Treatment interventions to maintain abstinence from alcohol in primary care: systematic review and network meta-analysis

Abstract Objective To determine the most effective interventions in recently detoxified, alcohol dependent patients for implementation in primary care. Design Systematic review and network meta-analysis. Data sources Medline, Embase, PsycINFO, Cochrane CENTRAL, ClinicalTrials.gov, and the World Health Organization’s International Clinical Trials Registry Platform. Study selection Randomised controlled trials comparing two or more interventions that could be used in primary care. The population was patients with alcohol dependency diagnosed by standardised clinical tools and who became detoxified within four weeks. Data extraction Outcomes of interest were continuous abstinence from alcohol (effectiveness) and all cause dropouts (as a proxy for acceptability) at least 12 weeks after start of intervention. Results 64 trials (43 interventions) were included. The median probability of abstinence across placebo arms was 25%. Compared with placebo, the only intervention associated with increased probability of abstinence and moderate certainty evidence was acamprosate (odds ratio 1.86, 95% confidence interval 1.49 to 2.33, corresponding to an absolute probability of 38%). Of the 62 included trials that reported all cause dropouts, interventions associated with a reduced number of dropouts compared with placebo (probability 50%) and moderate certainty of evidence were acamprosate (0.73, 0.62 to 0.86; 42%), naltrexone (0.70, 0.50 to 0.98; 41%), and acamprosate-naltrexone (0.30, 0.13 to 0.67; 17%). Acamprosate was the only intervention associated with moderate confidence in the evidence of effectiveness and acceptability up to 12 months. It is uncertain whether other interventions can help maintain abstinence and reduce dropouts because of low confidence in the evidence. Conclusions Evidence is lacking for benefit from interventions that could be implemented in primary care settings for alcohol abstinence, other than for acamprosate. More evidence from high quality randomised controlled trials is needed, as are strategies using combined interventions (combinations of drug interventions or drug and psychosocial interventions) to improve treatment of alcohol dependency in primary care. Systematic review registration PROSPERO CRD42016049779.

• Major concerns: downgrade the evidence one level • Some concerns: downgrade the evidence one level with 2 or more some concerns in other judgements Reporting bias Reporting bias was evaluated by non-statistical consideration of likelihood of non-publication of evidence.
We selected "suspected" among all comparisons but did not downgrade the confidence by this judgement. Indirectness As outcome (continuous abstinence) has consistent, clear definition, indirectness was only evaluated by majority of populations within each comparison.
Populations among studies were assessed by distributions of age, gender and comorbidities (refer to S4) • Major concerns: downgrade the evidence one level • Some concerns: downgrade the evidence one level with 2 or more some concerns in other judgements Imprecision Imprecision was focused on width of confidence interval (CI) based on a clinically important odds ratio of 1.2 for abstinence and 0.8 for dropouts.
We increased the concern one level if the width of CI is between 4 times and 10 times of lower limit. The concern level was increase two levels if the width of CI is above 10 times of lower limit.
• Major concerns: downgrade the evidence one level • Some concerns: downgrade the evidence one level with 2 or more some concerns in other judgements Heterogeneity Heterogeneity was evaluated according to the CINeMA documentation by variability of effects in relation to the clinically important size of effect and between-study variance for the network meta-analysis.
We increased the concern one level if there is no information regarding between-study heterogeneity for each direct comparison or I 2 > 60% in the direct comparison.
• Major concerns: downgrade the evidence one level • Some concerns: downgrade the evidence one level with 2 or more some concerns in other judgements Incoherence Incoherence was evaluated by the design-bytreatment intervention model globally and sidesplitting approach locally.
• Major concerns: downgrade the evidence one level • Some concerns: downgrade the evidence one level with 2 or more some concerns in other judgements Quality of the evidence (GRADE): High quality: Further research is very unlikely to change our confidence in the estimate of effect. Moderate quality: Further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate. Low quality: Further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate. Very low quality : We are very uncertain about the estimate.

MATCH project
Aim assignment to intervention (the 'intention-totreat' effect) Source Journal article(s) with results of the trial; Trial protocol; Grant database summary (e.g. NIH RePORTER, Research Councils UK Gateway to Research)

Domain Response Comments
Y Y PN "This procedure was successful, in that there were no significant differences across treatments on the matching variables assessed at baseline"

Low
This study employed adequate randomisation methods in the trial and represents low risk of bias in the randomisation process.

Some concerns
It was impossible to blind participants in this trial due to nature of interventions employed, which potentially induce deviations from intended interventions. On the other hand, the authors applied ITT analyses. Together, these contributed to "some concerns" in this domain.
Y Whole data were supplied by the committee.

Low
The MATCH project commeittee provided the data and nearly all participants were followed during 9 months period in the main analysis.

PY
It was impossible to blind outcome assessors.
PN "Laboratory tests are used to screen subjects for exclusion criteria (e.g., unreported drug use), monitor changes in alcohol consumption..."

Low
Although it was impossible to blind outcome assesser(patients), the outcome (abstinence) was confirmed by laboratory tests, which put low risk of bias in this domain.
N All time points were reported N

Low
The reviewer re-analysed total abstinence, aligning to the common definition. This was not included in the protocol so we rated "Low" in selection of the reported results.

Some concerns
Some concerns in deviations from the intended interventions contributed to "some concerns" in overall bias.

Angelone 1998
Aim assignment to intervention (the 'intention-totreat' effect) Source Journal article(s) with results of the trial

Domain Response Comments
NI NI PN A significant difference among groups, 23 vs 25 vs 33 but it was done deliberately, "The citalopram group was deliberately made larger a priori to achieve more experience with this drug, which is relatively new in Italy". No significant difference among characteristics of the study sample besides age and M:F ratio in the citalopram group.

Some concerns
No details were given regarding randomisation process. Although there were some difference between aga and M:F ratio among trials, it might due to small number of participants in this trial as by chance. Together these contributed to "some concerns" in this domain.

PN
One of the groups did not receive interventions, which might lead to deviations.
NA N NA

Some concerns
There was no complete information regarding blinding and one group did not receive any interventions (treated as usual), which might lead to deviation. On the other hand, the authors applied ITT analyses. Together, these contributed to "some concerns" in this domain.
N Some drop-outs were seen owing to side effects or moved were not included n=3 of 25 in fluvoxamine group; n=5 of 33 in citalopram group. PN 0%; 12%; 15% respectively -not clear if reasons for missing outcome data are similar or not.

PY
The authors did not perform sensitivity analyses but both on-treatment and ITT analyses led same results.

Low
Although there were some missing data, results still stood in consideration of missing data. "Low" risk of bias in this domain was rated.

PN
Did not address blinding of participants in the trial and the outcome was self-assessed and confirmed by relatives.
"The presence of relatives, or other key individuals for the patient, was required at each assessment, to confirm the patient's report and to obtain additional information about their alcohol intake."

Low
Abstinence was confirmed by relatives or key individuals, which strengthend the reliability of results.

Bias due to deviations from intended interventions
2.1 Were participants aware of their assigned intervention during the trial?
Interventions were different and MET had only 4 sessions so it was impossible to blind participants.

Signalling question
Bias arising from the randomization process 1.1 Was the allocation sequence random?
"The randomization process is centrally controlled by the (Yale) CC." "To ensure consistent delivery of treatments across sites, training, supervision, and certification of therapists are centralized at the Yale CC." 1.2 Was the allocation sequence concealed until participants were recruited and assigned to interventions?
1.3 Were there baseline imbalances that suggest a problem with the randomization process?

Signalling question
Bias arising from the randomization process 1.1 Was the allocation sequence random?
Only stated "random" 1.2 Was the allocation sequence concealed until participants were recruited and assigned to interventions?
1.3 Were there baseline imbalances that suggest a problem with the randomization process?

Bias in selection of the reported result
Are the reported outcome data likely to have been selected, on the basis of the results, from…

Low
No protocol or statisitical analysis was found but nature of outcome, abstinence, presents low risk of selected reported results and the authors described the method section clearly. Thus, we rated "Low" risk of bias in this domain.

Some concerns
Lack of detailed randomisation process and potential deviations from the intended interventions due to difference among interventions, together, these contributed to "some concerns" in overall bias for this trial.

Baltieri 2004
Aim assignment to intervention (the 'intention-totreat' effect) Source Journal article(s) with results of the trial

Domain Response Comments
NI NI PN There were 35 and 40 patients in placebo and acamprosate groups; The average daily alcohol intake was slight higher in acamprosate group than placebo group (370.1 (164.91) vs 348.5 (132.46)) but not statistically significant.

Some concerns
No details were given regarding randomisation process and no significant difference between groups suggesting "Some concerns" in this domain. Y "Ten patients who were receiving acamprosate and seven patients who were receiving placebo dropped out." "The reasons for dropping out were unwillingness to continue the treatment (two patients of the acamprosate group and two of the placebo group); "protocol violation," which was defined as the use of other psychopharmacological drugs during the study (one patient of the acamprosate group and one of the placebo group); and unavailability for follow-up (seven patients of the acamprosate group and four of the placebo group)." PN Conservative "Patients who missed a visit or withdrew from the study were deemed to be nonabstinent at the time those data were not available" and no sensitivity analysis.

Low
Although there were some missing data, results still stood in consideration of balanced missing data. "Low" risk of bias in this domain was rated.
PN "double-blind" "Major variables recorded at each visit included clinical examination results, patients' self-reported quantity and frequency of alcohol consumption and drug side effects. The patients' declaration of drinking behavior was verified by the results of γglutamyltransferase (GGT) levels in every case and by interviewing a family member if possible." NA Low Double-blind design and self-reporting outcome (confirmed by relatives and biochemistry results) put this domain as low risk of bias.
PN "The patients' declaration of drinking behavior was verified by the results of γ-glutamyltransferase (GGT) levels in every case and by interviewing a family member if possible."

Low
No protocol or statisitical analysis was found but nature of outcome, abstinence, presents low risk of selected reported results and the authors described the method section clearly. Thus, we rated "Low" risk of bias in this domain.

Some concerns
Lack of detailed randomisation process contributed to "some concerns" in overall bias for this trial.

Baltieri 2008
Aim assignment to intervention (the 'intention-totreat' effect) Source Journal article(s) with results of the trial Outcome Abstinence (12 weeks) Results

Domain
Response Comments Y PY PN "As shown in Table 2, therewere no significant differences among the groups at baseline on any socio-demographic, drug use, hepatic function or psychometric variables measured."

Low
This study employed adequate randomisation methods in the trial and represents low risk of bias in the randomisation process.

Bias in selection of the reported result
Are the reported outcome data likely to have been selected, on the basis of the results, from… Only stated "random" 1.2 Was the allocation sequence concealed until participants were recruited and assigned to interventions?
1.3 Were there baseline imbalances that suggest a problem with the randomization process?

Risk of bias judgement
Signalling question Bias arising from the randomization process 1.1 Was the allocation sequence random?
"patients were assigned randomly to one of the three medication conditions through a random number list" "Medication was dispensed under double-blind conditions. Only two pharmacists from the pharmacy sector at the Psychiatric Institute of the Clinical Hospital of the University of São Paulo knew which medication corresponded to the specific code. The packages containing the capsules were distributed to patients by two blinded research assistants, who also assessed patient outcomes throughout the study." 1.2 Was the allocation sequence concealed until participants were recruited and assigned to interventions?
1.3 Were there baseline imbalances that suggest a problem with the randomization process?

Bias in selection of the reported result
Are the reported outcome data likely to have been selected, on the basis of the results, from… "double-blind"; "Validity of the double-blind procedure was verified by obtaining a prediction from each patient and staff member as to whether a given individual had received active or placebo medication during the study." "Overall, researchers were able to differentiate active treatment (naltrexone or topiramate) correctly from placebo treatment in 33.6% of cases. Among subjects, 27% were able to differentiate active treatment (naltrexone or topiramate) correctly from placebo treatment." "Differences between conditions in overall dropout rates approached significance (X2 = 5.10, P < 0.07) and were statistically significant within the lost-to-follow-up category (X = 7.723, P < 0.02) with a significant difference between topiramate and placebo in post-hoc analysis.

PN
Conservative "Patients who missed a visit or withdrew from the study were deemed to be non-abstinent at the time of missed visits." and no sensitivity analysis

High
High porportion of missing data and imbalanced missing data without sensitivity analysis put "high" risk of bias in this domain.
PN "two blinded research assistants, who also assessed patient outcomes throughout the study."

Low
Double-blind design in the outcome assessors put low risk of bias in this trial.
PN "Alcohol consumption during the treatment was determined using a dailymonitoring card and compliance was evaluated by self-report, capsules count of the returned medication package and the dailymonitoring card."

Low
No protocol or statisitical analysis was found but nature of outcome, abstinence, presents low risk of selected reported results and the authors described the method section clearly. Thus, we rated "Low" risk of bias in this domain.

Overall bias High
High porportion of missing data and imbalanced missing data without sensitivity analysis put "high" risk of bias.

Bender 2007
Aim assignment to intervention (the 'intention-totreat' effect) Source Journal article(s) with results of the trial

Low
This study employed adequate randomisation methods in the trial and represents low risk of bias in the randomisation process. PY They did not give reasons for mssing outcome data but the reasons should be expected to be similar across intervention groups.
"The number of dropouts due to adverse events or intercurrent illnesses was comparable in both groups (tiapride, n=10; placebo, n=9)." PY "In the worst-case analysis, all dropouts for unknown reasons (lost to follow-up: tiapride, n=12; placebo, n=24) were considered as relapse.
In this analysis, the relapse rate was 62% in the tiapride group (93 patients) and 57% in the placebo group (85 patients)." "The difference in relapse rates in the worst-case analysis was not statistically significant ( x2 test, p=0.31)."

Low
Although there were some missing data, results still stood in consideration of balanced missing data after sensitivity analysis. "Low" risk of bias in this domain was rated.
PN "...double-blind study" might suggest that outcome assessors were not aware of the intervention.

Low
Double-blind design that outcome assessers (patients) were not influenced by the knowledge of intervention received puts this domain as low risk of bias.

Low
No protocol or statisitical analysis was found but nature of outcome, abstinence, presents low risk of selected reported results and the authors described the method section clearly. Thus, we rated "Low" risk of bias in this domain.

Overall bias Low
Overall low risk of bias

Reference
Besson 1998 Aim assignment to intervention (the 'intention-totreat' effect) Source Journal article(s) with results of the trial

Bias in selection of the reported result
Are the reported outcome data likely to have been selected, on the basis of the results, from…

Risk of bias judgement
2.6 If Y/PY/NI to 2.5: Was there potential for a substantial impact (on the estimated effect of intervention) of analysing participants in the wrong group?

Risk of bias judgement
Bias due to deviations from intended interventions 2.1 Were participants aware of their assigned intervention during the trial?
"This multi-centre, randomized, double-blind, placebocontrolled, parallel-group study was conducted at 11 centres in Germany (six psychiatric university hospitals, three non-academic psychiatric hospitals, one day-clinic and one private practice)." "Study medication was administered in tablets indistinguishable in colour, size, form, smell, taste, consistency, and packaging." "...patients were randomly assigned to one of the two treatment groups according to a predefined random code." "Eligible patients were chronologically randomized by assigning them the lowest unassigned treatment number available at the study centre." "...Study medication was administered in tablets indistinguishable in colour, size, form, smell, taste, consistency, and packaging" 1.2 Was the allocation sequence concealed until participants were recruited and assigned to interventions?
1.3 Were there baseline imbalances that suggest a problem with the randomization process?

Bias in selection of the reported result
Are the reported outcome data likely to have been selected, on the basis of the results, from…

Risk of bias judgement
Outcome Abstinence (360 days) Results

Domain Response Comments
PY NI N No significant difference between groups was found.

Some concerns
No details were given regarding randomisation process, contributed to "some concerns" in this domain.
Double-blind design and ITT analysis employed in this trial put "Low" risk of bias in deviation from intended interventions.
PN 36/55 participants in each group were nonattendant at 1y follow-up.
In the analysis, non-attenders were assumed to have relapsed.

PY
Proportions of missing data identical in the two groups. Reasons for missing data approximately balanced between groups, although slightly more patients in the placebo group were nonattendant due to relapse requiring rehospitalization, and slightly more patients were nonattendent due to concurrent illness in the acamprosate group.

Low
Although there were some missing data, results still stood in consideration of missing data. "Low" risk of bias in this domain was rated.

PN "double-blind"
NA Low Double-blind design and self-reporting put this domain as low risk of bias.

Low
No protocol was found but nature of outcome, abstinence, presents low risk of selected reported results and the authors described the method section clearly. Thus, we rated "Low" risk of bias in this domain.

Some concerns
Lack of detailed methods for the randomisation process contributed to "some concerns" in overall bias for this trial.

Burtscheidt 2002
Aim assignment to intervention (the 'intention-totreat' effect) Source Journal article(s) with results of the trial

Domain Response Comments
NI NI PN there were no significant differences in age, sex, severity and duration of illness, and sociodemographic data in terms of education, employment, and familial status between the three therapy groups.

Some concerns
No details were given regarding randomisation process, contributed to "some concerns" in this domain.

PY NI
No details on deviations from expected practice NA N No evidence that patients recieved a treatment other than the one they were assigned to.

Some concerns
The open label design in this trial might prompt deviations from the intended interventions, contributing to "some concerns" in this domain.

PN
There were some missing data due to loss to follow-up (~15%).
PY "There were no significant differences between the three treatment groups in this respect." = missing data rate.
No information provided on whether reasons for missing data were comparable between intervention groups NI Unclear how missing data were handled. No sensitivity analysis

Low
Although there were some missing data, results still stood in consideration of balanced missing data. "Low" risk of bias in this domain was rated.
PY Self report and reports from family/friends, who are aware of the treatment received.

PY
Outcomes are subjective and could have been influenced by awareness of the treatment

Some concerns
Although this is an open study, the outcome (abstinence) was confirmed by family member and saliva test, which put "some concerns" risk of bias in this domain. "randomized, parallel, double-blind, placebo-controlled study" "Balanced randomization" 1.2 Was the allocation sequence concealed until participants were recruited and assigned to interventions?
1.3 Were there baseline imbalances that suggest a problem with the randomization process?

Bias in selection of the reported result
Are the reported outcome data likely to have been selected, on the basis of the results, from… "Preferences of the patients for any of the three treatment approaches did not emerge; there were no significant differences in age, sex, severity and duration of illness, and sociodemographic data in terms of education, employment, and familial status between the three therapy groups." 1.2 Was the allocation sequence concealed until participants were recruited and assigned to interventions?
1.3 Were there baseline imbalances that suggest a problem with the randomization process?

Low
No protocol was found but nature of outcome, abstinence, presents low risk of selected reported results and the authors described the method section clearly. Thus, we rated "Low" risk of bias in this domain.

Some concerns
Lack of detailed methods for the randomisation process and blinding of participants and personnel, together, these contributed to "some concerns" in overall bias for this trial.

Caputo 2003
Aim assignment to intervention (the 'intention-totreat' effect) Source Journal article(s) with results of the trial Outcome Abstinence (3 months) Results

Domain Response Comments
NI NI N "At the time of admission to the study (Table 1), the two groups did not differ in terms of demographic data, education, employment, marital status, duration of alcohol addiction, time of abstinence, alcohol craving scale and alcohol dependence degree."

Some concerns
No details were given regarding randomisation process, contributed to "some concerns" in this domain. PY % comparable across arms, and reasons seem broadly similar.

PY
No sensitivity analysis but results still stood by considering the missing data.

Low
Although there were some missing data, missing data presented equally in both groups and results stood the same in consideration of missing data. "Low" risk of bias in this domain was rated.

Low
Although this is an open study, the outcome (abstinence) was confirmed by family member and saliva test, which put low risk of bias in this domain.

Low
No protocol or statisitical analysis was found but nature of outcome, abstinence, presents low risk of selected reported results and the authors described the method section clearly. Thus, we rated "Low" risk of bias in this domain.

Some concerns
Lack of detailed randomisation process contributed to "some concerns" in overall bias for this trial.

Caputo 2007
Aim assignment to intervention (the 'intention-totreat' effect) Source Journal article(s) with results of the trial

Domain Response Comments
NI NI PN 20 vs 18 vs 17.

Some concerns
No details were given regarding randomisation process, contributed to "some concerns" in this domain.

Low
Although there were some missing data, results still stood in consideration of missing data. "Low" risk of bias in this domain was rated. PN "These parameters were assessed on the basis of participant self-evaluation, the interview of a family member and the determination of alcohol concentrations in blood and saliva" -robust outcome

Low
Although this is an open study, the outcome (abstinence) was confirmed by family member and blood/saliva test, which put low risk of bias in this domain.

Low
No protocol or statisitical analysis was found but nature of outcome, abstinence, presents low risk of selected reported results and the authors described the method section clearly. Thus, we rated "Low" risk of bias in this domain.

Some concerns
Lack of detailed randomisation process, deviations from the intended interventions, together, these contributed to "some concerns" in overall bias for this trial.

Chick 2004
Aim assignment to intervention (the 'intention-totreat' effect) Source Journal article(s) with results of the trial

Domain Response Comments
Y Y N "At entry, there were no differences that reached statistical significance in the characteristics measured of patients allocated to the two treatment groups."

Low
This study employed adequate randomisation methods and represented low risk of bias in the randomisation process. See Table 2 -propotions are similar but reasons different between treatement arms.

PN
No sensitivity analysis but used a conservative approach -"...in which all drop-outs were to be regarded as treatment failures."

High
High porportion of missing data and imbalanced missing data without sensitivity analysis put "high" risk of bias in this domain.
PN "Patients were assessed, usually by the same rater at each occasion, after detoxification on the day of randomisation, and after 2, 4, 6, 8, 12, 16, 24, 32, 40 and 52 weeks of treatment." Robust double blind protocol means assessor would not know which treatment was received."

Low
Double-blind design put this domain as low risk of bias.

Low
No protocol or statisitical analysis was found but nature of outcome, abstinence, presents low risk of selected reported results and the authors described the method section clearly. Thus, we rated "Low" risk of bias in this domain.

Overall bias High
High porportion of missing data and imbalanced missing data without sensitivity analysis put "high" risk of bias.

Low
Although there was no information regarding allocation concealment, this study employed adequate randomisation methods and identical placebo in the trial, which represents low risk of bias in the randomisation process.
PY PY PN "It was intended that the medication would be used as an adjunct, not an alternative, to the clinic's usual psychosocial out-patient treatment programme."

Signalling question
Bias arising from the randomization process 1.1 Was the allocation sequence random?
"Randomisation after meeting inclusion and exclusion criteria was within centres, in blocks of eight, four patients per block to each treatment. At randomisation, patients were given the next sequential number at that centre, and received the trial supplies for that patient number." "The randomisation code was provided by the department of statistics and data management at Solvay-Duphar B.V." 1.2 Was the allocation sequence concealed until participants were recruited and assigned to interventions?
1.3 Were there baseline imbalances that suggest a problem with the randomization process?

Bias in selection of the reported result
Are the reported outcome data likely to have been selected, on the basis of the results, from…

Low
Although no blinding procedures were employed, this trial used identical placebo and aimed to be an adjunct, not an alternative, to the clinic's usual practice, we rated this domain as "Low" risk of bias.
N "Only 203 patients completed the study [A: 100 (35%), P: 103 (35%)]." PY "There were no statistically significant differences in attendance between the treatment groups at any time point in the study." data appear similar per arm PN "It was assumed that all patients who terminated treatment before the end of the study, including those experiencing adverse events, were treatment failures."

Low
High proportion of missing data but balanced missing data, suggesting "Low" risk of bias in this domain.

PN
Although there is no blinding, this trial employed identical placebo and uses of an alcolmeter.

Low
Although this study did not emphasise blinding, the outcome (abstinence) was confirmed by biochemistry test, which put low risk of bias in this domain.

Low
No protocol or statisitical analysis was found but nature of outcome, abstinence, presents low risk of selected reported results and the authors described the method section clearly. Thus, we rated "Low" risk of bias in this domain.

Overall bias Low
Low risk of bias overall

Reference
Cornelius 1997 Aim assignment to intervention (the 'intention-totreat' effect) Source Journal article(s) with results of the trial

Domain Response Comments
NI NI N "No significant differences were seen between treatment groups for sex, race, age, or marital or employment status."

Some concerns
No details were given regarding randomisation process, contributed to "some concerns" in this domain. Double-blind design and ITT analysis employed in this trial put "Low" risk of bias in deviation from intended interventions.
Y "Forty-six of these patients (90% of those randomized) completed the pharmacotherapy study; the other 5 patients (10%) dropped out before the end of the trial." NA NA

Low
Almost all outcome data were available, thus this domain was rated "Low" risk of bias.

PN
The one year "evaluations of current symptoms were conducted by an interviewer who had been kept blind to the original assessment of protocol medication and to any subsequent medication use."

Low
Double-blind design put this domain as low risk of bias.

Low
No protocol was found but nature of outcome, abstinence, presents low risk of selected reported results and the authors described the method section clearly. Thus, we rated "Low" risk of bias in this domain.

Some concerns
Lack of detailed methods for the randomisation process contributed to "some concerns" in overall bias for this trial.

Some concerns
No details were given regarding randomisation process, contributed to "some concerns" in this domain.

Some concerns
The open label design in this trial might prompt deviations from the intended interventions, contributing to "some concerns" in this domain.

PN
Only 10/30 completed the study (24 weeks) and 4 lost without any information supplied.

Some concerns
High porportion of missing data and no detailed reasons for missing data put "some concerns" in this domain.
PY Drinking data collection was performed by a trained research assistant who was unblinded to treatment assignment, but not involved in the patient treatment.'

Low
Although this is an open study, the outcome (abstinence) is not influenced by knowledge of intervention received.

Low
No protocol was found but nature of outcome, abstinence, presents low risk of selected reported results and the authors described the method section clearly. Thus, we rated "Low" risk of bias in this domain.

Some concerns
Lack of detailed randomisation process, potential deviations from the intended interventions due to difference among interventions and missing data, together, these contributed to "some concerns" in overall bias for this trial.

Reference
Favre 1997 Aim assignment to intervention (the 'intention-totreat' effect) Source Journal article(s) with results of the trial Outcome Abstinence (9 months) Results

Domain Response Comments
NI NI PN A difference in previous alcohol withdrawals was noted but might due to chance

Some concerns
No details were given regarding randomisation process, contributed to "some concerns" in this domain. No reasons of missing data were given.

PY
They presented results by per protocol and ITT analyses, which led to the same results.

Some concerns
High porportion of missing data and imbalanced missing data without proper sensitivity analyses put "some concerns" risk of bias in this domain.

PN
Self-reported and checked by biological results: Abstinence was determined if the patient said he had no more than one drink since the last visit, the GGT level and the mean corpuscular volume were normal, or had not increased since the last examination, and the blood alcohol was lower than 0.10 g/ l.
NA Low Double-blind design and self-reporting outcome (confirmed by biochemistry results) put this domain as low risk of bias.

Low
No protocol was found but nature of outcome, abstinence, presents low risk of selected reported results and the authors described the method section clearly. Thus, we rated "Low" risk of bias in this domain.

Some concerns
High proportion of missing data and lack of detailed randomisation process contributed to "Some concerns" risk of bias in overall bias.

Risk of bias judgement
Bias arising from the randomization process only stated "random" 1.2 Was the allocation sequence concealed until participants were recruited and assigned to interventions?
1.3 Were there baseline imbalances that suggest a problem with the randomization process?

Signalling question
Bias arising from the randomization process

Low
Almost all outcome data were available, thus this domain was rated "Low" risk of bias.
Y "Open-label" and self-reported outcome PN Abstinence was assessed by the participants and significant one with a clear definition. "Alcohol intake was assessed at each treatment session. Both the patient and the significant other were interviewed and the highest intake level reported was used."

Low
Although this is an open study, the outcome (abstinence) was confirmed by family member, which put low risk of bias in this domain.

Low
No protocol was found but nature of outcome, abstinence, presents low risk of selected reported results and the authors described the method section clearly. Thus, we rated "Low" risk of bias in this domain.

High
High risk of bias in deviations from the intended interventions contributed to "High" risk of bias in overall bias.

Friedmann 2008
Aim assignment to intervention (the 'intention-totreat' effect) Source Journal article(s) with results of the trial Outcome Abstinence (6 months) Results

Domain Response Comments
Y NI PN "The trazodone (N = 88) and placebo groups (N = 85) did not differ at baseline on any measured characteristic (Table 1)."

Some concerns
No details were given regarding randomisation process, contributed to "some concerns" in this domain. PY "Mixed linear regression analyses with full information maximum likelihood (FIML) estimation compared the other drinking outcome trajectories by treatment condition across baseline, 1-, 3-, and 6-month intervals." "For the measure of complete abstinence, those who did not complete all follow-ups were assumed to have resumed drinking, and were therefore coded as not having achieved complete abstinence by the end of the study."

Low
Although there were some missing data, the authors used mixed linear regression analyses and conservative approach to analysis their results, resulting "Low" risk of bias in this domain. "Treatment assignment was done by opening sequentially numbered envelopes based on a randomization list."

Bias due to deviations from intended interventions
1.2 Was the allocation sequence concealed until participants were recruited and assigned to interventions?
1.3 Were there baseline imbalances that suggest a problem with the randomization process?

Bias in selection of the reported result
Are the reported outcome data likely to have been selected, on the basis of the results, from…

Low
Double-blind design and ITT analysis employed in this trial put low risk of bias in deviation from intended interventions.

PN
There were around 10% of patients in each group with insufficient data to evaluate abstinence ( Figure 1)

NI
The proportion of drop-outs seems to be similar between groups but no details were given.

PN
Conservative analysis by assumming dropped out patients as "failed"

Some concerns
Although there were some missing data, the proportion of drop-outs seemed to be similar but no details were given. The authors did not run sensitivity analyses but they assummed drop-outs as failed. "Some concerns" risk of bias in this domain was rated.
N Abstinence (drinking) was reported by relative's/friend's interviews or from the urine or blood specimens.

NA
Low Double-blind design and self-reporting outcome (confirmed by relatives and biochemistry results) put this domain as low risk of bias.

Low
No protocol was found but nature of outcome, abstinence, presents low risk of selected reported results and the authors described the method section clearly. Thus, we rated "Low" risk of bias in this domain.

Overall bias
Some concerns Some concerns overall due to the way of authors handling with missing data.

Geerlings 1997
Aim assignment to intervention (the 'intention-totreat' effect) Source Journal article(s) with results of the trial

Domain Response Comments
Y PY N Participants in placebo group seemed to be able to stay alcohol-free longer than participants in acamporsate group, which might be resulted from chance

Low
This study employed adequate randomisation methods in the trial and represents low risk of bias in the randomisation process.

PN
Drop-outs were assumed as relapsed.

Some concerns
High porportion of missing data without sensitivity analysis put "some concerns" risk of bias in this domain.

PN "double-blind"
NA Low Double-blind design and self-reporting outcome put this domain as low risk of bias.

Low
No protocol was found but nature of outcome, abstinence, presents low risk of selected reported results and the authors described the method section clearly. Thus, we rated "Low" risk of bias in this domain.

Overall bias
Some concerns Some concerns overall due to high proportion of missing outcomes without sensitivity analysis

PN
Craving for alcohol higher in the atenolol group, but no other differences so probably due to chance.

Some concerns
No details were given regarding randomisation process, contributed to "some concerns" in this domain. N N NA NA N Doesn't specifically state ITT but no patients were analysed in a group different to the one to which they were assigned. "Patients who did not return for follow-up and could not be contacted were presumed to have returned to drinking and were counted as treatment failures."

Low
Double-blind design and ITT analysis employed in this trial put "Low" risk of bias in deviation from intended interventions.

N
The authors reported drinking status in all participants PN Numbers who dropped out, presumed drinking slightly higher in the placebo group. Numbers who withdrew whilst not drinking: higher in the atenolol group (17 vs. 13).

PN
May have led to a conservate estimate of the effect of atenolol.

High
High porportion of missing data and imbalanced missing data without sensitivity analysis put "high" risk of bias in this domain.
N "double-blind" NA Low Double-blind design and self-reporting outcome put this domain as low risk of bias.

Low
No protocol was found but nature of outcome, abstinence, presents low risk of selected reported results and the authors described the method section clearly. Thus, we rated "Low" risk of bias in this domain.

High
High risk of bias in missing data and some concerns in randomisation process contributed to "High" risk of bias in overall bias.

Gual 2001
Aim assignment to intervention (the 'intention-totreat' effect) Source Journal article(s) with results of the trial

Domain Response Comments
NI NI N

Some concerns
No details were given regarding randomisation process, contributed to "some concerns" in this domain. N More participants lost to follow-up and refused to comtinue in the placebo group. And more participants dropped before started the trial.

PN
Appears that % and reasons balanced between groups, with sole exception of "No data after baseline"

PY
The authors presented both ITT and pre-protocol analysis results, which led same conclusion.

Low
Although there were some missing data, results still stood in consideration of missing data. "Low" risk of bias in this domain was rated.
PN "double-blind" NA Low Double-blind design and self-reporting outcome put this domain as low risk of bias.

Low
No protocol was found but nature of outcome, abstinence, presents low risk of selected reported results and the authors described the method section clearly. Thus, we rated "Low" risk of bias in this domain.

Some concerns
Lack of detailed methods for the randomisation process these contributed to "some concerns" in overall bias for this trial.
the randomization process 1.3 Were there baseline imbalances that suggest a problem with the randomization process?

Risk of bias judgement
Bias due to missing outcome data 3

Signalling question
Bias arising from the randomization process 1.1 Was the allocation sequence random?
Only stated "random", "double-blind" 1.2 Was the allocation sequence concealed until participants were recruited and assigned to interventions?
1.3 Were there baseline imbalances that suggest a problem with the randomization process?

Bias in selection of the reported result
Are the reported outcome data likely to have been selected, on the basis of the results, from…  Table 2 doesn't show any difference.

Some concerns
No details were given regarding randomisation process, contributed to "some concerns" in this domain. N The attrition rate was higher in the placebo group (Table 4).

PN
Imbalanced reasons between groups

PY
The authors performed analyses included and excluded patients dropped out, leading to the same conclusion.

Low
Although there were some missing data, results still stood in consideration of missing data. "Low" risk of bias in this domain was rated.

PN "double-blind"
NA Low Double-blind design puts this domain as low risk of bias.

Low
No protocol was found but nature of outcome, abstinence, presents low risk of selected reported results and the authors described the method section clearly. Thus, we rated "Low" risk of bias in this domain.

Some concerns
Lack of detailed methods for the randomisation process contributed to "some concerns" in overall bias for this trial.

Gustafson 2014
Aim assignment to intervention (the 'intention-totreat' effect) Source Journal article(s) with results of the trial

Domain Response Comments
Y Y

Low
This study employed adequate randomisation methods in the trial and represents low risk of bias in the randomisation process.

Some concerns
It was impossible to blind participants in this trial, which might prompt deviations from the intended interventions, contributing to "some concerns" in this domain.

PN
There were around 15% of lost to follow-up NI No reasons were given for missing data but the proportions were similar.

Low
Although there were some missing data, results still stood in consideration of missing data and the authors analysed results using mixed effect models. "Low" risk of bias in this domain was rated.
Y Outcome assessors were participants.

High
Lack of blinding to outcome assessors (participants themselve) and self-reporting outcomes, which put "High" risk of bias in this domain.

Signalling question
Bias arising from the randomization process "patients were randomized to the control group or A-CHESS in a 1:1 ratio using a computer-generated random allocation sequencewith blocks of 8 .Randomizationwasimplemented usingsequentiallynumbered containers." 1.2 Was the allocation sequence concealed until participants were recruited and assigned to interventions?
1.3 Were there baseline imbalances that suggest a problem with the randomization process?

Bias in selection of the reported result
Are the reported outcome data likely to have been selected, on the basis of the results, from…

Low
No protocol was found but nature of outcome, abstinence, presents low risk of selected reported results and the authors described the method section clearly. Thus, we rated "Low" risk of bias in this domain.

Overall bias High
The nature of interventions made it was impossible to blind participants, contributing deviations from the intended interventions and outcome measurements. Overally, these result "High" risk of bias overall.

Huang 2005
Aim assignment to intervention (the 'intention-totreat' effect) Source Journal article(s) with results of the trial Outcome Abstinence (14 weeks) Results

Domain Response Comments
NI NI N "As shown in Table 1, there were no significant differences in the distribution of demographic characteristics between the naltrexone and placebo-treated groups."

Some concerns
No details were given regarding randomisation process, contributed to "some concerns" in this domain. N "nine (45%) did not complete the study, whereas in the placebo-treated group seven (35%) of 20 subjects failed to complete the study." PY "Four of the nine non-completers in the naltrexone-treated group and three of the seven noncompleters in the placebo-treated group dropped out (p = 0.671) because of alcohol relapse, as defined in this study. However, the rest of the non-completers in both groups were reluctant to continue the study."

Low
Although there were some missing data, results still stood in consideration of missing data, which was balanced in both groups. "Low" risk of bias in this domain was rated.

PN "double-blind"
NA Low Double-blind design and self-reporting outcome (confirmed by biochemistry results) put this domain as low risk of bias.

Low
No protocol was found but nature of outcome, abstinence, presents low risk of selected reported results and the authors described the method section clearly. Thus, we rated "Low" risk of bias in this domain.

Some concerns
Lack of detailed methods for the randomisation process contributed to "some concerns" in overall bias for this trial.

Huang 2002
Aim assignment to intervention (the 'intention-totreat' effect) Source Journal article(s) with results of the trial

Domain Response Comments
Y NI N No significant difference in outcomes measured at baseline, including alcohol consumption and craving.

Some concerns
No details were given regarding randomisation process, contributed to "some concerns" in this domain. Although a large proportion of the study group, only one patient dropped out in the naltrexone grouo and two in the placebo group. If these drops outs were all abstinent it would not change the result that natrexone is superior.

Signalling question
Bias arising from the randomization process 1.1 Was the allocation sequence random?
Only stated "random", "double-blind" 1.2 Was the allocation sequence concealed until participants were recruited and assigned to interventions?
1.3 Were there baseline imbalances that suggest a problem with the randomization process?

Risk of bias judgement
Bias in selection of the reported result 5

Signalling question
Bias arising from the randomization process

Was the allocation sequence random?
Block randomisation, no information on allocation concealment 1.2 Was the allocation sequence concealed until participants were recruited and assigned to interventions?
1.3 Were there baseline imbalances that suggest a problem with the randomization process?

Bias in selection of the reported result
Are the reported outcome data likely to have been selected, on the basis of the results, from… NA NA Low PN Self-reporting and "single-blind" NA Low Single-blind design and self-reporting outcome (confirmed by relatives and biochemistry results) put this domain as low risk of bias.

Low
No protocol was found but nature of outcome, abstinence, presents low risk of selected reported results and the authors described the method section clearly. Thus, we rated "Low" risk of bias in this domain.

Some concerns
Lack of detailed methods for the randomisation process contributed to "some concerns" in overall bias for this trial.

Some concerns
The open label design in this trial might prompt deviations from the intended interventions, contributing to "some concerns" in this domain.
N More than half participants dropped out NI Don't know drop-out rates in each group.

Some concerns
High porportion of missing data and no detailed reasons for missing data put "some concerns" in this domain.

PY
Open study, self-reported outcomes PY

High
Open study and self-reported outcome put this domain as "High" risk of bias NI NI

Some concerns
No protocol was found and insufficient information available. Thus, we rated "Some concerns" risk of bias in this domain.

Overall bias High
Lack of information as a conference abstract put this trial as "high" risk of bias

Low
This study employed adequate randomisation methods in the trial and represents low risk of bias in the randomisation process.

Signalling question
Bias arising from the randomization process

Was the allocation sequence random?
Only stated "random" 1.2 Was the allocation sequence concealed until participants were recruited and assigned to interventions?
1.3 Were there baseline imbalances that suggest a problem with the randomization process?

Bias in selection of the reported result
Are the reported outcome data likely to have been selected, on the basis of the results, from…

Risk of bias judgement
2.1 Were participants aware of their assigned intervention during the trial?
"Group allocation was blind for both the participants and the researchers or care providers, who enrolled, treated, or assessed the patients." 2.2 Were carers and trial personnel aware of participants' assigned intervention during the trial?

Signalling question
Bias arising from the randomization process 1.1 Was the allocation sequence random?
"A stratified, permuted block randomization was used with gender as the only stratum and blocks contained random sizes of 2, 4 or 6 allocations for males, and 2 or 4 allocations for females. Personnel, not associated with the wards involved in the study, generated the allocation sequence by using 'Random Allocation Software' (Saghaei, 2004) and assigned the patients to one of the 2 treatment groups. Only these persons and the involved pharmacists were aware of the medication assignment." 1.2 Was the allocation sequence concealed until participants were recruited and assigned to interventions?
1.3 Were there baseline imbalances that suggest a problem with the randomization process?

Bias in selection of the reported result
Are the reported outcome data likely to have been selected, on the basis of the results, from… NI Did not mention how to handle missing data.

High
No information how to handle missing data and contradictory in reported drop-outs in Figure 2 and contexts put this domain as "High" risk of bias.
N double-blind NA

Low
Double-blind design puts this domain as low risk of bias.

Some concerns
Abstence rate was not pre-specified in the methods section and not all results reported (follow-up 1), leading "some concerns" risk of bias in this domain.

Overall bias High
High risk of bias overall due to imcompleted outcome reporting (missing follow-up time point 1)

Kampman 2007
Aim assignment to intervention (the 'intention-totreat' effect) Source Journal article(s) with results of the trial

Domain Response Comments
NI NI PN 29 q vs 32 p (combined); some evidence of imbalances among participant characteristics, which might be by chance

Some concerns
No details were given regarding randomisation process, contributed to "some concerns" in this domain. PN PN NA NA N No evidence that patients recieved a treatment other than the one they were assigned to.

Low
Double-blind design employed in this trial put low risk of bias in deviation from intended interventions. PN There were 6/29 q and 8/32 p missing data in both groups and reasons were not given.
PY "there were no significant differences between medication and placebo groups in treatment completion (23/29, 77% for the quetiapine group; and 24/32, 75% for the placebo group)"

Low
Although there were some missing data, results still stood in consideration of missing data. "Low" risk of bias in this domain was rated.
PN "double-blind"; self-report NA Low Double-blind design and self-reporting outcome (abstinence) put this domain as low risk of bias.

Low
No protocol was found but nature of outcome, abstinence, presents low risk of selected reported results and the authors described the method section clearly. Thus, we rated "Low" risk of bias in this domain.

Some concerns
Lack of detailed methods for the randomisation process contributed to "some concerns" in overall bias for this trial.

Signalling question
Bias arising from the randomization process

Was the allocation sequence random?
Only stated "random" 1.2 Was the allocation sequence concealed until participants were recruited and assigned to interventions?
1.3 Were there baseline imbalances that suggest a problem with the randomization process?

Bias in selection of the reported result
Are the reported outcome data likely to have been selected, on the basis of the results, from…

Low
This study employed adequate randomisation methods in the trial and represents low risk of bias in the randomisation process.

Low
Although some missing data in this trial, the authors applied multivariate analyses of covariance and survival analyses on the results. So we rated "Low" risk of bias.
N Self-report and confirmed by laboratory results. "At each assessment, the patient was classified by the therapist as abstinent or relapsed according to his or her self-report." "Drinking diary, laboratory measures, and interviews of collaterals were compared for consistency and were used to justify abstinence, lapses, and relapses (D.N., K.W.)." NA Low Double-blind design and self-reporting outcome (confirmed by relatives and biochemistry results) put this domain as low risk of bias.

Low
No protocol was found but nature of outcome, abstinence, presents low risk of selected reported results and the authors described the method section clearly. Thus, we rated "Low" risk of bias in this domain.

Overall bias Low
Low risk of bias in all domains resulting low risk of bias overall.

Ladewig 1993
Aim assignment to intervention (the 'intention-totreat' effect) Source Journal article(s) with results of the trial

Domain Response Comments
NI NI NI No baseline characteristics of participants was given by groups.

Some concerns
No details were given regarding randomisation process, contributed to "some concerns" in this domain. PN PN NA NA N Analysis was intention to treat.

Low
Double-blind design and ITT analysis employed in this trial put low risk of bias in deviation from intended interventions.

PY
The authors followed all participants and reported their outcomes (not by group) NA NA

Low
Although there were some missing data, the authors followed all participants and reported their abstinence status during the trial. Therefore, we rated "Low" risk of bias in this domain.
N "Double-blind" NA Low Double-blind design and self-reporting outcome put this domain as low risk of bias.

Low
No protocol was found but nature of outcome, abstinence, presents low risk of selected reported results and the authors described the method section clearly. Thus, we rated "Low" risk of bias in this domain.

Risk of bias judgement
Bias arising from the randomization process 1.1 Was the allocation sequence random?
"Allocation codes were provided in sealed envelopes for each patient at the pharmacy of the University Hospital of Hamburg, where formulation and blinding was conducted. The randomization was organized by a computer-generated list (M.B.)." 1.2 Was the allocation sequence concealed until participants were recruited and assigned to interventions?
1.3 Were there baseline imbalances that suggest a problem with the randomization process?

Signalling question
Bias arising from the randomization process 1.1 Was the allocation sequence random?
Only stated "random" 1.2 Was the allocation sequence concealed until participants were recruited and assigned to interventions?
1.3 Were there baseline imbalances that suggest a problem with the randomization process?

Bias in selection of the reported result
Are the reported outcome data likely to have been selected, on the basis of the results, from…

Risk of bias judgement
Overall bias

Some concerns
Lack of detailed methods for the randomisation process contributed to "some concerns" in overall bias for this trial.

Reference
Landabaso 1999 Aim assignment to intervention (the 'intention-totreat' effect) Source Journal article(s) with results of the trial

Domain Response Comments
NI NI PN Same size between two groups and no significant difference in characteristic between groups.

Some concerns
No details were given regarding randomisation process, contributed to "some concerns" in this domain.
NI NI NI No information on deviations from intended interventions.

NA PN
No evidence of patients being analysed in the incorrect group.

Some concerns
There was no complete information regarding blinding, contributing to "some concerns" in this domain.
PY Very small number of missing data ( Table 2) and contexts NA NA

Low
Almost all outcome data were available, thus this domain was rated "Low" risk of bias.

NI
Didn't address the method of assessing abstinence but it seemed to be self-report.
PY Potential for assessor bias in prompting of patients to accurately or not recall the number of drinks taken

High
No texts mentioned blinding outcome assessor (patient themselves) and potential bias, contributing to "High" risk of bias.

NI
No detailed of evluation.

Some concerns
No protocol was found and the authors did not describe the method for abstinence clearly. Thus, we rated "Some concerns" risk of bias in this domain.

High
High risk of bias due to unclear methods in blinding and methods for assessing outcomes.

Mann 2006
Aim assignment to intervention (the 'intention-totreat' effect) Source Journal article(s) with results of the trial Outcome Abstinence (24 weeks) Results

Domain Response Comments
NI NI PN "With regard to sociodemographic and prestudy data, both GAL and placebo groups were homogenous at baseline (Table 1)."

Some concerns
No details were given regarding randomisation process, contributed to "some concerns" in this domain.

NI
The authors did not publish the details of drop-outs.

PN
"If appropriate, we calculated last-observation-carried-forward (LOCF) analyses. A two-tailed P value less than 0.05 was considered to be significant.Missing data were not replaced." "Survival analyses carried out for each trial center yielded no between-center differences."

Some concerns
High proportion of drop-outs and lack of details of drop-outs put this domain as "some concerns".

Risk of bias judgement
Signalling question Bias arising from the randomization process 1.1 Was the allocation sequence random?
Only stated "random" 1.2 Was the allocation sequence concealed until participants were recruited and assigned to interventions?
1.3 Were there baseline imbalances that suggest a problem with the randomization process?

Signalling question
Bias arising from the randomization process 1.1 Was the allocation sequence random?
Only stated "randomized" 1.2 Was the allocation sequence concealed until participants were recruited and assigned to interventions?
1.3 Were there baseline imbalances that suggest a problem with the randomization process?

Bias in selection of the reported result
Are the reported outcome data likely to have been selected, on the basis of the results, from…

Low
No protocol was found but nature of outcome, abstinence, presents low risk of selected reported results and the authors described the method section clearly. Thus, we rated "Low" risk of bias in this domain.

Some concerns
Lack of detailed methods for the randomisation process and high proportion of missing data without details, together, these contributed to "some concerns" in overall bias for this trial.

Marra 2002
Aim assignment to intervention (the 'intention-totreat' effect) Source Journal article(s) with results of the trial Outcome Abstinence (12 months) Results

Domain Response Comments
NI NI PN Except VAS results, there were not significant difference between groups.

Some concerns
No details were given regarding randomisation process, contributed to "some concerns" in this domain.

Low
No protocol was found but nature of outcome, abstinence, presents low risk of selected reported results and the authors described the method section clearly. Thus, we rated "Low" risk of bias in this domain.

Overall bias High
Lack of detailed methods for the randomisation process and missing data, together, these contributed to "High" in overall bias for this trial.

Martinotti 2009
Aim assignment to intervention (the 'intention-totreat' effect) Source Journal article(s) with results of the trial

Domain Response Comments
Y PY PN No detailed characteristic of participants but the authors stated that there were no significant differences betweene the baseline characteristics of patients.

Low
This study employed adequate randomisation methods in the trial and represents "Low" risk of bias in the randomisation process.  "Random assignment was achieved in a non-centre-specific manner with an interactive voice-response central randomisation service." 1.2 Was the allocation sequence concealed until participants were recruited and assigned to interventions?
1.3 Were there baseline imbalances that suggest a problem with the randomization process?

Bias in selection of the reported result
Are the reported outcome data likely to have been selected, on the basis of the results, from… N "Primary and secondary efficacy analyses were performed on the intent-to-treat population, which included all randomly assigned patients who took at least one dose of study medication."

Low
Double-blind design and ITT analysis employed in this trial put "Low" risk of bias in deviation from intended interventions.

PN
Completion: 22/29 in ARI group and 21/28 in NAL group PN There were higher proportion of discontinuation due to adverse events in NAL group (17.8%), compared with (6.9%) the ARI group.

Some concerns
High porportion of missing data and no detailed reasons for missing data put "some concerns" in this domain. N double-blind; self-evluation and family member interview NA Low Double-blind design and self-reporting outcome (confirmed by relatives) put this domain as low risk of bias.

Low
No protocol was found but nature of outcome, abstinence, presents low risk of selected reported results and the authors described the method section clearly. Thus, we rated "Low" risk of bias in this domain.

Overall bias
Some concerns Some concerns overall due to missing data in this trial.

Martinotti 2007
Aim assignment to intervention (the 'intention-totreat' effect) Source Journal article(s) with results of the trial

Domain Response Comments
Y PY PN There was no significant difference between groups except OCDS between groups, which might result from chance.

Low
This study employed adequate randomisation methods in the trial and represents low risk of bias in the randomisation process.

Some concerns
The open label design in this trial might prompt deviations from the intended interventions, contributing to "some concerns" in this domain.

High
High porportion of missing data and imbalanced missing data without sensitivity analysis put "high" risk of bias in this domain.

Low
Although this is an open study, the outcome (abstinence) was confirmed by a family member, which put low risk of bias in this domain.

Low
No protocol was found but nature of outcome, abstinence, presents low risk of selected reported results and the authors described the method section clearly. Thus, we rated "Low" risk of bias in this domain.

Overall bias High
Open-label design and missing data put this trial "High" risk of bias overall

Signalling question
Bias arising from the randomization process 1.1 Was the allocation sequence random?
"Random assignment was achieved in a non-centre-specific manner with an interactive voice-response central randomisation service." 1.2 Was the allocation sequence concealed until participants were recruited and assigned to interventions?
1.3 Were there baseline imbalances that suggest a problem with the randomization process?

Bias in selection of the reported result
Are the reported outcome data likely to have been selected, on the basis of the results, from… "...randomisation was performed using a common computer-generated system." "All study personnel in contact with the participants wre unaware of the randomisation sequence" 1.2 Was the allocation sequence concealed until participants were recruited and assigned to interventions?

Bias in selection of the reported result
Are the reported outcome data likely to have been selected, on the basis of the results, from…

PN
No characteristic of participants but the authors stated that no significant differences between two groups.

Low
This study employed adequate randomisation methods in the trial and represents low risk of bias in the randomisation process. NI "The overall rate of study discontinuation due to adverse event was 3.2% in the PRE group and 17.8% in the NAL group."

Some concerns
High porportion of missing data and no detailed reasons for missing data put "some concerns" in this domain. N double-blind; abstinence was evaluated by self-evaluation and family member interview.
NA Low Double-blind design and self-reporting outcome (confirmed by relatives) put this domain as low risk of bias.

Low
No protocol was found but nature of outcome, abstinence, presents low risk of selected reported results and the authors described the method section clearly. Thus, we rated "Low" risk of bias in this domain.

Overall bias
Some concerns Some concerns overall due to lack of details in missing data.

Moncini 2000
Aim assignment to intervention (the 'intention-totreat' effect) Source Journal article(s) with results of the trial Outcome Abstinence (6 months) Results  Table 3; the drop-out rates were low (2/9, group A and 2/8, group B) but did not indicate the reasons.

NI
The proportions of missing data were similar but no information regarding their reasons PN Small study

Some concerns
There were some missing data. Although the numbers of missing data were similar, no details and small study effects put "some concerns" in this domain.

Low
Double-blind design put this domain as low risk of bias.

Low
No protocol was found but nature of outcome, abstinence, presents low risk of selected reported results. Thus, we rated "Low" risk of bias in this domain.

Some concerns
Some concerns with missing data due to no details of missing data and a small study.
the randomization process 1.3 Were there baseline imbalances that suggest a problem with the randomization process?

Risk of bias judgement
Bias due to missing outcome data 3

Signalling question
Bias arising from the randomization process 1.1 Was the allocation sequence random?
Only stated "random"; "When discharged from the hospital, the patients were randomly divided into two groups, A and B" "Might indicate a sign of sequence concealment: "They were randomly divided into two groups (group A and group B) and, when the code was opened, group A proved to have been treated with GHB (mean daily oral dose 50 mg/kg) and group B with placebo." "when the code was opeened, A provided to have been treated with GHB, B with placebo" suggests allocation concealment but no details 1.2 Was the allocation sequence concealed until participants were recruited and assigned to interventions?
1.3 Were there baseline imbalances that suggest a problem with the randomization process?

Bias in selection of the reported result
Are the reported outcome data likely to have been selected, on the basis of the results, from…

Bias in selection of the reported result
Are the reported outcome data likely to have been selected, on the basis of the results, from…

Some concerns
No details were given regarding randomisation process, contributed to "some concerns" in this domain. PN PN NA NA N "Analyses were conducted for all subjects based on treatment assignment regardless of whether or when they stopped taking the study drug (intent-to-treat analysis)."

Low
Double-blind design and ITT analysis employed in this trial put "Low" risk of bias in deviation from intended interventions.

PN
The number of missing data was similar between groups but there could be more participants dropped in crbamazepine group due to medication toxicity.

High
High porportion of missing data and imbalanced missing data without sensitivity analysis put "High" risk of bias in this domain.

Low
Double-blind design put this domain as low risk of bias.

Signalling question
Bias arising from the randomization process 1.1 Was the allocation sequence random?
"...with the use of a table of random numbers." "To avoid selection bias, the randomization was carried out by a UNIAD employee not involved in the study, and who did not have access to any information regarding the research or the patients." 1.2 Was the allocation sequence concealed until participants were recruited and assigned to interventions?
1.3 Were there baseline imbalances that suggest a problem with the randomization process?

Signalling question
Bias arising from the randomization process 1.1 Was the allocation sequence random?
"Once meeting eligibility criteria and providing signed consent, subjects were randomized to either carbamazepine or placebo" No detailed info on randomisation or on allocation concealment 1.2 Was the allocation sequence concealed until participants were recruited and assigned to interventions?
1.3 Were there baseline imbalances that suggest a problem with the randomization process?

Bias in selection of the reported result
Are the reported outcome data likely to have been selected, on the basis of the results, from… "...accoring to a computer-generated randomization list (in blocks of 4; straftification with regard to sex)." "The ranomization list was kept by the biometrician and the study pharmacist who prepared the study medication packages. The study pharmacist did not have any further role in the trial." "Sealed envelopes containing study medication details were kept at the outpatient unit to be opened by a staff member in case of a study drug-related emergency...." 1.2 Was the allocation sequence concealed until participants were recruited and assigned to interventions?
1.3 Were there baseline imbalances that suggest a problem with the randomization process?

Bias in selection of the reported result
Are the reported outcome data likely to have been selected, on the basis of the results, from…

Signalling question
Bias arising from the randomization process 1.1 Was the allocation sequence random?
"Randomisation was stratifited by gender and recruitment site in a block design." 1.2 Was the allocation sequence concealed until participants were recruited and assigned to interventions?
1.3 Were there baseline imbalances that suggest a problem with the randomization process?

Bias in selection of the reported result
Are the reported outcome data likely to have been selected, on the basis of the results, from… 5.1. ... multiple outcome measurements (e.g. scales, definitions, time points) within the outcome domain?
5.2 ... multiple analyses of the data?

Signalling question
Bias arising from the randomization process 1.1 Was the allocation sequence random?
"According to a predetermined randomization list, eligible patients were assigned treatment with either 333 mg Acamprosate tablets on a dosage of four or six tablets per day in divided doses, or a matching placebo tablet." 1.2 Was the allocation sequence concealed until participants were recruited and assigned to interventions?
1.3 Were there baseline imbalances that suggest a problem with the randomization process?

Bias in selection of the reported result
Are the reported outcome data likely to have been selected, on the basis of the results, from…

Signalling question
Bias arising from the randomization process

Some concerns
No details were given regarding randomisation process, contributed to "some concerns" in this domain.

High
High porportion of missing data and no detailed methods for missing data put "High" risk of bias in this domain.

PN
Did not indicate the methods for assessing abstinence but the authors stated this was a double-blind study NA Low Double-blind design put this domain as low risk of bias.

Some concerns
No protocol was found and the authors did not describe the method for abstinence clearly. Thus, we rated "Some concerns" risk of bias in this domain.

High
High risk of bias in missing data, some concerns in randomisation process and little information regarding the methods and analyses contributed to "High" risk of bias in overall bias.

Pelc 2005
Aim assignment to intervention (the 'intention-totreat' effect) Source Journal article(s) with results of the trial

Domain Response Comments
There were a few differences in the educational status among two groups and marital status but these should be caused by chance rather than a problem with randoimization process.

Low
This study employed adequate randomisation methods in the trial and represents low risk of bias in the randomisation process.
There was a high attendence in the self-group participantion in the "no nurse follow-up" group.

PY
The subgroup analysis showed that self-help group participation had interaction with treatment "Although the results should be interpreted cautiously on account of the relatively low number of patients, significant treatment interactions were observed between gender and participation in self-help groups." N ITT analysis NA High Open-label design and possible deviations from intended interventions contributed to "High" risk of bias N High proportion of missing data (Figure 1) PN Numbers and reasons were different between groups.

High
High porportion of missing data and imbalanced missing data without sensitivity analysis put "high" risk of bias in this domain.
Y Should be patient self-report and this trial is an open trial so outcome assessors were aware of the intervention received by study participants.

High
This is an open study and the outcome (abstinence) could be influenced by the knowledge of interventions. Therefore, we rated "High" in this domain.

Low
No protocol was found but nature of outcome, abstinence, presents low risk of selected reported results and the authors described the method section clearly. Thus, we rated "Low" risk of bias in this domain.

High
High risk of bias in the deviations from the intended interventions, some concerns in missing data and outcome measurements put "High" risk of bias in overall bias.

Signalling question
Bias arising from the randomization process 1.1 Was the allocation sequence random?
"Investigators telephoned the centre before the inclusion of each subject, to obtain a randomization number defining the group to which the patient was to be assigned." 1.2 Was the allocation sequence concealed until participants were recruited and assigned to interventions?
1.3 Were there baseline imbalances that suggest a problem with the randomization process?

Bias in selection of the reported result
Are the reported outcome data likely to have been selected, on the basis of the results, from…

Domain Response Comments
NI NI N Equal number of pariticipants in each group and no difference among characteristics of participants. "A total of 188 patients were included in the trial: 62 were randomised to placebo (placebo), 63 to acamprosate 1332 mg/day (acamp. 1332) and 63 to acamprosate 1998 mg/day (acamp. 1998). No statistical difference was present for any criterion at inclusion."

Some concerns
No details were given regarding randomisation process, contributed to "some concerns" in this domain. PN loss to follow up 24% control, 9.5% acamp 1332, 9.5% acadmp 1998. so uneven and reasons not given. total 32% lost from placebo (excl relapse) ~20% acadmp 1332 and 17% from acamp 1998 PN Difference in the proportions of missing data was observed and no details were given PN No sensitivity analysis was performed. The authors treated participants dropped out as non abstinent

High
High porportion of missing data and imbalanced missing data without sensitivity analysis put "high" risk of bias in this domain.
N "double-blind"; alcohol consumption was assessed by review of patients' diary consumption cards and confirmed by urine test at each test.
NA Low Double-blind design and self-reporting outcome (confirmed by biochemistry results) put this domain as low risk of bias.

Low
No protocol was found but nature of outcome, abstinence, presents low risk of selected reported results and the authors described the method section clearly. Thus, we rated "Low" risk of bias in this domain.

High
Lack of detailed methods for the randomisation process and imbalanced missing data, together, these contributed to "High" in overall bias for this trial.

Poldrugo 1997
Aim assignment to intervention (the 'intention-totreat' effect) Source Journal article(s) with results of the trial

PN
More participants in placebo group refused to continue and severe relapsed.
N Although the authors used ITT analyses, the imbalanced missing data could not premit robust results.

High
Imbalanced missing outcome data across the groups led to high risk of bias

Risk of bias judgement
Signalling question Bias arising from the randomization process 1.1 Was the allocation sequence random?
Only stated "Random"; "The patients were randomly assigned to one of the three treatment groups..." 1.2 Was the allocation sequence concealed until participants were recruited and assigned to interventions?
1.3 Were there baseline imbalances that suggest a problem with the randomization process?

Signalling question
Bias arising from the randomization process 1.1 Was the allocation sequence random?
Did not stated the method but "Patients were randomized by individual subject randomization to the acamprosate group..." 1.2 Was the allocation sequence concealed until participants were recruited and assigned to interventions?
1.3 Were there baseline imbalances that suggest a problem with the randomization process?

Bias in selection of the reported result
Are the reported outcome data likely to have been selected, on the basis of the results, from…

Low
No protocol was found but nature of outcome, abstinence, presents low risk of selected reported results. Thus, we rated "Low" risk of bias in this domain.

Overall bias High
High risk of bias due to imbalanced missing outcome data.

Some concerns
It was impossible to blind participants in this trial due to nature of interventions employed, which potentially induce deviations from intended interventions. On the other hand, the authors applied ITT analyses. Together, these contributed to "some concerns" in this domain.

High
High porportion of missing data and imbalanced missing data without sensitivity analysis put "high" risk of bias in this domain.

PN
Participants was blinded NA Low Single-blind design and self-reporting outcome (confirmed by relatives and biochemistry results) put this domain as low risk of bias.

Low
No protocol was found but nature of outcome, abstinence, presents low risk of selected reported results. Thus, we rated "Low" risk of bias in this domain.

Overall bias High
High risk of bias derived from the missing data, deviations from the intended interventions and lack of details in the randomisation process, together, these put "High" risk of bias.

Richter 2012
Aim assignment to intervention (the 'intention-totreat' effect) Source Journal article(s) with results of the trial

Domain Response Comments
Y Y N No significant difference between groups (Table 1).

Low
This study employed adequate randomisation methods in the trial and represents low risk of bias in the randomisation process. Double-blind design and ITT analysis employed in this trial put "Low" risk of bias in deviation from intended interventions.

Bias due to deviations from intended interventions
2.1 Were participants aware of their assigned intervention during the trial?
"single-blind" but it was impossible to blind participants as this trial did not provide placebo. Only stated "random" 1.2 Was the allocation sequence concealed until participants were recruited and assigned to interventions?
1.3 Were there baseline imbalances that suggest a problem with the randomization process?

Bias in selection of the reported result
Are the reported outcome data likely to have been selected, on the basis of the results, from…

Risk of bias judgement
Bias arising from the randomization process 1.1 Was the allocation sequence random?
"Randomization was computerized, central and independent of the center, and blinded for physician and participants." 1.2 Was the allocation sequence concealed until participants were recruited and assigned to interventions?
1.3 Were there baseline imbalances that suggest a problem with the randomization process?

Bias in selection of the reported result
Are the reported outcome data likely to have been selected, on the basis of the results, from… Signalling question N "Overall, 79% patients (n = 158) completed the trial per protocol: 80 (75%) of 106 in the placebo and 78 (82%) of 95 in the LEV group." PN "Dropout reasons in the placebo group (n = 26) were depression with suicidal tendency (severe AE), a panic attack (severe AE), 9 side effects, a refusal of participation, 8 noncompliances, 2 other causes, 4 unknown reasons. In the LEV group (n = 12), we counted 2 side effects, a refusal of participation, 8 noncompliances, and 1 unknown reason as dropouts."

High
High porportion of missing data and imbalanced missing data without sensitivity analysis put "high" risk of bias in this domain.
N "double-blind" NA Low Double-blind design and self-reporting outcome (confirmed by relatives and biochemistry results) put this domain as low risk of bias.

Low
No protocol was found but nature of outcome, abstinence, presents low risk of selected reported results and the authors described the method section clearly. Thus, we rated "Low" risk of bias in this domain.

Overall bias High
High risk of bias in overall bias domain due to imbalanced missing data.

Rubio 2005
Aim assignment to intervention (the 'intention-totreat' effect) Source Journal article(s) with results of the trial

Domain Response Comments
NI NI N No significant difference between groups (Table 1).

Some concerns
No details were given regarding randomisation process, contributed to "some concerns" in this domain.
Y Y NI NA PN seems to be ITT analysis NA

Some concerns
The open label design in this trial might prompt deviations from the intended interventions, contributing to "some concerns" in this domain.
N high rate of drop outs (27.98% in NAL and 34.52% in control)

NI
The proportion of drop-outs did not differ between groups but there was no reasons reported.

Some concerns
There were some missing data, which were not evenly distributed in both groups and no details reported. "Some concerns" was rated in this domain.

Low
Although this was an open-label study, outcome was confirmed by biochemistry methods.

Low
No protocol was found but nature of outcome, abstinence, presents low risk of selected reported results and the authors described the method section clearly. Thus, we rated "Low" risk of bias in this domain.

Overall bias
Some concerns Some concerns due to lack of details in randomisaion process, open-label design and no details in missing data.

Risk of bias judgement
Bias arising from the randomization process 1.1 Was the allocation sequence random?
"Sealed envelope randomization with balance by blocks of 8 (4 per study medication) was used to obtain equal numbers per treatment group at each center." 1.2 Was the allocation sequence concealed until participants were recruited and assigned to interventions?
1.3 Were there baseline imbalances that suggest a problem with the randomization process?

Bias in selection of the reported result
Are the reported outcome data likely to have been selected, on the basis of the results, from…

Low
This study employed adequate randomisation methods in the trial and represents low risk of bias in the randomisation process. Double-blind design and ITT analysis employed in this trial put "Low" risk of bias in deviation from intended interventions.

PN
As above

PN
The authors presented PP and ITT results. No sensitivity analysis but treated drop-outs as non abstinent

High
High porportion of missing data and imbalanced missing data without sensitivity analysis put "high" risk of bias in this domain.
N "Double-blind"; "A patient's declaration of drinking behavior was verified by the results of a breathalyzer test and GGT levels in every case and by interviewing a family member if possible." NA Low Double-blind design and self-reporting outcome (confirmed by relatives and biochemistry results) put this domain as low risk of bias.

Low
No protocol was found but nature of outcome, abstinence, presents low risk of selected reported results and the authors described the method section clearly. Thus, we rated "Low" risk of bias in this domain.

Overall bias High
High risk of bias in overall bias due to imbalanced missing data.

Schmidt 2002
Aim assignment to intervention (the 'intention-totreat' effect) Source Journal article(s) with results of the trial Outcome Abstinence (12 months) Results

Domain Response Comments
NI NI N No evidence of differences in baseline characteristics.

Some concerns
No details were given regarding randomisation process, contributed to "some concerns" in this domain. N There were some missing data ( Table 2) N Similar proportions of missing outcome data but more participants in lisuride group (n = 9) dropped out due to adverse event compared to placebo group (n = 3)

High
The authors removed 16 patients from the ITT analysis and imbalanced missing data in adverse events put this domain in "High" risk of bias.
PN "double-blind" NA Low Double-blind design and self-reporting outcome put this domain as low risk of bias.

Low
No protocol was found but nature of outcome, abstinence, presents low risk of selected reported results and the authors described the method section clearly. Thus, we rated "Low" risk of bias in this domain.

High
Lack of detailed methods for the randomisation process contributed to "High" risk of bias in overall bias for this trial.

Risk of bias judgement
Bias due to missing outcome data 3

Signalling question
Bias arising from the randomization process 1.1 Was the allocation sequence random?
Only stated "random" 1.2 Was the allocation sequence concealed until participants were recruited and assigned to interventions?
1.3 Were there baseline imbalances that suggest a problem with the randomization process?

Bias in selection of the reported result
Are the reported outcome data likely to have been selected, on the basis of the results, from…

Domain Response Comments
NI NI PN Same number in each group. Some difference in education status and employment but might be from by chance

Some concerns
No details were given regarding randomisation process, contributed to "some concerns" in this domain.
Y PY NI They all received pharmacological and psychological interventions.
NA N Seemed to be ITT analysis NA

Some concerns
The open label design in this trial might prompt deviations from the intended interventions, contributing to "some concerns" in this domain.

PY
Did not provide information regarding missing outcome data.

Low
The authors did not provide missing data so the reviewer assumed there was no missing data, especially this trial involved small numbers of participants. "Low" risk of bias Y Open study

High
Open study and the outcome can be influenced by the knowledge of interventions -"High" risk of bias PN PN

Low
No protocol was found but nature of outcome, abstinence, presents low risk of selected reported results and the authors described the method section clearly. Thus, we rated "Low" risk of bias in this domain.

High
Lack of information in randomisation process and open-label design prompted "High" risk of bias in this trial.

Tempesta 2000
Aim assignment to intervention (the 'intention-totreat' effect) Source Journal article(s) with results of the trial

Domain Response Comments
PY PY PN Placebo group had higher proportion of patients with high consumption awareness and previous treatment for alcoholism, which might result by chance.

Low
This study employed adequate randomisation methods in the trial and represents "Low" risk of bias in the randomisation process.

PY
Reasons and % seem similar across arms (Table 2) PN No sensitivity analysis but the authors treated "drop-out" as non abstinent

Low
Although there were some missing data, results still stood in consideration of balanced missing data. "Low" risk of bias in this domain was rated.
N Double-blind design NA

Low
Double-blind design put this domain as low risk of bias.
Bias arising from the randomization process 1.1 Was the allocation sequence random?
Only stated "random", "They were randomized into four groups." 1.2 Was the allocation sequence concealed until participants were recruited and assigned to interventions?
1.3 Were there baseline imbalances that suggest a problem with the randomization process?

Risk of bias judgement
Bias due to missing outcome data 3

Low
This study employed adequate randomisation methods in the trial and represents "Low" risk of bias in the randomisation process. Y Y NI NA N "All data were used on an intention-to-treat basis."

Some concerns
The open label design in this trial might prompt deviations from the intended interventions, contributing to "some concerns" in this domain.
N "Whereas 17 patients in the disulfi ram group started group treatment, only seven completed it (41%). In the control group fewer started, i.e. 15, but 10 of these completed group treatment (67%)." NI There was no statistically significant difference in the proportions of missing outcomes. No reasons were given.

PN
No information regarding sensitivity analysis but the authors treated all drop-outs as relapsed.

Some concerns
High porportion of missing data and no detailed reasons for missing data put "some concerns" in this domain.

PY
No information regarding the method of assessing abstinence but since this is an open-trial, patients were aware of the intervention received.

High
Open-label design and outcome can be influenced by the knowledge of interventions -"High" was rated.

Low
No protocol was found but nature of outcome, abstinence, presents low risk of selected reported results and the authors described the method section clearly. Thus, we rated "Low" risk of bias in this domain.

Overall bias High
High in overall bias due to open-label design and missing data without details.

Reference
Volpicelli 1997 Aim assignment to intervention (the 'intention-totreat' effect) Source Journal article(s) with results of the trial Outcome Abstinence (12 weeks) Results

Domain Response Comments
Y NI N "The baseline sociodemographics of the 2 study groups of the sample did not differ (Table 1 )."

Some concerns
No details were given regarding allocation concealment, contributed to "some concerns" in this domain. Bias arising from the randomization process 1.1 Was the allocation sequence random?
"A nurse from the department not participating in the study performed randomization using sealed envelopes containing a label for one of the two treatment conditions." "In order to ensure a balanced number of patients in each group the envelopes were arranged in blocks of 6-10. In each block, half were labelled "disulfiram" and half were labelled "control".
"The envelopes were arranged and sealed by a secretary not participating in the study who was instructed to arrange the labels in a random order." 1.2 Was the allocation sequence concealed until participants were recruited and assigned to interventions?
1.3 Were there baseline imbalances that suggest a problem with the randomization process?

Risk of bias judgement
Bias in selection of the reported result 5

Low
No protocol was found but nature of outcome, abstinence, presents low risk of selected reported results. Thus, we rated "Low" risk of bias in this domain.

Overall bias
Some concerns Some concerns due to lack of details in randomisation process and missing data

Low
This study employed adequate randomisation methods in the trial and represents "Low" risk of bias in the randomisation process.

Some concerns
High porportion of missing data and no detailed reasons for missing data put "some concerns" in this domain.
N double-blind NA Low Double-blind design and self-reporting outcome put this domain as low risk of bias.

Low
No protocol was found but nature of outcome, abstinence, presents low risk of selected reported results and the authors described the method section clearly. Thus, we rated "Low" risk of bias in this domain.

Overall bias
Some concerns Some concerns due to lack of details in missing data.

Reference
Whitworth 1996 Aim assignment to intervention (the 'intention-totreat' effect) Source Journal article(s) with results of the trial

Domain Response Comments
Y Y PN "The groups were well matched in terms of demographic and alcoholrelated baseline variables on the day of selection and on day 0 (table 1)"

Low
This study employed adequate randomisation methods in the trial and represents "Low" risk of bias in the randomisation process.

PN
Bias arising from the randomization process 1.1 Was the allocation sequence random?
"Randomization followed a centralized assignment procedure independent of responsible or treating clinicians and hospitals." 1.2 Was the allocation sequence concealed until participants were recruited and assigned to interventions?
1.3 Were there baseline imbalances that suggest a problem with the randomization process?

Bias in selection of the reported result
Are the reported outcome data likely to have been selected, on the basis of the results, from… "The duration of double-blind treatment was 360 days."

Signalling question
Bias arising from the randomization process 1.1 Was the allocation sequence random?
"Rnadomisation was by computer-generated list organised in blocks of eight." "Allocation codes were provided in sealed envelopes for each patient." 1.2 Was the allocation sequence concealed until participants were recruited and assigned to interventions?
1.3 Were there baseline imbalances that suggest a problem with the randomization process?

Bias in selection of the reported result
Are the reported outcome data likely to have been selected, on the basis of the results, from…

Low
Although there were some missing data, results still stood in consideration of balanced missing data between groups. "Low" risk of bias in this domain was rated.
PN "double-blind" NA Low Double-blind design and self-reporting put this domain as low risk of bias.

Low
No protocol was found but nature of outcome, abstinence, presents low risk of selected reported results and the authors described the method section clearly. Thus, we rated "Low" risk of bias in this domain.

Overall bias
Low

Low
Double-blind design and mITT analysis employed in this trial put "Low" risk of bias in deviation from intended interventions.
N "Of 281 patients enrolled, 91 (32.4%) completed the trial (6 months treatment, 6 months follow-up)" N More patients in FLUX group dropped out due to severe relapse.

Some concerns
High porportion of missing data and imbalanced missing data without sensitivity analysis put "high" risk of bias in this domain.

PN
"double-blind"; "Outcome variables were based on absolute abstinence, which was defined as no alcohol consumption. To be considered abstinent, the patient's self-report had to be in accordance with the investigator's clinical assessment and the result of a breath analyser." NA Low Double-blind design and self-reporting outcome (confirmed by biochemistry results) put this domain as low risk of bias.

Low
No protocol was found but nature of outcome, abstinence, presents low risk of selected reported results and the authors described the method section clearly. Thus, we rated "Low" risk of bias in this domain.

Some concerns
Lack of detailed methods for the randomisation process and substantial difference in the reasons for missing data, together, these contributed to "some concerns" in overall bias for this trial.

Some concerns
The open label design in this trial might prompt deviations from the intended interventions, contributing to "some concerns" in this domain.
Y "The 2 treatments did not differ with respect to treatment adherence, which was high in both groups." >90% for both at 3 months NA NA

Low
The proportions of missing data were small, suggesting "Low" risk of bias.
Y "Alcohol intake was assessed at each treatment session. Both the patient and the significant other were interviewed and the higher reported intake level was used." PY Outcomes might be influenced by the knowledge of interventions.

High
High risk of bias due to outcomes can be influenced by the knowledge of interventions.
PN "Alcohol intake was assessed at each treatment session. Both the patient and the significant other were interviewed and the higher reported intake level was used."

Low
No protocol was found but nature of outcome, abstinence, presents low risk of selected reported results and the authors described the method section clearly. Thus, we rated "Low" risk of bias in this domain.

Overall bias High
Lack of details in randomisation and open-label design contributed to "High" risk of bias in this trial. Only stated "random" 1.2 Was the allocation sequence concealed until participants were recruited and assigned to interventions?

Reference
1.3 Were there baseline imbalances that suggest a problem with the randomization process?

Risk of bias judgement
Bias arising from the randomization process

Some concerns
No full study report put this domain as "some concerns"

Overall bias
High "High" risk of bias due to lack of full reports, details of randomisation process and missing data.  (Table 3)

Some concerns
No details were given regarding randomisation process, contributed to "some concerns" in this domain. PY Table 1 and 5, % and reasons seem comparable between treatment arms.

Low
Although there were some missing data, results still stood in consideration of missing data balanced betwee groups. "Low" risk of bias in this domain was rated.
PN "double-blind" NA Low Double-blind design and self-reporting outcome put this domain as low risk of bias.

Low
No protocol was found but nature of outcome, abstinence, presents low risk of selected reported results and the authors described the method section clearly. Thus, we rated "Low" risk of bias in this domain.

Some concerns
Some concerns from lack of randomisation process details.

Reference
PREDICT study Aim assignment to intervention (the 'intention-totreat' effect) Source Journal article(s) with results of the trial; Non-commercial trial registry record (e.g. ClinicalTrials.gov record); Research ethics application Outcome Abstinence (90 days) Results

Domain Response Comments
Y PY N No significant difference between groups was found

Low
This study employed adequate randomisation methods in the trial and represents "Low" risk of bias in the randomisation process.

Low
Double-blind design and ITT analysis employed in this trial put "Low" risk of bias in deviation from intended interventions.
N There were some drop-outs among groups (more than 10%)

PY
The proportions of missing data among groups were similar between groups.

Low
Although there were some missing data, results still stood in consideration of balanced missing data among groups. "Low" risk of bias in this domain was rated.

Risk of bias judgement
Signalling question Bias arising from the randomization process 1.1 Was the allocation sequence random?
Only stated "random" but no methods were mentioned 1.2 Was the allocation sequence concealed until participants were recruited and assigned to interventions?
1.3 Were there baseline imbalances that suggest a problem with the randomization process?

Risk of bias judgement
Bias arising from the randomization process 1.1 Was the allocation sequence random? "...using an imbalanced block-randomization algorithm..." "the trial register stated "triple" masking, which suggest possible allocation concealment." 1.2 Was the allocation sequence concealed until participants were recruited and assigned to interventions?
1.3 Were there baseline imbalances that suggest a problem with the randomization process?

Bias in selection of the reported result
Are the reported outcome data likely to have been selected, on the basis of the results, from… N Pre-registered on clincial trials.gov, but no definitely given for abstinence, as continuous abstinence not one of the outcomes (data retrieved from personal communication). Based on this, it is unlikely that selective reporting has occured. N Pre-registered on clincial trials.gov, but no definitely given for abstinence, as continuous abstinence not one of the outcomes (data retrieved from personal communication). Based on this, it is unlikely that selective reporting has occured.

Low
Protocol was found on ClinicalTrial.gov. No evidence of selection of reporting due to the nature of outcome (abstinence).

Overall bias Low
Low risk of bias

Low
Although it was impossible to blind participants in this trial due to nature of interventions employed, this was expected in psychosocial interventions and would not lead to deviation. On the other hand, the authors applied ITT analyses. Together, these contributed to "low" in this domain.
N There were some drop-outs among groups (more than 10%)

PN
The proportions of missing data among groups were not similar between groups.

High
Disproportional missing outcome data led to high risk of bias.
Y It was impossible to blind outcome assessors (participants).
PY "Blood alcohol levels were measured in all participants by using Alcoscan AL7000" every month. However, the test only provides daily measurement and hard to know the drinking habit throughout the trial period.

Some concerns
Lack of blinding to outcome assessors (participants themselve) and self-reporting outcomes, which put "some concerns" in this domain.

PY
Describes itself as a 2y follow-up study but only has data for the first year. There were other follow-up time points at 45 days, 90 days, and 180 days. Continuous abstinence data is not reported for these timepoints.

Some concerns
The authors described itself as a 2y follow-up study but only has data for the first year. However, continuous abstinence data were not reported for all the timepoints, leading to some concerns.

High
Disporportional dropouts and unclear methods in randomisation process led to high risk of bias.

Reference
Jirapramukpitak 2020 Aim assignment to intervention (the 'intention-totreat' effect) Source Journal article(s) with results of the trial; Non-commercial trial registry record (e.g. ClinicalTrials.gov record)

Low
Although it was impossible to blind participants in this trial due to nature of interventions employed, this was expected in psychosocial interventions and would not lead to deviation. On the other hand, the authors applied ITT analyses. Together, these contributed to "low" in this domain.
Y Only 1 dropout in each group (1/80 vs 1/42+37) NA NA Low Nearly all patients were followed during the trial.
Y It was impossible to blind outcome assessors.
PN "All negative samples were confirmed by reports of no drinking provided by participants themselves and their informants."

Low
Although there were no blinding to outcome assessors, the authors sought different sources to confirm patients' abstinence status.
N Pre-registered on Thailand National trial register, but no definitely given for abstinence nor patient criteria.

Low
Protocol was found on ClinicalTrial.gov. No evidence of selection of reporting due to the nature of outcome (abstinence).

Overall bias
Some concerns Some concerns in overall bias due to no adequate information in allocation concealment.

Bias in selection of the reported result
Are the reported outcome data likely to have been selected, on the basis of the results, from… 5.1. ... multiple outcome measurements (e.g. scales, definitions, time points) within the outcome domain?
5.2 ... multiple analyses of the data?

Risk of bias judgement
Risk of bias judgement 2.6 If Y/PY/NI to 2.5: Was there potential for a substantial impact (on the estimated effect of intervention) of analysing participants in the wrong group?

Risk of bias judgement
Bias due to missing outcome data 3

Figure 1. Forest plot
Results are displayed as a point estimate and 95% confidence interval. The blue colour represents direct evidence from each study, grouped by design. The green colour represents pooled treatment effect in each design, estimated by the inconsistency model. The red colour represents overall treatment effect, estimated by the consistency model.   Results are displayed as a point estimate and 95% confidence interval. The blue colour represents direct evidence from each study, grouped by design. The green colour represents pooled treatment effect in each design, estimated by the inconsistency model. The red colour represents overall treatment effect, estimated by the consistency model.

SUPPLEMENT 8. ADDITIONAL ANALYSES
In the additional analyses, we performed meta-regression on the main outcome data. We found no convincing causes of heterogeneity in intervention effects across the five variables we examined (Table 1 and Table 2). The few regression coefficients (3 out of 116) that suggested associations were compatible with chance.
Results of further sensitivity analyses focussing on different time points are presented in Figures S1-S6 for short term, Figures S7-S18 for medium term and Figures S19-S24 for long-term time points. In the medium-term analysis, there was no connected network so two subset network analyses were conducted using TAU (subset 1) and PLA (subset 2) as references.
Results for analyses based on type of interventions appear in Figures S25-S30 for psychotherapy and Figures S31-S36 for pharmacotherapy, with outcomes up to 12 months. Results were broadly in agreement.
The The following captions for all network plots throughout the documents: The size of circles is proportional to the number of randomised patients and the width of lines is proportional to the number of studies in each direct comparison. The colour of lines represents the overall risk of bias in the majority studies (green: low risk of bias; yellow: some concerns; red: high risk of bias).  Figure S1. Network plot of abstinence analysis in short-term (