Using methods of the U.S. Preventive Services Task Force (USPSTF),46 we developed an analytic framework and 7 key questions (KQs) to guide the review process (Appendix Figure 1). KQ1 assessed direct evidence indicating that behavioral counseling interventions reduce morbidity or mortality. KQ2 focused on methods used to identify appropriate target populations for alcohol-related behavioral counseling interventions in primary care. KQ3 concentrated on adverse effects associated with alcohol use screening and screening-related assessment. KQ4 addressed the impact of primary care identification and behavioral counseling interventions on risky or harmful alcohol use, and the essential elements of efficacious interventions. KQ5 sought to identify other positive outcomes from behavioral counseling interventions to reduce risky/harmful alcohol use. KQ6 addressed harms associated with behavioral counseling interventions. KQ7 examined the context for interventions to reduce risky/harmful drinking by examining health care system influences present in effective screening, screening-related assessments, and behavioral interventions.
As a result of the emphasis of the last USPSTF recommendations on alcohol-related screening, and the availability of new research on brief counseling interventions, we concentrated our efforts on effects of brief primary care interventions to reduce alcohol use, on other intermediate outcomes of such interventions, and on associated health outcomes (KQs 1 and 4). We did not systematically review the evidence on the efficacy of screening tools, nor did we look for direct evidence that screening alone improves outcomes.
We searched the Cochrane Database of Systematic Reviews and Database of Research Effectiveness (DARE) (2001, issues 2 and 3; 2002 issue 1). We used an inclusive search strategy (alcohol* or drink*) to identify recent systematic reviews addressing brief interventions in primary care to reduce risky/harmful alcohol use. We found 5 recent systematic reviews of interventions.35-39 None of these adequately addressed our key questions:
More details on these reviews are available elsewhere.40 To identify relevant primary literature, we searched MEDLINE®, Cochrane Controlled Clinical Trials, PsychInfo, HealthSTAR, and CINAHL databases from 1994 through April 2002, using search strings detailed in Appendix Table 1. Results from this search (integrated with articles retrieved from outside sources) are shown in Appendix Figure 2.
We conducted searches in MEDLINE® and PsychInfo from 1994 through April 2002, combining the terms described in Appendix Table 1 with "adverse effects of screening" and "adverse effects of counseling" to identify any literature on the harms of alcohol screening, screening-related assessment, or intervention; none was found.
To be eligible for inclusion, a study had to be a randomized or controlled clinical trial of behavioral counseling interventions in risky/harmful drinkers conducted in a primary care setting, as defined in a recent Institute of Medicine report. "Primary care is the provision of integrated, accessible health care services by clinicians who are accountable for addressing a large majority of personal health care needs, developing a sustained partnership with patients, and practicing in the context of family and community" (p. 31).43 We excluded from this review other clinical settings, such as emergency departments and hospitals, specialty treatment, behavioral health, or community or school settings without clinics, to maximize the applicability of the review findings to primary care. Other exclusion criteria included non-English abstract, non-controlled-trial study designs, population characteristics (age <12 years, primarily dependent drinkers, comorbid populations such as dual diagnosis patients), or interventions without a behavioral intervention component.
Investigators reviewed 4,331 non-duplicative titles and abstracts. A second investigator reviewed a random 35% of titles/abstracts for concordance. We found about 95% agreement in this dual review, and none of the 75 articles that met initial inclusion criteria was discrepantly coded. We also identified 62 outside source articles by contacting experts for unpublished studies, by reviewing bibliographies of all reviews and primary research articles located through database searching, and by retrieving all intervention trials cited in the alcohol screening chapter in the 1996 Guide to Clinical Preventive Services.34 Titles and abstracts of all potentially included studies were re-reviewed for eligibility, and full-text articles retrieved from database searches and outside sources were assessed using the same criteria. Quality of the articles was graded using the USPSTF criteria,46 supplemented by guidelines on evaluating study randomization, attrition, and intention-to-treat analyses from the Cochrane Drug and Alcohol Group. A second investigator reviewed all included articles to confirm setting eligibility and quality ratings.
After we reapplied initial inclusion and quality criteria to the full text of 137 trials, 44 remained eligible. Of these, 27 were excluded; reasons for their exclusion are listed in Appendix Table 2.83-110 One trial that met final setting and quality criteria was not published in time for inclusion in this review.48 Twelve of the included studies addressed adults and are the basis for this paper; the 3 included studies that addressed pregnant women and 1 that addressed adolescents are reviewed elsewhere.40 An updated database search through February 2003 revealed no new trials.
For included studies from any source, 1 of the authors abstracted relevant information using data-abstraction forms. A second author checked all key data that appear in the evidence tables. All studies were abstracted onto standardized data-abstraction forms developed for this review (available elsewhere40). Data-abstraction forms addressed 3 main issues: 1) study recruitment, randomization, and attrition (adapted from CONSORT);111 2) study design, conduct, and results; and 3) study quality.46,47 Selected abstraction details are described in the next section. A separate audit of available study outcomes was conducted by 1 of 2 research assistants. At least 2 authors conducted a quality review audit of each study (Appendix Figure 3), emphasizing the key aspects of quality in this literature (allocation concealment, attrition and replacement of missing values, baseline and final comparability of groups, adequate intervention delivery, and masking of patients and outcome assessment). The final quality rating for each study (Good, Fair, Poor) reported in the evidence tables was assigned by consensus of the investigator team.
We abstracted author, year published, type of trial, setting, and definition of a standard drink used in each study.
We abstracted the total number randomized to the study, the gender and racial distribution of subjects, participants' baseline alcohol consumption, the proportion that was alcohol-dependent, and the proportion that was motivated or help-seeking or thought they had a problem with alcohol. When sociodemographic or other baseline information on total study participants was unavailable, we reported intervention group characteristics in evidence tables.
Information abstracted included whether alcohol screening or screening-related assessment was masked within a more general lifestyle assessment; intervention and control protocols; whether the intervention involved personal contact; the intensity of personal contact (number of contacts and contact minutes); whether intervention delivery was measured and what percent of participants received it; all provider types involved; the use of usual care or other clinical or research personnel; components of the intervention (detailed below); and reported adverse effects of screening or intervention. The intervention components we abstracted are considered important elements in brief interventions:25,31,34,49
These components map to the 5A's (assess, advise, agree, assist, arrange/adapt) adopted by the USPSTF for reporting the results of behavioral counseling interventions.69 We also evaluated the presence or absence of tailoring described as part of the intervention.
A large variety of alcohol use variables and measures reflecting different definitions of problematic use were reported in different studies (for more information, go to Table 1 in full systematic evidence review40). To facilitate comparison between studies, we chose 3 primary alcohol use outcome categories commonly reported in epidemiologic and intervention literature: average consumption, binge use, and safe/moderate/recommended use. We preferentially report the outcome measures as reported by the study authors, or where necessary, we used established methods to recalculate reported study data into more comparable outcome measures.112 For average use outcomes, we abstracted the absolute followup levels (or change from baseline to 6-12 months' followup) in mean drinks per week for each arm of the study. If these were not available, we abstracted any other recent average consumption measures and converted them to mean drinks per week where possible. For binge use, we abstracted the percentage or proportion bingeing or not bingeing, or the reduction in either of these measures.
We then calculated the converse (1 minus reported percentage), where necessary, to convert all measures to percentage not bingeing. Studies varied in their definition of safe, moderate, or recommended use of alcohol (based on attaining each study's recommended limits on average consumption and/or binge use). To consider what percentage of study participants attained these recommendations, we created or adapted the within-study definition that best fit each study's intervention rationale and focus.
For each study, we then abstracted or calculated from reported data the percentage achieving "recommended" levels or patterns of alcohol use by group. For all the outcome categories, we recorded outcomes for all groups when studies had more than 1 intervention or control group. Similarly, we abstracted outcomes by gender where possible. We also examined health outcomes in the original intervention study (or subsequent reports) when available.
We recorded study recruitment (including whether patients were or were not primarily opportunistically recruited as part of routine care), provider support and training, use of research vs. clinical personnel, and use of incentives for providers or patients to participate in the study.
Where necessary, we used the converse (1 - proportion) to convert data for comparability. If these latter 2 outcomes were not directly reported but other relevant data were available, we calculated the relevant percentages from the number reported as meeting the criterion at followup divided by the number randomized to that group. Where available, outcomes were reported separately by gender. We examined results for all intervention arms (n=15) for which these outcomes were reported or could be calculated.
We used the USPSTF approach to grade the overall quality of evidence for each key question;46 this summary information is reported elsewhere.40