Cochrane Column
South African Cochrane Centre, Medical Research Council, PO Box 19070, Tygerberg 7505, South Africa. Email: taryn.young{at}mrc.ac.za
This month we highlight a review which assessed the effects of brief interventions for alcohol misuse and, as the review includes a meta-regression, we include a commentary on the use of meta-regression.
The aim of the Column is to highlight Cochrane Reviews of relevance to public health, and to stimulate debate on relevance, feasibility and acceptability. The Cochrane Collaboration (http://www.cochrane.org) is an international, non-profit organization that prepares and disseminates up-to-date systematic reviews on the effects of healthcare interventions in order to help people make well-informed decisions. Systematic reviews aim to answer focused healthcare questions by systematically identifying and evaluating all relevant research studies and synthesizing their results. If you are interested in contributing to the Cochrane Column or The Cochrane Collaboration, contact me at the South African Cochrane Centre.
Brief interventions in primary care for alcohol misuse: Cochrane review
 |
Background |
|---|
 Top Background MethodsResults and discussionConclusionsReferencesReferences References |
|---|
Excessive drinking is a significant cause of mortality, morbidity and social problem in both developed and developing countries with the global cost to health being above that of tobacco.1 Early identification and the secondary prevention of alcohol problems with the use of screening and brief intervention (SBI) in primary care, has increasingly been advocated as the way forward. The large body of research on this topic 2,3 has been criticized for being clinically unrepresentative and unable to inform routine practice.5,6
The aim of this review was to assess the effectiveness of SBI at reducing excessive drinking in primary care and to investigate whether SBI outcomes differ between efficacy (ideal world) and effectiveness (real world) trials.
|
Methods |
|---|
|
TopBackgroundMethodsResults and discussionConclusionsReferencesReferences References |
|---|
We searched eight databases up to 2006 (Cochrane Register of Controlled Trials, MEDLINE, EMBASE, CINAHL, PsycINFO, SCI, SSCI, ETOH) and the reference lists of key articles. The selection criteria were: (i) randomized controlled trials (RCTs); (ii) patients presenting to primary care not specifically for alcohol treatment; and (iii) brief intervention of one to four sessions. Two authors independently abstracted data, assessed trial quality and scored trials on eight criteria relating to an efficacy/effectiveness dimension (total score range 0–12). Random effects meta-analyses, sub-group and sensitivity analyses and meta-regression were conducted.
|
Results and discussion |
|---|
|
TopBackgroundMethodsResults and discussionConclusionsReferencesReferences References |
|---|
The primary meta-analysis included 21 RCTs involving 7286 patients. Patients receiving SBI significantly reduced alcohol consumption compared with controls (mean difference –41 g/week, 95% CI: –57 to –25) although there was substantial heterogeneity between trials (I2 = 52%). Sub-group analysis (eight studies, 2307 patients) confirmed the benefit of SBI in men (mean difference –57 g/week, 95% CI: –89 to –25, I2 = 56%), but not in women (mean difference: –10 g/week, 95% CI: –48 to 29, I2 = 45%). Meta-regression showed a non-significant trend of increased reduction in consumption of 1.1 g (95%CI: –0.05 to 2.2 g/week, P = 0.06) for each extra minute of treatment exposure. Trials fell onto a continuum in efficacy/effectiveness scores which ranged from 4.5 to 12 (median 9, interquartile range 7.5–10.5); though they were heavily skewed towards the effectiveness dimension. When trials were dichotomized (median cut-off point), there was no significant difference in effect-sizes between the efficacy and the effectiveness trials. The reduction in drinking was similar in the normal clinical setting as in a research setting with greater resources.
|
Conclusions |
|---|
|
TopBackgroundMethodsResults and discussionConclusionsReferencesReferences References |
|---|
Brief interventions results in significant reductions in weekly consumption for men, with an average drop of about six standard drinks per week in patients compared with controls. When data were available by sex, the effect was clear in men but unproven in women. This may be partly due to low statistical power (as trials reporting outcomes from women only enrolled 499 participants), and thus brief interventions for women are not yet justified. Longer duration of counseling had little additional effect. The lack of differences in outcomes between efficacy and effectiveness trials suggests that the current literature has clear relevance to ‘real world’ practice in primary care. Future research should focus on women and on delineating the most effective components of SBIs. There is some suggestion that screening alone may result in alcohol consumption reduction, and this should be investigated further. Finally, future research direction should focus on implementation issues including a more precise specification of brief intervention components.
The full text of the Cochrane Review is available in The Cochrane Library. Kaner EFS, Beyer F, Dickinson HO, Pienaar E, Campbell F, Schlesinger C, Heather N, Saunders J, Burnand B. Effectiveness of brief alcohol interventions in primary care populations. Cochrane Database of Systematic Reviews 2007, Issue 2. Art. No.: CD004148. DOI: 10.1002/14651858.CD004148.pub3
This version first published online: 18 April 2007 in Issue 2, 2007.
|
References |
|---|
|
TopBackgroundMethodsResults and discussionConclusionsReferencesReferences References |
|---|
1 World Health Organization. Global Status Report on Alcohol (1999) Geneva: World Health Organization.
2 Anderson P. Alcohol and primary health care (1996) No. 64. Copenhagen: WHO Regional Publications.
3 Wutzke S, Congrave K, Saunders J, Hall W. The long-term effectiveness of brief interventions for unsafe alcohol consumption: a 10-year follow-up. Addiction (2002) 97:665–75.[CrossRef][Web of Science][Medline]
4 Holder H, Flay B, Howard J, Boyd G, Voas R, Grossman M. Phases of alcohol problem prevention research. Alcoholic, Clinical and Experimental Research (1999) 23:183–94.
5 Kaner EF, Heather N, Brodie J, Lock CA, McAvoy BR. Patient and practitioner characteristics predict brief alcohol intervention in primary care. Br J Gen Pract (2001) 51:822–27.[Web of Science][Medline]
Commentary: Need to role out brief interventions in primary care settings to address problem drinking while expanding research in developing and transitional countries
Director: Alcohol & Drug Abuse Research Unit, Medical Research Council, PO Box 19070, Tygerberg (Cape Town), 7505, South Africa and Professor (Extraordinary), Department of Psychiatry, Stellenbosch University
There has been renewed recognition of the burden of alcohol following studies which estimated alcohol's burden at 9.2% of disability adjusted life years lost in developed sub regions and somewhat less in other sub regions (1.6–6.2%).1 This has been accompanied by calls from various parties, including the World Health Organization,2 for action to reduce this burden. Given competing priorities there is especially a need for cost-effective interventions that can reach large numbers of at risk drinkers.
Kaner et al.'s review confirms the contention of Babor et al.3 that brief interventions in primary care settings is one of the more effective interventions for addressing problem drinking, especially among men. Across the subgroup of studies among male participants, the mean difference in weekly drinking between participants in the intervention arm compared with controls was just under six standard drinks. However, the difference for females (one standard drink) was not significant.
It is disappointing that the review did not identify any RCTs in developing or transitional countries. However, given the number of trials of brief alcohol intervention showing a positive impact in men in developed countries and the growing burden of alcohol-related problems in developing and transitional countries,4 stepping up brief interventions in such countries cannot be delayed. Binge drinking is often a predominant mode of alcohol consumption in such countries and it was encouraging to note that Kaner et al.'s review reported on three trials that showed a significant reduction in the percentage of binge drinkers in the brief intervention arms compared with controls.
The review has also uncovered a number of gaps in the research which need to be addressed as a matter of urgency. These include the need for more RCTs in this area in developing and transitional countries. In such countries it is less important to focus on weekly drinking amounts given the high levels of weekend binge drinking that have been reported in some countries,5 but rather to look for effects in the area of drinking intensity per drinking day, something that was, however, not found to be significantly reduced in the four trials reviewed. There is also a need for more research among women and younger populations, something acknowledged in the review. Also needed is more research looking at the effect of interventions on total health care costs given the importance of showing such effects in influencing policy. Only one study6 was identified as having looked at this.
|
References |
|---|
|
TopBackgroundMethodsResults and discussionConclusionsReferencesReferences References |
|---|
1 Rehm J, Room R, Monteiro M, et al. Alcohol use. In: Comparative Quantification of Health Risks: Global and Regional Burden of Disease Attributable to Selected Risk Factors, Volume I—Ezatti M, Lopez AD, Rodgers A, Murray CJL, eds. (2004) Geneva: World Health Organization. 959–1108.
2 World Health Organization. Public health problems cause by harmful use of alcohol (2005) Geneva: World Health Organization.
3 Babor T, Caetano R, Casswell S, et al. Alcohol: No Ordinary Commodity. Research and Public Policy (2003) Oxford: Oxford University Press.
4 Room R, Carlini-Cotrim B, Gureje O, et al. Alcohol and the Developing World: A Public Health Perspective (2002) Helsinki: Finnish Foundation of Alcohol Studies in Collaboration with the World Heath Organization.
5 Parry CDH, Plüddemann A, Steyn K, Bradshaw D, Norman R, Laubscher R. Alcohol use in South Africa: findings from the first demographic and health survey (1998). J Stud Alcohol (2005) 66:91–97.[Web of Science][Medline]
6 Lock CA, Kaner E, Heather N, et al. Effectiveness of nurse-led brief alcohol intervention: a cluster randomized controlled trial. J Adv Nurs (2006) 54:426–39.[CrossRef][Web of Science][Medline]
Commentary: Use of meta-regression in the review on brief alcohol interventions in primary care populations
Institute for Maritime Technology, PO Box 181, Simon's Town 7995, South Africa
Meta-analysis traditionally combines results of different randomized controlled trials into a single estimate of treatment effect. Nevertheless, possible causes of heterogeneity (between trial variation1) in a meta-analysis should be explored.2 Common approaches include subgroup analysis, a random effects model to incorporate within- and between-trial variability, meta-regression and ideally2 individual patient data analysis.
Meta-regression is an extension to meta-analysis.3 Instead of estimating a single treatment effect, meta-regression explores the relationship between the treatment effect and a trial-level covariate4 (as opposed to individual patient-level outcomes and covariates in regression). Even if an overall test for heterogeneity is non-significant, meta-regression is appropriate to explore possible sources of heterogeneity.4 Through meta-regression we can investigate: (i) the influence of one key characteristic on size of treatment effect, (ii) important clinical differences between trials (to supplement MA) or (iii) causes of variation in results of different trials.4 Thompson and Higgens4 discussed statistical aspects, limitations, pitfalls and interpretation of meta-regression and is referenced throughout.
In this review, meta-regression (with continuous covariates) was used to confirm relationships (observed in forest plots) within/between subgroups. Ordering of trials by their efficacy/effectiveness score (or treatment exposure) in forest plots (as intended) would be useful, but ordering seem to have defaulted to alphabetic order. Meta-regression is a generalisation of subgroup analyses;3 and with a categorical trial-level covariate, it is equivalent to subgroup analysis.4 To view subgroup analysis formally as meta-regression (with a categorical trial-level covariate) is beneficial by allowing for residual heterogeneity not explained by subgroup analysis and by focusing on between subgroup differences as opposed to separate effects in each subgroup.4 To this purpose, other variables e.g. cluster/individual randomisation could also have been conducted as meta-regression instead of subgroup analysis.
Using meta-regression, the effect of brief intervention versus control had a non-significant increased reductions of 1.4 g/week (95%CI: –6.0 to 8.9 g/week, P = 0.71) in alcohol consumption associated with one-point increase in the efficacy/effectiveness score. The authors acknowledged that this lack of evidence for a trend may indicate insensitivity in their classification tool. They did not conduct post hoc analysis of individual factors in the efficacy score, reasoning this was not specified in the protocol. Although this is laudable, meta-regression using individual efficacy/effectiveness factors could be more meaningful than a combined score.
Meta-regression showed an increased reduction of 1.1 g/week (95%CI: –0.05 to 2.2 g/week, P = 0.06) in alcohol consumption associated with each increase of 1 min in the treatment exposure. When the three trials with extended interventions were included, a less marked effect of 0.3 g/week (95%CI: –0.2 to 0.8 g/week, P = 0.24) was found. The authors reported that, for trials classified as high treatment exposure, meta-regression showed little evidence for a trend of a greater reduction in alcohol consumption in trials with greater treatment exposure.
The relatively large range of the efficacy/effectiveness scores and treatment exposure across trials makes meta-regression suitable for investigating these covariates. Nevertheless, ecological confounding with regard to treatment exposure cannot be investigated without IPD.4
It is not reported whether fixed or random-effects meta-regression was conducted and if and how the analysis was weighted. As often, the linearity of the regressions is assumed, without comment.4 Diagrams would be more helpful if symbol size related to the precision of each treatment effect estimate.4 Multiple regressions were not attempted, though the number of trials may be too small to warrant multiple regression.
‘Data dredging’, thus over-fitting the available data, is the ‘principal pitfall’ of meta-regression.4 Only two covariates were used in meta-regression (specified in the protocol, although for subgroup analyses) and the review included a generally ‘large’ number of trials in meta-regression (n = 21–25). Thus, data dredging seems unlikely.
The authors concluded that the lack of significant difference in outcomes on the efficacy/effectiveness dimension suggests that current literature had clear relevance to routine primary care; and that longer treatment appeared to have little effect in significantly improving outcomes. As always, it is tempting to accept lack of evidence for an effect as evidence of no effect!4 It is important to remember that meta-regression is limited—it describes observational association across trials and is vulnerable to confounding.4 When a relationship between the treatment effect and a covariate is statistically non-significant it should not be interpreted as absence of true relationship.
|
References |
|---|
|
TopBackgroundMethodsResults and discussionConclusionsReferencesReferences References |
|---|
1 Thompson SG, Sharp SJ. Explaining heterogeneity in meta-analysis: a comparison of methods. Stat Med (1999) 18:2693–708.[CrossRef][Web of Science][Medline]
2 Sharp SJ, Thompson SG. Analysing the relationship between treatment effect and underlying risk in meta-analysis: comparison and development of approaches. Stat Med (2000) 19:3251–74.[CrossRef][Web of Science][Medline]
3 Higgens JP, Thompson AG, Deeks JJ, Altman DG. Statistical heterogeneity in systematic reviews of clinical trials: a critical appraisal of guidelines and practice. J Health Serv Res Policy (2002) 7:51–61.
4 Thompson SG, Higgins JP. How should meta-regression be undertaken and interpreted? Stat Med (2002) 15:1559–73.
CiteULike 
Connotea 
Del.icio.us What's this?
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||