Revealing a safer sex option to reduce HIV risk: a cluster-randomized trial in Botswana

The study was implemented in Botswana across four regions in a third of the nation. The regions were: Kgatleng, Kweneng, South East and Southern. These regions were selected to be representative across urban and rural settings and across socioeconomic background. South East is the region where the capital, Gaborone, is located and is most urbanized within the country.

Primary and junior secondary schools were randomized to treatment or control from a list of schools provided by the Ministry of Education. In primary schools, all grade 6 students were reached or surveyed. In junior secondary schools, all grade 8 and 9 students were reached or surveyed. Primary school students progress from grade 6 to 7 in the same school while Junior secondary school students transition to a new senior secondary school after completing grade 10. This enabled tracking of students a year later. Intervention took place at the class-level at an hour convenient to the school. In junior schools, this often occurred during study hour. The ratio of instructor to student was on average 1:30 in primary schools and 1:40 in junior schools.

Baseline data and follow-up data was collected, respectively, from August to November, 2014 and from September–December, 2015 using standard pretested data collection forms. This enabled a 12-month follow-up. A random sample of 231 schools was selected from the Ministry of Education list within each region. Sample size was determined through power calculations in Stata 15 (STATACORP. College Station, TX). Two schools initially recorded as public schools were later verified as private schools and excluded since the study focused on government schools. The remaining sample of 229 schools were randomized to intervention or control. All schools and grades were successfully followed up. The baseline sample includes 229 schools (47 junior secondary, 182 primary). The follow-up sample includes all 229 schools (47 junior secondary, 182 primary). The final sample of analysis included only girls (229 schools, n = 15,335) as summarized in Fig. 1. We analyze outcomes for girls only since they are the primary beneficiary and where the pregnancy outcome is relevant. Some differences in the samples are due to idiosyncratic factors such as daily attendance variation.

The intervention was a 1-h class revealing the relative HIV risks of older versus younger partners. At baseline, over 90% of students guessed this wrong as they thought older sex partners were safest. Thus, at the outset, the students did not realize dating same-age partners was a viable safer sex alternative. The intervention revealed this safer sex strategy through five segments: ice-breakers to generate comfort talking about sex; a puzzle activity where youth guessed in groups which age group of men was most likely to have HIV; a reveal of the true HIV graph disaggregated by age with statistics pulled from the Botswana AIDS Impact Survey (2013); a 14-min video produced by UNICEF showcasing a young girl overcoming an older partner in a rural setting; and finally, a short group discussion on the risks of intergenerational sex as well as actionable lessons learned to avoid risky older partners, including community support to draw on.

A related effect of the intervention is to change the perceived cost-benefit of dating older partners. At the outset, the benefits of older partners are known, visible and immediate, such as gifts in the form of airtime for cell phones. However, the costs, such as HIV, are underestimated and long-term. The intervention makes the HIV cost known and salient, thus potentially changing the cost-benefit of dating older partners.

Participants in the class intervention included both girls and boys. While the potential beneficiaries of the intervention were young girls, boys were considered critical counterparts and germane to the cycle of intergenerational relationships as they get older.

The intervention was adapted from a trial in Kenya which showed a reduction in pregnancy of 28% [19] to the Botswana context over the course of six to 9 months. This process included multiple pilots, focus group discussions, and contextual adaptation in partnership with the Ministry of Education, the Ministry of Health, National AIDS Coordinating Agency, the University of Botswana, the Jameel Poverty Action Lab (J-PAL), Evidence Action, the largest HIV clinic in the country (Botswana-Baylor Children’s Clinical Center of Excellence), and a youth-led NGO (Young 1ove). J-PAL also conducted the randomization and Young 1ove and Baylor enrolled participants. The class was designed to be short and cost-effective to facilitate scale-up and integration into existing curricula. A Project Review Committee (PRC) chaired by the Ministry of Education and consisting of key stakeholders across the government, UN and civil society met before the intervention and periodically thereafter to advise the trial.

Fifteen youth facilitators were trained for 5 days to deliver the intervention. The facilitators were selected out of a competitive pool of young people with experience and the ability to address sensitive topics using a youth-friendly approach. Facilitator preparation included class and pilot training. One facilitator was allocated to each classroom during intervention, ranging from 31 to 40 students on average. Random monitoring visits were conducted during implementation to provide quality assurance.

Students randomized to the control group received the standard-of-care for youth in schools, typically a life skills program delivered by the school’s Guidance and Counseling teacher. The curriculum covers sexual education and emphasizes abstinence messaging.

Schools were randomized using Stata 15 (STATACORP. College Station, TX). A random seed was set to enable replication of the randomization. The unit of randomization was the school. Randomization was stratified by region and school level (primary and secondary) to minimize the risk of contamination. All students in the designated grades (6, 8 and 9) were reached with either the intervention or a survey at baseline and a follow-up survey 12 months later. Study sensitizations were conducted with school heads to ensure community buy-in and minimize attrition. School heads were not blinded for logistical purposes and program planning. However, since schools are remote and far apart, the likelihood of cross-school contact is low. The average population density of Botswana is 3.5 people per square kilometer, the 11th lowest density country in the world.

All procedures were approved by the Botswana Ministry of Health IRB, the Botswana Ministry of Education Research Department and the MIT IRB. All data analysts have a human subject certificate. Since both the intervention and survey were non-invasive in nature, the Ministry of Education provided a waiver to collect data with verbal informed consent from students and school administrators. Prior to data collection, surveyors made clear participation was voluntary and confidential. It was explained that consent was provided by voluntarily agreeing to fill in the relevant data forms. Parent consent was sought out and obtained where appropriate. The trial was registered with the Pan African Clinical Trials Registry (PACTR201901837047199) in December 2018. The trial registry did not occur in advance of participant enrollment due to ambiguity of eligibility given the applied nature of the study.

Surveys were paper-based and self-administered in each class. The survey was a short, anonymous, two-page questionnaire where students were asked to answer questions covering school information, student demographic information and questions about knowledge, attitudes and self-reported sexual behavior. Surveyors provided a short introduction, handed out the surveys, and were available to answer questions during the survey. The survey was written in English, the language of instruction in schools after grade 4. However, instructions were given in a combination of Setswana and English to ensure comprehension. Questions were also answered in Setswana when necessary.

In addition to a paper-based survey, a roll call was taken at baseline of all students who were present. These roll call registers were taken back to schools at follow-up and read out loud. Students were marked as missing if they were absent and classmates provided one of six rationales: pregnant, dropout, absent for two or more weeks, transfer, repeating a grade or unknown. This approach was used instead of the daily school roll calls conducted by school teachers since pilots revealed these rosters were completed inconsistently.

A survey was also administered to school administrators. At primary schools, this was conducted with the guidance & counseling teacher. At junior schools, this was conducted with one of the Heads of Departments (HOD). Guidance teachers and HODs were chosen as the administrators with the most accurate knowledge of missing girls. School heads are removed from day-to-day activity, and individual teachers are only knowledgeable about their individual classrooms. Guidance teachers and HODs are most likely to have a complete picture of the girls missing and why.

The school administrator survey asked the administrator to identify all missing girls and the rationale for their being missing along the same six sub-categories as the roll call: pregnancy, dropouts, absent for more than 2 weeks, transfer, repeated or unknown.

Administrator surveys were triangulated real-time with the roll call data. The lead surveyor at each school was instructed to check if both surveys aligned. If discrepancies existed, the administrator was asked to identify missing girls picked up by the roll call.

Administrative data was further collected from the Ministry of Education on baseline enrollment by gender as well as dropouts due to pregnancy. This data was used to compare characteristics across intervention and control groups and as a component of a vector of baseline controls in the analysis.

Our primary outcome measure is pregnancy. The pregnancy measure was a maximum of pregnancies identified through the roll call and through the school administrator. This measure is an objective behavioral measure of risky sex, and is analogous to the measure used in a similar trial in Kenya enabling comparability.

Secondary outcomes were self-reported measures of sexual behavior at a 12-month follow up including: “Have you ever had sexual intercourse?” “How many people have you had sex with in the last 12 months?” and “Have you ever used a condom while having sex?” These measures were piloted in Botswana, build on a similar trial in Kenya, and relate to large-scale surveys conducted on sexual behavior such as the Demographic Health Survey and the Botswana AIDS Impact Survey.

Data were digitized from paper-based forms using Captricity software (Captricity. Oakland, CA). The data was manually checked. In only 1% of cases there were errors, such as text that was not legible by the software. These were manually corrected. Data was cleaned by a second reviewer. Intention to treat analysis was done using generalized linear multilevel models with a binomial distribution and log link at the individual level, clustering standard errors at the school level. Analyses compared the intervention and control groups at follow-up, adjusting for a vector of school-level baseline covariates to enhance statistical precision. Results are stratified by school level (primary and secondary) and by region. All statistical analysis was conducted in Stata 15 (STATACORP. College Station, TX). The analysis reported is conducted for girls, the primary beneficiary of the intervention. This provides an analogous comparison for self-reported measures of sexual behavior since the pregnancy outcome applies only to girls. The sample size fluctuates across variables given varying response rates.

Table 1 presents summary statistics by intervention group and school type for all students. The mean age of primary school students was 12.3 and 14.7 for junior secondary school students. The average class size in primary school was 31 and 40 in junior secondary school. Only 9–11% of students could identify 40-year old men as most likely to have HIV and only 42–45% of students could identify older partners as riskier in general. Faithfulness was the factor most salient in junior school in choosing a sexual partner at 50%, followed by condom use at 20%. In primary school, there was an even split between these two factors at around 30% of students choosing each. A partners’ HIV status was the top factor for roughly 14% of students. The percentage of students who had sex to date was 10–12%. Pregnancy rates at baseline were relatively low according to administrative data at .5% for junior school and nearly 0% at primary school.

Table 2 reports results at the 12-month follow up for girls. The intervention reduced pregnancy with an adjusted Relative Risk Ratio (aRRR) of .657 [95% CI .433–.997] significant the 5% level. Effects were larger at junior secondary school (aRRR = .575 [95% CI .394–.841]) and in rural areas (aRRR = .518 [95% CI .323–.831]) significant at the 1% level. There were no significant effects for primary school students. Self-reported outcomes reveal delay in sexual debut and fewer partners. Effects are largest in junior school. The biggest shift occurred in delaying sexual debut (aRRR = .786 [95% CI .641–.963]) significant at the 5% level. The second largest margin of sexual behavior change was fewer partners (aRRR = .793 [95% CI .643–.979]) significant at the 5% level. There is no significant reduction in an attempt to use a condom.

Table 3 reports exploratory analysis of results by baseline beliefs on which partner students think is most likely to have HIV before the intervention. These are school-level averages split above or below the median. We see that in schools where students think 10–19-year-olds are lower risk to begin with, there is no effect (aRRR = .972 [95% CI .707–1.336]. For schools that think 10–19-year-olds are higher risk at baseline, there is a large and significant reduction in pregnancy (aRRR = .375 [95% CI .210–.669]. This reinforces the notion that the largest impact occurs for students who initially over-estimate the risk of 10–19-year-olds. This is likely since they learn new information and update their beliefs that similar-age partners are in fact a lower risk partner choice than older partners. We see a corresponding larger effect where students think the risk of older 20–29-year-old partners is low to begin with, since they are most likely to learn older partners are riskier.