Analytic Framework and Key Questions

Using methods of the US Preventive Services Task Force (USPSTF) ¹⁰⁰ we developed an analytic framework and 7 key questions (KQs) (see Appendix: Figure 1) to guide this review process. KQ1 addressed the direct evidence that behavioral counseling interventions reduce morbidity or mortality. KQ2 concerned the screening methods used to identify the target populations for alcohol-related behavioral counseling interventions in primary care. KQ3 concerned harms and costs of identification of alcohol use disorders. KQ4 addressed the impact of primary care identification and intervention on risky or harmful alcohol use, and any other benefits (KQ5) or harms (KQ6) from such interventions. KQ7 examined the context for these interventions by looking at health care system influences that help or hinder appropriate patient identification and intervention delivery.

Figure

Figure 1. Analytic Framework & Key Questions. Does Behavioral Counseling Intervention for Alcohol-Related Risks and Problems In Adults Improve Health Outcomes and Decrease Morbidity and Mortality?

Given the emphasis of the last USPSTF recommendations on screening, and the availability of new research on brief counseling interventions, we concentrated on the effects of brief interventions in primary care on alcohol use, on other intermediate outcomes, and on health outcomes (KQs 1 and 4). We did not systematically review the evidence on the efficacy of screening tools nor look for direct evidence that screening alone improves outcomes.

A team from the Cost Working Group of the USPSTF conducted searches of MEDLINE, HealthSTAR, PsychInfo, and the British National Health Service's Economic Evaluation Database (see Appendix: Table 6). Titles and abstracts were reviewed by two investigators, and studies were excluded if they were clearly not original economic analyses reporting both costs and effects of primary-care based counseling interventions for risky/harmful alcohol use. All other studies were retrieved and reviewed, along with articles identified by experts and manual searching of reference lists. Included studies were critically appraised by all cost work group team members and abstracted using a standard form. Quality was rated according to previously developed criteria (see Appendix: Form 1), and the entire working group discussed study features, key parameters, assumptions, results and quality before making its general conclusions.

Table 6

Literature Search Terms.

Search Strategy

Key questions 1, 4, 5, and 7. We searched the Cochrane Database of Systematic Reviews and Database of Research Effectiveness (DARE) (2001, issues 2 and 3; 2002 issue 1) using an inclusive search strategy (alcohol* or drink*) to identify recent high-quality systematic reviews addressing brief interventions to reduce risky/harmful alcohol use in primary care. We found five well-conducted recent systematic reviews of interventions ⁹⁶ – ^{99, 101} and reviewed their bibliographies to ascertain any studies not identified through our separate searches of other primary databases (described below). Unpublished studies were located through expert contacts and referrals. We also retrieved relevant literature reviewed in the 1996 Guide to Clinical Preventive Services. ⁶⁴ We searched MEDLINE, Cochrane Controlled Clinical Trials, PsychInfo, HealthSTAR, and CINAHL databases from 1994 through April 2002, using search strings detailed in Table 6 of the Appendix, and results from this search (integrated with articles retrieved from outside sources) are shown (see Appendix: Figure 2). One investigator reviewed 4331 non-duplicative titles and abstracts, and a second investigator reviewed a random 35% of titles/abstracts for concordance. Approximately 95% agreement was found, and no articles that met review inclusion criteria were discrepantly coded by the two reviewers. For included studies, no matter the source, one primary reviewer abstracted relevant information using data-abstraction forms. All key data that appear in the evidence tables (see Appendix: Tables 9 & 10) were checked by a second reviewer and articles excluded for quality or relevance appear in Table 10 of the Appendix. Quality of the articles was graded using the USPSTF criteria, ¹⁰⁰ supplemented by guidelines on evaluating study randomization, attrition, and intention-to-treat analyses from the Cochrane Drug and Alcohol Group. ¹⁰²

Figure

Figure 2. Literature Search and Retrieval Results for Intervention Trials.

Table 9

Evidence for Pregnant Women.

Table 10

Excluded Adult Intervention Studies (Includes those excluded for quality and other reasons).

Key questions 3 and 6. We conducted searches in MEDLINE and PsychInfo from 1994 through April 2002 combining the terms described in Table 6 of the Appendix, 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, but none was found.

Inclusion and Exclusion Criteria

Figure 3 of the Appendix, includes a list of the eligibility criteria used to select articles for inclusion in evidence tables. To be eligible, studies had to be conducted in a primary care setting, as defined in a recent Institute of Medicine report. ¹⁰³ Other clinical settings, such as emergency departments and hospitals, were excluded from this review to maximize the applicability of the review findings to primary care. Excluded studies are listed in Table 10 of the Appendix. Titles and abstracts were reviewed for eligibility, and full-text articles of potentially eligible studies were re-assessed using the same criteria. Included articles were re-reviewed by a second investigator to confirm eligibility and quality ratings. Studies receiving a rating of “poor-quality” according to the USPSTF criteria ¹⁰⁰ were excluded from the review (see Appendix: Table 10).

Figure

Figure 3. Abstract and Article Review Process for Problem Drinking Interventions.

Data Extraction, Reliability, and Validity Assessments

All studies were abstracted by one of the study investigators into standardized data-abstraction forms developed for this review. Data-abstraction forms addressed three issues: 1) study recruitment, randomization, and attrition (adapted from CONSORT ¹⁰⁴ ), 2) study design, conduct, and results, and 3) quality (see Appendix: Forms 1–4). A separate audit of study outcomes was conducted by one of 2 research assistants, and all items abstracted into final evidence tables were double-checked by a second reviewer. A second investigator also conducted a quality review audit of each study, 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 based on the USPSTF criteria ¹⁰⁰ and assigned by consensus of the investigator team.

Study Characteristics Extracted During Review

Study Identification. We abstracted author, year published, type of trial, setting, and definition of a standard drink used in the study.

Study Participants. We abstracted the total number randomized to the study, the proportion of female and non-White participants, participants' baseline alcohol consumption, the proportion that was alcohol dependent, and the proportion that was motivated, help-seeking, or thought they had a problem with alcohol.

Intervention and Control Conditions. 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 total contact minutes); whether intervention delivery was measured and found to be above 75%; all provider types involved; the use of usual care, other clinical or research personnel, and intervention elements (advice, feedback, goal setting, assistance, followup, matching, and tailoring); and whether harms were reported.

We abstracted other intervention components besides intensity cited as important elements in brief interventions. ^{40, 55, 64, 105} These included:

1.

advice to reduce current drinking and/or about guidelines for low-risk use;

2.

feedback about current drinking patterns, problem indicators (such as laboratory results), and/or medical consequences of current use patterns;

3.

explicit goal-setting, usually for moderation and not abstinence;

4.

assistance in achieving the goal, including providing a menu of options for change, helping patients develop skills for managing high-risk drinking situations (e.g., recognizing antecedents, planning ahead, pacing drinking), coping with problems without drinking, and providing self-help materials.

5.

providing followup in the form of telephone calls, repeat visits, and/or repeat monitoring of lab test, physical examinations, etc.

These components map to the 5As (Assess, Advise, Agree, Assist, Arrange/Adapt) adopted by the USPSTF for reporting the results of behavioral counseling interventions, ⁵⁶ which are described in Discussion. We also evaluated the presence or absence of described tailoring in the intervention.

Outcomes. A large variety of alcohol use variables and measures reflecting different definitions of problematic use were reported in different studies (for more information, see Table 1). 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 ¹⁰⁶ . For average use outcomes, we abstracted the absolute follow-up levels (or change from baseline to 6–12 months' follow-up) 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.

Generalizability. 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.

Data Synthesis. To synthesize the results of the studies, we qualitatively compared results for the 3 selected alcohol use outcomes reported in the evidence tables and created 5 graphical outcome summaries (Figures 4–8). We examined the consistency and direction of the evidence for the effect of brief interventions on net (i.e., intervention minus control) change in mean drinks/week (n=8), on net percentage point decrease in average consumption from baseline (n=8), on the proportion not reporting binge drinking at follow-up (n=7), and on the proportion of subjects achieving recommended drinking levels/patterns after brief interventions (n=10). To calculate the net reduction in mean drinks/week, we used reported mean differences in between-group changes from baseline to follow-up or calculated group means, changes in group means, and between-group net mean differences. To calculate the percentage point reduction in average consumption from baseline for each group, we divided the reduction in mean drinks/week by the baseline mean drinks/week and multiplied by 100. We then subtracted the percentage reduction in the control group from the percentage reduction in the intervention group to obtain the net percentage point reduction in mean drinks per week. For the remaining 2 outcomes, we used directly reported percentages or proportions of those in intervention and control groups 1) not bingeing and 2) achieving recommended drinking levels or patterns. 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 follow-up 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.

Figure

Figure 4. Net reduction in mean drinks per week .

Figure

Figure 8. Percent Achieving Recommended Drinking Levels/Patterns at Followup (By Gender).

Evidence Synthesis. We used the USPSTF approach to grade the overall quality of evidence for each key question. ¹⁰⁰