Which interventions for alcohol use should be included in a universal healthcare benefit package? An umbrella review of targeted interventions to address harmful drinking and dependence

Abstract

Background

This study aimed to identify targeted interventions for the prevention and treatment of harmful alcohol use. Umbrella review methodology was used to summarise the effectiveness across a broad range of interventions, in order to identify which interventions should be considered for inclusion within universal health coverage schemes in low- and middle-income countries.

Methods and findings

We included systematic reviews with meta-analysis of randomised controlled trials (RCTs) on targeted interventions addressing alcohol use in harmful drinkers or individuals with alcohol use disorder. We only included outcomes related to alcohol consumption, heavy drinking, binge drinking, abstinence, or alcohol-attributable accident, injury, morbidity or mortality. PubMed, Embase, PsycINFO, Cochrane Database of Systematic Reviews, and the International HTA Database were searched from inception to 3 September 2021. Risk of bias of reviews was assessed using the AMSTAR2 tool. After reviewing the abstracts of 9,167 articles, results were summarised narratively and certainty in the body of evidence for each intervention was assessed using GRADE. In total, 86 studies met the inclusion criteria, of which the majority reported outcomes for brief intervention (30 studies) or pharmacological interventions (29 studies). Overall, methodological quality of included studies was low.

Conclusions

For harmful drinking, brief interventions, cognitive behavioural therapy, and motivational interviewing showed a small effect, whereas mentoring in adolescents and children may have a significant long-term effect. For alcohol use disorder, social network approaches and acamprosate showed evidence of a significant and durable effect. More evidence is required on the effectiveness of gamma-hydroxybutyric acid (GHB), nalmefene, and quetiapine, as well as optimal combinations of pharmacological and psychosocial interventions. As an umbrella review, we were unable to identify the extent to which variation between studies stemmed from differences in intervention delivery or variation between country contexts. Further research is required on applicability of findings across settings and best practice for implementation.

Funded by the Thai Health Promotion Foundation, grant number 61–00-1812.

Introduction

Alcohol use is the leading risk factor for premature death and disability among adults aged 15–49 years globally, contributing to 2.8 million deaths in 2016 [1]. While targeted policies and interventions over the past decade have successfully reduced population-level alcohol consumption in high-income countries, alcohol use continues to rise in low-income and middle-income countries (LMICs) [1, 2], contributing to a 5% increase in the number of disability-adjusted life years (DALYs) attributable to alcohol use globally between 2007 and 2017 [3]. Risk of all-cause mortality and cancers has been shown to increase proportionately with alcohol consumption. Hence, without effective measures to curb its use, alcohol is likely to place an increasing burden on LMIC health systems [1].

Harmful alcohol use refers to people experiencing detrimental health and social consequences because of their drinking, which extends to the people around them and society as a whole [4], whereas alcohol use disorders represent a sub-set of harmful drinkers, who experience chronic relapsing brain disorder with an impaired ability to stop or control alcohol use despite adverse social, occupational, or health consequences [5]. At the population level, there is a considerable body of evidence to suggest that alcohol taxation and pricing control are the most effective interventions to decrease alcohol consumption and heavy drinking, and this finding appears to be consistent across geographies and socio-economic groups [6,7,8]. Such policies are long-term measures to prevent initiation of drinking and reduce alcohol consumption, whereas short-term measures seek to reduce adverse health and social consequences among current drinkers [6]. The effectiveness of many short-term measures is likely to be context-specific, especially when applied to LMICs with varying health systems and cultural differences [9, 10]. This can present challenges for policymakers in interpreting the often conflicting body of evidence to identify which interventions to fund in their setting.

For countries with universal health coverage (UHC) schemes, policymakers must decide on the package of interventions to provide for harmful alcohol use. In many high-income countries, health technology assessment agencies employ horizon scanning to identify new technologies and treatments to provide to the population [11]. However, for LMICs with nascent UHC schemes or looking to improve the package of services provided under UHC, there may be a need to scope all available intervention options that are available and shortlist those that merit further evaluation for inclusion under UHC. Umbrella reviews, or overviews of systematic reviews, have a broader scope than individual systematic reviews or meta-analyses, and can therefore examine a broad range of interventions aligned with choices facing policymakers [12]. Moreover, umbrella reviews require fewer resources to conduct, compared with undertaking a series of systematic reviews for a clinical area, which makes umbrella reviews more feasible for informing policy choices in resource-constrained settings. In response to a request from the Thai Health Promotion Foundation to review effective interventions to address harmful alcohol use, we therefore employed an umbrella review methodology to scope interventions that may warrant further assessment for inclusion or removal from the Thai universal health coverage scheme.

The aim of this review was to identify targeted interventions for the prevention and treatment of harmful alcohol use, and to summarise the evidence of their effectiveness. We defined targeted interventions as those that were based on prior risk screening [13]. We had the following two research questions:

• 1. Which interventions are effective at reducing alcohol use among individuals identified as having, or being at risk of, harmful alcohol use?

• 2. Which interventions are effective for the treatment of alcohol use disorder?

Although the main purpose of this study was to inform policy in Thailand, we expect that our review will prove generalisable to other settings, especially as we explore umbrella review methodology as a pragmatic approach by which to shortlist interventions for further assessment during the benefit package selection process in LMICs.

Methods

We conducted an umbrella review, or review of systematic reviews. The protocol for this study was developed a priori, registered in the International Prospective Register of Systematic Reviews (PROSPERO) (CRD42017083412) and published as a pre-print [14]. The study was developed following the Cochrane Handbook for Systematic Reviews of Interventions [15] and adheres to the Preferred Reporting Items for Overviews of Reviews (PRIOR) guideline (Supplement 1) [16] and PRISMA-S checklist for reporting literature searches (Supplement 2) [17].

Eligibility criteria and search strategy

We searched MEDLINE via PubMed, Embase, PsycINFO, Cochrane Database of Systematic Reviews, and the International HTA Database from inception to 3 September 2021. Search strategies for each database are provided in Supplement 3. No language restrictions were applied and no published search filters were used. We further searched the reference lists of identified studies.

We included articles that met the following criteria: (1) participants were identified as risky drinkers, harmful/hazardous drinkers, or diagnosed with alcohol use disorder or dependency; (2) interventions were targeted interventions (i.e. based on prior risk screening [13]) and primarily intended to address alcohol use; (3) outcomes were related to alcohol consumption, heavy drinking, binge drinking, abstinence, or alcohol-attributable accident, injury, morbidity or mortality; (4) studies were systematic reviews with meta-analysis of randomised controlled trials (RCTs), for which the results from alcohol RCTs were presented separately and not combined with other interventions or study types. We chose to only include meta-analyses to facilitate comparison of quantitative data across reviews (especially given the very broad scope of the study), and we restricted inclusion to RCTs to reduce risk of bias [18]. We excluded any articles that were universal preventive interventions, which are provided to everyone regardless of individual risk [13], and meta-analyses that either did not meet the criteria for a systematic review [19] or for which only one RCT was included in the analysis. We accepted any comparator or setting.

Selection and data extraction

Reviews retrieved from the database search were imported into Covidence (systematic review management software). After removal of duplicates, articles were screened independently and in duplicate according to the eligibility criteria, first by title/abstract and then by full text. Eligibility criteria were piloted by all authors and refined before screening. Any conflicts were resolved by a third reviewer.

A data extraction form was developed in Excel and piloted by all authors before starting data extraction. One-third of papers were extracted independently in duplicate. Since there was good concordance among the articles extracted in duplicate, the remaining articles were extracted by a single reviewer and validated by a second reviewer. The data extraction form included citation details, purpose, eligibility criteria, meta-analysis methods, summary of meta-analysis results (including effect size, heterogeneity, publication bias, risk of bias), and main conclusions. Missing or unclear data were left blank.

Quality assessment

Methodological quality of included reviews was assessed using the A MeaSurement Tool to Assess systematic Reviews (AMSTAR) 2, which comprises 16 domains and provides a rating from ‘High’ to ‘Critically low’ [20]. For each article, the domains of the AMSTAR 2 tool were assessed independently by two reviewers, with conflict resolution by a third reviewer. Before carrying out the quality assessment, authors came to a consensus on the AMSTAR 2 domains that were considered critical/non-critical, in order to enhance consistency in the final ratings across reviewers. Domains that were considered critical are highlighted in bold in Fig. 2 and comprise the following: protocol established a priori, comprehensive search strategy, excluded studies listed and justified, satisfactory technique to assess risk of bias (RoB), appropriate methods for meta-analysis, RoB of individual studies discussed in results, publication bias investigated and discussed.

Synthesis of results

Outcomes from each meta-analysis were grouped by type of intervention (Table 1), using an adapted version of the classification from the International Standards for the Treatment of Drug Use Disorders [21], and further stratified by population, comparator, and outcome for narrative analysis. Risk of bias from primary studies was reported without re-assessment by study authors. Similarly, we reported discrepant data without seeking to manage discrepant results and reported publication bias where assessed in the meta-analysis, but did not undertake additional analysis. Forest plots were not used to summarise results because of the heterogeneity in effect sizes, populations, methods to deliver interventions, and outcomes across studies. Overlap between studies was not assessed using a citation matrix because the primary purpose of this review was intended to identify interventions for which there is evidence of an effect, and not to assess relative effectiveness of interventions. Certainty in the body of evidence for each intervention category was assessed using GRADE criteria modified for an umbrella review [22, 23].

Table 1 Classification of interventions within the scope of this review, adapted from the World Health Organization (WHO)/United Nations Office on Drugs and Crime (UNODC) International Standards for the Treatment of Drug Use Disorders [21]

Validation of results

To identify whether the umbrella review approach provided a comprehensive overview of the effectiveness of alcohol interventions, we compared our results with the National Institute for Health and Care Excellence (NICE) clinical guideline for harmful drinking and alcohol dependence in the United Kingdom [24]. We chose to use a clinical guideline for comparison in order to cover the full scope of our umbrella review, whilst recognising that clinical practice guideline development follows a different process to our review, most notably by including extensive stakeholder consultation with clinicians, patients, the public, and other groups. NICE was selected as an agency recognised for its methodological rigour. We additionally compared results against guidelines from Australia, Canada, Germany, and the USA (other countries with publicly available guidelines accompanied by evidence and evidence ratings for each recommendation) to verify whether the NICE guideline is similar to other countries [25,26,27,28,29].

Results

The following sections present the characteristics of included studies and assessment of methodological quality, followed by findings for each type of intervention and a comparison of findings with clinical practice guidelines.

Characteristics of included studies

The initial search yielded 9,167 articles, of which 1,413 were duplicates. Following title and abstract screening, 829 studies underwent full text review, yielding 86 studies for the review (Fig. 1). Most of the studies excluded during title/abstract screening did not report on an intervention primarily aimed at reducing or preventing alcohol use (Supplement 7). The characteristics of the included studies are detailed in Table 2 (with full details available in Supplement 4) and studies excluded during full text screening are listed in Supplement 5.

Fig. 1

PRISMA diagram of study selection process

Table 2 Characteristics of included studies

Almost all studies were direct meta-analyses, with only 6 studies (7%) using network meta-analysis. Most reviews were on brief interventions (30 studies, 35%), followed by pharmacological interventions (29 studies, 34%), multiple interventions (14 studies, 16%), and psychosocial interventions (10 studies, 12%). There was additionally one study on each of the following: physical activity [100], bibliotherapy [99], and home visits during pregnancy [101]. A third of studies did not specify setting of the review. General practice or primary healthcare settings were the most common, at 14 studies (16%), followed by educational institutions (8 studies, 9%) and emergency departments (7 studies, 8%).

Quality assessment

The average AMSTAR rating of studies was low (Fig. 2). Among the 11 studies published before 2009, all but two were rated as very low quality. More than 80% studies were assessed as having included PICO components (population, intervention, comparator, outcome), explained the selection of study designs, conducted data extraction in duplicate, selected appropriate methods for meta-analysis, and provided a satisfactory explanation of heterogeneity. However, 66 of 86 studies (77%) did not report funding sources of included studies. Less than half of studies reported a comprehensive search strategy or assessed the impact of risk of bias on the meta-analysis.

Fig. 2

Quality assessment of included studies (AMSTAR2 rating)

Screening, brief intervention, and referral to treatment (SBIRT) outcomes

From the review, 29 studies were identified for screening, brief intervention and referral to treatment and 1 network meta-analysis considered the impact of brief intervention combined with other interventions (Tables 3 and 4).

Table 3 Summary results of included systematic reviews, for screening, brief intervention, and referral to treatment (SBIRT)

Table 4 Certainty in evidence (GRADE) for screening, brief intervention, and referral to treatment (SBIRT)

Eight studies reviewed the efficacy of brief interventions across all hazardous and harmful drinkers [30, 31, 33, 38, 39, 46, 49, 115]. One high quality review of brief interventions totalling up to five sessions with total duration less than 60 min found a small but significant reduction across seven measures of alcohol consumption, heavy drinking, and binge drinking after a period of 12 months [46]. Five studies on interventions provided in primary care or general practice settings consistently found a significant effect beyond 6 months follow-up, across heavy drinking (absolute risk reduction 10.5%, 95% CI 7.1% to 13.9%) [30], frequency of hazardous drinkers (OR 1.6, 95% CI 1.33, 1.93) [31], net change in ethanol consumed (OR -37.87, 95% CI -51.13 to -24.61) [33], and grams of alcohol consumed per week (MD -30.86, 95% CI -46.49 to -15.23, at 12 months [38]. Among the studies on brief interventions delivered following opportunistic screening among non-treatment seeking drinkers in the emergency department, one found a significant reduction in grams of alcohol consumed per week (MD -18.21, 95% CI -26.71 to -9.70 at 12 months), while the other found a very small effect on alcohol consumption up to 5 months (SMD -0.15, 95% CI -0.24, -0.07), at 6 months (SMD -0.10, 95% CI -0.17 to -0.02), and at 12 months follow-up (SMD -0.06, 95% CI -0.13 to 0.02), but not among trials specifically considering injury patients [39]. The single study on heavy users admitted to general hospital inpatient units reported a significant effect on self-reported alcohol consumption at 6 months (WMD -69.43, 95% CI -128.14 to -10.72) and 9 months follow-up, but not at 12 months (WMD -33.62, 95% CI -82.27 to 15.03) and no effect for laboratory markers at either 6 months (WMD 7.00, 95% CI -33.77 to 47.77) or 12 months follow-up (WMD -5.05, 95% CI -36.82 to 26.73) [49]. A study considering a single brief intervention delivered by primary health professionals and community professionals in LMICs found no significant effect on relative risk reduction at 1–6 months (RR 0.93, 95% CI 0.77 to 1.12) or more than 6 months follow-up (RR 0.88, 95% CI 0.73 to 1.06) [115]. Overall confidence in the evidence for brief interventions in hazardous and harmful drinkers was graded as low. Although the sample size in all studies was very large and most studies showed an effect over the mid-long term, methodological quality of RCTs and meta-analyses was low, two reviews identified publication bias (out of three studies reporting on publication bias), and there was high heterogeneity among RCTs in many of the reviews.

Among the studies considering effect on hazardous/harmful drinking in specific sub-populations, a low-quality meta-analysis among concurrent illicit drug users with a small total sample (170 participants) found no effect on alcohol use after 3 months (SMD 0.07, 95% CI -0.24 to 0.37) [64]. A study among military and veterans did not identify a significant effect on self-reported alcohol consumption at 6–20 months, although confidence intervals were very wide (WMD 0.16, 95% CI -0.88 to 1.20, face-to-face delivery) [36]. However, a review encompassing different types of brief interventions delivered to pregnant women and mothers with alcohol problems (including single session motivational interview, computer-delivered screening and brief intervention, and self-help cognitive behavioural therapy) found a significant effect when combining all interventions on abstinence in pregnant women (OR 2.31, 95% CI 1.61 to 3.32) and alcohol consumption among women with dependent children (SMD 0.20, 95% CI 0.02, 0.38) as compared with usual care [114]. Similarly, a review of group sessions and web-based follow-up delivered through the workplace found a significant reduction in alcohol consumption compared with usual care (MD -2.25, 95% CI -4.20 to -0.30) [59].

Six studies considered brief interventions among adolescents, students and young adults [34, 40, 41, 52, 55, 102]. A high quality review of assessment and feedback among young people with at-risk drinking found a small but significant effect on alcohol consumption compared with no intervention at 0–6 months (SMD –0.17, 95% CI –0.27 to –0.08) and more than 6 months (SMD –0.17, 95% CI –0.3- to –0.04), but there was no significant effect when compared with education [55]. The same review found a significant reduction in blood alcohol content at 0–6 months for feedback combined with moderation skills when compared with feedback alone (SMD -0.26, 95% CI -0.49 to -0.03) [55]. Another study of adolescents in school or college facing negative consequences due to their alcohol use did not find a significant effect on alcohol consumption frequency in the short or intermediate term, nor on heavy drinking at 3–12 months follow-up (SMD -0.01, 95% CI -0.20 to 0.18) [34]. Conversely, studies in college/university students found a small but significant effect, for self-reported drinks per week at 12 months follow-up (MD -1.50, 95% CI -3.24 to -0.29) [40], frequency of heavy drinking days after 3 months (MD 0.16, 95% CI 0.07 to 0.25) [102], and alcohol consumption after 3–7 months (SMD 0.19, 95% CI 0.11 to 0.27) [102]. Among the studies of social norms interventions among university and college students, one high quality study found a small but significant effect across all delivery modes at more than 4 months follow-up for mean peak blood alcohol concentration (SMD -0.08, 95% CI -0.17 to 0.00) and drinks per week (SMD -0.08, 95% CI -0.12 to -0.05), although there was suspected publication bias [41]. Sub-group analyses for face-to-face and group feedback were not significant, aside from drinking days per week for face-to-face feedback. A low quality study across all modes of social influence interventions for any length of follow-up found a significant decrease across all outcomes of alcohol consumption (SMD 0.29, 95% CI 0.22 to 0.37) but noted very high heterogeneity and publication bias [52]. The body of evidence for social norms interventions was downgraded to very low due to the presence of publication bias and small effect size.

Regarding length of the intervention, a single study considered minimal intervention, defined as lasting 3–5 min, compared to control in hazardous drinkers and found no significant effect at 6–12 months (OR 0.95, 95% CI 0.72 to 1.25) [31]. Although the study found a significant decrease in the frequency of hazardous drinkers for extended brief interventions after 6–12 months (OR 1.5, 95% CI 1.12 to 1.95), the effect was similar to the combined effect for brief intervention and extended brief intervention studies (OR 1.55, 95% CI 1.27 to 1.90) [31]. It should also be noted that the study defined an extended brief intervention as having reinforcement sessions of 10–15 min, which overlaps with the definition of brief intervention in certain other studies [31, 38,39,40, 46, 49]. Another study found a very small reduction for extended brief intervention (defined as either more than 5 sessions or combined session durations totalling more than 60 min) in drinking days per week at 12 months follow-up, as compared to minimal or no intervention (MD -0.45, 95% CI -0.81 to -0.09), but no effect for binge drinking or quantity of alcohol consumption, or when compared with brief intervention [46]. A study comparing multi-dose assessment and feedback with single dose assessment and feedback among at-risk young drinkers found a small reduction in alcohol consumption at 0–6 months post-intervention (RR 0.84, 95% CI 0.78 to 0.91) [55].

Among the studies considering a specific type of brief interview, three considered brief motivational interviewing [9, 47, 58], three brief counselling [44, 51], one lay health-worker delivered interventions [115], and one brief interventions targeting multiple risk behaviours simultaneously [48]. All brief counselling interventions found a small but significant effect on drinks per week (MD -3.57, 95% CI -4.76 to -2.39, 12 months [44]; MD -1.59, 95% CI -2.15 to -1.03 [51]; MD 1.73, 95% CI 0.03 to 3.5, 6–12 months [57]) and heavy drinking (risk difference 0.12, 95% CI 0.07 to 0.16, at 12 months) [44]. Certainty in the evidence was graded as very low due to the presence of publication bias and very low AMSTAR rating of meta-analyses. For brief motivational interviews, one study in heavy drinkers found a large improvement in drinking moderation at 6–12 months (OR 1.95, 95% CI 1.66 to 2.30) [58] and a study of motivational interviewing delivered to young people with alcohol problems in emergency care found a significant effect at 3–12 months (SMD -0.17, 95% CI -0.32 to -0.02) [47]. However, a study across LMICs (primarily South Africa and India) found no effect at 3 months, 6 months, or 12 months across four different outcomes, except screening risk score after 3 months (SMD -0.34, 95% CI -0.67 to -0.01) [9]. There was very high heterogeneity for all of the outcomes based on more than 2 RCTs, suggesting significant variation in either context or how the intervention was delivered across RCTs. A review of lay health-worker led interventions in LMICs, including counselling, brief intervention, and brief motivational intervention, did find a significant improvement in screening score at 1 to 6 months (SMD -0.22, 95% CI -0.32 to -0.11) [115]. Certainty in the evidence for brief motivational interviewing was rated as very low due to very low AMSTAR ratings and inconsistency across both the meta-analyses and the RCTs within each meta-analysis. A high quality review of interventions targeting multiple risk behaviours in people aged 0–18 years found no effect on alcohol consumption for individual level (OR 1.02, 95% CI 0.80 to 1.31), family level (OR 0.83, 95% CI 0.47 to 1.46), or school level (OR 0.73, 95% CI 0.52 to 1.03) interventions, or on binge drinking [48].

Two studies were identified that included individuals with alcohol dependency or AUD. One study assessing injury-related deaths among individuals with alcohol dependency, abuse, or hazardous use found no significant effect (RR 0.65, 95% CI 0.21 to 2.00) [105]. One network meta-analysis on the effect of brief interventions on abstinence among individuals with AUD found no effect during treatment (OR 0.81, 95% CI 0.59 to 1.12) or after treatment (OR 0.76, 95% CI 0.46 to 1.22) [107]. The same study found that brief interventions in combination with either pharmacological interventions or psychological interventions had a lower probability of abstinence than pharmacological interventions. The certainty of evidence for brief interventions targeting AUD, either alone or in combination, was rated as very low due to low study quality and imprecision.

Psychosocial intervention outcomes

A total of sixteen publications reported outcomes for psychosocial interventions [42, 46, 56, 60, 62,63,64,65,66,67,68,69, 101, 103, 107, 115]. Tables 5 and 6 summarise the outcomes for psychosocial interventions.

Table 5 Summary results of included systematic reviews, for psychosocial interventions

Table 6 Certainty in evidence (GRADE) for psychosocial interventions

Two studies reported outcomes for cognitive behavioural therapy (CBT) [103, 115]. There was a significant decrease in alcohol consumption among hazardous/harmful drinkers in LMICs receiving CBT and counselling interventions from lay health workers up to 6 months post-intervention (SMD -0.23, 95% CI -0.56 to -0.09) [115], but the effect was not significant beyond 6 months (SMD -0.22, 95% CI -0.26 to 0.06). Among recently detoxified, alcohol dependent patients, one network meta-analysis found no effect on continuous abstinence (OR 0.53, 95% CI 0.23 to 1.22) [103]. We identified no studies reporting short-term outcomes among alcohol-dependent populations. The certainty of evidence was graded as very low due to risk of bias and imprecision.

For contingency management, two network meta-analyses considered abstinence among people with AUD or alcohol dependency [103, 107]. Neither found a significant effect of contingency management and all outcomes had very wide confidence intervals. Although one of the network meta-analyses showed considerable benefit of CM combined with psychotherapy (defined in the paper as BI, CBT, or motivational interviewing) relative to control (OR 0.20, 95% CI 0.10—0.40) and all other interventions, the effect was not maintained after treatment (OR 0.52, 95% CI 0.07 to 3.05) [107]. The body of evidence for contingency management was graded as very low due to high risk of bias and imprecision.

Only one meta-analysis was identified for the community reinforcement approach [63]. The study reported a very high reduction in abstinence among alcohol dependents after more than a year of follow-up (OR 3.97, 95% CI 2.26–6.95) for social network support. Although the long-term effect size was large and the meta-analysis was rated as moderate quality, the certainty of this finding was graded as low, since findings were based on two RCTs at high risk of bias with a total sample size of 210 participants. No studies for other types of community reinforcement approach were identified.

Six meta-analyses were identified for motivational interviewing, two of which were focussed on adolescents/young adults [56, 60], two on alcohol use in the general population [65, 68], one on people with AUD [103], and one on concurrent problem alcohol and illicit drug users [64]. Among adolescents and young adults, there were small reductions in alcohol consumption and alcohol problems, but not for binge drinking, over the intermediate term (over 4 months) among adolescents and young adults with risky drinking [60] and in days of alcohol use among adolescents with AUD or problematic alcohol use (MD -1.1, 95% CI -2.2 to -0.3) [56]. In the general population, one very low quality meta-analyses showed a considerable decrease in self-reported consumption for MI delivered in medical care settings (OR 2.31, 95% CI 1.75–3.06) [65] and among biochemically confirmed alcohol use after 4–6 months follow-up (Cohen’s d 0.30, 95% CI 0.03–0.57) [68]. However, no significant effect was found for abstinence maintenance after a year of follow-up among people with AUD (OR 0.45, 95% CI 0.19–1.11) [103] or for alcohol use among problem alcohol and illicit drug users after 3 months follow-up (SMD 0.04, 95% CI -0.29 to 0.37) [64]. The body of evidence for motivational interviewing in adolescents and young adults was graded as moderate due to good precision and consistency between studies, but very low among other population groups due to high risk of bias and small sample sizes.

One study was identified for family-oriented approaches, which reported on measures of alcohol use among school-aged children following family-based prevention programmes, administered to either the children themselves or to family members [42]. There was no significant effect across prevalence of alcohol use (SMD -0.16, 95% CI -0.36 to 0.05), volume of alcohol use (SMD 0.06, 95% CI -0.15 to 0.27), or frequency of alcohol use (SMD -0.65, 95% CI -1.64 to 0.33). Although the meta-analysis was rated as high methodological quality, the body of evidence was graded as very low due to high heterogeneity and limited generalisability to population groups outside of school children.

No reviews of mutual help groups met our inclusion criteria.

Among other psychosocial interventions identified, two reported on home visits [101, 103], one on coping skills training [103], and one on mentoring for children or adolescents with alcohol use [69]. For home visits, no significant effect was found in reducing continued alcohol use among pregnant women with alcohol problems (RR 0.18, 95% CI 0.96 to 1.46) [101] or in maintaining abstinence among recently detoxified, alcohol dependent patients (OR 0.95, 95% CI 0.32 to 2.85) [103], although the body of evidence was graded as very low due to high risk of bias and imprecision. For coping skills training, no significant effect was observed for promoting abstinence in recently detoxified, alcohol dependent patients (OR 0.95, 95% CI 0.32 to 2.85) [103]. Certainty in the evidence for coping skills training was graded as very low. For mentoring, there was a moderate reduction in alcohol use among children and adolescents after 9 to 15 months [69]. The body of evidence for mentoring among children and adolescents was graded as moderate due to low risk of bias, good precision, low heterogeneity, and the long timeframe.

Three reviews reported outcomes for specific psychosocial techniques, namely cue exposure therapy [67], forming implementation intentions [66], and controlled drinking non-abstinent strategies [62]. There was no evidence that cue exposure therapy affects alcohol consumption at 3- or 6-months follow-up among adults diagnosed with AUD or sub-clinical AUD, although the body of evidence was graded as very low due to high risk of bias, inconsistency in RCTs, publication bias and a very small sample size. Forming implementation intentions showed a large reduction in self-reported alcohol consumption in the general population over a short timeframe (up to 3 months) (OR 0.31, 95% CI 0.21, 0.42) [66]. Although risk of bias was low and there was good precision, certainty in the evidence was graded as moderate due to the short timeframe and single meta-analysis. Controlled drinking non-abstinent strategies were shown to be inferior to abstinence-based strategies among alcohol dependent adults at 6 to 42 months follow-up (OR 1.32, 95% CI 0.51 to 3.39) [62]. Body of evidence was downgraded to low due to wide confidence intervals and very small sample size.

Two reviews considered outcomes across psychosocial interventions. One review with very low AMSTAR rating found a small effect beyond 6 months for psychosocial interventions (primarily CBT and MI) delivered to emerging adults aged 18–25 years outside of college settings (Cohen’s d 0.18, 95% CI 0.07 to 0.29) [104], while a network meta-analysis considering the effect of CBT, MI and BI on promoting abstinence among people with AUD found a significant effect during treatment (OR 1.51, 95% CI 1.04 to 2.21) but not after (OR 0.99, 95% CI 0.57 to 1.70) [107]. Body of evidence was rated as very low.

Digital intervention outcomes

Eighteen studies analysed the effect of digital interventions (encompassing internet-, computer-, and mobile-based interventions) (Tables 7 and 8). Of the two very low quality studies considering impact among individuals at high risk of AUD or diagnosis with AUD, one found a decrease in alcohol consumption up to 9 months following internet-based interventions (Hedge’s g 0.44, 95% CI 0.17 to 0.71) but noted very high heterogeneity (I² = 81%) [110], while a review of e-interventions, including web-based and mobile applications, did not find a significant longer-term reduction in alcohol consumption at 6 months or 12 months, but did find a large improvement in abstinence (OR 1.94, 95% CI 1.14 to 3.31) [35]. Overall confidence in the evidence for digital interventions for alcohol dependent populations was graded as very low given risk of bias from study methods and high inconsistency.

Table 7 Summary results of included systematic reviews, for digital interventions

Table 8 Certainty in evidence (GRADE) for digital interventions

Of the studies on hazardous/harmful drinkers in the general population, a review of electronic screening and brief intervention found a significant reduction in weekly alcohol consumption up to 3, 6 and 12 months (MD (grams) -14.91, 95% CI -25.56 to -4.26, 6–12 months) but not beyond 12 months (MD (grams) -7.46, 95% CI -25.34 to 10.43) [37]. Similarly, a review of internet interventions (including computer and mobile based interventions) found a small reduction in mean weekly alcohol consumption (MD -5.02, 95% CI -7.57 to -2.48) at 1–12 months follow-up, although heterogeneity among studies was very high (I² = 90%) and there was evidence of publication bias [112]. A meta-analysis of low-intensity self-help interventions performed on a computer or mobile phone found a reduction in quantity of alcohol consumption immediately following delivery of the intervention (SMD 0.2, 95% CI 0.13 to 0.27) but not after 6–12 months (SMD 0.06, 95% CI -0.14 to 0.25) [111]. A single study focussed on mobile and text messaging interventions did not find a significant reduction in weekly alcohol consumption (MD (grams) -18.62, 95% CI -39.61 to 2.38) or heavy drinking episodes per month (MD -0.33, 95% CI -0.79 to 0.12), although confidence intervals were very wide [32]. Overall certainty in the evidence for digital interventions among hazardous and harmful drinkers was graded as very low due to risk of bias in study methods, presence of publication bias and inconsistency across studies.

Three studies evaluated personalised digital interventions delivered via web or mobile [45, 53, 54]. One study found a significant improvement across various measures of alcohol consumption, heavy drinking and binge drinking across a follow-up period of 1–12 months when compared with no intervention; moreover there was no significant difference in grams alcohol consumed per week (MD 0.52, 95% CI -24.59 to 25.63) or frequency of binge drinking (MD 0.04, 95% CI -0.15 to 0.22) when compared to face-to-face interventions for up to 24 months follow-up [45]. While another study also found a small improvement in drinking behaviour up to 9 months (SMD 0.22, 95% CI 0.16 to 0.29) [54], a review of remotely delivered personal normative feedback interventions found no improvement in drinking days (SMD -0.02, 95% CI -0.15 to 0.11) or symptom severity score (SMD 0.13, 95% CI -0.01 to 0.26) [53]. The body of evidence for personalised digital interventions was graded as very low certainty due to methodological risk of bias in the reviews and RCTs, as well as the presence of publication bias.

Among reviews of digital interventions in specific populations, 5 considered college and university students [35, 41, 43, 102, 113], 1 young people [55], 1 women aged 18–45 years [108], 1 military and veterans [36], and 1 cancer survivors [50]. A high-quality review found no significant difference between assessment and feedback delivered by computer as compared with assessment and feedback delivered by a counsellor (AMD -0.11, 95% CI -0.53 to 0.32, > 6 months) [55]. Although one review of computer-delivered interventions among college and university students found a decrease in alcohol consumption at 1 week to 3 years follow-up (SMD 0.22, 95% CI 0.14 to 0.29) [113], another found a very small reduction in alcohol consumption (WMD 0.14, 95% CI 0.03 to 0.24) and frequency of heavy drinking (WMD 0.13, 95% CI 0.02 to 0.24) up to 13 weeks post-intervention, but no significant effect at longer follow-up [102]. A review of social norms interventions delivered via web/computer, including tens of thousands of participants, found a small reduction in number of drinking days per week at more than 4 months follow-up (SMD -0.12, 95% CI -0.18 to -0.05) but noted suspected publication bias [41]. One network meta-analysis of digital intervention programmes found a significant reduction in quantity of alcohol consumed among all programmes (AlcoholEdu, BASICS, e-CHUG, THRIVE) at 0–3 months, but only BASICS showed continued benefit at 3–6 months (SMD -0.07, 95% CI -0.55 to 0.42) and the review found evidence of publication bias [43]. A single study considered e-interventions among college students at risk of, or diagnosed with, AUD and found no significant effect at 6 months follow-up for either alcohol consumption (MD (grams) -12.4, 95% CI -26.6 to 1.9) or binge drinking episodes (MD -0.1, 95% CI -1.0 to 0.9) [35]. Certainty in the evidence for digital interventions aimed at young people and college/university students was graded as low, as there was good precision and consistency, but poor methodological quality and publication bias.

Among military and veterans, web-based brief interventions showed no reduction in self-reported alcohol consumption at 6–20 months (WMD 1.81, 95% CI -0.06 to 3.68) but there was a significant reduction in deaths at 6 months and 12 months (RR 0.60, 95% CI 0.40 to 0.91, 12 months) [36]. Web or text-based interventions in women aged 18–45 years did not show a significant reduction in alcohol consumption at 1–6 months post-intervention (SMD 0.13, 95% CI -0.03 to 0.29), although it should be noted that any type of control intervention (inactive and active) was included [108]. Among cancer survivors, distance-based moderation showed no effect (SMD 0.12, 95% CI -0.08 to 0.31) [50]. Certainty in the evidence for military and veterans, women aged 18–45 years, and cancer survivors was downgraded to very low due to high risk of bias in review methods [36, 108] or RCTs [50].

Pharmacological intervention outcomes

Outcomes for pharmacological interventions were reported by 30 studies [70,71,72,73,74,75,76,77,78,79,80,81,82,83,84,85, 87,88,89,89,90,91,92,93,94,95,96,97,98, 103, 107] and four additional studies reported outcomes for a combination of psychosocial and pharmacological interventions [86, 106, 107, 109]. Tables 9 and 10 summarise the outcomes for pharmacological interventions.

Table 9 Summary results of included systematic reviews, for pharmacological interventions

Table 10 Certainty in evidence (GRADE) for pharmacological interventions

Five studies reported outcomes for anticonvulsants compared to placebo in individuals with AUD or alcohol dependency [73, 80, 89, 90, 103]. A moderate quality review considering all anticonvulsants found a small but significant effect after 12 weeks in terms of drinks per drinking day (MD -1.49, 95% CI -2.32 to -0.65) and mean heavy drinking (MD -0.35, 95% CI -0.51 to -0.19), but the effect for abstinence at 15 weeks was not significant (MD 1.21, 95% CI 0.97 to 1.52) [90]. A low quality network meta-analysis found no significant effect on continuous abstinence (although confidence intervals were very wide) for carbamazepine (OR 0.55, 95% CI 0.08, 3.90), levetiracetam (OR 1.03, 95% CI 0.46 to 2.34), oxcarbazepine (OR 2.46, 95% CI 0.91 to 6.61) or pregabalin (OR 1.97, 95% CI 0.58 to 6.47) [103]. Two low quality reviews of gabapentin with high heterogeneity found no significant effect across measures of alcohol consumption, heavy drinking, and abstinence, except for a very small effect on percentage of heavy drinking days in both studies (Hedge’s g 0.55, 95% CI 0.01, 1.08 [73]; MD -0.64, 95% CI -0.64, 95% CI -1.22 to -0.06 [80]). For topiramate, one low quality network meta-analysis found a significant increase in continuous abstinence at 84–365 days (OR 1.88, 95% CI 1.06 to 3.34) [103] and a high quality network meta-analysis found a significant reduction in total consumption at 3–52 weeks (SMD -0.77, 95% CI -1.12 to -0.42; NMD -0.79, 95% CI -1.21 to -0.36), as well as for drinking days and heavy drinking days, but drinks per drinking day and mortality outcomes were not significant, and the study noted very low quality of RCTs alongside presence of publication bias [89]. Certainty in the evidence for anticonvulsants was downgraded to very low, due to high risk of bias, high heterogeneity in RCTs, low precision, and presence of publication bias.

Four studies evaluated antidepressants for AUD, of which three considered individuals with comorbid depression [70, 76, 96]. No significant effect was observed on alcohol abstinence among people with comorbid major depressive disorder for nefazodone compared with placebo (OR 2.18, 95% CI 0.68 to 7.07) [96]; AUD remission rate compared to control among individuals with comorbid depression for mirtazapine (SMD -0.78, 95% CI -1.69 to 0.13) [76], SARI (OR 1.85, 95% CI 0.62 to 5.66) [76], tricyclic antidepressants (OR 1.65, 95% CI 0.57 to 4.73) [76]; or continuous abstinence when compared with placebo at 84–365 days for citalopram or escitalopram (OR 1.03, 95% CI 0.33 to 3.16), nefazodone (OR 0.57, 95% CI 0.19 to 1.76), tianeptine (OR 1.22, 95% CI 0.58 to 2.57), or tiapride (OR 0.56, 95% CI 0.3 to 1.05) [103]. For NRI, a very low quality meta-analysis found a reduction in AUD remission rate compared to control in individuals with comorbid depression for SMD (-2.44, 95% CI -3.53 to -1.36) but not OR (1.15, 95% CI 0.21 to 6.31) [76]. All studies considering any type of SSRI were conducted among people with comorbid depression. Whilst a moderate quality study found no significant effect on abstinence (OR 1.26, 95% CI 0.06 to 2.56) [96] and a very low quality network meta-analysis found no significant effect on AUD remission rate (OR 1.21, 95% CI 0.78 to 1.92) [76], a low quality study reporting publication bias reported a significant effect on drinks per drinking day (MD -1.42, 95% CI -2.58 to -0.26) and number of abstinent participants (RR 1.66, 95% CI 1.02 to 2.68) [70]. The same study identified a reduction in drinks per drinking day for 5-HT2 antagonists compared to placebo (MD -1.06, 95% CI-2.00 to -0.11) [70]. Another study reviewing the effect of specific SSRIs in promoting abstinence did not identify a significant effect for fluoxetine (OR 2.97, 95% CI 0.97 to 9.05), fluvoxamine (OR 1.03, 95% CI 0.57 to 1.88), or trazodone (OR 0.61, 95% CI 0.2 to 1.84) [103]. Certainty in the evidence for antidepressants was rated as very low due to low methodological quality, imprecision and publication bias.

For antipsychotics, three studies reported outcomes [76, 79, 103]. No significant effect was found compared to placebo for preventing relapse with aripiprazole (RR 1.07, 95% CI 0.92 to 1.24), quetiapine (RR 0.87, 95% CI 0.65 to 1.17), or tiapride (RR 1.07, 95% CI 0.67 to 1.71) [79], nor for maintaining abstinence with amisulpride (OR 0.39, 95% CI 0.09 to 1.64) or aripiprazole (OR 1.49, 95% CI to 0.43 to 5.18) [103]. Flupentixol was found to be inferior to placebo in maintaining abstinence in a network meta-analysis (OR 0.44, 95% CI 0.2 to 0.95), whereas quetiapine had a very large positive effect (OR 6,75, 95% CI 1.2 to 38.05) [103]. It should be noted that total number of participants for both flupentixol and quetiapine was below 100. For antipsychotics as a whole, a network meta-analysis of AUD remission rate among people with comorbid depression found no significant effect compared with control (OR 0.97, 95% CI 0.30 to 3.22) and suggested that antipsychotics are inferior to disulfiram (OR 0.19, 95% CI 0.04 to 0.9) [76]. However, whilst there was no effect on relapse, another review found a significant improvement in ratio of abstinent to drinking days when compared with placebo at 3–12 months follow-up (RR 0.17, 95% CI 0.01 to 0.33) [79]. Certainty in the evidence for antipsychotics was rated as very low due to risk of bias, heterogeneity among RCTs and poor precision.

Of the five studies on disulfiram, three very low qualities found a significant effect, for increase in AUD remission rate among people with comorbid depression compared with control (OR 5.00, 95% CI 1.97 to 12.95), acamprosate, antipsychotics, bromocriptine, lithium, naltrexone, and SSRI [76]; abstinence at 12 months (OR 2.24, 95% CI 1.69 to 2.27, compared to other or no treatment), with greater improvement for supervised administration of disulfiram (OR 3.89, 95% CI 2.66 to 5.58 [78]; and combined measures of alcohol use at 8–52 weeks (Hedge’s g 0.58, 95% CI 0.35 to 0.82), although the latter study identified potential publication bias [94]. In contrast, one low quality network meta-analysis found no effect on abstinence compared to placebo at 84–365 days (OR 0.93, 95% CI 0.48 to 1.79) [103] and a moderate quality review of moderate quality RCTs found no effect on return to drinking compared with placebo or another medication (risk difference -0.04, 95% CI -0.11 to 0.03) [77]. Evidence for disulfiram was rated as very low because of a high level of inconsistency and risk of bias in many reviews.

There was considerable variation in outcomes among studies comparing baclofen to control. One very low quality study on people with comorbid anxiety and depression identified no significant reduction in heavy drinking days (SMD -0.26, 95% CI -0.68 to 0.15) or improvement in abstinent days (SMD 0.03, 95% CI -0.10 to 0.15) [91]. Conversely, one low quality network meta-analysis with a very small sample identified a significant improvement in abstinence (OR 4.63, 95% CI 1.00 to 21.48) [103] and a moderate quality review with high quality RCTs found a significant decrease in amount of drinking (SMD 0.28, 95% CI 0.00 to 0.56) [71]. A very low quality network meta-analysis among people with comorbid depression did not find a significant effect in AUD remission rate aside from mean difference compared to control (the odds ratio and standardised mean difference measures were not significant) [76]. A low quality review found a significant increase in percentage abstinent days at 1–3 months (RR 2.79, 95% CI 1.79 to 4.34) but not for mean abstinent days (SMD 3.69, 95% CI 0.74 to 8.11) [82], while a high quality review found a small but significant effect for drinks per drinking day (MD 1.55, 95% CI 1.32 to 1.77) but not relapse (RR 0.88, 95% CI 0.74 to 1.04), percentage of heavy drinking days (MD 0.25, 95% CI -1.25 to 1.76), or percentage days abstinent (MD 0.39, 95% CI -11.51 to 12.29) [85], and a high quality network meta-analysis with a very small sample and suspected publication bias identified a significant reduction in total alcohol consumption at 3–52 weeks follow up (SMD -1.00, 95% CI -1.80 to -0.19; NMD -1.00, 95% CI -1.86 to -0.13), but not for non-drinking days or heavy drinking days [89]. Overall evidence for baclofen was graded as very low due to the inconsistency, small sample size, and presence of publication bias in many reviews.

No reviews were identified for benzodiazepines that met our inclusion criteria.

Nine reviews assessed the efficacy of acamprosate [72, 74, 76, 77, 83, 84, 92, 95, 103]. Seven of the studies showed acamprosate to be effective relative to placebo for promoting abstinence (OR 1.88, 95% CI 1.57 to 2.25, 1–12 months [72]; RR 0.83, 95% CI 0.78 to 0.89, 6 months [74]; RR 1.95, 95% CI 1.58to 2.42, 12 months [83]; OR 1.86, 95% CI 1.49 to 2.33, 84–365 days [103]; OR 1.87, 95% CI 1.57 to 2.23 [84]), cumulative abstinence duration (WMD 26.55, 95% CI 17.56 to 35.54, 3–12 months [72]; RR 1.11, 95% CI 1.01 to 1.21, 12 months [95]; MD 10.94, 95% CI 5.08 to 16.81, 12 months [92]), and mean percentage abstinent days (MD 10.38, 95% CI 7.10 to 13.65) [84]. A moderate quality review of acamprosate compared to either placebo or another medication found a small but significant effect for return to any drinking (risk difference -0.09, 95% CI -0.14 to -0.04) and percentage drinking days (WMD -8.8, 95% CI -12.8 to -4.8) but not for return for heavy drinking (risk difference -0.01, 95% CI -0.04 to 0.03) [77]. A very low quality network meta-analysis of AUD remission rates among individuals with comorbid depression did not find a significant effect relative to control (OR 1.66, 95% CI 0.89 to 3.05) and found acamprosate to be inferior to disulfiram (OR 0.33, 95% CI 0.1 to 0.9) [76]. Acamprosate combined with a nurse visit showed a significant improvement in continuous abstinence as compared to placebo (OR 4.59, 95% CI 1.47 to 14.36) [103]. Although methodological quality of reviews was low and two studies identified publication bias [72, 74], the evidence was graded as low certainty due to good agreement across a large number of studies, each with thousands of participants.

Reviews evaluated two opioid antagonists: nalmefene and naltrexone. Both studies comparing nalmefene to placebo found a small but significant decrease in alcohol consumption (SMD -0.2, 95% CI -0.3 to -0.1 [88]; SMD -0.19, 95% CI -0.29 to -0.10 [89]) and heavy drinking days (MD -1.65, 95% CI -2.41 to -0.89 [88]; SMD -0.22, 95% CI -0.32 to -0.12 [89]). However, only one of the studies found a significant decrease in mortality and no effect was found on non-drinking days (SMD 0.09, 95% CI-0.01 to 0.19) [89]. Certainty in the evidence for nalmefene was graded as low, since there was a small sample size and presence of publication bias.

Among the nine reviews of naltrexone, three network meta-analyses found no significant effect on abstinence (OR 1.36, 95% CI 0.97 to 1.91) [103], AUD remission rate (OR 1.38, 95% CI 0.88 to 2.18), or across five outcomes of alcohol use [89]. A low quality study found naltrexone to be significantly inferior to GHB (OR 2.31, 95% CI 1.22 to 4.36) [81]. Four studies did find a significant effect across mean percentage of drinking days (MD -2.8, 95% CI -5.8 to -0.2) and abstinence rate (RR 1.28, 95% CI 1.08 to 1.52) [97]; drinking days (MD-3.89, 95% CI -5.75 to -2.04) and heavy drinking days (MD -3.25, 95% CI -5.51 to -0.99) [93]; cumulative abstinence duration (RR 1.23, 95% CI 1.00 to 1.78) and relapse rate (RR 1.2, 95% CI 1.17 to 1.47) [95];and five measures of alcohol use, but not abstinence rate (OR 1.26, 95% CI 0.97 to 1.64) [72]. A moderate quality review found a small but significant effect on 50 mg oral naltrexone for five alcohol use outcomes, but most outcomes were not significant for 100 mg oral naltrexone and naltrexone injection, aside from percentage heavy drinking days (WMD -3.1, 95% CI -5.8 to -0.3 and WMD -4.6, 95% CI -8.5 to -0.56 respectively) [77]. Certainty of evidence was rated as very low due to low methodological quality, inconsistency of results and presence of publication bias.

Among other pharmacological interventions, a low quality meta-analysis found no significant effect compared to placebo on continuous abstinence for atenolol (OR 0.85, 95% CI 0.25 to 2.95), lisuride (OR 0.38, 95% CI 0.13 to 1.12), or modafinil (OR 2.48, 95% CI 0.72 to 8.53) [103]. The same study found galantamine to be inferior to placebo in promoting abstinence (OR 0.31, 95% CI 0.11 to 0.87). A very low quality network meta-analysis on AUD remission rate among people with comorbid depression did not find a significant effect relative to control of bromocriptine (OR 1.70, 95% CI 0.50 to 5.65), buspirone (OR 0.37, 95% CI 0.07 to 1.89), or memantine (OR 0.70, 95% CI 0.29 to 1.52) [76]. The same meta-analysis found bromocriptine to be inferior to disulfiram (OR 0.07, 95% CI 0.01 to 0.5) and naltrexone combined with SSRI (OR 0.16, 95% CI 0.03 to 0.96). Neither network meta-analysis identified a significant effect of lithium (AUD remission rate among individuals with comorbid depression OR 0.70, 95% CI 0.29 to 1.52 [76]; continuous abstinence OR 1.43, 95% CI 0.39 to 5.23 [103]). A comparison of paroxetine with placebo among patients with comorbid anxiety did not find a significant effect of paroxetine for reducing drinks per drinking day (MD –2.42, 95% CI –4.97 to 0.14) or proportion of days abstinent (MD 0.08, 95% CI –0.26 to 0.43) [75]. Of the studies on GHB, the effect on relapse to heavy drinking compared to naltrexone was not significant (RR 3.23, 95% CI 0.57 to 18.33) but there was a significant improvement in abstinence compared to naltrexone at 3 months (2.59, 95% CI 1.35 to 4.98) [81] and a significant effect on abstinence relative to placebo at 84–365 days (OR 2.31, 95% CI 1.22 to 4.36) [103]. For alpha blockers, a very low quality study showed significant reduction in drinks per day or week at 6–13 weeks follow-up (SMD –0.32, 95% CI –0.56 to –0.07) but not for heavy drinking days (SMD –0.44, 95% CI –0.94 to 0.06) [98]. Similarly, evidence from was inconclusive for antiepileptics, which showed a significant improvement in AUD remission rate among individuals with comorbid depression when compared to control, bromocriptine, or lithium in a network meta-analysis of odds ratios, but not when compared to control using standardised mean difference (SMD –0.70, 95% CI –2.05 to 0.65) [76], and for varenicline, which showed a significant effect for alcohol consumption (SMD –0.37, 95% CI –0.66 to –0.07) but not heavy drinking days (SMD -0.14, 95% CI –0.33 to 0.05) [87]. Certainty in the evidence for each of the other pharmacological interventions was rated as very low, due to high risk of bias and imprecision in many studies, resulting from very wide confidence intervals and/or sample sizes less than 100 (Supplement 6).

Two studies evaluated pharmacological interventions as a whole. A review of pharmacological treatments for people with concurrent bipolar disorder or depression found no effect on alcohol consumption (SMD –0.10, 95% CI –0.24 to 0.04) but did find an improvement in abstinence (OR 1.46, 95% CI 1.02 to 2.11) [96], unlike a network meta-analysis of pharmacotherapy for AUD which found no improvement in abstinence rate (OR 0.68, 95% CI 0.4 to 1.16, after treatment) [107]. Due to the variation in efficacy across different types of pharmacological intervention, any analysis of pharmacological interventions is likely to be very dependent on the specific interventions included.

Two network meta-analyses assessed combinations of pharmacological interventions [76, 103]. There were six combinations including naltrexone. Among individuals with comorbid AUD and depression, there was no significant effect on AUD remission rate as compared to control for naltrexone combined with disulfiram (OR 2.60, 95% CI 0.71 to 10.15) and although naltrexone combined with SSRI showed a significant effect compared to control (OR 2.24, 95% CI 1.15 to 4.50) and SSRI (OR 1.86, 95% CI 1.0 to 3.6) when odds ratios were considered, standardised mean difference was not significant (SMD –0.19, 95% CI –1.07 to 0.68) and there was no significant effect compared with naltrexone alone (OR 0.62, 95% CI 0.3 to 1.3), although confidence intervals were very wide [76]. Among people with alcohol dependence or AUD, there was no significant change in continuous abstinence relative to placebo for naltrexone combined with escitalopram (OR 2.57, 95% CI 0.25 to 25.85), but there was a significant effect for naltrexone combined with acamprosate (OR 3.68, 95% CI 1.5 to 9.02), GHB (OR 12.64, 95% CI 2.77 to 57.78), and GHB with escitalopram (OR 25.65, 95% CI 2.13 to 309.46) [103]. The effect on continuous abstinence of GHB combined with escitalopram was not significant (OR 5.13, 95% CI 0.53 to 49.92) [103]. Certainty in the evidence for all combinations of pharmacological interventions was graded as very low due to considerable imprecision and low methodological quality of meta-analyses and RCTs (Supplement 6).

Three reviews considered combinations of psychosocial and pharmacological interventions. Extended-release naltrexone combined with psychosocial interventions had a small but significant effect up to 6 weeks in reducing drinking days per month (MD -2.00, 95% CI -3.39 to -0.61) and heavy drinking days per month (MD -1.16, 95% CI -2.1 to -0.23) among patients with AUD, compared to psychosocial interventions alone [86]. Naltrexone (50 mg) combined with psychotherapy was found to be more effective than placebo and psychotherapy in reducing alcohol consumption and improving abstinence in alcohol dependent patients at 3–4 months (OR 1.46, 95% CI 1.07 to 2.00), but the effect was not significant at 6–9 months follow-up [109]. The body of evidence for naltrexone combined with psychosocial interventions was rated as very low. One review reported outcomes for intensive perioperative cessation programmes, which provided disulfiram, chlordiazepoxide, motivational counselling, brief interview, and B vitamins to hazardous drinkers undergoing surgical procedures [106]. There was a high improvement in self-reported abstinence between 1 and 3 months (RR 8.22, 95% CI 1.67 to 40.44) [106]. Although there was high methodological quality, as well as large effect size, the meta-analysis only incorporated 70 participants, so the certainty was downgraded to low (Supplement 6).

Miscellaneous interventions

The literature search yielded one very low quality study on physical activity and one very low quality study on bibliotherapy (Table 11). The meta-analysis of bibliotherapy (intervention in a written format) found a mean weighted effect size for alcohol consumption of 0.21 when compared with no intervention, but confidence intervals were not reported [99]. Certainty of evidence was graded as very low due to poor methodological quality of a single study with heterogeneous results. For physical activity, no significant effect was found on abstinence among individuals with AUD or those seeking treatment for alcohol use (RR 1.56, 95% CI 0.78 to 3.14) [100]. The certainty of evidence was graded as very low due to poor precision and very low quality RCTs.

Table 11 Summary results of included systematic reviews, for miscellaneous interventions

Comparison against clinical practice guidelines

A summary of results from the umbrella review are presented in Table 12. For screening, brief interventions and referral to treatment, the NICE clinical practice guidelines (CPG) recommend that all individuals identified with harmful drinking or dependence receive motivational interviewing [24]. Whilst our review found motivational interviewing (and brief motivational interviewing) to be effective, we identified a greater level of variation than for brief counselling or general brief interventions, and no evidence to support delivery of motivational interviews to those with alcohol dependence. However, it should be noted that findings from our review considered motivational interviewing delivered independently, not at the start of treatment initiation as with the NICE CPG.

Table 12 Summary of findings table

For harmful but not moderately or severely dependent drinkers, the NICE CPG recommends cognitive behavioural therapy, behavioural therapies, social network, or environment-based therapies for adults, with couples-based therapy provided when relevant, and cognitive behavioural therapy for children, potentially combined with family-based therapy in the presence of comorbidities or limited social support. CPGs from Australia, Canada, Germany, and the USA equally highlighted the importance of behavioural therapies in harmful drinkers [25, 26, 27, 28, 29]. In line with our findings, the NICE CPG recommends treating alcohol use and comorbidities separately. In our review we identified very limited evidence on the effect of cognitive behavioural therapy and no studies on social network interventions among this population. Furthermore, the only review identified for family-based approaches in children found no evidence of the effectiveness of this intervention, and from our review mentoring showed the greatest promise in addressing alcohol use among this population of adolescents and children. The umbrella review found non-inferiority of digital interventions, but this was not discussed in any of the CPGs.

For moderately or severely dependent individuals, the NICE CPG recommends focussing on a goal of abstinence, which is in line with findings from our review and guidelines from Australia, Canada, Germany, and the USA. The CPGs recommend the provision of either acamprosate or naltrexone in combination with psychosocial interventions, after assisted withdrawal with benzodiazepines and concurrent psychosocial support has taken place. NICE explicitly recommends against the use of GHB or antidepressants, including SSRIs. While the umbrella review showed good effectiveness of acamprosate, we found no outcomes for acamprosate delivered in combination with psychosocial interventions. Moreover, our review found conflicting evidence to support the use of naltrexone and suggested that GHB merits further research and review. No study identified in our review examined withdrawal.

Discussion

To our knowledge, this study is the first umbrella review to provide an overview of all available interventions to address harmful alcohol use and dependence. This review demonstrates an approach by which to summarise evidence across a clinical area in settings with limited resources for health technology assessment, in order to prioritise interventions for further assessment in UHC policy decisions.

We examined the effectiveness of: (1) interventions for non-dependent individuals either experiencing or at risk of harmful drinking; and (2) interventions for individuals with alcohol use disorder. For our first research question on harmful alcohol use, brief interventions showed a small but significant effect up to 12 months. In general, brief interventions consisting of a single session, however, did not show a significant effect. Further investigation is required on the relationship between efficacy and number/duration of sessions. This finding is consistent with two umbrella reviews of brief interventions, which found a moderate effect on alcohol consumption but noted the need for further research on sub-group differences and the optimum length and frequency of sessions [116, 117]. Although brief interventions have been shown to be effective, it will be important to consider appropriate settings in which to conduct opportunistic screening, as the additional workload for personnel should be balanced against potential health gains [30]. Findings from this review suggested that provision of screening and brief interventions in a digital format could be as effective as face-to-face interventions, which aligned with the conclusions from a review of systematic reviews of computer-based interventions for problematic alcohol use [118]. This finding could provide an opportunity to extend the reach of screening and brief intervention services without contributing to the burnout of personnel. However, operational research would be required to identify strategies to reach groups with traditionally lower use of digital technologies, such as the elderly, people with disabilities, and socio-economically disadvantaged groups [119].

We did not identify any meta-analyses on referral mechanisms to treatment, which may indicate a gap in the literature, but could also be an artefact of our inclusion criteria, which required studies to report outcomes on alcohol use or health outcomes. Among psychosocial interventions, there is evidence to suggest that cognitive behavioural therapy and motivational interviewing could slightly reduce alcohol consumption in the intermediate term (up to 6 months), but further review is required, especially since it has been highlighted that contextual and health systems factors may limit transferability of findings to LMICs [61]. Forming implementation intentions demonstrated a large effect after a 3-month follow-up, but further research is required on the durability of the response. Similarly, there was promising evidence to suggest that mentoring may be able to significantly reduce alcohol use among children and adolescents, with effects lasting beyond one year, but further research is warranted to consider the applicability of findings across settings and to identify best practice for implementation. The evidence identified in this review for contingency management came from two network meta-analyses with wide confidence intervals, making it difficult to draw conclusions on its effectiveness, and we identified no reviews on community reinforcement approaches.

For our second research question on interventions for alcohol use disorder, abstinence-based strategies were found to be more effective than controlled drinking in reducing alcohol consumption. The only non-pharmacological intervention to show a significant effect was social networking, for which there is low certainty evidence of a large, long-term benefit for maintaining abstinence. Among pharmacological interventions, acamprosate showed good evidence of effectiveness over a 12-month timeframe. Although there were many reviews identified for disulfiram, baclofen, and naltrexone, there was considerable inconsistency between meta-analyses and high heterogeneity across the RCTs in each study, making it difficult to draw meaningful conclusions. Further review is required to identify whether this heterogeneity could be arising from sub-group differences or different practice in delivering the intervention. Real-world evidence could provide large datasets of patients with different characteristics to partially address this question. The umbrella review also highlighted a potential benefit of nalmefene, topiramate, quetiapine, and GHB, although further evidence review is required. Overall, the review highlighted the need for more studies of combinations of psychosocial and pharmacological interventions, since combining naltrexone with psychosocial interventions showed an improvement relative to psychosocial interventions alone, but no evidence was found relative to providing naltrexone alone or for other drugs. Early health technology assessment and expected value of perfect information analysis may identify combinations of interventions for which RCTs should be conducted [120]. We did not identify any studies on recovery management, or long-term interventions to maintain abstinence in previously dependent individuals.

Although there was moderate agreement between our results and the NICE CPG (as well as CPGs from other countries), the comparison highlighted that our umbrella review, by limiting the focus to meta-analyses of RCTs, may not have captured the full body of evidence on appropriate pathways or combinations of interventions. However, it should be highlighted that this review did show non-inferiority of digital interventions, which was not mentioned in any of the other clinical practice guidelines, and may be important for implementation of interventions in resource-constrained settings. It is also important to note that beyond results from meta-analyses of clinical benefit, NICE guidelines incorporate inputs from a series of stakeholder consultations, as well as socio-ethical and economic considerations, which may explain some of the differences in findings. In parallel to conducting an umbrella review, it may be beneficial to conduct a review of clinical practice guidelines from internationally recognised institutions to supplement and compare with findings from the review.

There are a number of limitations to this umbrella review. Firstly, since we defined our search terms and inclusion criteria based on two broad research questions (as opposed to defining research questions from the clinical pathway), we inadvertently excluded reviews of referral to treatment and withdrawal, since these interventions are commonly measured by drop-out rates and measures of craving/alcohol withdrawal respectively, which fell outside the scope of our inclusion criteria. Secondly, to facilitate comparison across types of intervention, we restricted the inclusion criteria to meta-analyses of RCTs only; subsequently, we could only extract very limited information on the delivery of interventions, which meant that we were not able to assess whether variation in delivery led to some of the discordance in results. Although methodologically rigorous [15], restricting analysis to RCTs has been criticised for leading to pharmacological bias, particularly for mental health treatments, as psychological interventions are often influenced to a greater extent by relationships, trust, and socio-cultural context, thus harder to control [121,122,123]. This may explain the reason why we identified very few eligible studies for many types of psychosocial intervention. For example, although a Cochrane Review of mutual support groups exists [124], the results for RCTs were combined with quasi-randomised studies. We suspect this is also the reason that our review did not identify recovery management interventions, which aim to support recovered individuals over the long term. Since we had to balance the feasibility of conducting the review with overly-restricting the scope, basing our analysis on the WHO/UNODC framework was helpful to at least identify gaps in the evidence from our review.

A further two limitations arise from the studies included in this review. The first is the very limited number of RCTs within the meta-analyses that were conducted in LMICs. Appropriate interventions for the prevention and treatment of alcohol use disorder are expected to vary across different health and social care services and socio-cultural contexts. Whilst we do not expect that findings will be more transferable between LMICs than between HICs and LMICs, our findings highlight a gap in the contexts and healthcare settings in which intervention studies are conducted. Similarly, we found limited investigation into the reasons for high heterogeneity across or within RCTs. Beyond conducting trials outside of high income settings, we recommend that RCTs adhere to intervention standards such as the TIDieR checklist [125], and that both RCTs and meta-analyses make greater effort to investigate the reasons behind heterogeneity in results. The second limitation is the low certainty in most of our findings. Meta-analyses are already prone to bias because they pool studies of different quality (internal bias) and relevance (external bias) [126]; certainty in the evidence for most of our conclusions is low because of very low methodological quality of the meta-analyses themselves (brief interventions), limited relevance due to very specific target populations (psychosocial interventions), and the presence of publication bias (pharmacological interventions). This highlights the need for better adherence to reporting guidelines for systematic reviews and RCTs. In particular, we recommend that meta-analyses are more transparent in publishing a protocol before conducting the review and in detailing their search strategy, as well as investigating the potential impact of risk of bias and publication bias on results. As noted above, the considerable heterogeneity across RCTs and meta-analyses suggests that there is a need to shift the focus of research from the interventions that are effective to a better understanding of which interventions work for whom, at which time, and in which context.

Despite these limitations, we believe that this study has been successful in describing the landscape of interventions to prevent and treat alcohol use disorder. The umbrella review methodology allowed us to succinctly summarise a large body of literature using rigorous methods, which can be used as a starting point for benefit package policy discussions.

Conclusion

For people with harmful drinking or at-risk of harmful drinking, brief interventions show a small effect over a 6–12 month period, but further research is required to identify the appropriate frequency/duration of sessions, appropriate settings in which to conduct screening and brief intervention, and when digital brief interventions may be more appropriate, before identifying the scenarios for which brief interventions are a good use of resources. We also found a significant long-term effect of mentoring in adolescents and children, as well as limited evidence of a small effect of CBT and motivational interviewing in all populations, but further research on transferability to other contexts outside of high-income settings is required. For people with alcohol use disorder, there was consistent evidence of an effect for social network approaches and for acamprosate. Further review of community reinforcement approaches, including social network approaches, is recommended, particularly to identify any differences in effectiveness or contextualisation required for different cultures and health system settings, above all in LMICs. There may potentially be a benefit of GHB, nalmefene, and quetiapine, but further research is required on this topic, as well as for combinations of pharmacological and psychological interventions. The umbrella review methodology provided a practical and rigorous approach by which to summarise the effectiveness of interventions across a whole clinical area, for a research question with a very broad scope, and use of established frameworks and CPGs was important to identify gaps in the literature.

Registration and protocol

The protocol is registered in PROSPERO (CRD42021275471) and available as a pre-print [14]. The following five amendments were made to the original protocol. Firstly, to facilitate comparison of effect sizes, we restricted our analysis to systematic reviews with meta-analysis, whereas in the protocol it states that narrative systematic reviews would also be included. Secondly, we did not use a citation matrix to exclude systematic reviews with overlapping RCTs, since it was considered unnecessary given that the purpose of the review is to highlight interventions for further investigation in health benefit package decisions. Thirdly, we did not use elements of the TIDieR checklist [131] for data extraction, as during piloting of the data extraction form, it was found that systematic reviews included insufficient information on the methods of included RCTs. On a related note, since we extracted limited information on how the interventions were delivered, we were unable to evaluate the feasibility of delivering interventions in LMICs in our results. Finally, we did not detail methods to assess certainty in the body of evidence in the protocol but applied GRADE in this study.