We identified 21 SLCDIs (Fig. 1; Table 2) [18–35]. Two-thirds of the studies were conducted in international settings (k = 14) [24–35], including 5 in Africa (2 in Tanzania, 2 in Cameroon, 1 in Ghana), 4 in East & Southeast Asia (2 in China, 1 in Indonesia, 1 in Thailand), 2 each in the Caribbean and Mexico, and 1 in Central America. Among the 7 studies conducted in the U.S. [18–23], two either specifically targeted (100%) or consisted of majority of African Americans (71.5%) and one specifically targeted Latinos; 2 studies consisted of majority of whites (75.4 and 51%) and the remaining 2 studies included mixed ethnic/racial groups.
Over one-third of the studies targeted commercial sex workers (CSWs) and either their clients or brothel managers (38%; k = 8) and all of these were implemented in international settings. Over one-third targeted youth and young adults (38%, k = 8), with an equal number of these studies conducted in the U.S. and international settings. The remaining studies targeted clinic patients (10%, k = 2), individuals in high STI areas (4%, k = 1), or adults in the general population (10%, k = 2). Among those studies that reported the participants’ age (k = 17), the median age across all study samples was 22 (range 15–65 years). Among the studies that reported participant gender, the majority either targeted (100%) or consisted of a majority of women (65%, k = 13). The majority of these were implemented in international settings (62%, k = 8).
Regarding intervention components, SLCDIs were implemented in the communities for an average of 10 months (range 4–24 months) for the U.S.-based studies and 15 months (range 6–42 months) for international studies. The majority of the U.S.-based studies typically addressed availability (k = 4), whereas the international studies typically implemented a combination of two or more SLCDI components within the intervention (k = 10). Additionally, the majority of both U.S. and international studies (14 out of 21) had additional ILI, GLI or CLI components. Reported outcomes of interest included condom use (k = 20), number of sex partners (k = 7), condom acquisition/condom carrying (k = 6), sexual initiation or abstinence (k = 5 youth studies), and STI infection (k = 5).
With regard to study design and quality, 10 studies (4 RCTs, 6 non-RCTs) evaluated SLCDI intervention effects by including a comparison arm not receiving a SLCDI. The median retention rate for these studies was 85.9% (range 40–99%) at the longest follow-up assessment (median: 6 months post-baseline for behavioral outcomes; 9.5 months post-baseline for biologic outcomes). All of these studies, except two, used an intent-to-treat approach for analysis (i.e., analyzed participants as originally assigned regardless of exposure). The remaining 11 studies evaluated intervention effects by comparing data from independent cross-sectional samples assessed before and after implementation of a SLCDI. For this subset, the median follow-up assessment time for behavioral outcomes was 14 months post-baseline and the median follow-up time for biologic outcomes was 17.5 months post-baseline.
Overall Effect Sizes for HIV-Risk Sex Behaviors and STI Outcomes
Before we combined effect sizes, we tested whether study design and study location (U.S. vs. international) were associated with the effect size estimates. There was no evidence of such association as significant intervention effects were observed for both U.S. and international studies and regardless of the type of study design. Thus, the effect sizes across studies were consequently aggregated.
As seen in Table 3, significant intervention effects were found for the following outcomes: increased condom use (OR = 1.81; 95% CI = 1.51, 2.17; P < .01; 20 studies; N = 23,574; Fig. 2); increased condom acquisition/condom carrying (OR = 5.40; 95% CI = 1.86, 15.66; P < .05; 6 studies; N = 3,304), delaying sexual initiation/abstinence among youth (OR = 1.43; 95% CI = 1.01, 2.03; P < .05; 5 studies; N = 6,692); and reduced incidence of STIs (OR = 0.69; 95% CI = 0.53, 0.91, P < .01; 5 studies; N = 2,196). There was no significant intervention effect observed for the number of sex partners (OR = 1.28; 95% CI = 0.89, 1.85; P > .18; 7 studies; N = 4,660). Both the I2 statistic and the Q statistic indicate considerable heterogeneity among studies. However, the sensitivity test of each outcome did not reveal any individual effect size that exerted influence on the overall heterogeneity.
To evaluate the presence of publication bias, we used linear regression methods to investigate funnel plot asymmetry. There was no evidence of publication bias for any of the relevant outcomes (all Ps > .05).
Stratified Analyses for Intervention Effects on Condom Use
To provide an overview of the pattern of findings, we present the findings on condom use from the combined data as well as findings stratified by study location (U.S.-based vs. international) in Table 4. Interventions conducted in U.S.-based and international settings were both efficacious in improving condom use behaviors (Table 4), however, significantly greater efficacy was found among interventions that were implemented in international settings (QB = 6.81, P < .05). Since the pattern of findings tended to be qualitatively similar across these settings, the summary below focuses on the combined data which can also allow for greater power in detecting an intervention effect.
The tests described below were based on a-prior hypotheses for the potential moderator effects. The I2 values for most of the stratified analyses (Table 4) remained in the moderate to high range. When comparing the I2 values reported in Table 4 and the I2 value for condom use reported in Table 3, some moderators were associated with reduced heterogeneity (e.g., combined availability, acceptability, and accessibility; specifically targeting high STD and clinic participants).
Several significant findings emerged when comparing intervention groups to comparison groups within each of the stratified analyses shown in Table 4. The significant intervention effects were seen in trials regardless of participants’ characteristics (i.e., specifically targeting youth, commercial sex workers, adults, high STD and clinic populations or males).
With regard to type of structural component (e.g., that which increased the availability, acceptability or accessibility of condoms), the marginal analyses revealed a significant positive effect on condom use for interventions that increased the availability of condoms (OR = 1.70; 95% CI = 1.39, 2.07; P < .01; k = 14). Similarly, significant positive intervention effects were also observed for interventions that increased acceptability of condoms (OR = 1.63; 95% CI = 1.33, 2.00; P < .01; k = 11) and for interventions that increased accessibility to condoms (OR = 2.30; 95% CI = 1.67, 3.17; P < .01; k = 11).
Assessing the combinations of structural components revealed a significant effect for interventions that implemented availability as a sole strategy (OR = 1.43; 95% CI = 1.21, 1.69; P < .01; k = 3), and also when availability was coupled with messages that promote the acceptability of condom use (OR = 1.64; 95% CI = 1.21, 2.21; P < .01; k = 6) or when availability was coupled with methods that increase accessibility to condoms (e.g., massive distribution of free condoms (OR = 6.68; 95% CI = 3.46, 12.90; P < .01; k = 2). Significant effects were also observed for interventions that focused on increasing accessibility to condoms as a sole strategy (OR = 2.70; 95% CI = 1.29, 5.63; P < .05; k = 4).
With regard to level of implementation (e.g., individual, organizational or environmental level), the marginal analyses revealed a significant positive effect on condom use for interventions implemented at the individual level (OR = 1.87; 95% CI = 1.55, 2.25; P < .01; k = 19). Significant positive intervention effects were also observed for interventions implemented at the organizational level (OR = 1.77; 95% CI = 1.34, 2.34; P < .01; k = 9) and for interventions implemented at the environmental level (OR = 1.87; 95% CI = 1.44, 2.42; P < .01; k = 6). When assessing the combination of the different levels of implementation, interventions implemented at the individual level only (OR = 2.01; 95% CI = 1.41, 2.88; P < .01; k = 8) or at the individual + organizational + environmental level (OR = 2.18; 95% CI = 1.28, 3.71; P < .01; k = 3) were both effective in increasing condom use behaviors.
An overall significant intervention effect was observed for solely structural-level (SLI only, with no ILI, GLI, or CLI components) interventions (OR = 1.35; 95% CI = 1.09, 1.69; P < .05; k = 7), as well as for structural-level interventions that also included either ILI/GLI (OR = 2.61; 95% CI = 1.76, 3.86; P < .05; k = 6), CLI (OR = 1.83; 95% CI = 1.72, 1.95; P < .05; k = 1), both ILI/GLI and CLI components (OR = 1.89; 95% CI = 1.07, 3.33; P < .05; k = 6), or any ILI/GLI or CLI components (OR = 2.13; 95% CI = 1.71, 2.66; P < .05; k = 13) (Table 4).
When comparing across the different subgroups of interventions to identify factors related to greater efficacy, several important factors emerged. Interventions that coupled SLCDIs with any additional ILI, GLI, or CLI components were associated with significantly greater efficacy than SLCDIs that solely implemented a structural-level component (QB = 8.10, P < .01). With regard to type of structural component, interventions that implemented availability or accessibility as sole strategies, as well as interventions that implemented availability + acceptability messages or availability + accessibility demonstrated significantly greater efficacy compared to all other types of SLCDIs (QB = 22.05, P < .01). With regard to level of implementation, SLCDIs that were implemented at the individual level or the individual + organizational + environmental level demonstrated significantly greater efficacy compared to all other types of SLCDIs (QB = 10.94, P < .05).