Sexually transmitted infections (STIs) and unintended pregnancies are significant problems that are directly related to sexual behaviors.

The World Health Organization estimates that the incidence of STIs increased 40% between 1995 and 2012.1

Epidemiologic data and observational studies suggest that there has also been no improvement in the rates of unintended pregnancies. It was estimated that 49% of all pregnancies in the United States in 2006 were unintended and that the prevalence of such pregnancies rose to 80% in the age bracket of 19 years and under.2,3 One observational study covering 20 years of data for the US showed that the overall unintended birth rates were not falling.4

We note that the deterioration in these statistics has been accompanied by ever more widespread implementation of primary preventive measures, especially behavioral interventions.

In addition, spending on such programs in countries with low-to-middle income levels per capita, with the sole purpose of preventing human immunodeficiency virus (HIV) infections, was estimated to reach US $23000 million in 2015.5

Given this scenario, some health authorities have come to realize that so-called sexual education has not had, nor will it ever have, the desired results, and so they now recommend long-acting contraceptive methods such as intrauterine devices.6 The situation is the same for STIs: more and more institutions recommend so-called biomedical prophylaxis (preexposure antibiotics and/or antiretroviral drugs), suggesting loss of faith in the efficacy of behavioral interventions alone.7

The approach of evidence-based medicine should assist with decision-making by providing an assessment of available evidence for the true efficacy of these preventive strategies,8 and taking into consideration the special problems of trials to determine the impact of behavioral interventions on STIs and unintended pregnancies:

• -

  Double-blind trials are not undertaken because of the nature of the clinical context.

• -

  The widespread use of subjective outcomes to assess the effect of behavioral interventions is a well recognized source of bias. In one study, 16% of participants who initially reported having had sexual relations at baseline stated 6 months later that they had never had any sexual intercourse.9 For this reason study designers increasingly insist on the inclusion of biological markers of sexual activity (tests to confirm STIs, pregnancy, and semen exposure, notably detection of plasma prostate antigen or Y chromosome DNA.10

• -

  Some trials report irrational contradictions: it is common to see results like those described by Lu et al.,11 in which condom use and knowledge of HIV increased significantly in intervention groups while the number of sexual partners decreased, yet the groups’ prevalences of HIV infection and syphilis did not fall or may have even risen.

These difficulties have enormous social and economic impact. This overview of systematic reviews provides a clinically relevant and coherent description of all the information available on the efficacy of interventions targeting sexual behavior. The overview is intended to aid physicians, policy-makers, educators, and patients.

Main objective: To review systematic reviews of the efficacy of behavioral interventions to prevent STIs and unintended pregnancies.

Secondary objectives: To assess differences in behavioral intervention results according to their use of objective and subjective outcome measures; to study differences in Cochrane and non-Cochrane Collaboration review findings.

As suggested in the PRISMA guidelines (listing the Preferred Reporting Items for Systematic Reviews and Meta-Analyses), we prepared a flow chart of the literature search that specifies excluded articles (see supplementary material). Of the 9765 titles initially found, we finally included 55 systematic reviews.

Our findings cannot be analyzed quantitatively, given that the interventions, measurement intervals, and populations in the studies included in the reviews were heterogeneous. Many studies were included in several systematic reviews. Table 1 shows the outcomes of behavioral interventions for each systematic review; statistical results can be found in the supplementary material. We can say that 29 of the 40 reviews (72.5%) that evaluated objective data for pregnancies and/or STIs found no positive effect of the behavioral interventions in exposed vs control subjects. Two reviews concluded that an effect could not be determined (“indeterminate”), and 9 (22.5%) detected some effect but it was generally marginal.

No clear evidence emerged in 26 of the 54 reviews that looked at whether behavioral interventions modified sexual conduct with regard to subjective (self-reported) outcomes (e.g., condom use, number of partners, age of onset of sexual relations). Thirteen reviews (24%) concluded that the results of interventions could not be determined, but 16 (29.6%) did find evidence that the behavioral interventions had had an effect.

The absence of effect was most marked in young people and adolescents, who are the population groups most often named in these systematic reviews. Only 1 review among those that examined objective outcomes detected an effect.15 Eleven reviews could find none.16–26 Among the reviews that examined subjective outcomes, 9 out of 14 found that the interventions had no effect,16,17,20,23,24,26–29 3 found a positive effect,21,30,31 and 2 could come to no conclusion (“indeterminate”).22,25

Moreover, in the systematic reviews that did find an effect, which were a minority, the effect size was small. The authors of 6 of the 9 reviews reporting an effect estimated effect sizes that were near the statistical limit of significance or found that the intervention affected only some of the outcomes.32–37

The effect size detected in the relatively few systematic reviews that found an effect on subjective outcomes was also low (see supplementary material for the full set of results). Johnson et al.38 reported a relative risk (RR) of 0.73 (95% CI, 0.63–0.85) for unprotected anal sex in the intervention group, a figure that reflected a mean overall reduction in number of unprotected sexual encounters relations (from 10.1 to 7.4 at 6 months) and a reduction in the mean number of partners from 1.2 to 0.9.

Moreno et al.39 reported a statistically significant effect for only 1 of their 16 analyses, specifically an increase in condom use during recent sexual intercourse with any type of partner (RR=1.20; 95% CI, 1.03–1.40). The statistical significance is attributable to the fact that 255 of 961 participants reported using condoms after the intervention (vs 207 of 937 in the control group).

A systematic review by Ward et al.40 reported that 1163 of 2761 participants used condoms after behavioral interventions (vs 966 of 2700 participants in the control group). These figures mean that there was a RR of 1.17 (95% CI, 1.10–1.25) for this outcome and that the interventions had not changed the self-reported behaviors of 57% of the participants.

Oringanje et al.23 found no evidence of lower rates of unintended pregnancies or STIs in a meta-analysis of 41 RCTs enrolling a total of 95 622 adolescents. They did detect a certain degree of postponement of the onset of sexual relations (RR=0.86; 95% CI, 0.77–0.96), based on finding that 308 of 844 adolescents initiated sexual relations after the behavioral intervention in contrast with 331 out of 702 in the control group. Thus, two thirds of the participants did not change their behavior, according to self reports; moreover, the pregnancy and STI rates were similar in the intervention and control groups.

We observed marked differences between the findings of studies that used objective outcome measures and those that relied on self reports. Sixteen systematic reviews found evidence that behavioral interventions are effective when outcome data were from participants’ reports (Table 1). In contrast, 10 of the 16 reviews did not detect effects of the same interventions on the basis of objective outcomes; the difference is probably attributable to self-report bias.

A trend toward absence of evidence of effect is greater in reviews that included higher-quality trials. The results of Cochrane reviews (or others that applied the same method) showed no evidence of effect on objective outcomes (13/15 reviews; 86.6%) or subjective ones (12/17; 70.5%). The remaining reviews reported uncertain results (3 references) or positive ones (2 references) (Table 2).

The tendency to find evidence of effect in poorer quality systematic reviews holds true when the results are analyzed by population subgroups receiving the behavioral interventions:

• -

  Young people, adolescents, and students: None of the Cochrane (or similar) reviews detected an effect on interventions in this population.23–25,28

• -

  African Americans: The higher quality reviews (Cochrane method, which only included RCTs with lower degrees of heterogeneity) found clearly negative outcomes of interventions.41 The other 2 systematic reviews relevant to this population included nonrandomized trials (1 in women and 1 in heterosexuals in general), finding that behavioral interventions did have an effect.34,35

• -

  Patients attended in STI units: One systematic review that only included RCTs found no statistically significant effect of interventions overall.40 In contrast, a review that included studies without controls did find an effect on both objective and subjective results.42

• -

  Drug addicts: A Cochrane review by Meader et al.43 found an absence of effect of interventions targeting risk behaviors related to STIs and HIV. When the same group published another review that included quasi-experimental studies, they did find a certain degree of evidence supporting an effect.44

We were able to ascertain that 72.5% of the systematic reviews meeting our minimum quality criteria and published since 2000 failed to find evidence that behavioral interventions modify sexual behavior in general if the outcome is evaluated based on objective measures such as serology or pregnancy tests. A higher rate of Cochrane reviews (86.6%) failed to find such evidence.

When outcomes were assessed on the basis of self-reports, 48.1% of all systematic reviews and 70.5% of Cochrane reviews failed to find evidence that outcomes were better in the behavioral intervention groups than in controls.

If we consider only reviews that did find evidence of a statistically significant effect, we find that the results are far from optimal for effective prevention. The Cochrane Collaboration urges reviewers to select interventions to study with a view to providing information of value in making decisions about whether to use a particular intervention or not.50 An effect that is discretely superior to the control intervention's effect (which in any case is a result found in only a few studies and which in absolute terms can only be seen in a third of participants) does not seem to offer an optimal way to prevent STIs or unintended pregnancies.

The results of this overview are consistent with actual epidemiological figures, which do not show improvements in STI and unintended pregnancy rates in many countries in spite of the intense implementation of various ways of imparting sex education in schools and other settings.

We observed large discrepancies in findings according to the method used to assess the effect of the behavioral interventions chosen. In 10 of the 16 systematic reviews that did find evidence that an intervention affected sexual behaviors, the results based on subjective measures contradict the results based on objective ones. This is to say that when participants in the intervention group state that they used a condom or had fewer sexual partners than participants in the control group, their actual STI or pregnancy rate did not decrease below the level of the controls’. This pattern seems to show that their sexual behaviors did not in fact change and that self-reports are unreliable.

Self-reports are a source of bias for many reasons. For example, there is the propensity to give socially desirable responses, the influence of monetary incentives, availability of free condoms, use of psychoactive substances, and simple forgetfulness.31 For members of socially vulnerable groups (such as homeless persons, prisoners, sex workers, or students) there can be a motivation to offer positive answers to the researchers’ questions.51 For these and other reasons, results based on self-reports can be very different from those based on objective biological outcomes.

A telling example of this situation can be found in the systematic review of Ross et al.,52 which carried considerable weight in the various reviews we included. A systematic review by Fonner et al.30 included 21 RCTs evaluating condom use after various behavioral interventions, but data from only 8 of the 21 RCTs could be meta-analyzed, given the authors’ inclusion criteria. Only 3 of the included RCTs reported a statistically significant effect, yet in the meta-analysis showed that the intervention did have an effect. It was attributable, they stated, to the aforementioned systematic review of Ross et al., which had included 9645 randomized young people in 20 communities in Kenya between 1999 and 2002. That trial found noteworthy increases in condom use among men exposed to the intervention (RR, 1.47) but no effect on the incidence of HIV infection, 5 other STIs, or the rate of unintended pregnancy. Furthermore, when 7040 participants were reevaluated between 2005 and 2007, the previously identified effect had been eradicated.53 Therefore, a single finding in a subjective outcome in 1 RCT accounted for the positive result that was reported in the systematic review of Fonner et al., even though no objective positive outcomes were found and even though the effect disappeared over time.

These findings are consistent with a study that found that 42% of sex workers reported not having had sexual relations or having used a condom in their most recent intercourse, yet the presence of plasma prostate antigen in a vaginal smear demonstrated semen exposure.54

Quality control tools for studies do not currently distinguish between evidence from trials that assessed self-reporters rather than biological outcomes.

We conclude that it would be useful to consider possible steps to take in the interest of making real progress that will have an impact on the population:

• 1)

  Designers of RCTs should use objective outcome measures, such as pregnancy tests or analyses for STIs, to study behavioral interventions.

• 2)

  Developers of tools to evaluate the quality of RCTs and systematic reviews should consider distinguishing between subjective and objective outcome assessment measures.

• 3)

  Educational policy-makers should promote interventions that are supported by evidence from high-quality RCTs showing a real (and large) effect based on objective measures of sexual conduct; this suggestion is consistent with recommendations from other overviews.46,55

• 4)

  We should study interventions and policies implemented to manage other health problems for which real effects have been demonstrated; we should also examine educational strategies used in social groups with lower STI and unintended pregnancy rates.

The authors declare that they have no conflicts of interest.