Study design and population
The paper uses data from a large cross-sectional study conducted among married and cohabiting individuals in Rakai district, south western Uganda between November 2013 and February 2014 [22]. The dataset contains 1,834 unique individuals with known HIV status. These data were merged using partner identification information to form 664 complete couples. Individuals were sampled from three study regions of differing HIV prevalence (range: 9-43%) within the Rakai Community Cohort Study (RCCS) enumeration area. The RCCS has been previously described [23, 24].
In brief, the RCCS is a population-based study with approximately annual surveys of 14,000 consenting persons aged 15-49 years, resident in 50 communities, and has been ongoing since 1994 and has been described elsewhere. Census is done prior to each survey round to identify eligible participants who are then contacted in their homes or invited to attend at central locations (“hubs”) for interview and provision of blood for HIV diagnosis. Interviews are then done to ascertain information on socio demographic characteristics, sexual behaviors and health every 12 to 18 months using structured questionnaires administered in private by same sex interviewers. The large study from which these data have been drawn was conducted within the context of the RCCS. Based on available data, the pooled estimate of HIV prevalence across the three study regions was 23.2%.
Sampling procedures
Initially, all the eleven study regions that form the RCCS enumeration areas were grouped into three categories based on HIV prevalence data from the RCCS. The decision on the lower and upper boundaries for each stratum was made by the study investigators at the time of study initiation. The lowest HIV prevalence was 9% while the highest was 43%. The study regions were thus grouped into low HIV prevalence (9-11.2%), medium prevalence (11.4-20%) or high HIV prevalence regions (21-43%). The grouping of study regions into the three strata was done in such a way as to ensure that each stratum had between 3-4 study regions. Within each stratum, one region was selected to participate in the study. Within each study region, four study communities were randomly selected using computer-generated random numbers for a total of 12 study communities. The study communities were already demarcated for their participation in the RCCS; so, there was no need for further demarcation. Residents in the selected communities who were aged 15-49 years and who were married or in a cohabiting relationship at the time of the study were eligible for inclusion in the study.
Data collection procedures and methods
Data were collected using interviewer-administered questionnaires. Data were collected on socio-demographic (age, sex, education, religion) and behavioural (condom use at last sex, non-marital sexual relationships, number of sexual partners in the past 12 months, and alcohol use before sex) characteristics. Prior to the interviews, individuals were invited to a “central hub” – a location within the community that individuals considered to be within easy reach by all participants. Individuals who did not turn up at the hub were followed up at home, and if available, they were interviewed. Data collection within each stratum took, on average, up to four weeks. All individuals gave informed written informed consent prior to participation in the study. Interviews, on average, lasted between 45-60 minutes. To ensure that individuals would be easily linked to their marital partners, partner identifying information (e.g. name) were obtained from each interviewed respondent. Individuals were then linked to their marital partners using study identifiers.
Measurement of variables
We used the term ‘married couples’ to refer to individuals who, in response to two pre-set questions (“are you currently married?”; if yes, “what type of marital union are you currently engaged in?”), responded that they were either ‘officially’ married in church or mosque; had a traditional introduction ceremony done; were married at the marriage registrar’s office; or were in a consensual union – defined as a union in which both members considered themselves as married and were also considered as ‘married’ by the community in which they lived. Our analysis focused on individuals who were in a heterosexual relationship; i.e. currently married individuals with an identifiable partner of the opposite sex. We used the term ‘marital order’ to refer to the number of times an individual has ever been married, counting from their current marital relationship. Individuals that reported that they had never been in any other marital relationship other than the current one were categorized as being in their ‘first’ union while those that had ever been in any previous relationship that ended were categorized as being in their ‘second’ or ‘third or higher-order’ union, depending on the number of times that they had been previously married. The outcome variable, the HIV infected couple, is hereby defined as a couple where one or both partners were HIV positive. An individual was classified as living as part of an HIV infected couple or HIV positive couple relationship if he or she was positive or the partner was positive status, obtained by linking individuals in a couple who were either in an HIV discordant relationship or HIV concordant relationship. A couple was defined as polygamous if a man indicated that he had more than one wife (married or cohabiting) or a woman indicated that her male partner had more than one wife.
Statistical analysis
The dependant variable was binary whether an individual is living as part of an infected couple or not. This was summarised using frequencies and percentages. Similarly, all categorical independent variables were summarized using frequencies and percentages. Unadjusted Odds ratios and their 95% Confidence intervals were used to assess the association between HIV infection and different potential risk factors at the bivariate analysis level. Only factors that had a likelihood ratio test p-value <0.02 were included in the multivariable logistic regression. Data were analysed using STATA statistical software (version 14.1).