Background
HIV infection rates among women increased greatly in the 1990s and have remained stable since then. In 2014, women accounted for half of the 36.9 million cases of HIV infection in the world, mainly as a result of heterosexual transmission 1. Southern Africa is the epicenter of an epidemic marked by profound gender inequality, with HIV infection rates in women up to eight times higher than in men 2. In Brazil, 35% of all AIDS cases reported in 2013 were in women (13,934 out of 39,501), or a female-to-male ratio of 1:1.8 for persons living with AIDS 3.
The large number of HIV cases in women has dramatic public health consequences. Women face challenges to interrupt the HIV transmission chain, prevent mother-to-child transmission, and maintain their own health. These challenges directly affect their sexual practices. Women living with HIV are at increased risk for sexually transmitted infections (STI) 4), (5, cervical dysplasia, and genital cancer 6), (7. Concurrent STI and HIV also increase the risk of female-to-male HIV transmission during sexual intercourse 5 and the risk of mother-to-child HIV transmission by pregnant women. Furthermore, some STI during pregnancy pose potential risk of congenital problems, for example maternal syphilis 8.
Healthy sexual behaviors are important buffers against the challenges faced by women living with HIV. Protected sexual intercourse through consistent condom use has been described as the most effective way to prevent both STI and HIV transmission 9. Despite positive advances (expansion of treatment and HIV testing) showed in the "15 by 15" report 10 of the Global AIDS Response, sex protected by condom use remains a challenge. The focus on treatment as prevention has achieved recognized success; however, detectable HIV viral loads and imperfect adherence to antiretroviral therapy (ART) were not uncommon. Barrier methods such as condoms remain necessary to achieve broad reduction in HIV and other STI 11), (12, considering that sexual transmission is responsible for most of new HIV cases 9.
Among women, although the prevalence of sexual risk behavior generally declines after HIV diagnosis, a substantial group (30-40%) continues to engage in unprotected sexual intercourse 13), (14. Women are also less likely to use a condom when they have an HIV-infected partner than when the partners are HIV-negative or with unknown HIV status. Findings from a cross-sectional study including 978 women showed that 51% of unprotected sexual intercourse among HIV-infected women in the previous three months involved an HIV-infected partner (30% of unprotected sexual intercourse occurred with an HIV-uninfected partner, and 26% with a partner of unknown status) 14. Studies also reported higher levels of unprotected sex among women after ART initiation when compared to women before treatment initiation, independently of therapeutic response 15, suggesting that beliefs regarding lower levels of infectivity under ART are also associated with less frequent condom use.
In addition to these findings, it is crucial to highlight the social determinants surrounding unprotected sexual behavior in women. Gender inequities have been cited as the main reason for unprotected sexual practices among women living with HIV 10), (12), (16, including precarious socioeconomic conditions, low educational level, gendered power imbalances, intimate partner violence, and partner alcohol abuse. Reproductive intentions have also been emphasized as a reason for not using condoms 17.
In order to avoid sexual risk behavior, programs for social and behavioral change have been encouraged and are seen as an essential component of HIV prevention approaches 18. The effectiveness of such prevention programs in increasing condom use and preventing HIV infection was demonstrated in a previous meta-analysis with HIV-seronegative participants 19. Considering people living with HIV, behavioral interventions have focused on reducing sexual risk behaviors and coping with other challenges of living with HIV 20), (21. Behavioral interventions to increase antiretroviral adherence 22, disclosure of HIV diagnosis to sex partners 23, and promotion of mental health 24 have shown promising results. Regarding sexual risk behaviors, studies showed that behavioral interventions can reduce unprotected sexual practices among people living with HIV 25), (26. The effectiveness is also attested by meta-analyses, especially when intervention includes skills training 21), (27. Interventions guided by longer and more intensive behavioral theories were found to be more effective at promoting protected sexual intercourse. One-to-one interventions by healthcare providers in medical care settings were also associated with a reduction in unprotected sex 6.
Women living with HIV should be the target population for behavioral interventions, especially in countries with high HIV prevalence. These interventions must be based on a clear understanding of the social and cultural contexts shaping sexual behavior 18. Findings thus suggest that interventions addressing gender-specific issues, power imbalances, and culturally relevant information are more successful in reducing HIV transmission risk behavior among women living with HIV 12), (28), (29. Although an emerging body of evidence indicates that behavioral interventions can be effective, few studies have focused on women living with HIV. The absence of solid evidence in this area presents a gap in the field of prevention for this population. A previous meta-analysis investigated the effectiveness of behavioral interventions on promoting condom use among women living with HIV 30. The pooled analysis of five randomized controlled trials found little effect on behavioral interventions at promoting consistent condom use among HIV-positive women. Since these findings were based on a few small trials, the present study aimed to update and broaden this previous systematic review and meta-analysis 30 in order to better understand the intervention effects on protected sexual behavior in women living with HIV.
Just three years after the original systematic review was performed, several relevant studies were published. Considering the low quality of the evidence found on behavioral interventions to promote condom use among women living with HIV, an update was expected to provide a more accurate picture of these interventions and their implications for public health research and decision-making. Some methodological shortcomings in the initial review could also be addressed (e.g., being based only on a dichotomous outcome, with few studies to conduct subgroup analyses), thereby expanding the findings. Finally, this update review aimed to critically describe and discuss all relevant behavioral interventions to promote condom use among women living with HIV, regardless of whether the study provided data for inclusion in the meta-analysis.
Methods
Since the current study aimed to update a previous review, we replicated the same methodological procedures described in Carvalho et al. 30 and summarized as follows. We searched the following electronic databases regardless of language or publication status using the optimal sensitive search strategy developed by the Cochrane Collaboration 31: (1) Cochrane Central Register of Controlled Trials (CENTRAL); (2) MEDLINE (PubMed); (3) EMBASE; (4) Latin American and Caribbean Health Sciences Literature (LILACS); (5) PsycInfo; (6) Social Science Citation Index (SocINDEX); and (7) Cumulative Index to Nursing and Allied Health Literature (CINAHL) through EBSCO. The previous search covered the period between January 1980 and May 2010, and this review used specific search terms to identify relevant studies from May 2010 to July 2014. We also examined reference lists from all pertinent reviews and published studies; we searched for unpublished studies in the WHO International Clinical Trials Registry Platform and the Clinicaltrials.gov database and in the International AIDS Conferences, National HIV Prevention Conference (through the Meeting Abstracts Database - NLM Gateway), the International Society for Sexually Transmitted Disease Research, and the Conference on HIV Pathogenesis and Treatment (through the conference websites). Finally, we contacted experts in the field for recommendations on additional intervention research reports and unpublished sources.
A systematic review and meta-analysis were conducted following the Cochrane Collaboration methods 31. Studies included were randomized trials (RCT) or controlled studies (CCT) investigating HIV or STI behavioral interventions designed for people living with HIV which: (1) included HIV-positive women; (2) focused on condom use promotion; (3) presented outcomes by gender or reported having performed analyses by gender; (4) provided sufficient data to calculate effect sizes (in the meta-analysis portion); and (5) used a 3-month follow-up or more. Interventions could focus on providing information, counseling, health education, emotional well-being, skills training, coping strategies, or peer education related to HIV risk behaviors. There were no restrictions to the intervention's theoretical approach, setting, frequency, or duration. Protected sexual intercourse (or consistent condom use) was described as use of condoms in all vaginal, anal, or oral sexual relations with casual and/or steady partners. All other situations were considered inconsistent condom use (e.g., condom use "almost always" or "sometimes"). This update included a new outcome (reduction in unprotected sexual intercourse - continuous outcome).
The search was completed by one of the authors (T.R.G.). Citations retrieved from electronic searches were inspected by two authors (F.T.C. and E.R.F.) who independently screened studies for inclusion. Uncertainties were resolved by consensus. Following an initial screening, all potentially eligible studies were independently read by at least two authors (E.R.F., F.T.C., or T.R.G.) who assessed in detail the study design, types of participants, types of interventions, and outcome measures. Kappa coefficient indicated good agreement between reviewers both in the first review version (k = 0.76) and in the updated version (k = 0.69).
Using a standardized data extraction form, E.R.F., F.T.C., and T.R.G. extracted the following characteristics from each study that met the inclusion criteria: (1) description of study participants (e.g., sample size, demographic characteristics, country where the study was performed); (2) eligibility criteria for enrolment (e.g., HIV+ diagnosis); (3) details of the intervention (e.g., length of intervention and follow-up, individual or group modality, behavioral techniques); (4) assessment of risk of bias (e.g., study design, generation of allocation sequence, allocation concealment, blinding, loss to follow-up, inclusion of all randomized participants, and incomplete outcome data addressed); (5) outcome measures (e.g., acquisition of STIs or hepatitis B, self-reported protected anal, vaginal, or oral sexual intercourse); and (6) data analysis strategy.
Methodological quality was assessed through RevMan 5 (Cochrane Tech; http://tech.cochrane.org/revman/download) in accordance with the recommendations of the Cochrane Handbook of Systematic Reviews of Interventions31. Risk of bias was categorized as "low risk", "high risk", or "unclear" and was listed in risk of bias tables separated by trial. We contacted the studies' authors to obtain additional information on missing data. Finally, the GRADE system was used to evaluate the overall quality of evidence for the outcome 32), (33.
Statistical analysis was conducted according to Cochrane guidelines for systematic reviews of distinct treatments 31. For dichotomous outcomes (consistent versus inconsistent condom use), the absolute numbers of participants reporting consistent condom use in each group (intervention and control) were extracted. The results for each intervention's effect were expressed as risk ratios (RR) with 95% confidence intervals (95%CI) and were combined for meta-analysis. For the continuous outcome (reduction in unprotected sexual acts), mean and standard deviations of condom use during sexual intercourse (relative frequency) for intervention and control groups were obtained. The difference in means (MD) and respective 95%CI were used to measure the absolute difference between the means in interventions and comparison groups. Since one included study was a cluster randomized trial, we used values of the intra-class correlation coefficient reported in a previous publication related to this intervention 34 to adjust trial analysis for clustering and enter its adjusted values in the meta-analysis. Consistent condom use (dichotomous outcome) was estimated using intention-to-treat analysis and included all subjects who were randomized, regardless of baseline condom use behavior. Reduction of unprotected sexual acts (continuous outcome) was estimated through an available case analysis, and the potential impact of missing data on the results was discussed.
The meta-analysis for both dichotomous and continuous outcomes used random-effects model in the RevMan 5.3 software. This strategy accounts for potential heterogeneity following unique intervention approaches developed in diverse study settings. Statistical heterogeneity between results of different studies was examined with the χ2 test. A p-value for a χ2 test of less than 0.10 indicated heterogeneity. An alternative approach to quantify the effect of heterogeneity is assessing the inconsistency (I2) among the results of studies with 95% uncertainty intervals. A value of 0% indicates no observed heterogeneity, and a value greater than 50% indicates substantial heterogeneity. Reporting bias was assessed by examining a funnel plot graph, which can detect small trial effects, even those resulting from publication bias.
The protocol for the first systematic review was planned to conduct subgroup analyses in order to explore possible sources of heterogeneity, but which were not performed because few studies were included. In the update, subgroup analyses were feasible for the dichotomous outcome (e.g., consistent condom use). We performed such analyses taking into account interventions that exclusively targeted (versus did not exclusively target) women living with HIV, and another considering the countries where the studies were done (North America versus Africa).
Results
In the first comprehensive search for this review 30, out of 3,046 citations, 35 potentially relevant studies and their full-text versions were extracted. The search in the update was performed in July 2014 and retrieved 1,721 citations. Thirty-four intervention studies were selected for full-text review in the update, including 24 articles, 6 conference proceedings, and 4 registered trials. For the sake of comprehensiveness, we reported the results of the selection process just to the period from 2010 to 2014, when 3 new studies were included in the meta-analysis. The flowchart in Figure 1 summarizes the study selection process. The quantitative syntheses jointly describe the results from studies included in the first version (5 studies) and in the update (3 studies). Likewise, the qualitative syntheses comprised 13 studies (5 from the first version plus 8 added in the update).
Included studies
Data from 8 RCT with a total of 1,355 women living with HIV were included in the meta-analyses. Of these, 5 were conducted in the United States 29), (35), (36), (37), (38 and 3 in South Africa 39), (40), (41. Five interventions were developed exclusively for women living with HIV 29), (37), (38), (40), (41 and 3 targeted both women and men living with HIV 35), (36), (39. None targeted couples or participants' sex partners. One intervention was developed for pregnant women in sub-Saharan African communities including those living with HIV 40. All interventions followed the initiation of ART.
Even after contact with authors by e-mail, another 3 studies that met the inclusion criteria could not be included in the meta-analysis, since no further information was obtained about either the absolute number of women living with HIV engaged in safe sex or the mean and standard deviation of protected sexual acts 42), (43), (44, besides another study due to lack of further information on analyses done on intervention effects by gender 45. We identified only 1 non-randomized study 46 which provided complete data on female participants, which was not included in the meta-analysis to avoid potential confounding. We decided to include the 5 interventions in the systematic review and describe the results reported narratively by the authors. All 5 studies were developed in the United States, and 3 involved interventions targeting women living with HIV 42), (43), (44. Kalichman et al. 45 conducted interventions for women and men living with HIV separately, and Fisher et al. 46 included both female and male participants in the intervention or comparison groups.
Considering the 13 studies included in the systematic review, most involved group-level interventions 29), (36), (37), (38), (41), (42), (43), (44), (45 and were delivered in settings for HIV care 29), (35), (38), (39), (42), (43), (44), (45), (46. One was home-based 40 and another 3 were developed in community settings 36,37,41. The intervention facilitators were only health workers in 6 interventions 36), (38), (39), (40), (43), (46, health workers plus peer leaders in 4 interventions 29), (41), (44), (45, and only peer leaders in 1 intervention 37. One intervention was computer-delivered 35 and another failed to report this information 42.
All interventions followed standardized procedures. Principles of cognitive-behavioral theory guided almost all of the interventions, focusing on motivational interview, stress management and coping strategies, and psycho-educational techniques. One intervention 37 involved narrative strategies and social support, and 4 added components of gender and power theories 29), (37), (41), (44. Only 6 studies explicitly mentioned that gender issues were included in the interventions 29), (37), (38), (41), (43), (44. Eight interventions focused primarily on changing sexual risk behavior and secondarily included such issues as HIV status disclosure, health care, and social support 29), (37), (39), (41), (42), (43), (44), (45. Four interventions also encompassed drug use 35), (38), (40), (46 and another covered mental health issues 36. Beyond sexual risk behavior, 1 intervention targeted infant growth and nutrition, breastfeeding, and medical adherence 40, while another aimed to improve adherence to antiretroviral therapy 38. Regarding total time of exposure to interventions, in 3 studies it was less than three hours 35), (39), (46, in 6 it ranged from 5 to 10 hours 37), (40), (42), (43), (44), (45, in another 3 it ranged from 13 to 16 hours 29), (38), (41, and in 1 study exposure lasted 22 hours and a half 36.
Table 1 shows details on methodology, participants, interventions, and outcomes for each included study.
Excluded studies
After assessment, 27 studies were excluded: 5 did not consider gender in their outcome analyses; 5 included strategies to promote condom use in control groups; 4 were not RCT or CCT studies; 4 included non-HIV-infected participants; 2 did not determine the study population's HIV status; 1 did not assess the target outcomes; 1 intervention did not aim to promote condom use; and 1 only presented results for couples. Another 4 studies were ongoing trials 47), (48), (49), (50 and may be eligible when the results become available.
Risk of bias in included studies
There was limited evidence of selection bias, selective reporting, and other biases across all the studies. It was not possible to clearly define the method used in random sequence generation 42), (45 or the allocation concealment strategy 37), (42 used in three of the studies, while another two studies used quasi-experimental designs 43), (46. All authors who were contacted responded with explanations for participants' dropout. Four studies did not report attrition 37), (41), (42), (43, however, authors of 2 studies included in the meta-analysis did describe the absolute number of participants for each target outcome 37), (41. Regarding loss to follow-up, the following studies found higher percentages of female participants dropping out before the last assessment, considering both control and intervention groups: Sikkema et al. 36 (41%); Gilbert et al. 35 (28.5%), and Saleh-Onoya et al. 41 (24.5%). The remaining studies had 15% loss to follow-up or less.
Most studies performed intent-to-treat analysis, but 3 of the included studies 37), (41), (42 and 1 that was not included 43 in the meta-analysis did not perform it. It was not possible to determine whether intent-to-treat analysis was used in another study 40. Figure 2 summarizes reviewers' assessment of the studies' methodological qualities.
Quality of the evidence
We performed a GRADE evaluation of the quality of evidence for all interventions included in this meta-analysis 32), (33. The GRADE system provided a structured approach to assess the quality of evidence across studies for each outcome included in the review according to risk of bias, publication bias, inconsistency, indirectness, and imprecision 32), (33. This classification indicated low quality of evidence for consistent condom use and very low quality of evidence for reduction in unprotected sexual acts. We downgraded the quality of evidence twice for both outcomes, based on the wide confidence intervals, indicating imprecision, and because results from different intervention protocols, with differences in duration, providers, and delivery cannot be generalized to diverse populations of women living with HIV (indirectness). We also downgraded the quality of evidence for the reduction in unprotected sexual relations, considering the inconsistency criteria, since it was not possible to conduct subgroup analyses to explore the sources of the moderate heterogeneity. The summary of findings can be accessed in Table 1 of the Supplementary Material: https://www.dropbox.com/s/tjhjhozyn8h235o/Supplementary%20Material%20reviewed.pdf?dl=0.
We specifically considered that the imprecision could also relate to the lack of sample size calculation in 3 36), (39), (41 out of 8 included studies 35), (38), (40 and to the fact that the overall sample size across studies for both outcomes (in each follow-up assessment) appears too small to estimate a relevant public health impact from behavioral interventions on condom use. Based on the results of the meta-analysis, the optimal information size for the dichotomous outcome - consistent condom use - would require 4,356 participants in each intervention arm (e.g., P1 expected proportion in the control group 555/1,103= 0.50; P2 expected proportion in the intervention group 612/1,162 = 0.53; lowest p = 0.75; difference between P1 and P2 = 0.03; alpha and beta of 0.05 and 0.20, respectively 51). Moreover, three of the studies included in the meta-analysis targeted both women and men, having lower percentages of female participants.
Assessment of publication bias
A meta-analysis can be vulnerable to publication bias if studies with less favorable results are not found. A useful test for publication bias is based on the funnel plot, which compares intervention effects estimated from individual studies against a measure of study size. In the absence of bias, the plot resembles a symmetrical inverted funnel 52. However, the test is underpowered to distinguish chance from real asymmetry when less than 10 studies are included in the meta-analysis 31, which was the case in this review. Even so, in order to explore evidence of publication bias, we performed funnel plots for both outcomes. For the dichotomous outcome, the funnel plot was quite symmetrical, while in the continuous outcome, which includes fewer studies, plot asymmetry could be observed (see Figures 1 and 2 in the Supplementary Material: https://www.dropbox.com/s/tjhjhozyn8h235o/Supplementary%20Material%20reviewed.pdf?dl=0). Considering the small number of studies included in both meta-analyses, we can hypothesize that asymmetry may be also related to true heterogeneity among studies with different sizes rather than indicating publication bias, since there were differences in the intensity of the interventions as well as among HIV+ women who lived in highly diverse contexts in terms of social and individual vulnerability.
Effects of interventions
In this updated review, we conducted meta-analyses for two different outcomes: consistent condom use and reduction in unprotected sexual intercourse, respectively. A meta-analysis of the 8 studies did not demonstrate an effect of behavioral interventions on consistent condom use among women living with HIV when compared to standard care or minimal HIV support interventions. No intervention effects on consistent condom use were noted at the 3-month follow-up (RR = 0.92; 95%CI: 0.73, 1.16; p = 0.48), 6-month follow-up (RR = 1.13; 95%CI: 0.96, 1.34; p = 0.15), and 12-month follow-up (RR = 0.91; 95%CI: 0.77, 1.08; p = 0.30) or the overall effect (RR = 1.02; 95%CI: 0.92, 1.12; p = 0.74).
The overall result for consistent condom use showed evidence of moderate heterogeneity, indicating some inconsistency across study results (τ2 = 0.08; χ2 = 19.92, df = 12, p = 0.07, I2 = 40%). As shown in Figure 3, the higher heterogeneity for consistent condom use was found at the 6 month follow-up, probably related to positive results found by one study 37. Subgroup analyses could not distinguish statistically significant differences in the intervention effect considering interventions that exclusively targeted (versus did not exclusively target) women living with HIV, or between the countries where they were conducted, i.e., in North America compared to Africa (see Figures 3, 4 and 5 in the Supplementary Material: https://www.dropbox.com/s/tjhjhozyn8h235o/Supplementary%20Material%20reviewed.pdf?dl=0).
A combination of particular characteristics of DeMarco & Chan's study 37 deserves to be mentioned, since it could be related to the positive results found in the meta-analysis. The Sistah Powah Structured Writing Intervention targeted aging, low-income, black women living with HIV, a group of women highly neglected by secondary prevention efforts because pregnancy is less likely and who face strong social vulnerability and stigma 53. The intervention was conducted in small groups and was the only one that was exclusively peer-led, involved culturally relevant narrative techniques, and helped to build a social support network in a community setting. Considering the population and the intervention's characteristics jointly, its success in promoting consistent condom use could be linked to a more effective approach to empower women living with HIV, stimulating their autonomy and the adoption of adapted strategies to deal with HIV according to gender and in socially challenging settings.

Figure 3: Effectiveness of behavioral intervention versus standard care or minimal HIV support treatment in consistent condom use.
As shown in Figure 4, the meta-analysis for reducing unprotected sexual intercourse (e.g., increased condom use - continuous outcome) showed no significant effects of the behavioral interventions when compared to standard care or minimal HIV support interventions at 6-month (MD = -1.80; 95%CI: -4.21, 0.62; p = 0.14) and 12-month follow-up (MD = -1.39; 95%CI: -2.29, 0.21; p = 0.09). However, significant heterogeneity was detected (τ2 = 1.19, χ2 = 9.37, df = 4, p = 0.05, I2 = 57%), probably reflecting the diversity of studies included in the 6-month follow-up as well as the high imprecision observed in 1 study 36. We were unable to further explore heterogeneity with subgroup analyses due to the small number of studies included for this outcome.

Figure 4: Effectiveness of behavioral intervention versus standard care or minimal HIV support treatment in reducing unprotected sexual intercourse.
Considering the statistical analyses conducted individually by the authors of the studies included in the meta-analyses, all but 1 found positive results for increased condom use 41. Eight studies considered a combined number of sexual events in which frequency rather than consistency of condom use was the primary outcome 29), (36), (39), (42), (43), (44), (45), (46. Another study originally used a composite safer sex score 37. Three other articles did not originally present data on intervention results by gender, even reporting that gender analyses were done and that no significant differences were found 35), (36), (39.
The current review also aimed to assess the effect of behavioral interventions on STI incidence, but only 2 studies 29), (41 assessed this outcome. The studies had different timing of follow-up assessments and 1 one 29 presented adequate data to be grouped by results related to STI. One study found significantly higher incidence of Chlamydia trachomatis (OR = 0.21; 95%CI: 0.07, 0.59; p < 0.05), Neisseria gonorrhoeae (OR = 0.10; 95%CI: 0.02, 0.49; p < 0.05), and Trichomona vaginalis (OR = 0.06; 95%CI: 0.01, 0.46; p < 0.05) in the control group than in the intervention at 3 months follow-up 41. No significant difference was found for incidence of bacterial vaginosis. Wingood et al. 29 reported that participants in the intervention group were not significantly less likely to have a Trichomonas infection any follow-up assessment but were significantly less likely to have an bacterial STI (Chlamydia or gonorrhea) at the 12-month follow-up (OR = 0.10; 95%CI: 0.01, 0.70; p = 0.023) and over the entire 12 months (OR = 0.20; 95%CI: 0.01, 0.06; p = 0.006). These findings indicate that behavioral interventions could be a promising strategy to reduce STI incidence even if it cannot be confirmed through meta-analysis.
Another 5 studies were included in this systematic review but did not provide sufficient data to combine their results, and we present them in a narrative review as follows. One non-randomized trial of an intervention delivered jointly for women and men living with HIV included 45% of female participants in the intervention group and 39% in the comparison 46. People living with HIV in the intervention group showed a significant reduction (b = 20.51; SE = 0.23, p < 0.05) in unprotected vaginal and anal intercourse and insertive oral sex over a follow-up of 18 months, while these behaviors increased among participants in the comparison arm (b = 0.51; SE = 0.19; p = 0.01). Interventions' effects remained significant when these analyses were performed separately by gender, and the effect was higher in men (p = 0.002) than in women (p = 0.04). In a more conservative measure (unprotected vaginal and anal sex only), authors found a trend toward a reduction in unprotected sex in the intervention group over time (b = 20.42; SE = 0.25; p = 0.09), and a significant increase among standard-of-care control participants (b = 0.61; SE = 0.21; p < 0.01). Similar results were found for both outcomes (broader and conservative) when considering only sexual acts with HIV-negative partners or those with unknown HIV status.
Another RCT 45 included 60 women living with HIV (32%) in the intervention arm and 38 (26%) in the comparison arm. Since the intervention content was tailored to gender, the latter was controlled throughout the analysis, although the results were not presented separately for women and men. The results did not suggest differences between intervention and control groups for number of sexual partners reported at 3 or 6 months follow-up or for rates of unprotected intercourse at the 3-month follow-up. However, there were significant differences between conditions for rates of unprotected sexual acts (OR = 2.2; 95%CI: 0.23, 4.17) and condom use rate at the 6-month follow-up (OR = -17.1; 95%CI: -27.94, -6.26) regardless of the partner's HIV status (OR and 95%CI calculated based on the reported means and standard deviations). Considering sexual practices only with HIV-negative partners, participants in the intervention group showed significantly lower mean rates of unprotected intercourse at 3 months (OR = 0.2; 95%CI: -0.22, 0.62) and 6 months (OR = 0.8; 95%CI: -0.34, 1.94), but no differences were found in the condom use rate (OR and 95%CI calculated based on the reported means and standard deviations). Potential moderators of the intervention effects were examined and no main interactions were found, including participant's gender.
Three other interventions were developed exclusively for women living with HIV and tested in the United States 42), (43), (44. In a quasi-experimental study, Jones et al. 43 reported that at the 3-month assessment, both groups increased male condom use reaching 40% of protected sexual events in the intervention group (p < 0.001) and 34% in the comparison arm (p < 0.01). Nine months post-intervention, no changes in male condom use were observed, but at both the 3- and 9-month assessments it was not clear if a inter-group analysis had been performed. Using a RCT design and applying only bivariate analysis, Echenique et al. 42 found women living with HIV in the intervention group significantly (p < 0.05) reduced inconsistent condom use with all partners from baseline (20%) to 6-month follow-up (9.2%), while women in the control group did not. Finally, another RCT 44 reported that intervention participants were more likely to report condom use at 6 months (AOR = 17.13; 95%CI: 2.96, 99.10; p < 0.01) and 18 months (AOR = 270.04; 95%CI: 24.53, 2,971.94; p < 0.01).
Discussion
A previous systematic review and meta-analysis on behavioral interventions to promote condom use among women living with HIV was not able to demonstrate positive effects 30. Considering the need for a more accurate picture of these interventions and their implications for research and decision-making in public health, this study aimed to update and broaden the previous review, including newly published studies and adding a continuous outcome to the pooled analysis.
Concerning consistent condom use (dichotomous outcome), this updated version was unable to identify effects of behavioral interventions even after including other studies, maintaining the result of the original meta-analysis 30. As suggested previously 30, we included a less rigorous outcome in the quantitative syntheses which was intended to identify a reduction in unprotected sexual acts and an increase in the frequency of condom use. Contrary to expectations, the inclusion of a continuous outcome in the updated review did not show positive intervention effects, thus corroborating a growing body of evidence showing low efficacy of behavioral interventios to prevent STI 54. Even adding some studies in this updated version, we found that data are still limited on the effects of behavioral inteventions on condom use among women living with HIV, and are based on small studies. Thus, our results might change when findings from larger studies are included.
Recent debates have questioned whether biomedical trial methods are the most fitting for assessing behavioral inteventions 18. Behavioral change is a complex and highly context-dependent phenomenon, in contrast to biomedical outcomes which are more objective and easier to measure and respond to - for example, morbidity assessed through blood tests or self-reported symptoms. According to Joint United Nations Programme on HIV/AIDS (UNAIDS) 18, behavioral change interventions appear to be difficult to evaluate through RCTs and can be erroneously interpreted as ineffective. Still, behavioral change has been responsible for important early declines in HIV incidence in several countries with generalized epidemics 18. Promoting consistent condom use is still a central public health strategy to stop ongoing transmission and avoid negative effects of multiple sexually transmitted infections and/or superinfection in people living with HIV, especially in resource-limited settings where antiretroviral therapy is not widely accessible. The potentially false idea that behavioral interventions do not work could be overcome by carefully examining contextual aspects and designing studies with better methodological quality.
Contrary to findings in the meta-analysis, a different picture emerged when we examined the qualitative syntheses. Different statistical approaches conducted individually by each study showed positive results in all but one intervention 41, either considering consistent condom use and/or reducing the proportion of unprotected sexual acts. These discordant findings are probably related to the intention-to-treat approach assumed in the pooled analyses instead of the per-protocol analysis adopted by most of the studies.
Regarding reduction of STI incidence in women living with HIV, two studies individually mentioned favorable effects of behavioral interventions, but the results could not be pooled in a meta-analysis. STI diagnosis is a critical public health outcome in behavioral intervention trials, since solid evidence backs the biological pathways by which STI facilitate HIV transmission and the negative effects of multiple sexually transmitted infections, especially in women living with HIV 4), (5), (53. We therefore emphasize that even small reductions in STI incidence could favor critical reductions in HIV morbidity and its associated treatment cost. Future behavioral interventions should include STI reduction among their main outcomes in order to better examine their effects.
Another key issue is that few RCTs were found with available data to evaluate the effects of interventions on condom use according to gender. Furthermore, behavioral interventions especially tailored for condom use promotion among women living with HIV are still scarce, which could explain why subgroup analyses could not find differences in intervention effects, based on whether they were exclusively tailored for women. Despite repeated research findings indicating the challenges faced by women living with HIV, more studies are needed to understand how gender-linked factors (e.g., skill-building to negotiate condom use, gender violence, and gender power imbalances regarding economic resources and employment) influence the effectiveness of behavioral interventions to promote condom use 16), (54), (55. A recent systematic review of behavioral interventions to improve contraceptive use for family planning among women living with HIV also found limited evidence to compare the interventions' results 56. Among seven included studies, all in Africa, new use of condoms was more likely in women attending services delivering multi-level and integrated care compared to those in routine care in one study. The authors highlighted the lack of effectiveness of delivering standardized information compared to interventions focused on the specific needs and concerns of women living with HIV regarding family planning.
In addition, none of the included studies explicitly mentioned promotion of female condom use, implying that male condoms were the intervention target. This could be seen as an important limitation to such interventions, since male condoms are not a female-initiated contraceptive method. Therefore, interventions involving female condoms should be promoted. We also endorse and advocate interventions to promote pleasant and empowered settings for female and/or male condom use, especially among women living with HIV. Future studies should also try to better elucidate the diversity across women living with HIV, dealing with heterogeneous intervention effects, for example, among women planning to become pregnant, women with casual versus steady partners, women with seroconcordant or serodiscordant partners, and women with multiple partners. Such research would be useful for understanding how specific groups of women could take advantage of interventions to promote condom use as well as for identifying new strategies for designing interventions.
We recommend that behavioral interventions focusing on condom use promotion among women living with HIV could have greater public health impact if they were part of broad positive prevention programs delivered with a gender perspective (including women-initiated prevention options) and developing actions that involve male partners, family planning counseling, community empowerment, mental health and drug abuse treatment, and safe serostatus disclosure settings 11), (55. As seen in the qualitative syntheses, few studies included issues beyond sexual risk behavior and none approached family planning or included women's partners. We agree that without addressing social and cultural norms that shape sexual behaviors, opportunities for individual-level behavioral change are limited 18. A new approach to assessment of interventions' effect on condom use in women living with HIV should thus include mediator and moderator factors that assess broad social norms and diversified cultural contexts where sexual practices take place as well as access to specialized and primary healthcare services.
The review process was minimally biased since searches were not restricted by language and included a comprehensive search of databases, conference proceedings, and trials registers as well as contact with experts. The same search strategy and inclusion criteria were applied in both versions. Still, the present systematic review has some limitations that need to be examined. The review's first version had a specific protocol 30 while the present version did not, which could introduce bias. Even so, some of the analyses in the present version were included in the original protocol (e.g., subgroup analysis) as well as in the discussion of results in the first review (e.g., inclusion of a continuous outcome). In addition, some eligible studies 42), (43), (44), (45 lacked data for inclusion in the meta-analyses, which may have affected the effect size estimates.
Most of the studies were based in clinics where participants were receiving regular HIV care and do not accurately reflect women living with HIV as a whole. No study included in the review had an adequate number of female participants, emphasizing the importance of further research with larger female sample sizes. Meta-analysis also assumes that interventions are sufficiently similar to be combined, while behavioral interventions entail some dissimilarity, which makes it more difficult to propose strict recommendations. Finally, condom use is a multi-causal outcome demanding multimodal and multilevel interventions 18. Evaluating behavior change interventions requires understanding how individuals perceive HIV risk and the ability to protect themselves, including knowledge of the available choices.
After assessing data on interventions in female participants following participation in behavioral interventions, we were unable to identify either increases in consistent condom use or reduction in unprotected sex among women living with HIV. In order to capture intervention effects on condom use among women, we strongly recommend multi-method evaluations involving HIV incidence modeling, meta-analyses, and surveys on behavioral and service utilization trends, as well as ethnographical and qualitative clinical trials.