The study population were male and female adolescents aged 10 to 24 years, living ‘on’ and ‘of’ the street. At each study location, places where SIYPs aggregate in large numbers such as major streets, market places and motor-parks were identified through a mapping exercise conducted by the research team along with officials of the State Ministry of Health. These locations were grouped as clusters. The clusters in Lagos State were Bariga and Ajah while that for Osun State were Oke-Baale, Olaiya and Sabo.
Study recruitment procedure: Following community engagement, the study participants were recruited using respondent-driven sampling and time-location sampling methods [23, 24]. The respondent-driven sampling developed by Heckathorn [25] is a sampling process whereby participants recruit their peers, in hard-to-reach study populations [26]. In this study, the first ten seeds recruited through the respondent-driven sampling were given a labelled tag with generated identity numbers. After their enrollment, they were given two additional tags to recruit friends/peers. Each referred respondent was checked for eligibility, enrolled and interviewed once the eligibility criteria were met. The eligibility criteria were: aged 10–24 years, living ‘on’ or ‘of’ the street in Lagos or Osun State and mentally stable.
Study participant recruitment through respondent-driven sampling was slow, as there were identified boundaries within street groups and low density of social networks. The research team then adopted time-location sampling to recruit the target population at specific times, days and venues where SIYP gather [27]. To limit selection bias with this method, we ensured that the venue-day-time options with possible large turnouts of SIYP were selected for the recruitment of participants.
Through the respondent-driven sampling, 34 seeds were recruited, and 465 coupons were given out over six recruitment waves within five days. Only 120 respondents were recruited and interviewed through this method. In contrast, the time-location sampling method resulted in 1,800 coupons given, and 1,385 respondents recruited and interviewed [See Supplementary File 1 for study recruitment characteristics].
The questionnaire was administered in a place the respondent identified to be most comfortable to respond to questions. Data were collected by the field worker electronically using REDcap - a secure web application used to build and manage online surveys/databases.
Further, respondents were asked ‘if they ever attended/or were given a talk on SRH’; a positive response was assigned “1”, and otherwise “0”. The respondents were also asked to state the contraceptive method used at last sexual activity by self or partner.
The variables adjusted for in this study were selected background characteristics of respondents: age [10–14, 15–19, 20-24years], educational status [none, primary, secondary] and employment status [not working, working]. The outcome variable was ‘sexual risk behavior’ proxied by three variables: (i) inconsistent use of condom, (ii) multiple sexual partners; and (iii) transactional sex. Both male and female respondents who were sexually active were asked if they used condom at the last sexual activity. An answer in the affirmative was assigned “1”and a negative response was assigned “0”. Respondents were also asked the number of sex partners they currently had. Those who responded they had one were assigned “1” while those with more than one sex partner were assigned “0”. Also, two questions were asked to probe for transactional sex: “have you ever paid or exchange gift for sex?”, and “have you ever been paid or receive gift in exchange for sex?”. An answer in the affirmative to either of the questions was assigned “1” and a negative response was assigned “0”.
Based on the Pearson chi-square significant association test, we set the p-value cut-off point of 0.20 (p ≤ 0.20) for the inclusion of confounders in the regression model. The Hosmer-Lemeshow goodness-of-fit test was conducted to ascertain all study variables fulfilled the underlying assumption of a univariate regression. Statistical significance was considered at p-value less or equal to 0.05.