OUP user menu


Tho Dinh-Zarr, Carolyn Diguiseppi, Elizabeth Heitman, Ian Roberts
DOI: http://dx.doi.org/10.1093/alcalc/34.4.609 609-621 First published online: 1 July 1999


To assess the effect of treatment of problem drinking on injury risk, we conducted a systematic review of randomized controlled trials by searching 12 computerized databases, cross-checking bibliographies, and contacting authors and governmental agencies. We identified 19 trials of interventions for problem drinking that measured injury outcomes. Treatment for problem drinking was associated with reduced suicide attempts, domestic violence, falls, drinking-related injuries, and injury hospitalizations and deaths, with reductions ranging from 27 to 65%. Interventions among convicted drunk drivers reduced motor vehicle crashes and injuries. The precision of all the point estimates was low, however. We did not combine the results quantitatively, because the interventions, patient populations, and outcomes were so heterogeneous. The results suggest that treatment for problem drinking may reduce injuries and their antecedents. Because injuries account for much of the morbidity and mortality from problem drinking, further studies are warranted to confirm these effects.


Alcohol consumption has been linked with injuries incurred through motor vehicle crashes, falls, drowning, fires and burns, and violence [National Committee for Injury Prevention and Control (NCIPC), 1989; US Preventive Services Task Force (USPSTF), 1996]. Compared to the general population, alcoholics have a 16 times greater risk of dying by falling and a 10 times greater risk of dying by fire or burns (National Committee for Injury and Prevention Control, 1989). A strong association has been found between alcoholism and domestic violence (O'Farrell and Murphy, 1995). Even moderate alcohol consumption has been associated with increases in deaths from trauma (Andreasson et al., 1988). Problem drinkers who do not meet definitions for alcohol dependence are responsible for the majority of alcohol-related morbidity and mortality in the general population (Institute of Medicine, 1990). In the US alone, half of the estimated 100 000 deaths attributed to alcohol each year are due to intentional and unintentional injuries (Stinson and DeBakey, 1992). Based on estimates of global injury mortality and its contributors (Murray and Lopez, 1997a,b), alcohol-related injuries worldwide may cause several million deaths each year.

Numerous randomized controlled trials have evaluated a diverse range of interventions to reduce alcohol dependence, misuse or consumption, e.g. pharmacotherapy, individual, couple, and group counselling, exercise, acupuncture, controlled drinking, brief educational interventions (alcohol intake assessment and provision of information and advice), and other in- and out-patient therapies and combinations of treatments. Most such trials have measured effects of treatment on alcohol consumption and maintenance of abstinence. Many trials have also evaluated the effects of treatment on a wide variety of negative consequences linked directly or indirectly to drinking (e.g. hospitalizations, social or occupational maladjustment) (Babor et al., 1994). Because of the increased risk of injuries associated with problem drinking, we undertook a systematic review to evaluate the effectiveness of interventions for problem drinking in preventing injuries.



Alcohol dependence, (i.e. ‘alcoholism’, ‘alcohol addiction’) involves impaired control over drinking, manifested by physiological addiction to alcohol and/or serious disturbances of health, work, social or recreational activities, or other areas of functioning related to alcohol use (American Psychiatric Association, 1994). Alcohol abuse (i.e. ‘harmful drinking’) involves serious disturbances of health, work, or other areas of functioning related to alcohol use, without satisfying the criteria for alcohol dependence (American Psychiatric Association, 1994). Hazardous use of alcohol, such as binge or chronic heavy drinking, places asymptomatic drinkers at risk for future health and other problems (US Preventive Services Task Force 1996). For the purposes of this review, we refer to alcohol dependence, alcohol abuse, and hazardous use of alcohol as ‘problem drinking’.

Inclusion criteria

Studies were included if: (1) the study population comprised people with alcohol dependence, alcohol abuse, or other problem drinking; (2) subjects were randomly assigned to experimental and control groups; (3) the intervention was designed to reduce or eliminate alcohol consumption, or to prevent injuries or their antecedents (e.g. falls, motor vehicle crashes, suicide attempts, aggressive/ violent behaviour); (4) outcome measures included injuries or their antecedents.

Data sources

Eligible trials were identified by searching relevant computerized medical databases (see below), reviewing reference lists of relevant trials, contacting national and international agencies for information about unpublished studies, and asking authors of relevant trials to identify additional published or unpublished trials.

Twelve electronic databases were searched: MEDLINE (1966–August 1996), EMBASE (1982– January 1997), the Cochrane Controlled Trials Register (The Cochrane Library 1997, issue No. 1), PSYCHINFO (1967–January 1997), the Cumulative Index to Nursing and Allied Health (CINAHL) (1982–October 1996), the Educational Resource Information Center (ERIC) (1966–December 1996), Dissertation Abstracts International (1861 –November 1996), International Road Research Documentation (IRRD) (1972–January 1997), TRANSDOC (a publication of the European Conference of Ministers of Transport) (1972–January 1997), Transportation Research Information Services (TRIS) (1968–January 1997), the International Bibliography of the Social Sciences (IBSS) (1961–January 1997), and the Index of Scientific and Technical Proceedings (ISTP) (1982–January 1997).

MEDLINE was searched by combining the Cochrane Collaboration's optimally sensitive search strategy for controlled trials (Dickersin et al., 1994), with a strategy developed to identify studies of interventions for problem drinking. Search terms included drink* or alcohol* near excessive, binge, heavy, hazard*, problem* or abuse; drink* or drunk* or influence near driv*; (accidents-traffic or automobile-driving) and alcohol*; alcoholi*; and the mesh headings (explode) ALCOHOLIC-INTOXICATION, (explode) ALCOHOLISM, ALCOHOL-DRINKING, and TEMPERANCE, with all subheadings. Similar search strategies were developed for the other databases. We also hand-searched abstracts from the Transport Research Laboratory Database of World-wide Published Information and relevant conference proceedings at the Transport Research Laboratory Library (United Kingdom).

To find other eligible published or unpublished trials, we contacted the National Highway Traffic Safety Administration and the National Institute for Alcohol Abuse and Alcoholism (United States), Federal Office of Road Safety (Australia), Addiction Research Foundation (Canada), Transport Research Laboratory (United Kingdom), University of Auckland's Injury Prevention Research Centre (New Zealand), and Väg-och Trafik-Institutet (Sweden).

Study selection

One author (TD) reviewed all titles and/or abstracts to exclude studies that clearly failed to meet our first three inclusion criteria (e.g. subjects without alcohol problems, observational studies or uncontrolled trials). The full texts of the remaining citations were reviewed to exclude studies that did not meet the first three inclusion criteria. We contacted the corresponding authors of all remaining studies to identify additional potentially relevant trials and to request further details if required to determine eligibility. If studies met the first three inclusion criteria but did not report collecting injury-related outcomes (criterion four), we asked the authors to provide any unpublished data on such outcomes. We attempted to contact additional authors (by mail, telephone, and Internet search) when corresponding authors were deceased or could not be traced.

Data extraction

Two of the authors (TD and CD) independ-ently extracted data and rated the quality of allocation concealment for each eligible study. We extracted data on the number and description of participants, type of intervention, duration of follow-up, method of allocation concealment, and outcomes evaluated. Differences were resolved by discussion. We assessed the quality of allocation concealment as follows: an ‘A’ rating signified adequate measures to conceal allocation (e.g. central randomization; serially numbered, opaque, sealed envelopes); a ‘B’ rating signified unclear adequacy of allocation concealment; and a ‘C’ rating signified inadequately concealed allocation (e.g. open list of random numbers) (Schulz et al., 1995). There was 100% agreement in the allocation concealment ratings. Studies that would have received a ‘C’ rating based on the use of quasi-random allocation (e.g. alternation) were ineligible under our inclusion criteria.


Of the 7014 published and unpublished studies identified by our search strategies, 569 (8.1%) were potentially relevant based on title or abstract. After full text review, nine trials met all four inclusion criteria (Brown, 1980; Reis, 1982a,b; Walsh et al., 1991; Anderson and Scott, 1992; Ojehagen et al., 1993; Mann et al., 1994; Barber and Crisp, 1995; Sitharthan et al., 1996). An additional 314 met the first three criteria. For 23 of these 314 trials (7%), we could not determine whether injury-related outcomes had been measured, because all authors were either untraceable or deceased. The authors of 119 (41%) of the remaining 291 studies responded to our requests for further information. From these responses, we identified an additional nine eligible, completed studies (Gallant et al., 1968; Landrum et al., 1981; McCrady et al., 1982; Fizgerald and Mulford, 1985; Potamianos et al., 1986; Kuchipudi et al., 1990, WHO Brief Intervention Study Group, 1996; Toteva and Milanov, 1996; Sitharthan et al., 1997). We also identified three eligible trials still in progress (written communications: M. Bohn, University of Wisconsin Medical School, Madison, WI, 29 March 1997; F. P. Rivara [for L. M. Gentilello], Harborview Injury Prevention and Research Center, Seattle, WA, 12 May 1997; E. Wells-Parker, Mississippi State University, Mississippi State, MS, 10 November 1997). In addition, we learned from one author (written communication, A. Ojehagen, Lund University Hospital, Lund, 20 October 1997) that long-term follow-up data from a previously published trial (Ojehagen et al., 1993) were being collected. Two trials (McCrady et al., 1982; Anderson and Scott, 1992) were subsequently excluded because their ‘injury’ outcome measures were found to include other disorders (criminal behaviour and alcohol-related illness, respectively) that could not be separated from the injury data. Thus, we identified a total of 19 randomized controlled trials that met all four inclusion criteria (Table 1). The injury-related data for three studies were published in government reports (Reis 1982a,b; Landrum et al., 1981) and we obtained unpublished injury data from the authors of four studies (Gallant et al., 1968; Fitzgerald and Mulford, 1985; Kuchipudi et al., 1990; Barber and Crisp, 1995). Allocation concealment ratings are shown in Table 1.

View this table:
Table 1.

Randomized controlled trials of problem drinking interventions containing injury-related outcomes

Injury-related outcomes are summarized in Table 2. Trials are grouped by type of outcome, and appear under more than one subheading if different types of outcomes were collected. Due to the heterogeneity of the patient populations, interventions, and types of injury outcomes reported, no attempt was made to combine the results quantitatively.

View this table:
Table 2.

Injury data from trials of interventions for problem drinking


All seven trials that collected injury mortality outcomes, including one trial in progress, received ‘B’ allocation concealment ratings. A total of 21 deaths were reported in the five trials for which data were available. The two trials comparing intervention to no intervention (Kuchipudi et al., 1990; Mann et al., 1994) reported a reduced risk of injury death in the intervention group, with imprecise effect estimates. Surprisingly, Kuchipudi et al. (1990) also reported slightly lower abstinence rates and higher rates of driving under the influence of alcohol (DUI) in the intervention group, although these effect estimates were imprecise. The other three completed trials reporting mortality data compared different treatment modalities. There were too few deaths in each of these three trials to identify differences in the effects of specific treatment modalities on injury deaths.

Non-fatal injuries and their antecedents

Seventeen trials collected data on non-fatal injuries and their antecedents. Those trials collecting outcomes specifically identified as non-fatal violence or motor vehicle crashes and related injuries are summarized separately below. There were nine trials that collected non-fatal injury outcomes (e.g. ‘accidents’), that combined injuries due to a variety of causes. Two of these are in progress. Relevant data were available from only two of the seven completed trials. Fitzgerald and Mulford (1985) received an ‘A’ rating and Kuchipudi et al. (1990) received a ‘B’ rating for allocation concealment. In both of these trials, the intervention for problem drinking reduced the risk of injuries and their antecedents, compared to no intervention, despite reporting no beneficial effects of treatment on abstinence. All these effect estimates were imprecise, however. In the trial by Fitzgerald and Mulford (1985), the number of subjects differed markedly between the two study groups at Center A, because 43 subjects originally assigned to a second experimental group of ‘patient-initiated’ aftercare contacts were combined with the ‘no aftercare’ control group in the analysis after only one patient initiated such contact. The authors stated that there was little effect on their results whether these subjects were included or excluded from the control group.

Non-fatal violence

Eight trials, two of which are still in progress, collected data on non-fatal violence. Data were available for three of the six completed trials. Fitzgerald and Mulford (1985) received an ‘A’ rating for allocation concealment and the other two (Barber and Crisp, 1995; Sitharthan et al., 1996) received ‘B’ ratings. Both trials evaluating intervention vs no intervention suggested a reduction in violence with intervention, while only Barber and Crisp (1995) showed a beneficial effect on drink-related outcomes. The third trial (Sitharthan et al., 1997) compared two different interventions and reported a reduced risk of committing assault after cognitive behavioural therapy, but a greater reduction in alcohol consumption with cue exposure therapy.

Motor vehicle crashes and related injuries

Seven trials assessed motor vehicle crashes and related injuries. Data were available from three of the four completed trials. The two trials by Reis (1982a,b) received ‘A’ ratings and the trial by Landrum et al. (1981) a ‘B’ rating for allocation concealment. In Landrum et al.'s (1981) trial of interventions for persons convicted of DUI, the monthly probation and the rehabilitation interventions were associated with reduced risks of both motor vehicle crashes and crash-related injuries. Effects were greater on injuries than on crashes, although all effect estimates were imprecise. The combination of probation and rehabilitation appeared to have less effect on motor vehicle crashes and a small adverse effect on crash-related injuries. Among persons convicted once of DUI (Reis, 1982a), in-class education reduced the cumulative accident rate, but there appeared to be little beneficial effect from home study. These results are consistent with the stronger effect of in-class education than of home study on alcohol consumption and the slightly stronger effect of the former on DUI arrest recidivism (see Table 1). In the study (Reis 1982b) of persons convicted more than once of DUI, only educational counselling combined with disulfiram therapy appeared to reduce the cumulative incidence of alcohol-related crashes and injuries compared to no intervention, despite the fact that all three intervention groups had less DUI arrest recidivism than did the control group receiving no intervention.


Injury is a major public health problem worldwide and alcohol is a significant contributor. The reduction of unintentional and intentional injuries due to problem drinking is therefore an important public health goal. Although our data are not conclusive, they do suggest that interventions for problem drinking may be effective in reducing injuries and injury deaths. In the seven trials that compared interventions for problem drinking to control interventions and which provided outcome data, nearly all interventions showed a beneficial effect on injury-related outcomes. The effect sizes were large, ranging from 27% reduction of ‘drinking-related injuries and accidents' to a 65% reduction in ‘accidental and violent deaths’. Because the trial sample sizes were generally small, however, the precision of these estimates was low. Nevertheless, the consistency of the results suggests that interventions to reduce problem drinking could have an important effect on the incidence of injuries and injury deaths.

Although reduced alcohol consumption would seem a likely mechanism for any beneficial effects of treatment on injuries, our review does not provide strong support for this mechanism. Among the seven trials reporting beneficial effects of treatment on injuries or injury antecedents, four reported on abstinence or reduced alcohol consumption and three on the incidence of driving under the influence of alcohol. Reis (1982a) and Barber and Crisp (1995) found a beneficial effect of intervention on drink-related outcomes, whereas Fitzgerald and Mulford (1985) and Kuchipudi et al. (1990) showed slight adverse effects. Similarly, Reis (1982a,b) found significant reductions in DUI recidivism rates in both of his trials, but Landrum et al. (1981) reported only a slight reduction in DUI incidence with intervention and Kuchipudi et al. (1990) found an adverse effect on DUI. We also found four trials that compared different treatment modalities and reported injury outcomes. In two of these, there were significantly greater declines in alcohol consumption with one therapeutic modality compared to the other(s) (Walsh et al., 1991; Sitharthan et al., 1997). In both trials, however, the treatments that reduced consumption had adverse effects on injury-related outcomes (although effect estimates in both cases were imprecise due to small sample sizes). It is possible that these paradoxical results can be explained by chance, reflecting the nearly universally imprecise effect estimates, or by measurement error in the assessment of the drink- or injury-related outcomes. It is also possible that any beneficial effect on injuries is mediated by other aspects of treatment for problem drinking (e.g. receipt of medical attention and social support).

The aim of our systematic review was to make explicit the totality of the randomized evidence on what appears to be a promising approach to tackling the problem of alcohol-related injuries. The fact that the trials that we found reported imprecise effect estimates and often had important methodological weaknesses is a key finding of the review, indicating that this promising approach requires further research. We considered the possibility of combining the available data from these trials in a meta-analysis to increase the precision of the effect estimates. However, this would have involved combining markedly heterogeneous groups of patients, interventions, and outcomes. In such circumstances, a meta-analysis can produce inappropriate, and even misleading, conclusions (Bailar, 1997; Editorial 1997).

We limited our critique of the quality of the included studies to an assessment of the quality of allocation concealment, because this is the most important criterion for assessing the validity of a trial (Schulz et al., 1995). Unfortunately, we were able to determine this criterion accurately in only a very small proportion of the trials reviewed. Few trials reported allocation concealment in detail, and among the others, very few researchers provided us with sufficient information to assess adequately this criterion. We cannot, therefore, make firm conclusions about quality for most of the trials.

Publication bias is an important threat to the validity of systematic reviews. Such bias may arise if outcome data are selectively omitted from published reports, because the results fail to reach significance. To avoid the effects of this type of bias, we wrote to the authors of all identified trials that met our first three inclusion criteria, asking them to provide any unpublished outcome data on injuries or their antecedents. Nine additional completed trials, and three trials still in progress, were identified by this approach. Unfortunately, we were able to obtain the unpublished injury-related data from only a few of the completed trials. The difficulties involved in extracting unpublished data and other information for systematic reviews have been reported previously (Roberts and Schierhout, 1997). Many of the authors of studies that met our first three inclusion criteria were untraceable or deceased (7%) or did not respond to our requests for information (55%). While it is likely that some did not respond because they did not measure injury-related outcomes, the inability to identify all unpublished data might have biased our results.

In conclusion, previous reviews have shown that interventions for problem drinking can reduce alcohol consumption (Freemantle et al., 1993) and driving under the influence of alcohol (Wells-Parker et al., 1995). This review suggests that interventions for problem drinking have the potential to reduce the incidence of injuries and their antecedents, but current data are insufficient to draw firm conclusions. Because injuries account for a large proportion of the morbidity and mortality due to problem drinking, further studies are warranted to evaluate the effect that treating problem drinking may have on injuries and to investigate how any beneficial effects on injuries are mediated.


We gratefully acknowledge the assistance of Drs Ralph Bloch and Ellen Ingham (translations), Dr Robert Zarr (data collection), and all the researchers who sent us information and unpublished data. Ms Dinh-Zarr was funded in part through a University of Texas–Houston Health Sciences Center Summer Internship. Dr DiGuiseppi was funded by the Camden and Islington Health Authority.


  • * Author to whom correspondence should be addressed.


View Abstract