Assessing the Impact of an Intervention to Prevent Intimate Partner Violence and HIV in Uganda

Research Team: Rakai Health Sciences Program Data collection for the proposed dissertation analysis was done by the research teams and investigators from Rakai Health Sciences Program (RHSP) which was established in 1988 with a focus on HIV/AIDS research, including evaluation of health education and condom promotion. Over time activities have increased in nature and volume and expanded to include laboratory and clinical research, randomized trials of new prevention strategies, health professional training, expanded community services including HIV voluntary testing and counseling, provision of HIV antiretroviral therapy (ART), general and HIV-related medical care, prevention of mother-to-child transmission of HIV (PMTCT), and family planning services. In 2002, the Rakai Program became an International Center for Excellence in Research (ICER), sponsored by the National Institute of Allergy and Infectious Diseases (NIAID). The core of the RHSP is the 18-year community-based longitudinal Rakai Community Cohort Study (RCCS), within which the RHSP has implemented multiple interdisciplinary studies. Over 150 peer-reviewed papers have been generated from the RCCS and its nested studies.

Data Source: Rakai Community Cohort Study All data for the proposed analysis were derived from the Rakai Community Cohort Study (RCCS) which conducts survey interviews every 12-18 months with all consenting people aged 15-49 in ~5,000 households in 47-50 communities throughout Rakai. These 47-50 communities represent 7% of the 720 communities situated in Rakai District. The RCCS includes participants residing in trading centers and agrarian villages, representative of rural southwestern Uganda. Each RCCS participant receives a unique, life-long study identification number used to link data over time and between RHSP studies. RCCS written informed consent forms cover study participation, sample archiving for future assays including genetic testing, linkage of RCCS data to other RHSP databases, and permission to re-contact participants for other studies.

Trial Design The SHARE intervention trial used a cluster randomized controlled design where study regions rather than individuals were selected to receive (or not receive) exposure to the intervention, with all consenting adults in the community being offered treatment (intervention). The community-based approach was selected because offering a partner violence prevention intervention to individuals without involving their intimate partners would not achieve holistic results. IPV involves both members of the couple and its prevention must likewise target both.

The SHARE intervention trial was done via the Rakai Community Cohort Study (RCCS) which operates in ~50 communities which were aggregated into 11 study regions (categorized as rural or semi-urban) for a prior trial of sexually transmitted disease control and HIV transmission prevention conducted between 1994 and 1999. A previous FP outreach trial (1999 and 2002) had randomized these communities,35 and the SHARE study built on this prior randomization whereby the SHARE intervention was implemented in 4 regions (Buyamba, Kabira, Kalisizo, Katana) of the total 11 RCCS research areas. . For the current trial, four of the six FP intervention clusters were randomly assigned to the IPV intervention arm.

The process used to select regions for exposure to the SHARE trial built on a previous cluster randomized trial (CRT) that was conducted between 1999 and 2002 to assess the impact of family planning outreach via social marketing in Rakai. For the family planning CRT, 5 clusters were randomly selected to receive standard family planning services while 6 clusters were assigned to the intervention arm and received additional family planning information, counseling, and contraceptive methods from government service providers and community-based volunteer agents using social marketing and other strategies. Condom use was promoted in both control and intervention areas. Findings indicated family planning outreach via social marketing can significantly increase hormonal contraceptive use and decrease pregnancy rates. No differentials were found, however, in condom use between study arms.

In 2005, 4 of the 6 family planning intervention clusters were randomly assigned to the intervention arm of the SHARE trial. The remaining 2 family planning intervention clusters, as well as all 5 family planning control clusters were assigned to the control arm of the SHARE intervention trial, making a total of 7 control regions in the SHARE trial. The decision to allocate only 4 communities to intervention was based on budgetary constraints which precluded a larger number of intervention areas.

RCCS rounds and intervention trial implementation

Data from 2 rounds of the RCCS, R1 and R2, will be analyzed for the proposed study:

R1 = baseline, pre-intervention (Feb 2005-Jun 2006) R2 = post-intervention follow-up (Aug 2006-Apr 2008) R3 = post-intervention follow-up (Jun 2008-Nov 2009)

Baseline interviews were conducted in all 4 intervention and 7 control regions via RCCS between February 2005 and June 2006. SHARE was rolled out in the 4 intervention regions between August 2005 and November 2006. SHARE activities were introduced in each intervention region only after all baseline data collection had been completed for the entire study area. Because regions are surveyed via RCCS in the same order during each annual/semi-annual round, we estimate that each intervention community had approximately the same length of exposure to SHARE before each point of follow-up.

Independent/Dependent Variables and Confounders

Independent (exposure) variables

There are 2 independent variables for the proposed study: RCCS round and arm of the trial

1. RCCS round:
2. Pre-intervention baseline (R1)
3. Post-intervention follow-up time 2 (R2)
4. Post-intervention follow-up time 2 (R3)
5. Arm of trial
6. Intervention arm
7. Control arm Dependent (outcome) variables The primary dependent variable is past year experience of IPV (victimization among women, perpetration among men). Secondary outcome variables include number of non-marital sexual partners, condom use in the past 6 months, alcohol use at last sex and HIV incidence. All dependent variables will be determined at baseline, post-intervention time 1 and time 2. Confounders Potential confounding variables will be assessed at baseline to determine comparability between arms. The association between potential confounders and outcomes of interest will be assessed by univariate analyses, and potential covariates will include; age, sex (gender), education, marital status, religion, pregnancy-status, reproductive/fertility preferences, HIV-status, use of VCT, household socio-economic status. We will also assess the associations between sexual behaviors and IPV/HIV acquisition. Intention to Treat An intention to treat approach will be used in which all participants with evaluable data for each outcome measure will be included in the analysis by randomization arm. Retention rates will be estimated by arm. Analysis Plans The proposed statistical analysis will follow four key steps. Step 1 - Establish baseline comparability First, the characteristics of study participants will be assessed at pre-intervention (baseline) in both the control and intervention regions. Comparability of the two arms at baseline is important for designing final statistical analysis models capable of adjusting for any differences and assessing the impact of the intervention with minimal or no confounding. Pre-intervention data on outcome variables, socio-demographic and covariates will be analyzed, using descriptive statistics and graphics to display in the control and intervention communities. Exploratory analysis will include frequencies for dichotomous and categorical variables, using X2 and Fisher's exact tests for differences in proportions. Continuous variables will be assessed by means and medians using t-tests and Wilcoxon rank test. Step 2 - Investigate associations between intervention exposure and primary/secondary outcomes The second analytic step will be to investigate the associations between intervention exposure and the primary and secondary outcomes (past year IPV, condom use, alcohol use with sex, number of non-marital partners and HIV incidence) using univariate regression analyses. Covariates (characteristics or behaviors) found to be correlated with outcomes of interest in the univariate models at the p<0.15 significance levels will be included in subsequent multivariate models. Step 3 - Construct adjusted multivariate models Adjusted multivariate regression models will be constructed for outcomes found to be significantly predicted by intervention exposure. Details are provided in the subsequent section on data modeling and power calculations for the 3 aims of the proposed dissertation research. Proposed Data Modeling and Power Calculations by Research Aim AIM 1 - Intimate Partner Violence Outcome To assess the impact of SHARE + RHSP community services on report of victimization from and perpetration of physical and/or sexual IPV in the past 12 months, compared to the impact of RHSP community services alone. Statistical Modeling for Aim 1 The measures of past year experience of IPV will be analyzed separately for physical and sexual IPV and both will be modeled categorically. Three categories of physical violence will be used: (1) None (referent); (2) Moderate physical IPV; (3) Severe physical IPV. Two categories of sexual violence will be used: (1) None (referent); (1) Any sexual violence. Participants who report any experience (victimization or perpetration) of physical or sexual IPV in the past 12 month will be counted as having experienced the outcome. The proportion of each type of IPV will be tabulated by follow-up visit, and unadjusted differences between trial arms will be assessed using χ2 and Fisher's exact tests. The proportion of each type of IPV will be calculated separately for men (perpetration) and women (victimization). Behavior changes will be evaluated from baseline to follow-up (i.e. the second post-trial follow-up) by which time individuals in intervention regions had been exposed to the SHARE intervention for more than one year. Linear regression models will be fitted to estimate the marginal distributions of repeated outcomes to compare the within-individual behavior changes in relation to exposure or non-exposure to the intervention. The following equation will be used where Yit denotes the absence/presence of each type of IPV for every unit i at each time t. Yit = β0 + β1G + β2T + β3(G×T) + β4X1... + βkXk +εit Covariates included in the multivariate models will be characteristics or behaviors found to be correlated with IPV outcomes in univariate analyses at p<0.15. Models will also adjust for any statistically significant lack of comparability between the trial arms at baseline. Statistical inference for individual covariates will be based on the Wald statistic, and model fit assessed by the log likelihood ratio test (LRT). The difference between follow-up and baseline measures of IPV will be calculated separately for intervention and control regions and the difference in changes will be assessed under the null hypothesis that there is no DiD: (IPV-IntFup - IPV-IntB) = (IPV-CF - IPV-Cb). The statistical significance of the interaction coefficient, β3, will be obtained using the Wald statistic, and model fit assessed by the log likelihood ratio test (LRT). All hypothesis tests will be two sided with significance levels of 0.05. Behavioral variables found to have changed significantly during the post-trial follow-up will be further analyzed through repeated fitting of the linear model analyses stratifying by age group, type of community (rural vs. urban), and other important variables to investigate if behavior changes are associated with specific characteristics of individuals or populations. Power Calculations for Aim 1 Data on the impact of SHARE + RHSP community services on past year IPV will be analyzed to present (1) the percent change (reduction or increase) or lack thereof of reports of the IPV outcome in both groups (intervention and control) from baseline to follow-up; (2) the difference in the percent change of the IPV outcome between intervention and control groups. For Aim 1, power estimates were calculated to assess the study's power to measure percent change (reduction or increase) of IPV from baseline to follow-up, assuming there will be comparability in the percentages reported pre-intervention in both arms. Thus, study power estimates were calculated for the primary outcome of past year IPV based on (1) the set sample size for each round of data collection, (2) baseline prevalence of each type (physical and sexual) of IPV (3) assumptions regarding the expected magnitude of decline in each type of IPV that might result from the SHARE intervention combined with RHSP HIV prevention activities. Two sets of power calculations were used for physical IPV (moderate and severe combined) and sexual IPV. The sample size is fixed as per Table 3. STATA 1059 was used to estimate power to detect a minimum percent reduction, assuming a two-sided α = 0.05 with an approximate intervention population of 3,500 and an approximate control population of 11,100 at both time points. Baseline prevalence of different IPV types are based on RCCS data suggesting 20% of women experienced physical IPV and 14.5% experienced sexual IPV during the year before the survey. It is thus estimated that the study has 80% power to detect as little as a 7% reduction in the proportion of individuals reporting physical IPV and as little as a 16% reduction in the proportion of individuals reporting sexual IPV in both arms between baseline and follow-up. AIM 2 - Sexual Risk Behavior Outcomes To assess the impact of SHARE + RHSP community services on report of sexual risk behaviors, including non-marital partnerships in the past year, condom use over the past 6 months and use of alcohol before sex among men and women compared to the impact of RHSP community services alone. Statistical Modeling for Aim 2 The three measures of past year sexual risk behaviors will be analyzed separately for non-marital partnerships, condom use in the past 6 months, and alcohol use at last sex. All outcomes will be modeled categorically. Non-marital partners Number of non-marital sex partners in the past year will be modeled as a categorical outcome: (1) no non-marital partner (referent); (2) one non-marital partner; (3) two or more non-marital partners. Condom use in the past 6 months Condom in the past 6 months will be modeled categorically as: (1) no condom use (referent); (2) inconsistent use; and (3) consistent use. Alcohol use at last sex Alcohol use at last sex will be modeled categorically as: (1) no alcohol use at last sex (referent); (2) any alcohol use at last sex. The proportion of each type of sexual risk behavior reported will be tabulated by follow-up visit, and unadjusted differences between trial arms will be assessed using χ2 and Fisher's exact tests. The proportion of each type of IPV will be calculated separately for men and women and behavior changes will be evaluated from baseline to follow-up. Linear regression models will be fitted to estimate the marginal distributions of repeated outcomes67 to compare the within-individual behavior changes in relation to exposure or non-exposure to the intervention. The following equation will be used where Yit denotes the absence/presence of each type of sexual risk behavior for every unit i at each time t: Yit = β0 + β1G + β2T + β3(G×T) + β4X1... + βkXk +εit The remainder of the description of statistical methods for Aim 2 is identical to that of Aim 1. Power Calculations for Aim 2 Estimates were calculated to assess the study's power to measure percent change (reduction or increase) of each sexual risk behavior from baseline to follow-up, assuming comparability at baseline in in both arms. Again, study power estimates were calculated were based on (1) the set sample size for each round of data collection and (2) baseline prevalence, derived from Rakai data, of each type of sexual risk behavior. Three sets of power estimates were calculated. Non-marital partners: Rakai data indicated approximately 61% of men and 10% of women had more than one non-marital sex partner in the past year,49 the proposed analysis has 80% power to detect as small as a 3% reduction in men and an 11% reduction in women from time one to time two in each arm. Condom use: Data on condom use in the last six months was not available. But it is estimated that 6.5% of men and 3% of women consistently used condom use in the past year. 49 Based on these findings, it is estimated that the proposed analysis has the power to detects as small as 15% increase in consistent condom use among women and as small as a 23% increase in consistent condom use among men. Alcohol use at last sex: Previous Rakai findings suggest that 75.5% of men and 37% of women drink alcohol before sex. Based on these findings, it is estimated that the proposed analysis has the power to detect as small as 2% and 5% decrease in alcohol use with sex among men and women, respectively. AIM 3 - HIV Incidence Outcome To assess the impact of SHARE + RHSP community services on HIV incidence. Statistical Modeling for Aim 3 For Aim 3, data from all 3 rounds of RCCS will be analyzed. Person years (PY) of exposure will be cumulated from baseline to the last negative HIV result if the person remained negative, or to the midpoint of the interval between the last negative tests and first positive tests for seroconverters. HIV incidence will be estimated per 100 PY. The primary analysis will be intention-to-treat, using a log link and Poisson distribution to estimate the incidence rate ratios (IRR) of HIV acquisition. Random effects models will be used to account for intra-region correlation. All models will be adjusted by study region. In univariate analyses we will assess covariates associated with HIV acquisition at p<0.15 to determine potential confounders for inclusion in multivariate models. Models will also adjust for any statistically significant baseline differences between randomization arms. Potential covariates include terms for sex, age, sexual risk behaviors, and community baseline prevalence of HIV. Homogeneity of the treatment effect in men and women will be assessed by testing for a sex by treatment interaction. Results of these models will be used to estimate the number of HIV infections prevented by the intervention over a ~4.5 period per 100 participants. Power Calculations for Aim 3 Assuming 75% annual retention and two follow up intervals post-intervention, we estimate we will have ~4,753 PY of exposure in the intervention group and ~16,969 PY of exposure in the control group. The power to detect minimal IRRs with this person time was estimated using STATA 1059 assuming a two sided α =0.05 and an HIV incidence estimate of 1.5/100 PY among women and 1.4/100 PY among men. The study has 80% power to detect a 36% lower HIV incidence in intervention compared to control communities over time of follow-up. ETHICAL CONSIDERATIONS Several steps were taken by the RHSP to ensure the safety and confidentiality of all research participants. Because this is secondary data analysis, the main risk is breach of confidentiality if participant records, interviews or lab results are revealed to third parties. Informed consent documents are retained in locked filing cabinets and store rooms, accessible only to senior investigators or designated staff. All questionnaires and samples are identified by pre-printed study ID numbers, and names are removed. All questionnaires are stored in secure, locked facilities in the field station in Kalisizo and permanent stores in Entebbe. Only designated staff members have access to these records. All computerized data bases contain study ID numbers and the lists linking study numbers to names is kept separately in a password protected computer accessible only to senior data managers. Files of lab results are maintained in a separate safe computer file, with study ID numbers, and contain no personal identifiers. The RCCS interviews for this study were conducted in the respondents' homes or at a central location (referred to as a "hub"). Interviews were done in complete privacy in the language of Luganda by experienced same-sex interviewers using structured questionnaires. Recognizing that research on IPV raises important ethical and methodological challenges beyond those posed by general population health research, RHSP investigators followed recommendations developed by the WHO regarding the safe and ethical conduct of domestic violence research. IRB approval for the original study data was granted under two studies "Rakai Community Cohort Study" and "Assessing the impact of community-based intervention designed to reduce levels physical and sexual domestic violence in Rakai District, Uganda." The "Rakai Community Cohort Study" received IRB approval from WIRB (WIRB PRO#20031318) and the study "Assessing the impact of community-based intervention designed to reduce levels physical and sexual domestic violence in Rakai District, Uganda" received approval from the IRB of the World Health Organization (WHO.A55085). Both studies also received ethical approval from the Uganda Virus Research Institute's Science Ethics Committee and the Ugandan National Council of Science and Technology. Ethical approval for the proposed secondary analysis was received by WIRB as a sub study to the ongoing Rakai Community Cohort study (WIRB PRO#20031318). All enrollment and data collection has ended.