South African men aged 15 to 26 (n = 1,368) were recruited from 70 villages near Mthatha in the Eastern Cape, South Africa to participate in a cluster randomized controlled trial of the Stepping Stones HIV prevention intervention [14].
Young women and men interested in participating in the study were invited to a detailed information session where they were encouraged to ask questions. Additionally, each potential participant received a Xhosa-language leaflet describing the study using terms understandable to a lay audience. The leaflet included the phone numbers of staff and toll free helplines [14]. Participants were requested to talk with their families before committing themselves to the study.
The young people who decided to participate in the study were asked to report at an assigned time anywhere from two to seven days after the information session. At that time, each participant provided written consent and study recruitment finalized. Ethical approval for the study and the consent process was given by the University of Pretoria Ethics Committee and the University of Witwatersrand Ethics Committee.
Approximately 20 male volunteers per cluster participated and each cluster received either the Stepping Stones intervention or a short control intervention on HIV prevention. The Stepping Stones trial was conducted between March 2003 and April 2006; results have been published elsewhere [15].
Detailed data were collected on men’s violence perpetration and sexual behavior at each of three time points: baseline (T0), first follow-up (T1; N = 1,034) which occurred approximately one year after baseline, and second follow-up (T2; N = 983) which occurred approximately two years after baseline. One-thousand one-hundred eighty-seven participants (85.3%) had data from at least one follow-up time point and were included in this analysis.
At each time point, face-to-face interviews were carried out by trained male interviewers. Fathering a pregnancy after baseline, the outcome of interest, was measured at both first follow-up and second follow-up with the question “Since the previous interview have you been told by a girlfriend that you made her pregnant?” An affirmative response to this question asked at T1 was categorized as fathering an incident pregnancy during the period between the baseline and first follow up and an affirmative response at T2 was classified as an incident pregnancy occurring between first and second follow up. To eliminate any misclassification of pregnancies during the follow up period, the dates of when their girlfriend became pregnant and baseline interview dates were compared to ensure that the pregnancy started within the period after the baseline interview and before the first follow up interview and not prior to baseline. Incident T1 and T2 reports of fathering a pregnancy were combined to create a dichotomous variable that represented having ever or never fathered a pregnancy over the follow-up period.
Perpetration of intimate partner violence by men was measured using an adaptation of the WHO Violence Against Women instrument [16]. This instrument consists of subscales measuring physical abuse (5 items) and sexual violence (4 items) directed towards an intimate partner. An example from the physical abuse subscale is “Since the first interview, did you hit [name of partner] or any other girlfriend with a fist or with something else which could hurt her? Did this happen many times, a few times, once or did it not happen?” An example item from the sexual violence subscale is “Since the first interview did you physically force [name of partner] or any other girlfriend to have sex with you when she did not want to? Did this happen many times, a few times, once or did it not happen?” Men who only responded ‘once’ to all queries about frequency were classified as perpetrating violence only once because even men who endorsed multiple items could have perpetrated only one multifaceted event. All questions were asked both for the past year and ‘before the past year’. Several variables were created to categorize type and intensity of intimate partner violence. A four-level variable classified IPV as no abuse, physical abuse only, sexual abuse only, and physical and sexual abuse perpetrated. Frequency of abuse was a three level variable with no abuse, one episode of physical or sexual abuse, and more than one episode of physical and or sexual abuse. Temporality of abuse was a four level categorical variable with no abuse ever, abuse that ceased before 12 months prior to baseline, abuse that first occurred within the 12 months prior to baseline, and abuse perpetrated both before and within the 12 months prior to baseline (ongoing abuse).
An eight item scale assessed relationship control with a man’s current or most recent partner (alpha = 0.73) [17]. A typical item was “I have more to say than [NAME OF GIRLFRIEND] does about important decisions that affect us”.
Alcohol use was measured using the AUDIT scale [18]. A score of 8 or higher was considered to be indicative of problem drinking. We also asked about use of dagga (cannabis), benzene, mandrax, injected drugs, or other drugs and dichotomized responses into ever and never drug users based on a positive response to having used any one of the listed drugs.
Partner numbers were calculated by summing the responses to questions about the number of main partners, khwapheni (hidden partners concurrent with main partners) and casual or “once off” partners reported in the past year. Time since last sex, a rough proxy for coital frequency, was calculated based on the response to the question “When was the last time you had sex?” [19]. Socio-demographic measures included age and completed years of schooling. Socio-economic status was measured on a scale derived for the study encompassing household goods ownership, food scarcity and perceived difficulty accessing a fairly small (but not trivial) sum of money for a medical emergency (R100 which was about $14).
Statistical analysis
Since the original study was a stratified, two stage survey with villages sampled from predefined strata based on geographical characteristics and participants clustered within villages, initial data analyses were carried out using the survey commands in Stata 10 (Stata Corp., College Station, Texas, USA). These procedures allowed us to account for the lack of independence in the observations (non-zero, positive intra-cluster correlation) because of the sampling design. Descriptive statistics were first calculated for all variables; and two-way associations were determined between categories of IPV perpetration and fathering a pregnancy.
Random effects Poisson models were built to test the hypothesis that baseline perpetration of partner violence and relationship control predicted the incident pregnancies fathered. Four models were built to investigate the association of type of abuse, frequency of abuse, temporality of abuse, and relationship control with fathering. The perpetration of abuse at the baseline (T0) was used as the primary exposure of interest, while the primary outcome was incident fathering of a pregnancy. Each model included variables for the study treatment arm, partner numbers, time since last sex, stratum, and person years of exposure. We tested for interactions between perpetration of IPV and the intervention treatment arm, and perpetration of IPV and substance use, neither were significant. We also assessed the models for confounding by age, education, SES, having a concurrent partner, substance use and duration of primary relationship, and having fathered a pregnancy prior to baseline. Any variable found to affect the point estimate for the main exposure of interest by more than 10% was included in the final model [20]. We tested goodness of fit. We confirmed the findings of associations by modeling survival time under observation using a Weibull model, with the same sets of variables.