Study design and context
We conducted a cross-sectional study as part of the baseline Indonesian arm of the longitudinal Global Early Adolescent Study (GEAS, www.geastudy.org). In Indonesia, the GEAS is part of the Explore4Action quasi-experimental trial evaluating the effect of a comprehensive sexuality education intervention on adolescent SRHR. The project is a collaboration between Rutgers, University Gadjah Mada and the Indonesia Planned Parenthood Association, with support from Johns Hopkins University and Karolinska Institutet. Baseline data were collected in 2018 at intervention and control schools across three sites: Bandar Lampung (Sumatra), Semarang (Java) and Denpasar (Bali), with two follow-ups planned for 2021 and 2022. The sites are predominately urban with different ethnic, cultural, religious and social-economic characteristics. Lampung is a multi-ethnic, Muslim majority city characterized by farming, forestry, mining, fishing, industry and retail. It is generally more Islam conservative than Semarang, which is a Muslim and Javanese majority city predominated by retail, construction and other industries. In contrast, Denpasar is mainly Balinese with Hinduism as main religion, and tourism dominating the economy .
The study received ethical approval from the Medical Research and Ethics Committee at the Faculty of Medicine, Public Health and Nursing, University Gadjah Mada, and the Johns Hopkins Bloomberg School of Public Health Institutional Review Board for secondary data analysis. We used the STROBE cross-sectional reporting guidelines to guide the development of the current manuscript .
Participants and sampling
For study inclusion, participants had to be aged 10-14 years, in grade 7 at one of the 18 selected schools and provide written parental/guardian consent and their own assent to participate. In each site, three schools were assigned to the intervention, and matched with three nearby control schools with similar characteristics. In total, 4684 boys and girls participated in the baseline survey: 1414 in Lampung, 1517 in Semarang, and 1753 in Denpasar (response rate 75.7%, 92.8% and 99%, respectively) .
Of the total sample, 2550 participants had complete data for all 18 sexual wellbeing outcomes assessed, meaning 46.5% of participants were missing one or more variables, mainly due to “don’t know” and “refuse to answer” responses. Most were missing 1 (21.3%) or 2-3 (12.6%) outcomes, wherefore we coded don’t know/refuse as “no” (0) to retain as many as possible. To assure against bias due to recoding, we conducted rigorous sensitivity analyses to compare the distribution in outcomes and covariates when restricting the sample to those with complete data on all outcomes (N=2550) versus coding don’t know/refuse as no (N=4309). The sensitivity analysis (available upon request) did not indicate any systematic differences in the sample characteristics, with the exception that everyone in the complete case had started puberty and scored higher on a couple of outcomes. This indicated that those participants might have been more experienced and comfortable responding to SRHR questions. Nonetheless, results from bivariate and multivariate analyses were similar for the two approaches. We therefore proceeded with the larger sample of N=4309.
Data for the baseline survey were collected at the schools between August and October 2018 by a team of trained data collectors using computer-assisted self-interviewing via tablets, and computer-assisted personal interviewing for adolescents with low literacy levels. The standardized survey covered sociodemographics, family, peer, school and neighborhood factors, gender attitudes, physical and mental health, puberty, empowerment, violence, romantic relationships, and SRHR. Data were uploaded to a secure server at Universitas Gadjah Mada with regular quality checks. A formal protocol ensured child protection and reporting of child abuse.
We included multiple outcome variables across different domains of sexual wellbeing, guided by the conceptual framework.
Sexual literacy and communication
HIV and pregnancy knowledge were assessed by aggregating the mean score of 10 correct responses, dichotomized into low and high SRHR knowledge. SRHR communication was assessed by combining five items related to ever having discussed body changes, sexual relationships, pregnancy/contraceptives with someone else.
Three scales developed and validated as part of the GEAS  were used to assess perceptions about gender norms, including: 1) sexual double standards (SDS); 2) gender stereotypical traits (GST); and 3) gender stereotypical roles (GSR). Response options ranged from 1 to 5 and were averaged into mean scores (higher = more gender equal attitudes) and dichotomized at the median. We also included a single item measuring agreement with gender-related teasing (e.g. ok to tease a boy who “acts like a girl”).
Comfort with body and emerging sexuality
Body satisfaction was assessed via a 4-item, 5-point scale , with mean scale scores dichotomized into low vs. high. Pubertal comfort was measured via the item “I like the fact that I am becoming a man/woman”, and feelings of guilt in relation to sexuality was assessed using a 3-item, 5-point scale (e.g. perceived guiltiness for looking at oneself naked, being attracted to someone else), with scores dichotomized into low vs. high. A single item measured agreement with whether it is normal for adolescents to be curious about love and sexuality.
Measured via two 5-point items about communication of romantic feelings vs. consent, and a 3-item scale assessed self-efficacy to prevent pregnancy with mean scores dichotomized into low vs. high.
Freedom from bullying and violence
Bullying by peers in the last 6 months was assessed using two items and categorized into no bullying, bullying by both boys/girls, and bullying by the opposite sex only. The same approach was used for physical peer violence. For bullying victims, a dichotomous variable explored whether they thought that this was due to gender or not.
These included study site, age, pubertal onset, religion, religiosity, and perceived agency, assessed using two validated 5-point scales  to measure voice and decision-making. Sex (boy, girl) was self-reported; for purposes of simplicity we use this interchangeably with gender. Three variables assessed adolescent’s relationship status and lifetime experiences of sexual activities. Family variables encompassed main caregiver, household structure and parental connectedness. School/community factors included educational aspirations, number of school days missed last month, whether the adolescent ever felt threatened at school, and access to social media.
(See Appendix Tables 1-2 for a detailed description of all measures).
Following exploratory data analysis, missing values on covariates were imputed using K-nearest neighbor (kNN). We then conducted bivariate analysis using Chi-square, Student t-test, and Wilcoxon rank-sum to compare the prevalence of outcomes by sex. Next, we selected one outcome variable from each sexual wellbeing domain for analysis in relation to covariates based on their relevance to the age group, variation in responses and validity. A series of multivariable logistic regression models were fitted for each of the five outcomes separately for boys vs. girls, guided by backward selection and Akaike’s Information Criterion (AIC). All analyses were conducted using STATA version 15.1 (StataCorp. 2017. Stata Statistical Software: Release 15. College Station, TX).