Characteristics and birth outcomes of pregnant adolescents compared to older women: An analysis of individual level data from 140,000 mothers from 20 RCTs

Background: Adolescence is a critical period of maturation when nutrient needs are high, especially among adolescents entering pregnancy. Using individual-level data from 140,000 participants, we examined socioeconomic, nutrition, and pregnancy and birth outcomes for adolescent mothers (10–19 years) compared to older mothers in low and middle-income countries. Methods: This study was conducted between March 16, 2018 and May 25, 2021. Data were obtained from 20 randomised controlled trials of micronutrient supplementation in pregnancy. Stratified analyses were conducted by age (10–14 years, 15–17 years, 18–19 years, 20–29 years, 30–39 years, 40+ years) and geographical region (Africa, Asia). Crude and confounder-adjusted means, prevalence and relative risks of pregnancy, nutrition and birth outcomes were estimated using multivariable linear and log-binomial regression models with 95% confidence intervals. Findings: Adolescent mothers comprised 31.6% of our data. Preterm birth, small-for-gestational age (SGA), low birthweight (LBW) and newborn mortality followed a U-shaped trend in which prevalence was highest among the youngest mothers (10–14 years) and then reduced gradually, but increased again for older mothers (40+ years). When compared to mothers aged 20–29 years, there was a 23% increased risk of preterm birth, a 60% increased risk of perinatal mortality, a 63% increased risk of neonatal mortality, a 28% increased risk of LBW, and a 22% increased risk of SGA among mothers 10–14 years. Mothers 40+ years experienced a 22% increased risk of preterm birth and a 103% increased risk of stillbirth when compared to the 20–29 year group. Interpretation: The youngest and oldest mothers suffer most from adverse pregnancy and birth outcomes. Policy and programming agendas should consider both biological and socioeconomic/environmental factors when targeting these populations. Funding: Bill and Melinda Gates Foundation (Grant No: OP1137750). The Global Young Women’s Nutrition Investigators Group was established in 2016 as a voluntary global adolescent nutrition interest consortium aiming to study key health and nutritional outcomes in this population. As part of the consortium, we identified and collated individual participant data (IPD) from individually- and cluster-randomised trials of the effects of micronutrient supplementation interventions among pregnant girls and women (Table 1). A total of 20 trials with IPD for 140,000 mothers were obtained. The main analysis examined the effect of antenatal multiple micronutrient (MMN) supplementation on pregnant adolescents as compared to older mothers. This analysis is published separately and also details the systematic review process used to identify relevant randomised controlled trials (RCT), along with the process of establishing the collaboration.18 Eligibility criteria required trials to have been conducted in an LMIC and to have included at least 100 adolescents (10–19.9 years) in their sample. For the current study, we have included all 20 trials in pregnant girls and women for which we were able to obtain IPD; an acceptable approach with minimal bias even when pooling individual and cluster randomised trials.19 Pooling individual-level trial data that include adolescents is the best means of assessing health and birth outcomes within this population subset, given the absence of trials that recruit only pregnant adolescents, and data limitations with observational studies. Analyses were conducted between March 16, 2018 and May 25, 2021. Trial characteristics. All outcomes, covariates of interest, and statistical methods were specified a priori. Maternal age groups were selected and categorised into biologically meaningful subgroups for adolescents (10–14 years, 15–17 years, 18–19 years) and women (20–29 years, 30–39 years, and 40+ years), as determined by global guidance and our study expert advisory group.20,21 Outcomes and covariates of interest were selected through consensus: based on expert opinion of the consortium co-investigators, advisory panel, and trial collaborators, with consideration given to feasibility, given the time and data management resources that a longer list would necessitate. Outcomes included birthweight (grams), LBW (<2500 gs), gestational age (weeks), preterm birth (<37 weeks), SGA (<10th centile, based on Intergrowth Standards), stillbirth, perinatal mortality, neonatal mortality, maternal hemoglobin (Hb), and maternal anemia (third trimester Hb <11.0 g/dL). Selected covariates measured at enrolment included gestational age at enrolment, maternal Hb at baseline, height and weight at baseline, parity, maternal education, and residential location of participant (urban or rural). We report individual trials’ method of assessment of gestational age in Table 1. Briefly, 7 out of 20 trials used ultrasonography, while the remainder calculated gestational age using the woman's first date of her last menstrual period. To categorize underweight/overweight and low stature among adolescents (up to age 19), the World Health Organization (WHO) age and sex-specific BMI-for-age and height-for-age growth charts were used as the reference. For women above age 19, we used the following cut-offs: (i) underweight: BMI <18.5, (ii) overweight: BMI ≥25.0, (iii) low stature: <152 cm.22 Selected covariates measured post-enrolment included number of antenatal care (ANC) visits and skilled birth attendance (SBA). For all analyses, participants from control and intervention arms were pooled (to ensure the largest possible sample size) and intervention status was included as a fixed effect. Sensitivity analyses conducted in duplicate by ECK and NA of control arm estimates only showed negligible differences (within 0.001) in parameter estimates when compared to the pooled control and intervention analyses. We thus opted for analyses based on the larger pooled sample size. Summary statistics (frequencies/proportions, means/standard deviation (SD)) were calculated to examine baseline characteristics by maternal age groups. Adjusted means and prevalence of maternal and newborn outcomes were estimated with generalised linear models or log binomial regression models, respectively, with appropriate standard errors (SE) and 95% confidence intervals (CIs). Estimates were adjusted for fixed study effects, intervention given, maternal education and parity. Within-study clustering was accounted for by including trial as fixed effect in the models. All analyses were also performed stratified by region (Africa (N = 8 trials) and Asia (N = 10 trials)) when possible; 2 calcium supplementation trials were conducted outside of these geographies (in Argentina) and thus were not included in the analyses stratified by region. Some outcomes were not reported separately for Africa and Asia due to small sample sizes which led to model convergence issues. Minimum criteria for stratified analyses included: (i) for continuous outcomes, sample size ≥30 women, given the behavior and distribution of continuous outcomes which were approximately normal beyond that threshold, (ii) for birth outcomes (preterm, LBW, SGA) and maternal anemia, sample size ≥200, given that the prevalence of these conditions was high (10%−30% across trials) and (iii) for mortality outcomes (stillbirth, perinatal and neonatal mortality), sample size ≥500, given the rarity of these outcomes. Unstable estimates are denoted by red text within the tables. Applying these criteria, the 10–14 and 40+ groups were excluded entirely from analyses by region. We fitted log binomial regression models using a log link function via a single-step model with age as a categorical variable and intervention type, maternal education, and parity as covariates to estimate adjusted relative risks (RR) of age effects on outcomes. Several other covariates were considered but not included due to extensive missing data (up to 73%) or a lack of notable variation across age groups (Appendix; Table S1). We conducted a complete case analysis without imputation. Model diagnostics were consulted as appropriate and parameters were estimated with SE and 95% CIs. For all analyses, HIV-positive women and multiple births were excluded. SAS version 9.4 and Stata version 15.1 were used to conduct analyses. Ethical approval of the study was obtained through the Hospital for Sick Children's Research Ethics Board. The funders had no role in the procurement of data, access to data, or decision to submit for publication. NA, ECK, and SC had access to the study dataset and ZAB decided to submit the study for publication.