Inequalities in early marriage, childbearing and sexual debut among adolescents in sub-Saharan Africa

Background Adolescent sexual and reproductive health (ASRH) is a major public health concern in sub-Saharan Africa (SSA). However, inequalities in ASRH have received less attention than many other public health priority areas, in part due to limited data. In this study, we examine inequalities in key ASRH indicators. Methods We analyzed national household surveys from 37 countries in SSA, conducted during 1990–2018, to examine trends and inequalities in adolescent behaviors related to early marriage, childbearing and sexual debut among adolescents using data from respondents 15–24 years. Survival analyses were conducted on each survey to obtain estimates for the ASRH indicators. Multilevel linear regression modelling was used to obtain estimates for 2000 and 2015 in four subregions of SSA for all indicators, disaggregated by sex, age, household wealth, urban–rural residence and educational status (primary or less versus secondary or higher education). Results In 2015, 28% of adolescent girls in SSA were married before age 18, declined at an average annual rate of 1.5% during 2000–2015, while 47% of girls gave birth before age 20, declining at 0.6% per year. Child marriage was rare for boys (2.5%). About 54% and 43% of girls and boys, respectively, had their sexual debut before 18. The declines were greater for the indicators of early adolescence (10–14 years). Large differences in marriage and childbearing were observed between adolescent girls from rural versus urban areas and the poorest versus richest households, with much greater inequalities observed in West and Central Africa where the prevalence was highest. The urban–rural and wealth-related inequalities remained stagnant or widened during 2000–2015, as the decline was relatively slower among rural and the poorest compared to urban and the richest girls. The prevalence of the ASRH indicators did not decline or increase in either education categories. Conclusion Early marriage, childbearing and sexual debut declined in SSA but the 2015 levels were still high, especially in Central and West Africa, and inequalities persisted or became larger. In particular, rural, less educated and poorest adolescent girls continued to face higher ASRH risks and vulnerabilities. Greater attention to disparities in ASRH is needed for better targeting of interventions and monitoring of progress. Supplementary Information The online version contains supplementary material available at 10.1186/s12978-021-01125-8.


Data analysis
This paper is based on secondary analysis of existing data from DHS and AIS, and all analyses were done by survey for each country. The study involves mainly two components of analysis: the survival analysis done by survey for each country and linear regression analysis where data from multiple countries were pooled to describe the overall distribution of key indicators at regional and sub regional level.
Our analyses to examine inequality was informed by our conceptual framework of looking at similarities and differences by cross-cutting variables: sex of the respondent, urban vs. rural as residential area, education level and household wealth. The equity analyses involve triple disaggregation: age, sex and socioeconomic information, as well as several age-dependent indicators such as education. Education -categorized in two main groups: none or primary education as one group and secondary or higher as a second group -was based on highest level of education completed during interview [1].

Survival analysis
Survival analysis (sts function in the statistical package of Stata version 15) was performed to examine the distribution of early marriage, childbearing, as well as timing of sexual debut among adolescents and young men and women at time survey. The Kaplan-Meier failure function was also utilized to estimate the distribution of the cumulative probability of experiencing any of these key life events (i.e., first sex, marriage or childbirth) by each single year, particularly to estimate the distribution of early marriage, childbearing and timing of sexual debut before age 15, 18 and 20 as appropriate. For each study variable (marriage, birth and sexual debut) separately, the data were transformed into person-years in which respondents contributed the number of person-years they lived before experiencing the specific life event of interest. Reported age at first sex among those who ever had sex is set as the failure event (coded as 1) and those who never had sexual intercourse were censored (coded as 0) at their current age. The analysis replicated for the timing of first marriage and age at first childbirth.
Analysis was performed by survey for each gender (male vs. female) and country. These data were also used as underlying input data points to fit linear regression models in order to explore further the levels and trends of these key indicators at regional and sub-regional level, and assess gender gaps, urban-rural differences and inequalities in socioeconomic characteristics.
We used data from respondents 15-24 years of age, to strike the best balance between sample size and limiting the recall period for the first events of marriage, sexual debut and childbearing. With most events occurring at ages 16-18 years the longest recall period would be 6-8 years for the 24 year olds. With an average age of the sample of 15-24 year olds below 20 years, the recall period is short.

Multi-level linear regression model analysis
Multilevel mixed-effects linear regression analysis was used to fit the underlying data generated from the survival analysis in order to describe the overall distribution of key indicators at a regional and sub-regional level [2]. In order to decide the choice between fixed and random effects model, we performed Hausman testconstructed as a function of the difference between the two estimators of the model parameters to test whether the unique errors in the linear regression are correlated with time (or the independent variable or regressor) -and suggested that random-effects model fits the data better than fixed-effects model [3,4]. For regional estimates, country-year estimates obtained from linear regression line were used to feed into the multilevel model. All analyses were performed for each sex (male vs. female), level of stratifying variable (residence: rural vs. urban, education status: primary education or less vs. secondary+, poorest vs. richest) and indicator (marriage, childbearing and sexual debut) as appropriate.
For each linear mixed model, we used maximum likelihood to fit each model, with an independent randomeffects variance structure -which allows a distinct variance for each random effect within a random-effects equation and assumes that all variances are zero -implemented to specify the structure of the covariance matrix for the random effects in order to obtain best linear unbiased predictions of random effects and of conditional means (fitted values).
The multilevel data used for analysis has mainly two levels; it is characterized by a hierarchical structure, where we have 37 countries -each with multiple date points measured from 1990 to 2018 -nested within the four sub-regions (Central, Western, Southern and Eastern) and regions nested with the bigger region of sub-Saharan Africa. The subregional classifications used in the analysis are widely used by UNDP, UNICEF and other international organizations. In order to estimate the levels and trends at regional and sub regional levels, data from multiple countries in the region and each subregion were pooled to better describe the overall distribution of key indicators.
Our focus on subregions is based on the assumption of greater homogeneity (sociocultural and socioeconomic) between countries within a subregion than between subregions. In the analysis the time variable, which is the survey year in our particular case, was centered at sub-regional mean and incorporated into the random part of the model. We have assessed the fit of the models by using residual diagnostics and plots to check for heteroskedasticity, as well as looking at graphs.
Regional and country-level annual rate of changes (AARC) in percentages of key indicators were computed from established trends using linear regression models fitted to all available data from survival analysis generated using surveys between 1990-2018 (Appendix Table S1). The use of all available data and consideration to take country variations into account provides a more accurate estimate to measure the trends in marriage and childbearing among adolescents compared to using two end data points on year 2000 and 2015. The changing patterns were examined using a gender lens and relevant urban-rural and sociodemographic characteristics to shed light on the inequality of progress.
In order to assess trend levels, directions and associated p-values for urban-rural differences and inequalities by socioeconomic characteristics were obtained from time series linear regression coefficients. For each ASRH indicator and stratifying variable (identified as place of residence, education level and wealth tertile), data on differences in proportions -obtained from 37 countries of DHS data collected between 1990-2018 -used for analyses. Differences in proportions data were used as dependent variable over time (independent variable), and difference by stratifying variable were computed only between two extreme subpopulations in a stratifying variable, and that is between rural and urban for place of residence, primary or less education and secondary+ for education level, and poorest (33.3%) and richest (33.3%) populations for wealth tertile.
For each country-year estimate of prevalence of the period 2000-2015 that was used to feed into regional level estimates, as well as to compute AARC, four scenarios were considered to obtain better estimates of the period 2000-2015, and/or missing years in between: 1) if pre-2000 data were available, all data points generated from survival analysis using surveys between 1990-2000 were used as a priori information to establish trends over the period 2000-2015. Out of the 36 countries with at least one post-2000 survey, a total of 21 countries (58%) has at least one pre-2000 data point (Table S1).  (Table S1). 4) For missing data points between 2000 and 2015 (excluding the two extremes), linear regression was used to estimate the prevalence of county-year estimates for each sex, level of a stratifying variable and indicator. This method was applied to all countries with at least two surveys of which one is on or after 2009/2010.          Figure S1. Levels and trends of child marriage (age 18), childbearing (age 20) and sexual debut (age 18) among adolescent girls and boys sub-Saharan Africa (early child marriage and birth (age 15) among early adolescent boys was found low and are not included in the figure) Figure S2. Levels and trends child marriage (age 18), childbearing (age 20) and sexual debut (age 18) among adolescent girls in sub-Saharan Africa, disaggregated by place of residence, education and wealth tertile Figure S3. Levels and trends of early marriage (age 15), childbearing (age 15) and sexual debut (age 15) among adolescent girls, in sub-Saharan Africa, disaggregated by place of residence, education and wealth tertile Figure S4. Percentages of countries with significant reduction overtime(p<0.05): adolescent girls child marriage, childbearing, and sexual debut, 32 countries with at least two surveys, sub-Saharan Africa. Figure S5. Overall trends in urban-rural differences and inequalities by socioeconomic characteristics in the proportions of adolescent girls' child marriage, childbearing and sexual debut, 37 countries, sub-Saharan Africa. Figure S6. Trends in urban-rural differences and gaps in socioeconomic characteristics in the proportions of adolescent girls' child marriage (age 18), childbearing (age 20) and sexual debut (age 18), 37 countries, sub-Saharan Africa. Figure S7. Levels and trends in urban-rural differences and gaps in socioeconomic characteristics in the proportion of adolescent girls' child marriage, childbearing and sexual debut before age 15, 37 countries, sub-Saharan Africa. . Levels and trends of child marriage (age 18), childbearing (age 20), and sexual debut (age 18) for each country, obtained from survival analysis using DHS surveys data from 37 countries between 1990-2018, 37 countries, sub-Saharan Africa. Figure S16. Levels and trends of early marriage, childbearing and sexual debut before age 15 for each country, obtained from survival analysis using DHS surveys data collected from 37 countries between 1990-2018, 37 countries, sub-Saharan Africa.