Since 1994, the Rakai Health Sciences Program has followed an open cohort of about 12,000 participants aged 15 to 49 years in 50 communities in the Rakai district of southwestern Uganda. The cohort has been described in detail elsewhere [9, 10]. In brief, participants provide consent and are interviewed privately by interviewers of the same gender every 10 to 14 months, using a standardized questionnaire. Venous blood is collected for HIV-1 testing. More than 90% of eligible individuals participate in any given survey round.
Women who participated in the study between 2000 and 2009 were included in analyses if they reported at least one sexual partner during the study period and if they had provided any data on whether they experienced IPV during the study period.
Intimate partner violence (IPV) was defined as any physical, sexual, or verbal violence by a partner in an intimate relationship. Questions on IPV were modified from the Revised Conflict Tactics Scales (CTS2), and, in some analyses, type of IPV was classified as minor or severe as per the CTS2 [11, 12]. Minor physical violence was defined as a husband or partner who had “pushed you, pulled you, slapped you or held you down,” and severe physical violence questions were having had a husband or partner who “punched you with a fist or with something that could hurt you,” “kicked you or dragged you,” “tried to strangle you or burn you,” or “attacked you with a knife, gun or other weapon.” Minor verbal violence was defined as a husband or partner having “verbally abused or shouted at you” and severe verbal violence was defined as a husband or partner who had “threatened you with a knife, gun, or other weapon.” Sexual violence was defined as a sexual partner having “used verbal threats to force you to have sex when you did not want to,” “physically forced you to have sex when you did not want to,” or “forced you to perform other sexual acts you did not want to do,” and all of these types of sexual violence were considered severe. Data were collected in all survey rounds on experiences of IPV in the current year, and in one survey round on experiences of all three forms of IPV ever. Data were collected on frequency of IPV in the past year for physical and verbal IPV in four survey rounds, and for sexual IPV in three survey rounds.
Potential risk factors for IPV were determined based on the literature and on which data were collected for the RCCS between 2000 and 2009. Since characteristics and experiences in early life may cause some of the characteristics and experiences in adulthood, and specifically, risk factors for IPV in adulthood may be on the causal pathway from early factors to IPV , risk variables were separated into early factors and contemporary factors, with early factors defined as variables that may affect women early in life (i.e. in childhood or in early adulthood) and contemporary factors defined as variables that may affect women in their current lives. These two groups of variables were analysed separately.
Early factors included sexual abuse in childhood or adolescence , age at first intercourse [14, 15], whether first intercourse was coerced, and education level [4, 6, 15–22]. Childhood or adolescent sexual abuse was defined as having ever been sexually abused by a male before the age of 18, and was only asked about in one survey round. Age at first intercourse was taken from the round in which this question was first answered. Coerced first intercourse was defined as force having been used the first time a participant had sex, and was coded as yes if the participant indicated in any survey round that their first sex had been coerced. Education level was taken from the baseline survey round, i.e. from the first round of participation during the period under study.
Contemporary factors included demographic variables such as age [4, 16–21, 23–25], marital status [4, 6, 20–25], religion [16, 19, 22, 25], occupation [4, 21, 25], partner’s occupation, and pregnancy status [5, 20, 23]; relationship variables such as type of relationship with partner , length of primary sexual partnership [20, 24], difference in age between the participant and her partner [17, 18, 20, 24, 25], number of sexual partners in the past year [14–16, 19, 21, 23, 26–28], alcohol use before sex by the participant [14, 18, 26], alcohol use before sex by the participant’s partner [6, 17, 25], attitudes toward violence, and HIV status [4, 6, 15, 17, 20, 21, 23–25, 29]. Pregnancy status was self-reported. A variable for attitudes toward violence was derived from a series of questions about whether a man is justified in beating his wife or partner in several situations, with acceptable defined as responding yes to any of these questions and not acceptable defined as responding no to all these questions, which were: she neglects household responsibilities, she disobeys the instructions of her husband/elders, she uses contraception without permission, she refuses her husband sex, he learns about his wife’s partner’s positive HIV serostatus, he learns about his positive HIV serostatus, argues over money, is unfaithful, or another reason. HIV status was defined as a positive result on two enzyme immunoassays, confirmed by Western blot or RT-PCR.
For participants who reported multiple partners in the past year, data about the partner with whom the participant reported having had sex most recently was used for the variables type of relationship with partner, alcohol use by partner, length of sexual partnership, and difference in age with partner; it was not possible to determine which specific partner (if any) had perpetrated violence.
The prevalence of IPV and of potential IPV risk factors was assessed. To identify early factors, logistic regression was used to estimate the bivariate and multivariable odds ratios (OR) and 95% confidence intervals (95% CI) associated with violence during the period of study participation. For contemporary factors, population-averaged logistic regression models were used to look at bivariate and multivariable associations , which account for repeated measures for each participant, using an exchangeable correlation matrix and a robust variance estimator, and modelling the associations between each variable and IPV in the subsequent year.
For each of early factors and contemporary factors, since there were multiple predictors of interest and to minimize the risk of Type I error of conventional backward selection models, an Allen-Cady modified backward selection procedure was used for the multivariable models . Candidate variables were identified a priori as being of greater importance on the basis of known associations with violence, including sexual abuse in childhood or adolescence, coerced first sex, and education for early factors, and age, marital status, pregnancy status, difference in age with partner, use of alcohol, number of partners in past year, and attitudes toward violence for contemporary factors [2, 27, 32]. Additional variables hypothesized to be relevant were then ranked in order of putative importance, which in ascending order of importance for contemporary factors were relationship type, length of relationship, woman’s occupation, partner’s occupation, religion, and HIV status. For early variables, this included only age at first sex. Variables from the second group were deleted in order of ascending importance, i.e. age at first sex for early factors and beginning with relationship type for contemporary factors, until the first variable was encountered with a p value of p < 0.1, either by Wald test or by likelihood ratio test, depending on whether the variable was continuous, binary, or categorical.
Separate models were run to look at the associations between contemporary factors and risk of violence in the same year, in consideration of the fact that certain associations, such as the temporal association between pregnancy and violence, might not be adequately captured when looking at exposure and outcome data from sequential years. Analyses were done using Stata 12.
During the period under study, the Rakai Community Cohort Study was approved by institutional review boards at the Scientific and Ethics Committee of the Uganda Virus Research Institute, the Uganda National Council of Science and Technology, Columbia University, Johns Hopkins University, Johns Hopkins Bloomberg School of Public Health, and the Western. Ethics approval was obtained for this analysis from the University of Toronto.