Study design and setting
We performed a secondary analysis of sexual health and wellbeing of adolescent and young adults project. The cross-sectional study was conducted among students aged 17-24 years in a South African university located in Eastern Cape province. We selected the university conveniently in a setting with high sexual violence and HIV prevalence . Our focus on AGYW is influenced by the fact that new HIV infection rates were disproportionately high among this cohort in South Africa. University students were recruited because of ease of accessibility and lack of funds for a household survey. The full details of the methodology for the study have been published elsewhere[45, 46]. Only unmarried male and female students aged 15 to 24 years were eligible for the study. Visiting students from another university, married students, and those aged over 24 years were ineligible for selection.
A total of 420 participants were estimated to be the appropriate sample size for this study based on ±5% precision level, a 95% confidence level, and female students' population of 4000 and 10% possible attrition, using MaCorr Sample Size Calculator. We employed stratified sampling to ensure representativeness. Stratification was based on the following characteristics of students, faculties, and years of study. We drew participants from all faculties based on probability proportionate to the size of the faculty, including social sciences and humanities (n=138), law (n=70), health sciences (55), management and commerce (92) and education (155). Students were also stratified by year of study to ensure the final sample is representative of the distribution of study by year of study. We recruited participants from their lecture halls because we were unable to get a comprehensive list of all students. Well-trained research assistants administered the survey using ODK collect installed on android devices. The research assistants were postgraduate students and were trained on using the ODK collect application, ethical considerations guiding the research, and participants' selection. We trained them to select a pre-specified number of students at a particular level and faculty of study. They were instructed to select every tenth student in the classroom and skip participants who refuse to participate. The ODK collect application ensured privacy. Participants were approached and informed about the study purpose. To minimise social desirability bias and further ensure privacy, consenting students completed the survey using either their personal mobile phones or the research assistants' mobile device in private spaces earmarked for the study on campus. No personal identifying information was collected, and they were shown how to operate the ODK App for android and navigate the survey questions. The study was conducted between June and November 2018. We conducted training of research assistants and pilot testing of study instruments among 30 students using a different university before the study commenced. The response rate was 88% among the female participants included in this study.
The University of Fort Hare ethical review body approved this study (Reference number: GON011). All participants gave written consent indicating that they voluntarily and willingly took part in the study and affirmed that they understood the study purpose, process, and usage of findings. Anonymity and confidentiality of the information provided were ensured throughout the study. We followed all the IRB guidelines for using human subjects in research.
Our dependent variable is the lifetime experience of unintended pregnancy, which we defined as becoming pregnant at a time a person is not prepared to become pregnant or intended to have children. We measured this by asking participants if they have ever been pregnant when they never wanted to get pregnant. We used a binary response of "yes" or "no" to categorise participants' responses. We followed the question up by asking what action was taken when they found that they were pregnant. The actions were classified as terminated the pregnancy and carried the pregnancy to term.
Our main independent variable of interest is sexual violence. Sexual violence is defined in this study as any sexual act or attempts to obtain a sexual act by violence or coercion by any person irrespective of their relationship to the victim. We asked respondents if they have ever experienced sexual abuse, such as forced sex or rape and touching of genitals without proper consent. We favoured a narrow definition of sexual violence in this study by excluding verbal sexual assault, which may not lead to unintended pregnancy. Also, we asked if they experienced this sexual abuse before the age of sixteen. The responses were classified as yes or no.
We included three sets of covariates, individual level, behavioural and household, and family level covariates based on existing literature[48-51]. The individual-level covariates include age, religion, and parity. Age was measured as a continuous variable by asking respondents to state their age at their last birthday. We later categorised the ages into late-adolescent girls (17-19 years) and young women (20-24 years). Religiosity was measured by asking participants to rate their level of religiosity out of ten. We classified those who rated themselves 8 to 10 as very religious, those who rated themselves from 5 to 7 as moderately religious, and 1 to 4 as not religious. Lastly, we asked respondents to state how many children they have ever had.
The behavioural factors include recreational drug use and relationship status. We asked participants to indicate if they ever used drugs such as dagga, codeine, cannabis, and/or tramadol for pleasure or to ease tension or stress. Also, we asked respondents if they are single or cohabiting. We classified their responses as yes or no. We included several family/household level factors, including family structure, family support, death of parents, communication of sexual encounters with parents, and parenting type. Family structure was classified as single-parent, polygamous, both parents, and foster family.
Family support was used as a proxy to measure parents' socioeconomic status. We asked participants to rate the level of support they received from their family as adequate, moderate, insufficient, and no support. Family support relates to financial support and not psychological support. Also, we measured communication with parents regarding sexual encounters by asking respondents if they have ever discussed sex with any or both of their parents. Lastly, we measured parenting style by asking respondents to describe their parents as strict and not strict.
The analytical sample of the current paper begins with 510 AGYW, who took part in the survey. From this number, we remove 59 respondents who refused to answer the questions on sexual behaviours, unintended pregnancy, sexual violence, and HIV testing. These sets of participants exercised their right to refuse to answer any questions they feel uncomfortable answering and to drop out of the study at any time. We examined the characteristics of these sets of respondents and did not find any significant difference between them and those included in our analysis. We, therefore, analysed data from 451 AGYW who returned with complete responses representing a response rate of 88.4% among female participants. We performed descriptive statistics of all variables included in this study. We present mean and standard deviation for age. To determine whether sexual violence was associated with a higher likelihood of unintended pregnancy, we fitted two logistic regression models. The first model was a baseline model with no covariates, which was used to estimate the unadjusted odds ratio of the association between the main independent variable and the dependent variables. The second model is a multivariable model, where we included all relevant covariates, including individual, behavioural and family, and household level factors. Alpha level of less than 0.05 was considered to be statistically significant.