Benzodiazepines for alcohol withdrawal

[Intervention Review] Benzodiazepines for alcohol withdrawal Christos Ntais 1, Emilios Pakos 2, Panayiotis Kyzas 2, John PA Ioannidis 3 1 Department of Hygiene and Epidemiology, University of Ioannina School of Medicine, Ioannina, Greece. 2 Department of Hygiene and Epidemiology, University of Ioannina, Ioannina, Greece. 3 Department of Hygiene and Epidemiology, Evidence-based Medicine Unit, University of Ioannina School of Medicine, Ioannina, Greece Contact address: Christos Ntais, Department of Hygiene and Epidemiology, University of Ioannina School of Medicine, 14 Ch Zoidi st, Ioannina, GR45444, Greece. md02802@cc.uoi.gr. (Editorial group: Cochrane Drugs and Alcohol Group.) Cochrane Database of Systematic Reviews, Issue 1, 2009 (Status in this issue: Unchanged) DOI: 10.1002/14651858.CD005063.pub2 This version first published online: 20 July 2005 in Issue 3, 2005. Last assessed as up-to-date: 18 May 2005. (Help document - Dates and Statuses explained) This record should be cited as: Ntais C, Pakos E, Kyzas P, Ioannidis JPA. Benzodiazepines for alcohol withdrawal. Cochrane Database of Systematic Reviews 2005, Issue 3. Art. No.: CD005063. DOI: 10.1002/14651858.CD005063.pub2. Background A B S T R A C T Alcohol withdrawal syndrome is a cluster of symptoms that occurs in alcohol-dependent people after cessation or reduction in alcohol use. This systematic review focuses on the evidence of benzodiazepines use in the treatment of alcohol withdrawal symptoms. Objectives To evaluate the effectiveness and safety of benzodiazepines in the treatment of alcohol withdrawal. Search strategy We searched the Cochrane Central Register of Controlled Trials (The Cochrane Library Issue 4, 2004), MEDLINE (1966 to October 2004) and EU-PSI PSI-Tri database with no language and publication restrictions. We also screened references of retrieved articles. Selection criteria All randomised controlled trials examining the effectiveness and safety of a benzodiazepine in comparison with a placebo or other pharmacological intervention or other benzodiazepine were considered. Data collection and analysis Two reviewers independently assessed trial quality and extracted data. Main results Fifty-seven trials, with a total of 4,051 people were included. Despite the considerable number of randomized controlled trials, there was a very large variety of outcomes and of different rating scales and relatively limited quantitative synthesis of data was feasible. Benzodiazepines offered a large benefit against alcohol withdrawal seizures compared to placebo (relative risk [RR] 0.16; 95% confidence interval [CI] 0.04 to 0.69; p = 0.01). Benzodiazepines had similar success rates as other drugs (RR 1.00; 95% CI 0.83 to 1.21) or anticonvulsants in particular (RR 0.88; 95% CI 0.60 to 1.30) and offered a significant benefit for seizure control against nonanticonvulsants (RR 0.23; 95% CI 0.07 to 0.75; p = 0.02), but not against anticonvulsants (RR 1.99; 95% CI 0.46 to 8.65). Changes in Clinical Institute Withdrawal Assessment for Alcohol (CIWA-Ar) scores at the end of treatment were similar with benzodiazepines versus other drugs, although some small studies showed isolated significant differences for other, less commonly, used scales. Data on other comparisons were very limited, thus making quantitative synthesis for various outcomes not very informative. 1

Authors conclusions Benzodiazepines are effective against alcohol withdrawal symptoms, in particular seizures, when compared to placebo. It is not possible to draw definite conclusions about the relative effectiveness and safety of benzodiazepines against other drugs in alcohol withdrawal, because of the large heterogeneity of the trials both in interventions and assessment of outcomes but the available data do not show prominent differences between benzodiazepines and other drugs in success rates. P L A I N L A N G U A G E S U M M A R Y Benzodiazepines are more effective than placebo against alcohol withdrawal seizures while they have variable profile against other commonly used treatments This Cochrane review summarizes evidence from fifty-seven randomised controlled trials evaluating the effectiveness and safety of benzodiazepines in the treatment of alcohol withdrawal symptoms. The available data show that benzodiazepines are effective against alcohol withdrawal seizures when compared to placebo. However, there are no prominent differences between benzodiazepines and other drugs in success rates. Data on safety outcomes are sparse and fragmented. There is a need for larger, well-designed studies in this field. 2

B A C K G R O U N D Alcohol abuse and dependence represents a most serious health problem worldwide with major social, interpersonal and legal interpolations. Dependence on alcohol is associated with both physiological symptoms such as tolerance and withdrawal, and behavioral symptoms such as impaired control over drinking (Hasin 1990). Alcohol withdrawal syndrome is a cluster of symptoms that occurs in alcohol-dependent people after cessation or reduction in alcohol use that has been heavy or prolonged. The clinical presentation varies from mild to serious and the onset of symptoms typically occurs a few hours after the last alcohol intake. The most common manifestations are tremor, restlessness, insomnia, nightmares, paroxysmal sweats, tachycardia, fever, nausea, vomiting, seizures, hallucinations (auditory, visual, tactile), increased agitation, and tremulousness. A minority of patients develop very severe alcohol withdrawal syndrome, including delirium tremens. These symptoms involve a wide range of neurotransmitter circuits that are implicated in alcohol tolerance and reflect a homeostatic readjustment of the central nervous system (De Witte 2003; Hall 1997; Koob 1997; Nutt 1999; Slawecki 1999). Long-term alcohol consumption affects brain receptors that undergo adaptive changes in an attempt to maintain normal function. Some of the key changes involve reduced brain gamma-aminobutyric acid (GABA) levels and GABA - receptor sensitivity (Dodd 2000; Gilman 1996; Kohl 1998; Petty 1993) and activation of glutamate systems (Tsai 1995), which leads to nervous system hyperactivity in the absence of alcohol. The advances in knowledge of neurobiology and neurochemistry have prompted the use of drugs in the treatment of alcohol dependence and withdrawal that act through these GABA pathways. Benzodiazepines in particular have been widely used for the treatment of alcohol withdrawal symptoms. An early meta-analysis of studies concerning pharmacological therapies of alcohol withdrawal (Mayo-Smith 1997) has suggested that benzodiazepines are effective in reducing withdrawal severity, incidence of delirium and seizures with a greater margin of safety and lower abuse potential when compared to other therapies. A more recent systematic review (Holbrook 1999) of randomised controlled trials (RCTs) comparing benzodiazepines to placebo or other therapy reached similar conclusions. However a large amount of evidence of benzodiazepine use has been published during the last years and it is important that an up-to-date systematic review is performed. Moreover not all patients may need pharmacological treatment and it is unknown whether different benzodiazepines and different regimens of administration (e.g. fixed versus symptom-triggered schedule) may have the same merits. The purpose of this systematic review was to examine the evidence on the effectiveness and safety of benzodiazepines in the management of alcohol withdrawal. O B J E C T I V E S The objectives of this systematic review were: (1) to evaluate the effectiveness of benzodiazepines in the treatment of alcohol withdrawal; (2) to evaluate the safety (potential harms) of benzodiazepines in the treatment of alcohol withdrawal. M E T H O D S Criteria for considering studies for this review Types of studies RCTs examining the effectiveness, safety and overall risk-benefit of a benzodiazepine in comparison to a placebo or other pharmacological intervention or another benzodiazepine in the treatment of alcohol withdrawal were considered. Non-randomized trials were excluded from this systematic review and meta-analysis. Types of participants We considered all patients with alcohol dependence diagnosed with appropriate standardized criteria (e.g. Diagnostic and Statistical Manual of Mental Disorders [DSM] criteria or equivalent) who experienced alcohol withdrawal regardless of the severity of the clinical manifestations (e.g. regardless of Clinical Institute Withdrawal Assessment for Alcohol [CIWA-Ar] score before treatment). All patients were included regardless of age, gender, nationality, and inpatient or outpatient setting. The history of previous treatments was considered, but it was not an eligibility criterion. Types of interventions All RCTs were considered that compare: (1) benzodiazepine versus placebo; (2) benzodiazepine versus other drug; (3) different benzodiazepines among themselves; (4) benzodiazepine combined with other drug versus that other drug; (5) fixed versus symptom-triggered schedule of a benzodiazepine. Types of outcome measures Information was collected on standardized forms for each treatment arm on a variety of outcomes. These included: Primary outcomes Severity of overall alcohol withdrawal syndrome as measured in prespecified scales (including CIWA-Ar 3

score, total severity assessment, the doctor s or patient s global assessment, therapeutic success (relief from alcohol withdrawal symptoms, and the doctor s or patient s global assessment of efficacy). For comparisons where there were non-overlapping studies that examined therapeutic success, global improvement (by doctor) or global improvement (by patient), we also examined the outcome of any success at the end of treatment using data on global improvement (by doctor) when therapeutic success data were not available and data on global improvement (by patient) when neither therapeutic success nor global improvement (by doctor) were available. This outcome allows the combination of information from more studies, using data on outcomes that are fairly comparable Alcohol withdrawal seizures Alcohol withdrawal delirium (2) Secondary outcomes Number of patients presenting side effects (with emphasis on severe, life-threatening, and serious events) Discontinuations due to side effects Number of withdrawals per arm Mortality Combined scores for psychiatric symptoms (illusions, hallucinations, clouding of sensorium, quality of contact) Combined scores for neurologic symptoms (tremor, agitation, headache) Combined scores for autonomic symptoms (paroxysmal sweats, flushing of face, nausea, vomiting, other gastrointestinal symptoms) Combined psychological measures of successful treatment (anxiety measures, depressive symptoms) Combined social measures of successful treatment (medical problems, legal problems, social and family relations, employment related problems) Quality of life Quality of sleep Cost of treatment Search methods for identification of studies Electronic searches We used detailed search strategies to identify the studies included in this review, with no language restriction. We searched:cochrane Central Register of Controlled Trials (CENTRAL - The Cochrane Library 4, 2004), MEDLINE database (OVID January 1966 to October 2004),see Appendix 1; Appendix 2. EU-PSI PSI- Tri database was also searched for additional relevant trials (www.psitri.helsinki.fin) Manual searches In addition to searching the reference lists of retrieved studies, we handsearched reviews. We also contacted experts in the field wherever possible to find out if they knew of any other published or unpublished controlled trials. Data collection and analysis Study selection Two authors independently screened the titles and abstracts of all publications, which were obtained by the search strategy. We obtained all potentially eligible studies as full articles and two authors independently assessed these for inclusion. In doubtful or controversial cases, the authors discussed all identified discrepancies and reached consensus on all items. If consensus was still not reached, they referred to a third author to solve the problem. Experts familiar to the language managed potential language problems in retrieved studies. We accepted all randomised trials of benzodiazepines in alcohol withdrawal regardless of type of benzodiazepine treatment, daily and total benzodiazepine dose, duration of therapy and type of therapy (inpatient, outpatient). Whenever reports pertained to overlapping patients, we retained only the largest study, to avoid duplication of information. However, we retained studies that involved patients with previously published data on different outcomes. Assessment of the methodological quality In order to limit bias, gain insight into potential comparisons and guide interpretation of findings, two authors, used the criteria described in the Cochrane Reviewers Handbook Alderson 2004, to assess the methodological quality of the eligible studies. Factors that warrant assessment are those related to applicability of findings, validity of individual studies, and certain design characteristics that may affect interpretation of the results. In the context of a systematic review, the validity of a study is the extent to which its design and conduct are likely to prevent systematic errors, or bias (Moher 1995). There are four sources of systematic bias in trials of the effects of healthcare: selection bias, performance bias, attrition bias, and detection bias. All of them were described in each eligible trial. Selection bias Judging the studies on the method of allocation concealment performed as an evaluation of selection bias, authors indicated whether allocation concealment was adequate (A), unclear (B), inadequate (C), or that allocation concealment was not used (D) (Schulz 1995): A. Low risk of bias: adequate allocation concealment, i.e. central randomisation (e.g. allocation by a central office unaware of subject characteristics), pre-numbered or coded identical bottles or containers which are administered serially to participants, drug prepared by the pharmacy, serially numbered, opaque, sealed envelopes, on-site computer system combined with allocations kept in a locked unreadable computer file that can be accessed only after the characteristics of an enrolled participant have been entered or other description that contained elements 4

convincing of concealment; B. Moderate risk of bias: unclear allocation concealment, in which the authors either did not report an allocation concealment approach at all or report an approach that did not fall in the category A or C; C. High risk of bias: inadequate allocation concealment, such as alternation or reference to case numbers, dates of birth, day of the week. Any procedure that is entirely transparent before allocation, such as an open list of random numbers or other description that contained elements convincing of not concealment; D. This score refers to studies in which the investigators have not used allocation concealment, i.e. for studies which are not randomised. Performance bias When examining performance bias, the authors considered the use of blinding as a criterion for validity. This could be assessed by the following questions: Were the recipients of care unaware of their assigned intervention? Were those providing care unaware of the assigned intervention? Attrition bias Attrition bias refers to systematic differences between the comparison groups in the loss of participants from the study. Due to inadequacies in reporting how losses of participants (e.g. withdrawals, dropouts, protocol deviations) were handled, the authors were cautious about implicit accounts of follow-up. What is reported in study reports on attrition after allocation has not been found to be consistently related to bias (Schulz 1995). Thus, authors did not use the reported follow-up as a validity criterion. Detection bias Detection bias refers to systematic differences between the comparison groups in outcome assessment. Authors considered specification of predefined primary outcomes and analyses by the investigators as indicators of validity. Selective reporting of particular outcomes was taken as evidence for potential bias. Overall quality When assessing overall quality, the authors used simple approaches that include individual criteria, such as adequacy of allocation concealment, randomisation method, blinding to the intervention allocation, and selective reporting of results. Authors avoided scales with multiple items and complex scoring system because there are not supported by empirical evidence ( Schulz 1995), they do not provide more reliable assessments of quality (Juni 1999), and may carry a greater risk of confusing the quality of reporting with the validity of the study. Data extraction Two authors independently extracted data on characteristics of studies and patients, measurements, and results (outcomes); any discrepancy was discussed until consensus was reached. We also recorded in more detail, the names of the authors, journal and year of publication, country of origin, years of patient enrolment, number of patients randomised, number of patients analysed, reasons for differences between randomised and analysed participants, and number of arms and type of intervention(s) per arm, demographic characteristics in each arm, history of withdrawal symptoms, previous detoxification or withdrawal treatment, severity of alcohol withdrawal symptoms before therapy, time of pre-therapy evaluation, type of benzodiazepine(s), dose (daily and total) of benzodiazepine(s), number of sessions, duration and class (short-acting or long-acting) of benzodiazepine therapy, time, dose and duration of other therapies, and any supplementary medications used. In addition we also recorded information on the safety (potential harms) of the compared interventions with emphasis on discontinuations due to toxicity and benzodiazepine dependence as well as all severe, life-threatening, and serious adverse events. Finally, we determined whether each trial specified the mode of randomisation, blinding, allocation concealment and withdrawals per arm as well as who assessed outcomes in each trial. Data synthesis We performed meta-analysis when appropriate for each of the pre-specified outcomes. Both fixed and random effects models were used and primary results are given for random effects calculations. Continuous measures were summarized using weighted mean differences and binary outcomes were summarized using relative risks. Between-study heterogeneity was estimated with the chi square based Q statistic (considered statistically significant at p < 0.10) and was also expressed by the I-square estimates (considered significant at > 75%). Also, we tried to separate symptomtriggered from fixed schedule regimens. Calculations were performed using RevMan 4.2. For continuous measures with missing estimates of standard deviation per arm, we imputed the median standard deviation from the other available studies. For continuous measures evaluated during follow-up, we compared changes from baseline values. Meta-analysis was performed only for studies that had directly comparable interventions and used the same dichotomous outcomes or exactly the same rating scales for continuous outcome measures. For outcomes and comparisons where data were available from only one study, information could not be synthesized. However, this potentially useful information is shown in the Analyses. Following results are derived from quantitative synthesis and are divided into five main comparisons: Benzodiazepine versus Placebo, Benzodiazepine versus Other Drug, Different Benzodiazepines among themselves, Benzodiazepine + Other Drug versus Other Drug, and Benzodiazepine (fixed schedule) versus Benzodiazepine (symptom-triggered). R E S U L T S Description of studies See: Characteristics of included studies; Characteristics of excluded studies. 5

The electronic search strategies resulted in 660 records, which were screened by reading the titles and abstracts. Seventy-seven reports were considered appropriate for full-text scrutiny. Excluded studies We excluded 16 potentially eligible trial reports that were ineligible upon full-text scrutiny (Benuzzi 1967; Bowman 1966; Gallant 1969; Gerra 1991; Huber 1990; Lazarova 2003; Malmgren 1967; Mery 1979; Muller 1969; Myrick 2000; Rothstein 1973; Sampliner 1974; Sereny 1965; Shaffer 1968; Silpakit 1999; Thompson 1975), four duplicate or preliminary presentations ( Baumgartner 1988; Gillman 2002; Klett 1971; Malcolm 2000), and one that could not be retrieved in full text and its abstract did not provide relevant data (McGrath 1975). See Characteristics of excluded studies Included studies Fifty-six reports were considered eligible. One eligible report included two separate trials (Burroughs 1985a; Burroughs 1985b) according to initial severity of alcohol withdrawal, thus a total of 57 different trials that involved 4,275 participants were considered. Two eligible studies (Malcolm 2002a; Pena-Ramos 1979) involved 224 participants with previously published data on different outcomes. Thus, a total of 4,051 different patients were included in the analysed trials. See Characteristics of included studies Number of studies per type of comparison (1) Benzodiazepine versus placebo (No. = 8 studies) (2) Benzodiazepine versus other drug (No. = 34 studies) (3) Different benzodiazepines among themselves (No. = 13 studies) (4) Benzodiazepine combined with other drug versus that other drug (No. = 3 studies) (5) Fixed versus symptom-triggered schedule of a benzodiazepine (No. = 3 studies) Risk of bias in included studies All included studies were randomised, controlled trials. The randomisation method was described in 21 studies; randomisation was performed centrally in only seven trials (Burroughs 1985a; Burroughs 1985b; Daeppen 2002; Jauhar 2000; Malcolm 2002a; Malcolm 2002b; Radouco-Thomas 1989). The efforts for allocation concealment were not clarified in the majority of the studies. According to the published reports it did not appear that allocation was adequately concealed in 11 trials (Baumgartner 1991; Burroughs 1985a; Burroughs 1985b; Gillman 2004; Gillmer 1973; Golbert 1967; Lepola 1984; Lucht 2003; Overall 1973; Radouco-Thomas 1989; Runion 1978), while allocation was adequately concealed in only nine trials (Brown 1972; Daeppen 2002; Jauhar 2000; Kaim 1969; Kaim 1972; Malcolm 2002a; Malcolm 2002b; Spies 2003; Tubridy 1988). All the trials were described as double-blind except for one single-blind trial (Addolorato 1999) and two open-label trials (Brown 1972; Lucht 2003) and six studies that did not comment adequately on masking (Funderburk 1978; Golbert 1967; Lenzenhuber 1999; Longo 2002; Overall 1973; Spies 2003). Selective reporting of particular outcomes was more than common in the included studies. Effects of interventions Benzodiazepine versus Placebo Therapeutic success : Three trials (graph 01-01-01), including a total of 112 randomised participants, addressed this outcome. Therapeutic success tended to be more common in the benzodiazepine-group (RR 1.40; 95% CI 0.87 to 2.27; p = 0.17; p = 0.03 for between-study heterogeneity). Alcohol withdrawal seizures: Three trials (graph 01-02-01), including a total of 324 randomised participants, addressed this outcome. Benzodiazepines offered a significant benefit against seizures (RR 0.16; 95% CI 0.04 to 0.69; p = 0.01; p = 0.88 for betweenstudy heterogeneity). Number of withdrawals per arm: Three trials (graph 01-03-01), including a total of 314 randomised participants, addressed this outcome. Number of withdrawals per arm tended to be less common among benzodiazepine-treated participants (RR 0.69; 95% CI 0.38 to 1.24; p = 0.22; p = 0.50 for between-study heterogeneity). Mortality : Of 8 trials (graph 01-04-01) including a total of 480 randomised participants, none died. Discontinuation due to side effects: Two trials (graph 01-06-01) addressed this outcome. No discontinuation due to side effects was seen among 53 benzodiazepine-treated participants, while one of 33 participants receiving placebo discontinued the treatment due to side effects. Benzodiazepine versus Other Drug CIWA-Ar score (change from baseline): (i) at 48 hours. Six trials (graphs 02-01-01 and 02-01-02), including a total of 403 randomised participants, addressed this outcome. Benzodiazepines were not more effective in reducing CIWA-Ar score at 48 hours compared to either other drugs (WMD -0.76; 95% CI -1.89 to 0.36; p = 0.18; p = 0.67 for between-study heterogeneity) or anticonvulsants (WMD -0.73; 95% CI -2.88 to 1.42; p = 0.51; p = 0.68 for between-study heterogeneity). (ii) at the end of treatment. Six trials (graphs 02-02-01 and 02-02-02), including a total of 403 randomised participants, addressed this outcome. No difference between benzodiazepines and other drugs was seen for CIWA-Ar score reduction at the end of treatment (WMD -0.05; 95% CI - 1.18 to 1.08; p = 0.93; p = 0.72 for between-study heterogeneity), while benzodiazepines tended to be less effective when compared to anticonvulsants (WMD -1.04; 95% CI -3.45 to 1.38; p = 0.40; p = 0.87 for between-study heterogeneity). Doctor s global assessment (change from baseline) at the end of treatment: Data from three trials (graph 02-03-01), including a total of 122 randomised participants, on this outcome showed that benzodiazepines were borderline less effective compared to other drugs (WMD -2.14; 95% CI -4.29 to 0.01; p = 0.05; p = 0.009 for between-study heterogeneity), but data were sparse. 6

Patient s global assessment (change from baseline) at the end of treatment: Data from three trials (graphs 02-05-01 and 02-05-02), including a total of 157 randomized participants, on this outcome showed that benzodiazepines tended to be more effective compared to either other drugs (WMD 4.95; 95% CI -10.60 to 20.51; p = 0.53) or anticonvulsants (WMD 10.76; 95% CI -11.29 to 32.81; p = 0.34). However, there was significant between-study heterogeneity in these analyses (p < 0.00001) and data were very sparse. Hamilton Anxiety Rating Scale (HARS) score (change from baseline): (i) at 48 hours. Two small trials (graph 02-06-01), including a total of 83 randomised participants, addressed this outcome. There was a significant inferiority of benzodiazepines compared to other drugs (WMD -1.57; 95% CI -2.54 to -0.60; p = 0.002; p = 0.74 for between-study heterogeneity). (ii) at the end of treatment. Three trials (graph 02-07-01), including a total of 130 randomised patients, addressed this outcome. Benzodiazepines were significantly less effective compared to other drugs (WMD -2.44; 95% CI -3.91 to -0.97; p = 0.001; p = 0.30 for between-study heterogeneity), but data were very sparse. Therapeutic success: Eight trials (graphs 02-09-01 and 02-09- 02), including a total of 384 randomised patients, addressed this outcome. There was absolutely no difference in therapeutic success between benzodiazepines and either other drugs (RR 1.00; 95% CI 0.83 to 1.21; p = 0.99; p = 0.004 for between-study heterogeneity) or anticonvulsants (RR 0.88; 95% CI 0.60 to 1.30; p = 0.51; p = 0.003 for between-study heterogeneity) while 95% confidence intervals were narrow enough to suggest any clinically meaningful effect. Global doctor s assessment of efficacy: Four trials (graphs 02-10- 01 and 02-10-02), including a total of 281 randomised patients, addressed this outcome. No difference between benzodiazepines and either other drugs (RR 1.04; 95% CI 0.98 to 1.10; p = 0.24; p = 0.69 for between-study heterogeneity) or anticonvulsants (RR 1.02; 95% CI 0.90 to 1.17; p = 0.74; p = 0.18 for between-study heterogeneity) was seen. Global patient s assessment of efficacy: Data from two trials (graph 02-11-01), including a total of 140 randomised patients, showed no difference between benzodiazepines and other drugs (RR 1.05; 95% CI 0.98 to 1.11; p = 0.16; p = 0.73 for betweenstudy heterogeneity). Any success: Twelve trials (graphs 02-12-01, 02-12-02, 02-12-03 and 02-12-04), including a total of 667 randomised patients, addressed this outcome. There was no difference in success between benzodiazepines and either other drugs (RR 1.02; 95% CI 0.92 to 1.12; p = 0.75; p = 0.02 for between-study heterogeneity) or anticonvulsants (RR 1.00; 95% CI 0.87 to 1.16; p = 0.99; p = 0.007 for between-study heterogeneity). No difference between long-acting benzodiazepines (RR 1.00; 95% CI 0.86 to 1.18; p = 0.95; p = 0.007 for between-study heterogeneity) or short-acting benzodiazepines (RR 1.04; 95% CI 0.98 to 1.11; p = 0.21; p = 0.70 for between-study heterogeneity) and other drugs was seen. Alcohol withdrawal seizures: Eleven trials (graphs 02-13-01, 02-13-02, 02-13-03, 02-13-04 and 02-13-05), including a total of 1,139 randomised participants, addressed this outcome. There was a non-significant benefit against seizures of benzodiazepines compared to other drugs (RR 0.54; 95% CI 0.21 to 1.42; p = 0.21; p = 0.46 for between-study heterogeneity), but this difference was in the opposite direction when benzodiazepines were compared to anticonvulsants (RR 1.99; 95% CI 0.46 to 8.65; p = 0.36; p = 0.96 for between-study heterogeneity). Analysis of benzodiazepines versus non-anticonvulsants showed a significant benefit (RR 0.23; 95% CI 0.07 to 0.75; p = 0.02; p = 0.70 for betweenstudy heterogeneity). Long-acting benzodiazepines offered a nonsignificant benefit against seizures (RR 0.41; 95% CI 0.13 to1.28; p = 0.12; p = 0.43 for between-study heterogeneity), but this was not seen for the comparison of short-acting benzodiazepines vs. other drugs (RR 1.09; 95% CI 0.18 to 6.46; p = 0.92; p = 0.36 for between-study heterogeneity). The confidence intervals for both subgroup estimates were nevertheless very large and overlapping. Alcohol withdrawal delirium: Seven trials (graphs 02-14-01, 02-14-02, 02-14-03, 02-14-04 and 02-14-05), including a total of 752 randomised participants, addressed this outcome. There was a non-significant benefit against delirium of benzodiazepines versus other drugs (RR 0.48; 95% CI 0.14, 1.61; p = 0.23; p = 0.04 for between-study heterogeneity), but this was in the opposite direction for the comparison of benzodiazepines vs. anticonvulsants (RR 1.37; 95% 0.25, 7.37; p = 0.71; p = 0.13 for between-study heterogeneity). Analysis of benzodiazepines versus non-anticonvulsants showed a non-significant benefit (RR 0.32; 95% CI 0.06, 1.76; p = 0.19; p = 0.03 for between-study heterogeneity). Longacting benzodiazepines offered a non-significant benefit against delirium (RR 0.35; 95% CI 0.09 to 1.36; p = 0.13; p = 0.02 for between-study heterogeneity), while data for short-acting benzodiazepines versus another drug were limited to a single small study. Number of withdrawals per arm: Nineteen trials (graphs 02-15- 01 and 02-15-02), including a total of 1,638 randomised participants, addressed this outcome. No difference in number of withdrawals per arm was seen when benzodiazepines were compared to either other drugs (RR 0.91; 95% CI 0.68, 1.23; p = 0.54; p = 0.90 for between-study heterogeneity) or anticonvulsants (RR 0.93; 95% CI 0.60 to 1.45; p = 0.75; p = 0.98 for between-study heterogeneity). Mortality: In 34 trials (graphs 02-16-01 and 02-16-02), two of 901 benzodiazepine-treated participants and five (one treated with an anticonvulsant) of 1,275 participants receiving other drug died. Side effects: Fifteen trials (graphs 02-17-01 and 02-17-02), including a total of 717 randomised participants, addressed this outcome. Benzodiazepine-treated participants had a slightly higher incidence of side-effects compared to patients receiving either other drugs (RR 1.23; 95% CI 0.92 to 1.63; p = 0.16; p = 0.69 for between-study heterogeneity) or anticonvulsants (RR 1.16; 95% CI 0.77 to 1.75; p = 0.47; p = 0.35 for between-study heterogeneity), but this didn t reach formal statistical significance. 7

Severe, life-threat ing side-effects: Eight trials (graphs 02-18-01 and 02-18-02), including a total of 386 randomised participants, addressed this outcome. Life-threating side-effects tended to be less common among benzodiazepine-treated participants compared to participants receiving other drugs (RR 0.85; 95% CI 0.19 to 3.82; p = 0.84; p = 0.38 for between-study heterogeneity), but this difference was in the opposite direction when benzodiazepines were compared to anticonvulsants (RR 1.45; 95% CI 0.19 to 11.24; p = 0.72; p = 0.65 for between-study heterogeneity). Discontinuation due to side effects: Ten trials (graphs 02-19- 01 and 02-19-02) addressed this outcome. Three of 252 benzodiazepine-treated participants discontinued the treatment due to side effects, while 11 (7 treated with an anticonvulsant) of 369 participants receiving other drug did so. Different Benzodiazepines among themselves Thirteen trials (Adinoff 1994; Brown 1972; Jauhar 2000; Kolin 1981; Martin 1975; McLendon 1980; Mendels 1985; Miller 1984; O Brien 1983; Ritson 1986; Saletu 1983; Solomon 1983; Wilson 1985), including a total of 571 randomised participants, examined eight different comparisons of benzodiazepines among themselves for various outcomes. Limited data on scattered comparisons of various drugs in small studies were available. There was no conclusive evidence or even hints for superiority of specific drugs, but modest differences could have been missed due to limited data. Benzodiazepine+Other Drug versus Other Drug Three trials (Dion 1968; Sellers 1977; Spies 1996), including a total of 207 randomised participants, compared the combination of a benzodiazepine with another drug (magnesium sulphate, propranolol, haloperidol) versus this other drug for various outcomes. Very limited data were available, thus making quantitative synthesis not applicable. Benzodiazepine (fixed schedule) versus Benzodiazepine (symptom-triggered) Three trials (Daeppen 2002; Saitz 1994; Spies 2003), including a total of 262 randomised participants, compared fixed versus symptom-triggered schedules of a benzodiazepine (chlordiazepoxide, oxazepam, flunitrazepam) for various outcomes. There was a small and non-significant benefit of symptom-triggered regimens regarding CIWA-Ar score (change from baseline) at the end of treatment (WMD -0.51; 95% CI -2.36 to 1.35; p = 0.59; p = 0.31 for between-study heterogeneity), therapeutic success (RR 0.88; 95% CI 0.52 to 1.52; p = 0.66; p = 0.002 for between-study heterogeneity) and number of withdrawals per arm (RR 1.96; 95% CI 0.50 to 7.70; p = 0.33; p = 0.63 for between-study heterogeneity). The CIWA-Ar difference was formally significant at 48 hours (WMD -5.70; 95% CI -11.02 to -0.38; p = 0.04) but was based on a single small study. Data on all other outcomes were very limited to perform any meaningful quantitative synthesis. D I S C U S S I O N This systematic review includes data from 57 RCTs with over 4,000 patients with alcohol withdrawal syndrome. Despite the considerable number of RCTs, the large variety of outcomes and rating scales limited considerably the ability to perform a quantitative synthesis of all available data. Benzodiazepines clearly offered a significant benefit against alcohol withdrawal seizures compared to placebo. This might suggest that their current status as first-line treatment for alcohol withdrawal syndrome is justified. Nevertheless, the available evidence did not suggest that benzodiazepines are clearly superior to other drugs with the exception of a possible superiority in seizure control when compared against non-anticonvulsants. Broadly defined success rates were very similar with benzodiazepines versus other drugs and the 95% confidence intervals even exclude any clinically meaningful differences in this regard. This was true for both short-acting and long-acting benzodiazepines. Differences between regimens in isolated small trials using more particular outcome measures should be interpreted very cautiously given the large number of outcomes in this field and the small sample size of the studies conducted therein. Nevertheless, even the limited data on such outcomes are not suggestive of any clear superiority of benzodiazepines. Benzodiazepines were less effective, if anything, compared to other drugs in reducing the severity of alcohol withdrawal symptoms, as indicated by change from baseline in doctor s global assessment score at the end of treatment and Hamilton Anxiety Rating Scale score at both 48 hours and the end of treatment in a few small trials. When compared to anticonvulsants, benzodiazepines tended to offer a non-significant benefit for change from baseline in the patient s global assessment score at the end of treatment, but also a non-significant increased risk for alcohol withdrawal seizures. Data on the comparisons of different benzodiazepines among themselves, benzodiazepine combined with other drug versus other drug, and fixed-schedule versus symptom-triggered regimens of benzodiazepines were very limited, thus making quantitative synthesis for various outcomes either not applicable or not very informative. Based on indirect comparisons, there is no strong evidence at the moment that particular benzodiazepines are more effective than others. The overall results need to be interpreted with caution, as bias cannot be excluded. Most trials were of very small sample size. Although randomisation was an inclusion criterion indicating some methodological quality for these studies, the method of randomisation was not usually described in sufficient detail in the published reports. Moreover, allocation concealment was usually unclear and information on follow-up was often missing. Furthermore, it was difficult to interpret comparisons of specific settings of treatment (inpatient or outpatient), because a large number of the analysed studies did not present sufficient background information so as to 8

allow a clear understanding of the treatment setting. We could not also examine dose-response effects since patients were not treated with even similar doses of various benzodiazepines across RCTs. Small trials are difficult to interpret in isolation, especially in the face of potential selective reporting of outcomes. Information on side-effects was not consistently reported in the trial reports. More detailed data on adverse effects would be important to record in these trials, since discontinuation due to sideeffects may affect the success of treatment. Moreover, data on side effects should be compared cautiously, as they were derived from patients with potentially different co-morbidity. Patients with severe medical conditions, such as hepatic, heart or lung disease, were often excluded from these trials. However, these patients may be more susceptible to various adverse effects of benzodiazepines. The extremely small mortality rate in all these studies is reassuring, but data on other harms-related outcomes are sparse and fragmented. This important deficiency of the benzodiazepine literature needs to be highlighted for improving future clinical research efforts in this field. Overall, the small sample size of published RCTs and lack of detailed information on various outcomes illustrate a need for larger, well-designed studies in this field. These studies should be limited to more important efficacy variables and consistency on rating continuous outcomes with the same scales should also be achieved among researchers. A U T H O R S C O N C L U S I O N S Implications for practice Benzodiazepines are effective against alcohol withdrawal symptoms, in particular seizures, when compared to placebo. It is not possible to draw very precise conclusions about the relative effectiveness and safety of benzodiazepines against other drugs in alcohol withdrawal, because of the large heterogeneity of the trials both in interventions and assessment of outcomes. Nevertheless, the available data do not show differences between benzodiazepines and other drugs in broadly defined success rates. Data on potential harms are sparse and fragmented. Implications for research Although a significant number of trends has emerged, most of these were small and the data for most outcomes did not reach statistical significance, indicating the need for larger, well-designed studies in this field. These studies should be limited to few, important efficacy variables such as severity of the alcohol withdrawal syndrome, incidence of seizures and delirium tremens, side effects and mortality. Consistency on rating continuous outcomes in the same scales should also be achieved in order to obtain comparable information from all relevant studies. A C K N O W L E D G E M E N T S We thank Dr. Despina Contopoulos-Ioannidis for providing helpful assistance with data extraction and Dr. Athina Tatsioni for article retrieval. R E F E R E N C E S References to studies included in this review Addolorato 1999 {published data only} Addolorato G, Balducci G, Capristo E, Attilia ML, Taggi F, Gasbarrini G, et al.gamma-hydroxybutyric acid (GHB) in the treatment of alcohol withdrawal syndrome: a randomised comparative study versus benzodiazepine. Alcoholism, Clinical and Experimental Research 1999;23(10):1596 604. Adinoff 1994 {published data only} Adinoff B. Double-blind study of alprazolam, diazepam, clonidine, and placebo in the alcohol withdrawal syndrome: preliminary findings. Alcoholism, Clinical and Experimental Research 1994;18(4): 873 8. Ansoms 1991 {published data only} Ansoms S, Lebon O, Pelc I, Cabri C, Poels R. Zopiclone or lormetazepam in the treatment of insomnia and the effect on behavior and mood in patients during the post alcoholism withdrawal period. Current Therapeutic Research, Clinical & Experimental 1991; 49(1):54 64. Anton 1997 {published data only} Anton RF, Kranzler HR, McEvoy JP, Moak DH, Bianca R. A doubleblind comparison of abecarnil and diazepam in the treatment of uncomplicated alcohol withdrawal. Psychopharmacology (Berl) 1997; 131(2):123 9. Bailly 1992 {published data only} Bailly D, Servant D, Blandin N, Beuscart R, Parquet PJ. Effects of beta-blocking drugs in alcohol withdrawal: a double-blind comparative study with propranolol and diazepam. Biomedicine and Pharmacotherapy 1992;46(9):419 24. Baumgartner 1987 {published data only} Baumgartner GR, Rowen RC. Clonidine vs chlordiazepoxide in the management of acute alcohol withdrawal syndrome. Archives of Internal Medicine 1987;147(7):1223 6. Baumgartner 1991 {published data only} Baumgartner GR, Rowen RC. Transdermal clonidine versus chlordiazepoxide in alcohol withdrawal: a randomised, controlled clinical trial. South African Medical Journal 1991;84(3):312 21. 9