Scope and Purpose

This systematic review addresses the benefits and harms of behavioral sexual risk reduction counseling interventions in primary care to prevent STIs among adolescents and adults. The USPSTF will use this review to update its 2008 recommendation on behavioral counseling to prevent STIs.⁵³

Key Questions and Analytic Framework

We developed an analytic framework (Figure 1) and four Key Questions (KQs) to guide our review, in consultation with the Agency for Healthcare Research and Quality (AHRQ) and members of the USPSTF. These KQs were adapted from the questions addressed in the 2008 review.⁵⁴

Figure 1

Analytic Framework. Abbreviation: STI = sexually transmitted infection.

1. Is there direct evidence that behavioral counseling interventions to reduce risky sexual behaviors and increase protective sexual behaviors reduce STI incidence and/or related morbidity and mortality?
   1. Are there populations or intervention characteristics that influence the effectiveness of the interventions?
2. Do behavioral counseling interventions to prevent STIs reduce risky sexual behavior or increase protective sexual behavior?
   1. Are there populations or intervention characteristics that influence the effectiveness of the interventions?
3. Are there other positive outcomes beside STI incidence and changes in risky or protective sexual behavior from behavioral counseling interventions to prevent STIs?
4. What adverse effects are associated with behavioral counseling interventions to prevent STIs in primary care?

Data Sources and Searches

We performed comprehensive literature searches in the following databases: PubMed, Database of Abstracts of Reviews of Effects, Cochrane Database of Systematic Reviews, British Medical Journal Clinical Evidence, Institute of Medicine, National Institute for Health and Care Excellence, MEDLINE, PsycINFO, Cochrane Central Register of Controlled Trials, and the Cumulative Index to Nursing and Allied Health from January 1, 2007 through November 4, 2013. Our research librarian designed this literature search to identify studies published since the 2008 review (Appendix A). We limited all searches to articles published in the English language. We managed literature search results using Reference Manager® version 12.0 (Thomson Reuters, New York, NY).

We also reviewed reference lists of included studies and relevant reviews and meta-analyses to identify potentially relevant studies that were published prior to our literature search dates or not identified in our literature searches. We obtained additional references from expert reviewers, members of the public, and bibliographies of other sources (e.g., clinical guidelines). We conducted grey literature searches of government agencies, professional organizations, and other organizations that may sponsor or publish relevant research. We also supplemented our searches with articles identified through news and table-of-contents alerts from Google (Google, Inc., Mountain View, CA), ScienceDirect (Elsevier B.V., Maryland Heights, MO), and HighWire Press (Stanford University Libraries, Palo Alto, CA).

Study Selection

Two investigators independently reviewed titles and abstracts. Investigators then reviewed the full-text articles against prespecified inclusion and exclusion criteria (Appendix A Table 1). Disagreements were resolved through discussion and consensus.

We included randomized, controlled trials and nonrandomized, controlled clinical trials with interventions targeting risky sexual behaviors to prevent STIs (alone or in combination with other behaviors) in adults and adolescents, including pregnant women, of any sexual orientation or level of sexual activity. We excluded studies limited to persons who were HIV positive or populations that had very high HIV prevalence, inmates and court-involved persons, and psychiatric inpatients (i.e., ≥50% of the study population was institutionalized or in supportive housing). Although these patient populations are seen in primary care, trials aimed at these populations were excluded from this review because their results may not be generalizable to general primary care populations, and the results of these trials do not help inform the question of whether counseling should be routinely offered in primary care settings.

We excluded studies that exclusively targeted unintended pregnancy or another behavior (e.g., drug use) associated with risky sexual behavior without directly addressing STI prevention in the intervention, even if they reported relevant biological or behavioral outcomes. We also excluded studies without control groups. Control groups could include usual care, attention control, minimal intervention (≤15 minutes), wait list, or no intervention.

Included studies had to be conducted in or recruited from primary care, mental health clinics, reproductive health clinics (including STI clinics), or from broader health care systems in developed countries (with “very high” human development according to the World Health Organization).⁵⁵ We excluded studies in worksites, schools (except school health clinics), or other settings that were not generalizable to primary care. We also excluded studies with interventions that included components that could not be implemented in primary care settings, such as communitywide media messages or interventions conducted within existing social networks (e.g., churches, social clubs).

We included studies that reported health outcomes (STI incidence or related morbidity), behavioral outcomes (changes in sexual behavior), or adverse effects of sexual risk reduction counseling (e.g., care avoidance, shame, guilt, stigma). Any additional reported outcomes that could be related to harms were also abstracted from studies meeting inclusion criteria (e.g., stress, acceptability ratings). Similarly, other positive outcomes (e.g., reductions in unintended pregnancy) were also abstracted, but were not sufficient to warrant inclusion in the review. We required at least 3 months postbaseline followup for all outcomes except harms.

Consistent with the 2008 review, we excluded trials published before 1988, as these were largely conducted before HIV/AIDS was identified and current approaches to medical management were developed.

Quality Assessment and Data Abstraction

Two investigators independently assessed the quality of included studies using predefined criteria from the USPSTF⁵⁶ and assigned each a final quality rating of “good,” “fair,” or “poor” (Appendix A Table 2). Investigators resolved disagreements through discussion. We excluded studies rated as poor quality (i.e., attrition >40%, differential attrition >20%, other “fatal flaws,” or the cumulative effects of multiple minor flaws or missing enough important information to limit our confidence in the validity of the results) (Appendix B). Good-quality studies had adequate randomization procedures, allocation concealment, blinding of outcome assessors, reliable outcome measures, comparable groups at baseline (with specified eligibility criteria) and followup, low attrition, acceptable statistical methods, and adequate and faithful adherence to the intervention. In addition, we also considered whether the study evaluated a biological outcome. We rated studies as fair quality if they did not meet most of the good-quality criteria.

One investigator abstracted data from all included studies into a standard evidence table and a second investigator checked the data for accuracy. We abstracted study design characteristics, population demographics, baseline sexual history, intervention details, biological outcomes, behavioral outcomes (e.g., condom use, sex partners, unprotected sex, and other sexual activity), other positive outcomes (e.g., reduction in unintended pregnancy), and adverse events.

Data Synthesis and Analysis

We created separate tables for the results for each KQ and created additional summary tables of key trial characteristics and results. We qualitatively examined these tables to identify the range of results and potential associations with effect size. We examined trials that targeted only adolescents separately from those that targeted adults (some of which also included adolescents). We also examined subgroup analyses limited to adolescents from three trials with a wide age range.^57-59 Because we planned a priori to present data separately for adolescents and adults in our report, we abstracted age-specific results when they were presented, even if there was no indication in the trial methods that subgroup analyses by age were planned a priori. We analyzed adolescents separately because the developmental differences between adolescents and adults may warrant different intervention approaches and targets (e.g., delayed sexual debut or reduced sexual activity may be appropriate only for adolescents).

Within age strata, we further stratified tables by intervention intensity: low (single session, estimated ≤30 minutes of intervention contact), moderate (30 to 120 minutes of intervention contact), or high (>2 hours of intervention contact). We also categorized studies into one of four risk strata: 1) mixture of sexually active and presexually active participants (relevant for adolescents only); 2) sexually active participants with no further risk factors required and not in an increased-risk setting (relevant for adults only); 3) participants with increased risk based on characteristics of the population (e.g., adolescents, low-income inner-city residents, African Americans, Latinos, American Indians/Alaska Natives, MSM, military recruits, persons with mental illness or disability, intravenous drug users, persons with a history of sexual abuse, sex workers, or persons reporting high-risk behaviors) or setting (e.g., STI clinics, clinics in low-income inner-city locations, or clinics serving a high proportion of patients at increased risk); and 4) persons at high risk based on a current or recent STI. Within risk strata, we grouped studies by setting (primary care, reproductive health clinic, STI clinic, other setting).

We conducted random-effects meta-analyses for STI incidence using the DerSimonian and Laird method, with separate pooling for adolescents and adults. Because some of our pooled estimates were derived from combining a small number of trials, we ran sensitivity analyses using the profile likelihood method, which is more conservative when the number of studies is small. All statistically significant results remained within the profile likelihood method. Results shown are those using the DerSimonian and Laird method.

We used Stata 11.2 (StataCorp LP, College Station, TX) for all statistical analysis. We analyzed odds ratios (ORs) based on reported adjusted ORs if available. We used unadjusted ORs if adjusted results were not available. We used the raw numbers of events and numbers of participants with followup if ORs were not available. For condom use outcomes, we included any measure of the proportion using condoms by group, including use at most recent intercourse, consistent use, and any use. If a trial reported the proportion reporting any unprotected sex, we subtracted the number from 100 percent and included those data in the condom use analysis.

We stratified our primary analysis by intervention intensity. If trials included more than one treatment arm and these arms fell into different intensity categories, we included both treatment arms but only pooled within the intensity categories. As such, trials never contributed more than one data point to a pooled estimate. If there were multiple intervention arms within the same intensity category, we included the arm that was most intensive or comprehensive in the meta-analysis.

We assessed the presence of statistical heterogeneity among the studies using standard chi-squared tests and estimated the magnitude of heterogeneity using the I² statistic.⁶⁰ We applied the Cochrane Collaboration's rules of thumb for interpreting heterogeneity: less than 40 percent likely represents unimportant heterogeneity, 30 to 65 percent represents moderate heterogeneity, 50 to 90 percent represents substantial heterogeneity, and greater than 75 percent indicates considerable heterogeneity.⁶¹ We also included prediction intervals in forest plots, which provide an estimate of where 95 percent of newly conducted trials would fall, assuming the between-study variability in the included trials held for new trials.⁶² Prediction intervals are shown with pooled estimates on forest plots, and the horizontal lines represent the 95 percent confidence interval (CI) of the pooled effect.

We examined small-study effects on STI outcomes using funnel plots, Peter's test for small-study effects, and Harbord's modified test for small-study effects, combining adult and adolescent trials. These tests examined whether the distribution of the effect sizes was symmetric with respect to effect precision. The results of these tests were contradictory; Peter's test suggested the presence of small-study effects, while Harbord's did not. Visual inspection of the funnel plot showed that two trials with large effects and few events seemed to be driving asymmetry in the funnel plot.^63,64 We tested this hypothesis by excluding these two trials and re-running the two tests of small-study effects. After excluding these studies, neither test indicated the presence of small-study effects. One of these trials was the only low-intensity trial in adolescents, so it was not combined with other trials in the meta-analysis and therefore does not put pooled estimates at risk for bias. The other trial was a high-intensity trial in adults, which had the smallest weight of all high-intensity trials in the meta-analysis. Therefore, we believe that the pooled effect estimate is no more than minimally biased by small-study effects, because of this study's small contribution to the pooled effect.

We calculated the number needed to treat (NNT) to prevent one STI and the number of cases prevented based on the pooled OR at three levels of baseline risk.⁶¹ For baseline risk, we chose incidence rates of 5, 15, and 25 percent. This same process was used to estimate CIs for the NNTs, using the CIs of the pooled ORs.

Expert Review and Public Comment

A draft research plan for this topic was available for public comment from January 29 to February 25, 2013. The draft version of this report was reviewed by experts and USPSTF federal partners and posted for public comment on the USPSTF Web site from April 29 to May 26, 2014. No new substantive issues were identified that were not previously considered, and no major changes were made to the final report. Comments received during any period were reviewed, considered, and addressed as appropriate.

USPSTF Involvement

This research was funded by AHRQ under a contract to support the USPSTF. We consulted with four USPSTF liaisons at key points in the review, including the development of the research plan (i.e., KQs, analytic framework, and inclusion/exclusion criteria), as well as finalizing the systematic review. An AHRQ Medical Officer provided project oversight, reviewed the draft and final versions of the report, and assisted with public comment on the research plan and draft evidence report. The USPSTF and AHRQ had no role in the study selection, quality assessment, or writing of the systematic review.