genotype determination in clinical practice, and no guideline statement was made. However, use of genotype to identify predictors of response remains a promising avenue for research.

Studies related to naltrexone are listed in Table B–10.

The AHRQ meta-analysis (Jonas et al. 2014) found no statistically significant difference between naltrexone and acamprosate on return to any drinking (RD, 0.02; 95% CI, –0.03 to 0.08; three trials), return to heavy drinking (RD, 0.01; 95% CI, –0.05 to 0.06; four trials), or drinking days (WMD, –2.98; 95% CI, –13.4 to 7.5) (Table B–11). Patient characteristics did not appear to be associated with a preferential response to either medication.

The COMBINE study (Anton et al. 2006, p. 2003) found that “patients receiving medical management with naltrexone, CBI, or both fared better on drinking outcomes than those who received placebo, but acamprosate showed no evidence of efficacy, with or without CBI.” Analyses of alternative summary measures of drinking, including drinks per drinking day (p = 0.03) and heavy drinking days per month (p = 0.006), were consistent with those for the co-primary end points (percentage of days abstinent from alcohol and time to first heavy drinking day) in showing a significant naltrexone by CBI interaction. Although the CBI and naltrexone treatment combination showed a statistically significant difference in quality of life measures, the AHRQ review noted that this was unlikely to be clinically significant (Jonas et al. 2014). In a subsequent analysis that examined predictors of abstinence from heavy drinking, assignment to a specific treatment was not a major contributor to outcome, but individuals with more consecutive days of abstinence, with a drinking goal of abstinence, or with a lesser frequency of alcohol use prior to treatment were more likely to achieve abstinence from heavy drinking (Gueorguieva et al. 2011, 2014). By 3 years, median but not mean costs (treatment cost plus social costs of AUD such as health care, arrests, and motor vehicle accidents) were diminished in the COMBINE study by a number of treatment combinations that included pharmacotherapy (Zarkin et al. 2010). Treatment arms that were cost-effective, from a policy (Dunlap et al. 2010) and patient-centered (Zarkin et al. 2008) standpoint, were MM with placebo, MM plus naltrexone therapy, and MM plus combined naltrexone and acamprosate therapy.

The medium-risk-of-bias German PREDICT study (total N = 426) is the only study identified in the updated literature search that included a head-to-head comparison of acamprosate and naltrexone (Mann et al. 2013) and is not included in Table B–11. This trial was modeled after the COMBINE study and found no difference among naltrexone, acamprosate, and placebo groups on the time to first heavy drinking. Point estimates for heavy drinking relapse free survival from the Kaplan-Meier curves were 48.3% for acamprosate, 49.1% for naltrexone, and 51.8% for placebo. A secondary analysis of adherent patients also showed no significant differences among the treatment groups.

In terms of adverse events, the risks of headache, nausea, and vomiting were noted to be slightly higher for individuals treated with naltrexone as compared with acamprosate in the AHRQ review (Jonas et al. 2014) (Table B–12). The number of deaths in head-to-head studies of naltrexone and acamprosate was extremely small, and no statistical comparison was possible (Jonas et al. 2014). In the PREDICT trial (not included in Table B–12), diarrhea was significantly greater with acamprosate, and nervousness/anxiety was greater in placebo subjects. Serious adverse events (9.9% of patients during active treatment and 17.4% during follow-up) and related dropouts (6.3%) did not differ among the treatment groups.

Studies comparing acamprosate and naltrexone are listed in Table B–13.

Evidence for the benefits of disulfiram comes from randomized controlled double-blind and open-label trials as well as expert opinion. The AHRQ review (Jonas et al. 2014) included four studies conducted at Veterans Health Administration Medical Centers and found no statistically significant difference between disulfiram 250 mg/day and sham comparators (i.e., placebo, disulfiram 1 mg/day, riboflavin). In the two trials included in the AHRQ review that assessed percentage of drinking days, one reported no significant difference among treatment groups. The other trial limited its reporting to a subset of subjects (those who drank during the trial and who also completed all assessments) and found that disulfiram was associated with fewer drinking days (p = 0.05) than for those who received a comparator (49% with disulfiram 250 mg/day vs. 75.4% with disulfiram 1 mg/day and 86.5% with riboflavin). In the two RCTs included in the AHRQ analysis that had a medium risk of bias (Fuller and Roth 1979, 1986), treatment adherence was associated with abstinence, regardless of whether the subject was assigned to active disulfiram or control treatment. Some shorter RCTs of disulfiram have also been associated with benefits on drinking-related outcomes (Jørgensen et al. 2011).

In a medium-risk-of-bias trial conducted in Japan (Yoshimura et al. 2014), subjects (total N = 109) were randomly assigned according to a 2 × 2 design with disulfiram 200 mg/day vs. placebo and receipt of educational material on drinking harms and craving management vs. no such education. At 26 weeks, there were no differences among groups in the percent of individuals who remained abstinent. However, this study may have limited generalizability because individuals were randomly assigned to disulfiram after a 2- to 3-month inpatient stay.

A single study in the AHRQ review (Petrakis et al. 2005) compared disulfiram, naltrexone, placebo, and the combination of disulfiram plus naltrexone for 12 weeks in Veterans Health Administration outpatient settings (Table B–14). Naltrexone was given in a double-blind fashion, but disulfiram was administered as an open-label medication. The trial found no statistically significant difference between disulfiram and naltrexone for number of subjects achieving total abstinence (51 vs. 38, p = 0.11), percentage of days abstinent (96.6 vs. 95.4, p = 0.55), or percentage of heavy drinking days (3.2 vs. 4, p = 0.65).

A meta-analysis (Skinner et al. 2014) differed from the AHRQ analysis in including RCTs that were open-label as well as randomized controlled double-blind trials. Skinner and colleagues (2014) hypothesized that in a double-blind trial, subjects in both disulfiram and placebo groups would avoid drinking because of having been warned of the potential for adverse events regardless of actual treatment assignment. They included 22 studies (2,414 subjects) and found a significant overall effect but no difference between disulfiram and control groups in the double-blind RCTs. When only open-label trials were considered, disulfiram was significantly better than controls on alcohol-related outcomes (Hedges’ g = 0.70; 95% CI = 0.46–0.93), for which control conditions included acamprosate, naltrexone, and no disulfiram. Individual comparisons for each of these control conditions were also statistically significant. In studies where medication adherence was assured through supervised administration, the effect of disulfiram was large. As with the double-blind RCTs, however, only a small proportion of women were included in the open-label trials, which limits generalizability.

The data on harms from the studies included in the AHRQ report were insufficient to conduct meta-analyses. One study showed a greater rate of drowsiness in patients receiving versus not receiving disulfiram (8% vs. 2%, p = 0.03). Several patients discontinued disulfiram because of increased levels of hepatic enzymes. A four-arm study (2 × 2, disulfiram vs. placebo, naltrexone vs. placebo) showed greater rates of specific side effects in patients taking any study medication but no differences between groups. In this study, patients taking disulfiram and placebo experienced 6 of 14 serious adverse events. In the study of Yoshimura and colleagues (2014), 1/53 disulfiram treated subjects had a dermatological problem, 2/53 had liver enzyme elevations, and 1/53 had renal dysfunction, whereas no adverse events were noted in placebo-treated subjects. In the study of Petrakis and colleagues (2005), which compared disulfiram, naltrexone, placebo, and the combination of disulfiram plus naltrexone, fever was more common in the disulfiram group than in the naltrexone group (p = 0.03), whereas nervousness (p = 0.005) and restlessness (p = 0.03) were more common in the naltrexone group than in the disulfiram group.

In the meta-analysis of Skinner et al. (2014), data from randomized, controlled, open-label trials showed considerable heterogeneity but showed a significantly greater number of adverse events with disulfiram as compared with control conditions.

Additional information on potential harms of disulfiram comes from the product labelling (Rising Pharmaceuticals 2016), which notes that disulfiram should not be given to individuals who have recently received metronidazole, paraldehyde, alcohol (within 12 hours), or alcohol-containing preparations. It is also noted to be contraindicated in the presence of severe myocardial disease or coronary occlusion. When alcohol is taken within 14 days of disulfiram ingestion, it can produce the following:

flushing, throbbing in head and neck, throbbing headache, respiratory difficulty, nausea, copious vomiting, sweating, thirst, chest pain, palpitation, dyspnea, hyperventilation, tachycardia, hypotension, syncope, marked uneasiness, weakness, vertigo, blurred vision, and confusion. In severe reactions, there may be respiratory depression, cardiovascular collapse, arrhythmias, myocardial infarction, acute congestive heart failure, unconsciousness, convulsions, and death.

Disulfiram is noted to be contraindicated in the presence of psychosis or in individuals with hypersensitivity to disulfiram or thiuram derivatives used in pesticides and rubber production. Hepatic toxicity is also reported to have occurred in individuals receiving disulfiram.

Studies related to disulfiram are listed in Table B–15.

Evidence for topiramate comes from multiple randomized controlled trials, some of which included subjects with co-occurring conditions. The AHRQ review (Jonas et al. 2014) included three studies of topiramate versus placebo and one study of topiramate versus naltrexone versus placebo. The latter study (Baltieri et al. 2008, 2009) was rated as having a high risk of bias and showed no significant differences in the two treatments on drinking outcomes. The two placebo-controlled trials (total N = 521) that had a low or medium risk of bias were included in the AHRQ meta-analysis (Johnson et al. 2003, 2007). These trials had a duration of 12–14 weeks and were both conducted in the United States. On the basis of this meta-analysis, the AHRQ review concluded that there was a moderate strength of evidence for topiramate efficacy on drinks per drinking days (WMD: –1.10, 95% CI –1.75 to –0.45), percentage of heavy drinking days (WMD: –11.53, 95% CI –18.29 to –4.77), and percentage of drinking days. For the latter outcome, it was not possible to combine the results of the two trials, but each showed a comparable mean difference (WMD: –8.5, 95%, CI –15.9 to –1.1; mean difference –11.6, 95% CI –3.98 to –19.3). Findings from sensitivity analyses were similar when high-risk-of-bias studies were included.

A number of subsequent randomized controlled trials (not included in Table B–16) have also examined effects of topiramate. In a low-risk-of-bias U.S. government–funded trial, topiramate in doses of up to 200 mg/day (N = 67) was compared with placebo (N = 71) and was associated with a larger (p = 0.001) and more rapid (p = 0.0001) reduction in heavy drinking and a larger (p = 0.03) and more rapid (p = 0.01) increase in the number of days abstinent (Kranzler et al. 2014a). Topiramate subjects were more likely to have had no heavy drinking days in the last 4 weeks of treatment (35.8% vs. 16.9% with placebo, OR = 2.75, 95% CI 1.24–6.10) and to have abstained from alcohol use at the end of treatment (OR = 2.57, 95% CI  1.13–5.84). The odds of a heavy drinking day were greater in the placebo group than the topiramate group (OR = 5.33, 95% CI  1.68–7.28) by the last week of treatment. These benefits of topiramate appeared to be limited to individuals who were homozygous for the rs2832407 C-allele of GRIK1 (which encodes the kainate GluK1 receptor subunit). However, at 3- and 6-month follow-up, the beneficial effects of topiramate on percent heavy drinking days and percent days abstinent were no longer significant (Kranzler et al. 2014c). Topiramate (300 mg/day; N = 21) was also one of the treatment arms in a 14-week medium-risk-of-bias, double-blind, randomized controlled trial of several other anticonvulsant agents that included levetiracetam (N = 21), zonisamide 400 mg/day (N = 19), and placebo (N = 24) (Knapp et al. 2015). For topiramate as compared with placebo, significant treatment effects were seen for weekly percent days drinking (P < 0.0001), percent days heavy drinking (P < 0.0001), and drinks consumed per day (P = 0.0007). A 12-week, medium-risk-of-bias, double-blind, randomized placebo-controlled trial of topiramate (260 mg/day average dose) conducted in Thailand (total N = 106) was limited by 50% attrition rates but showed no significant difference between the treatments in heavy drinking days, time to first heavy drinking day, or secondary drinking outcomes (Likhitsathian et al. 2013).

Several smaller studies of topiramate have been conducted in individuals with a co-occurring psychiatric disorder. A small (total N = 30) double-blind, randomized placebo-controlled trial of flexibly dosed topiramate (up to 300 mg/day) was conducted at a Veterans Affairs Medical Center in individuals with co-occurring PTSD (Batki et al. 2014) (Table B–16). This low-risk-of-bias study showed a 51% decrease in drinking days with topiramate as compared with placebo as well as reductions in standard drinks per week but no effect on the percent of heavy drinking days. Another U.S. government–funded, low-risk-of-bias, double-blind, randomized placebo-controlled trial of topiramate (300 mg/day) enrolled individuals with co-occurring cocaine dependence (Kampman et al. 2013). During the 13-week trial, 41/87 (47%) of placebo-treated subjects were lost to follow-up versus 29/83 (35%) with topiramate. However, on primary outcome measures of weekly differences in percent days drinking, percent days heavy drinking, and mean drinks per drinking day, there was no difference between the placebo and topiramate-treated groups. An additional study in individuals with co-occurring bipolar disorder reported the results of 12 randomly assigned participants but had difficulty recruiting subjects because of problems with topiramate tolerability (Sylvia et al. 2016).

Studies of topiramate in other disorders have reported a number of treatment-related side effects. In the studies of topiramate for AUD that were included in the AHRQ report (Jonas et al. 2014), the most notable side effects of topiramate as compared with placebo were cognitive dysfunction and numbness/tingling/paresthesias (Table B–17). In the study of Likhitsathian et al. (2013), paresthesias were more common in the topiramate group as compared with placebo (45.3% vs. 17%). Kampman et al. (2013) also found a greater frequency of paresthesias in topiramate-treated subjects as compared with placebo-treated subjects (20% vs. 3%). Knapp et al. (2015) also noted paresthesias in 19% of topiramate subjects and erectile dysfunction in 14% of topiramate subjects. In addition, Knapp et al. (2015) found a significant effect of topiramate on the mental slowing subscale of the A-B Neurotoxicity Scales relative to placebo (P = 0.008). Batki et al. (2014) found no significant differences in side effects between topiramate- and placebo-treated subjects.

Studies related to topiramate are listed in Table B–18.

The AHRQ review (Jonas et al. 2014) did not include any studies with a primary focus on gabapentin. In one included study (Anton et al. 2011), gabapentin was added in one treatment arm as an adjunct to naltrexone during the initial 6 weeks of the trial and was associated with improved outcomes at 6 weeks but not at the end of the trial. Several small randomized trials of shorter duration also showed benefit of gabapentin on alcohol-related outcomes (Anton et al. 2009; Furieri and Nakamura-Palacios 2007).

A government-funded low-risk-of-bias, double-blind, randomized controlled dose-ranging trial (Mason et al. 2014) compared gabapentin at 900 mg/day (N = 54) and 1,800 mg/day (N = 47) with placebo (N = 49). The primary study outcomes, which were rate of complete abstinence (chi square = 4.19; P = .04) and rate of no heavy drinking (chi square = 5.39; P = .02), increased linearly with the dose of gabapentin. Sustained 12-week abstinence was 4.1% (95% CI 1.1%–13.7%) with placebo, 11.1% (95% CI 5.2%–22.2%) with 900 mg/day of gabapentin, and 17.0% (95% CI 8.9% –30.1%; NNT = 8) with 1,800 mg/day gabapentin with a NNT of 8 for increased rate of abstinence at a dose of 1,800 mg daily. Corresponding rates of no heavy drinking were 22.5% (95% CI 13.6%–37.2%), 29.6% (95% CI 19.1%–42.8%), and 44.7% (95% CI 31.4%–58.8%; NNT = 5), respectively, with a NNT of 5 for reduction in heavy drinking days at a dose of 1,800 mg daily. Significant dose-dependent reductions were also noted in the prespecified secondary outcomes: levels of GGT, alcohol craving, sleep, and depression. For subjects who completed the trial, rates of complete abstinence, drinks per week, and number of heavy drinking days per week were sustained at 24-week follow-up. The most frequent adverse events were fatigue (23%), insomnia (18%), and headache (14%), but rates of these side effects did not differ among the three study arms. In addition, there were no differences in the number, severity, or type of reported adverse effects (Mason et al. 2014). Insufficient information was available on side effects of gabapentin to grade the overall supporting body of research evidence for harms.

Studies related to gabapentin are listed in Table B–19.

APA recommends (1B) that antidepressant medications not be used for treatment of alcohol use disorder unless there is evidence of a co-occurring disorder for which an antidepressant is an indicated treatment.

Evidence for this recommendation comes from a number of studies of serotonin reuptake inhibitors and tricyclic antidepressants that assessed alcohol-related outcomes in individuals with alcohol dependence and a depressive or anxiety disorder (Jonas et al. 2014). On the basis of a substantial number of trials that directly assess the efficacy of antidepressant medications in treating AUD, the strength of research evidence is rated as moderate.

The AHRQ review (Jonas et al. 2014) included seven trials comparing placebo with sertraline in doses of 50–200 mg/day and treatment durations of 12–26 weeks. Of the seven studies, five were done in the United States, three included only individuals with major depressive disorder and alcohol dependence, and one included individuals with PTSD and alcohol dependence. Meta-analysis did not show a benefit of sertraline on the alcohol-related outcomes, and for the outcome of percent of heavy drinking days the comparison favored placebo (low strength of research evidence; WMD: 1.85 [0.70 to 3.0]). An additional study (total N = 170) compared placebo with naltrexone alone, sertraline alone, or the combination of naltrexone and sertraline and reported no difference between sertraline and placebo conditions on abstinence rates (Pettinati et al. 2010). The combination of naltrexone plus sertraline showed greater abstinence rates than either treatment alone (p = 0.001) as well as a longer time to relapse to heavy drinking. A subsequent double-blind RCT of sertraline 200 mg/day (N = 32) versus placebo (N = 37) was conducted in individuals with co-occurring PTSD and alcohol dependence (Hien et al. 2015). Treatment in this low-risk-of-bias trial also included 12 sessions of a “Seeking Safety” intervention. At the end of treatment, at 6-month follow-up, and at 12-month follow-up, both sertraline and placebo subjects showed a decreased number of drinks per drinking day, a decrease in heavy drinking days, and an increase in 7-day abstinence rate. PTSD symptoms showed greater improvement with sertraline than placebo, but there was no specific effect of sertraline treatment as compared with placebo on alcohol-related outcomes.

The AHRQ review included two trials (Naranjo et al. 1995; Tiihonen et al. 1996) of 12–13 weeks duration that compared citalopram 40 mg/day with placebo. Both trials were rated as having a high risk of bias, and neither trial showed an effect of citalopram on drinking-related outcomes. A subsequent medium-risk-of-bias 12-week trial of citalopram 40 mg/day (N = 138) versus placebo (N = 127) found worse outcomes with citalopram than placebo in terms of the percentage decrease in the frequency of alcohol consumption (p = 0.016), the percentage decrease in the quantity of alcohol consumed per drinking day (p = 0.025), the average number of heavy drinking days (p = 0.007), drinks per drinking day (p = 0.03), and money spent on alcohol (p = 0.041) (Charney et al. 2015). When individuals with depression were compared with those without depression, the findings in both subgroups were consistent with findings for the overall sample. In another 12-week study in which all subjects (total N = 138) received naltrexone (up to 100 mg/day), there was no significant difference on alcohol use or depression-related outcomes between subjects who were randomly assigned to citalopram (up to 60 mg/day) and those assigned to placebo (Adamson et al. 2015).

The AHRQ review (Jonas et al. 2014) included three U.S. trials lasting 12–15 weeks and comparing placebo with fluoxetine in doses from 20 mg to 60 mg per day (Cornelius et al. 1995; Kabel and Petty 1996; Kranzler et al. 1995). In one of the trials, in which all subjects (N = 51) had major depressive disorder, subjects treated with fluoxetine had fewer drinking days (WMD –11.6; 95% CI –22.7 to –0.5) and fewer heavy drinking days (4.8 versus 16, p = 0.04) than those who received placebo (Cornelius et al. 1995). When the two medium-risk-of-bias trials were combined (Cornelius et al. 1995; Kranzler et al. 1995), meta-analysis found no difference between fluoxetine and placebo in drinking days (WMD –3.2; 95% CI –18.2 to 11.9) or heavy drinking days (WMD –1.2; 95% CI –4.6 to 2.2).

In a single European trial of fluvoxamine 100–300 mg/day as compared with placebo, there was no difference at 12 weeks of treatment or at 52 weeks of follow-up in the percent of subjects who had returned to drinking or the percent who returned to heavy drinking (Chick et al. 2004). At 12 weeks, fluvoxamine-treated patients had more drinking days in the prior month than placebo-treated patients, but the groups did not differ on this outcome at 52 weeks of follow-up.

One randomized trial compared paroxetine (10–60 mg/day, mean dose 45 mg/day) with placebo in individuals with social anxiety disorder, of whom 79% of 42 subjects also had a co-occurring diagnosis of alcohol dependence (Book et al. 2008; Thomas et al. 2008). After 16 weeks (12 weeks at final paroxetine dose), there was no difference in the mean number of drinks per drinking day or the proportion of drinking days or heavy drinking days for paroxetine-treated patients as compared with placebo-treated patients. In an additional high-risk-of-bias trial (Petrakis et al. 2012), paroxetine with and without naltrexone was compared with desipramine with and without naltrexone in subjects with co-occurring alcohol dependence and PTSD. Individuals who received paroxetine had more heavy drinking days (p = 0.009) and drinks per drinking day (p = 0.027) than those who received desipramine. Although all groups showed reductions in PTSD symptoms, combination treatment with naltrexone and an antidepressant was associated with reduced craving as compared with groups treated with antidepressant plus placebo (Petrakis et al. 2012).

Another U.S. study with a medium-risk-of-bias compared desipramine (median dose = 200 mg/day) with placebo. In this trial, 39% also had a diagnosis of depression (Mason et al. 1996). Although 12% of desipramine-treated patients returned to heavy drinking as compared with 32% of placebo-treated patients, this difference was not statistically significant. A medium-risk-of-bias-study of imipramine 50–300 mg/day (mean dose = 262 mg/day) as compared with placebo in individuals with depression and alcohol dependence found no significant difference between imipramine and placebo groups on percent return to any drinking, percent heavy drinking, or number of drinks per drinking day (McGrath et al. 1996).

Studies related to antidepressants are listed in Table B–20.

Evidence for this recommendation is indirect and is based primarily on expert opinion. Consequently, the strength of research evidence is rated as low. The systematic review of the literature did not yield any references that dealt directly with the use of a benzodiazepine to treat AUD, except in the context of alcohol withdrawal or alcohol detoxification. A Cochrane review of pharmacotherapy for co-occurring AUD and anxiety disorders also did not find any randomized trials of benzodiazepines for anxiety disorders in this population, although studies of naltrexone, acamprosate, and disulfiram were excluded from the review (Ipser et al. 2015). One small open-label study (Bogenschutz et al. 2016) assessed use of lorazepam in combination with disulfiram and manual-based MM in individuals with DSM-IV alcohol dependence and symptoms of anxiety. Subjects had reductions in anxiety, depression, and craving and had no signs of misuse or dose escalations for lorazepam, but two-thirds of the 41 subjects were no longer adherent to treatment at 16 weeks.

Evidence for this recommendation is indirect and is based on data from case reports, registries, case control studies of birth outcomes, and, in some instances, animal studies of teratogenicity and neurodevelopmental effects of medication exposure during pregnancy. Consequently, the strength of research evidence is rated as low. Additional evidence that was considered in making this recommendation was the relatively small effect sizes of these medications for treatment of AUD as discussed with Statements 9, 10, and 11.

Data in pregnant animals suggest a moderate risk for use of naltrexone, high risk for use of acamprosate, and possible risks for use of gabapentin and topiramate (Briggs and Freeman 2015). For disulfiram, Briggs and colleagues (Briggs and Freeman 2015) noted that there are no animal data available. Data for the use of these medications in pregnant women is limited (Briggs and Freeman 2015); however, an increased risk of malformation does appear to be associated with use of topiramate (Alsaad et al. 2015; Briggs and Freeman 2015; Tennis et al. 2015; Weston et al. 2016) but not gabapentin (Weston et al. 2016). No clustering of birth defects has been seen when disulfiram is taken by pregnant women, but samples have been small (Briggs and Freeman 2015).

Little data are available on the use of these medications in breastfeeding women, but there may be potential for toxicity with disulfiram and naltrexone (Briggs and Freeman 2015; Sachs and Committee on Drugs 2013) as well as topiramate (Briggs and Freeman 2015), whereas acamprosate and gabapentin are noted to be “probably compatible” (Briggs and Freeman 2015) with breastfeeding.

Evidence for this statement comes from a pharmacokinetic study (Sennesael 1992), which shows increases in terminal elimination half-life and peak plasma concentration with decreases in renal clearance of drug from plasma after a single dose of 666 mg of acamprosate. Individuals with moderate (CrCl of 1.8–3.6 L/h/1.73 m²) or severe (CrCl of 0.3–1.74 L/h/1.73 m²) renal impairment had a mean terminal elimination half-life of 33.4 hours and 46.6 hours, respectively, as compared with 18.2 hours for healthy volunteers (with CrCl of > 4.5 L/h/1.73 m²). Peak plasma concentrations were 198 mcg/L for healthy volunteers, as compared with 398 mcg/L and 813 mcg/L for individuals with moderate or severe renal impairment, respectively. On the basis of the significant curvilinear relationship between renal impairment and pharmacokinetic properties, the overall strength of research evidence was viewed as low.

Evidence for this statement also comes from a pharmacokinetic study (Sennesael 1992), as described in Statement 15 above. Evidence for reducing the dose of acamprosate, if it is used, comes from basic principles of pharmacokinetics.

Evidence for this recommendation is indirect. Direct data are not available for the conditions specified in this recommendation (i.e., acute hepatitis, hepatic failure) because individuals with these conditions were excluded from clinical trials. Consequently, the strength of research evidence is rated as low. In early studies of other conditions (e.g., obesity, dementia), some patients had severalfold elevations in hepatic transaminase levels with naltrexone treatment (Knopman and Hartman 1986; Malcolm et al. 1985; Mitchell et al. 1987; Pfohl et al. 1986; Verebey and Mulé 1986). The FDA initially included a black box warning on the package labeling discussing potential hepatotoxicity and recommending that naltrexone not be used in individuals with acute hepatitis or hepatic failure. However, subsequent studies suggested that elevations of hepatic enzymes in individuals treated with naltrexone occurred at about the same frequency as in individuals treated with placebo (Brewer and Wong 2004; Lucey et al. 2008; Vagenas et al. 2014; Yen et al. 2006). In addition, a small study suggested that hepatic enzymes did not change and that reducing the dose of naltrexone was not needed in individuals with mild to moderate hepatic impairment (Turncliff et al. 2005). Consequently, the FDA removed the black box warning (Stoddard and Zummo 2015), although the potential for adverse hepatic effects continues to be noted in the package labeling for naltrexone.

Evidence for this recommendation is indirect, and consequently, the strength of research evidence is rated as low. Multiple studies have used opioid antagonists to hasten opioid discontinuation in individuals with opioid use disorder (Gowing et al. 2009, 2010). Although opioid antagonist administration was reliable in producing opioid withdrawal, the extent of any benefit was unclear, and potential for complications was noted (Gowing et al. 2009, 2010). These findings suggest that naltrexone not be given to individuals who are currently using opioids unless there is a clinically appropriate period of opioid abstinence before naltrexone initiation. Expert opinion is consistent with this recommendation. Clinical experience also suggests a need for adjustment to typical regimens for pain management in individuals who are receiving naltrexone (R. Chou et al. 2016; Vickers and Jolly 2006) because of the effects of naltrexone in blocking opioid receptors.

APA recommends (1C) that in patients with alcohol use disorder and co-occurring opioid use disorder, naltrexone be prescribed to individuals who