The initial search of our chosen four electronic databases yielded 1499 articles; of which 274 duplicates were removed. Of the remaining 1225, 1064 were excluded due to lack of relevance to peer education, the target population i.e. not HIV risk groups, or because they were reviews. Full text screening of the remaining 131 papers led to the further exclusion of 64 papers for the following reasons: lack of information on target outcomes (n=26), not a peer-led intervention (n=36), and two articles could not be downloaded. Thus, 60 studies met our predefined inclusion criteria (Figure 1). One of these 60 studies was not included in the meta-analysis because it lacked quantitative data. The characteristics of each study are detailed in Table 1. The information of the selected studies can be found at Supplemental material 1.
Of these 60 studies, 34 articles employed randomized controlled trials or quasi-experiments, and 9 articles were cohort studies. The remaining 18 studies were serial cross-sectional studies. As shown in Table 1, 25 studies were conducted in East and Southeast Asia, 9 in Central Asia, 15 in North America, eight in Africa, 7 in Europe, and one in South America with some studies conducted across two countries.
Target populations included MSM (n= 18)[17-21, 28-40], injected drug users (n= 22)[41-62], female sexual workers (n= 20)[63-82]. The included studies were undertaken between 2001 and 2009, and the population number ranged from 69 to 7,015. Mean or median age varied from 16 to 43 years. Study quality assessment scores ranged from 2 to 8, with a mean score of 5.05 out of 8 which is the most rigorous (Supplemental Table 1). According to the sensitivity analysis, the results were robust after moving each study (Supplemental Figure 1-8) and the Egger tests indicated that there was no publication bias, which are shown in the supplemental material (Supplemental Table 2).
Impact of peer education on outcome measures
Table 2 presents a summary of the pooled effect sizes for the five outcomes, including overall effects, effects stratified by the three target populations, as well as the level of a country's economic/social development.
Fifteen studies[18, 19, 22, 28, 31, 32, 36, 40, 49, 65, 77, 78, 81, 82] reported the quantitative outcomes on HIV testing with a combined study population of 12775 and two studies did not show an increase rate of HIV testing. The outcome of the random effect model suggested that the effect was significant (OR: 3.19; 95%CI: 2.13-4.79) with substantial heterogeneity across studies (I2 = 92%). This thus indicates that the peer education was able to increase the rate of HIV testing among high risk HIV groups globally (Figure 2).
Sixteen studies[22, 43-45, 47-53, 55-57, 59, 60] generated 17 discrete effect sizes on equipment sharing with a combined study population of 13,830. Although seven of the sixteen studies reported non-significant changes in equipment sharing before and after intervention, the outcome of the random effect model indicated that the overall effect was significant (OR: 0.52; 95%CI: 0.35-0.76) with substantial heterogeneity across studies (I2 =93%). The meta-analysis of these 16 articles suggested that through peer education, IDUs would reduce equipment sharing (Figure 3)
Ten studies generated[17, 20, 39, 40, 44, 53, 55-57, 72] 11 independent effect sizes on unprotected sex with a combined study population of 6289. Four of the articles showed a significant reduction in unprotected sex, while four of the articles showed a non-significant reduction. Another three studies found no changes before and after peer education intervention. The fixed effect meta-analysis model showed that peer education lowered 18% of unprotected sex among high risk groups worldwide (OR: 0.82; 95%CI: 0.72-0.94; I2 =50%) (Figure 4).
Thirty-two studies[19, 28, 29, 31, 33-38, 41, 49, 51, 54, 61, 63, 64, 66, 68-71, 73-82] reported a condom use outcome after intervention with a population of 46,130. Results across these studies were mixed but most revealed an increase in condom use, and only six of the 32 studies showed insignificant condom use after intervention. In general, after the peer education intervention, condom use among the HIV risk population increased with a combined OR of 2.66 (95%CI: 2.11-3.36; I2=90%) (Figure 5). Subgroup analyses also demonstrated significant results among FSWs, MSM and IDUs, with a pooled OR effect of 3.19 (95%CI: 2.41-4.23; I2=88%), 1.76 (95%CI:1.37-2.26; I2=72%) and 2.84 (95%CI: 1.08-7.48; I2=95) respectively, which indicated a positive effect of peer education on condom use (Supplemental figure 9-11).
Subgroup analyses were also carried out by different partner types and using patterns, revealing that peer-led intervention increased condom use both with casual sexual partners (OR: 2.79; 95%CI: 2.13-3.66; I2=84%) and regular sexual partners (OR: 2.45 ; 95%CI: 1.64-3.66; I2=95%). Considering that consistent condom use had a more profound impact on preventing HIV, we also conducted a meta-analysis and found it increased after peer education (OR: 1.80; 95%CI: 1.47-2.21; I2=86%) (Supplemental figure 12-14).
Nine studies[34, 46, 58, 62, 63, 69, 70, 73, 74] generated 10 independent effect sizes on the HIV measure with a population of 28,061. Five showed an insignificant reduction in HIV measure after prevention, and one study found an increased odds of HIV infection. However, the overall results of the meta-analysis suggested 36% lower odds of HIV measure in high risk groups (OR: 0.64; 95%CI: 0.47-0.87; I2=83%) (Figure 6).
Duration effect of peer education
Figure 7 presents the effectiveness of peer education among high risk groups for the five outcomes at different time periods.
A. Unprotected sex
The follow-up time of the articles reporting the outcome of unprotected sex was mainly within 12 months of peer education, thus we only analyzed the time effect of unprotected sex at 3, 6, and 12 months. The results highlighted a non-significant effect after 3 and 6 months, with a pooled OR of 0.68 (95% CI:0.37-1.26) and 0.93 (95% CI:0.80-1.08) respectively. However, a significant effect was found after 12 months of peer education, justifying its ability to reduce the cases of unprotected sex in the long term (OR:0.64; 95% CI:0.52-0.80).
B. Equipment sharing
Peer education had a non-significant impact on reducing equipment sharing behaviors within 12 months however the pooled odds ratio showed a downward trend. After 24 months of intervention, the combined effect was 0.32 (95%CI: 0.16-0.63), suggesting that peer education was still valid in reducing equipment sharing over a long period.
C. HIV testing
The persistent effect of peer education on HIV testing was significant with an overall OR of 6.85 after 24 months of intervention, higher than the effects after 3,6,12 months of 2.54 (95%CI: 1.35-4.78), 2.36 (95%CI: 1.21-4.60), 1.78 (95%CI: 1.26-2.52) collectively. The general increasing trend indicated that peer education had a persistent positive impact on encouraging high risk groups to get an HIV test.
D. Condom use
The overall time effect of condom use was positive but variable. During the first year after the intervention, the significant impact decreased with a pooled OR of 1.98 (95%CI:1.25-3.13), 1.81 (95%CI:1.25-2.63), 1.54 (95%CI:1.35-1.76) after 3, 6,12 months respectively. However after 12 months, the effect began to increase from 1.81 (95%CI:1.22-2.69) in 24 months to 2.65 (95%CI:1.62-4.35) in 36 months, and finally reaching 2.86 (95%CI:2.2-3.71) in 48 months, suggesting a general persistent effect of peer education on condom use.
E. HIV measure
The follow-up time of the HIV measure in the studies focused mainly on 1 to 3 years, thus so we conducted the time analysis in 12, 24, 36 months. Results indicated that after 12 months and 24 months intervention, HIV measure significantly declined with a pooled effect of 0.56 (95%CI:0.43-0.73) and 0.33 (95%CI:0.20-0.55) respectively. Although it still reduced the HIV measure after 36 months, the intervention effect was not statistically significant (OR:0.77; 95%CI:0.55-1.09), implying that the preventive effect of HIV measure may have a slight decline over time.
For the HIV test, equipment sharing and condom use meta analyses whose heterogeneity were over 90%, a multivariate meta regression was conducted. The meta regression model quantified the impact of the follow-up time, study sites and high risk groups. Supplemental table 3 highlights that the follow-up time was the source of heterogeneity for both the meta analyses of HIV testing and equipment sharing, while the study sites and high risk groups did not show significant heterogeneity among these three meta analyses. Although the heterogeneity in the meta-analysis of condom use was relatively high, we had not found the source of heterogeneity. After adjusting for the impact of follow-up time, the pooled effect of the meta-analysis was still significant, which was shown in the results of the effect of duration.