Association between maternal age at childbirth and child and adult outcomes in the offspring: A prospective study in five low-income and middle-income countries (COHORTS collaboration)

Background: Both young and advanced maternal age is associated with adverse birth and child outcomes. Few studies have examined these associations in low-income and middle-income countries (LMICs) and none have studied adult outcomes in the offspring. We aimed to examine both child and adult outcomes in five LMICs. Methods: In this prospective study, we pooled data from COHORTS (Consortium for Health Orientated Research in Transitioning Societies)-a collaboration of five birth cohorts from LMICs (Brazil, Guatemala, India, the Philippines, and South Africa), in which mothers were recruited before or during pregnancy, and the children followed up to adulthood. We examined associations between maternal age and offspring birthweight, gestational age at birth, height-for-age and weight-for-height Z scores in childhood, attained schooling, and adult height, body composition (body-mass index, waist circumference, fat, and lean mass), and cardiometabolic risk factors (blood pressure and fasting plasma glucose concentration), along with binary variables derived from these. Analyses were unadjusted and adjusted for maternal socioeconomic status, height and parity, and breastfeeding duration. Findings: We obtained data for 22 188 mothers from the five cohorts, enrolment into which took place at various times between 1969 and 1989. Data for maternal age and at least one outcome were available for 19 403 offspring (87%). In unadjusted analyses, younger (≤19 years) and older (≥35 years) maternal age were associated with lower birthweight, gestational age, child nutritional status, and schooling. After adjustment, associations with younger maternal age remained for low birthweight (odds ratio [OR] 1·18 (95% CI 1·02-1·36)], preterm birth (1·26 [1·03-1·53]), 2-year stunting (1·46 [1·25-1·70]), and failure to complete secondary schooling (1·38 [1·18-1·62]) compared with mothers aged 20-24 years. After adjustment, older maternal age remained associated with increased risk of preterm birth (OR 1·33 [95% CI 1·05-1·67]), but children of older mothers had less 2-year stunting (0·64 [0·54-0·77]) and failure to complete secondary schooling (0·59 [0·48-0·71]) than did those with mothers aged 20-24 years. Offspring of both younger and older mothers had higher adult fasting glucose concentrations (roughly 0·05 mmol/L). Interpretation: Children of young mothers in LMICs are disadvantaged at birth and in childhood nutrition and schooling. Efforts to prevent early childbearing should be strengthened. After adjustment for confounders, children of older mothers have advantages in nutritional status and schooling. Extremes of maternal age could be associated with disturbed offspring glucose metabolism. Funding: Wellcome Trust and the Bill & Melinda Gates Foundation. COHORTS (Consortium for Health Orientated Research in Transitioning Societies) is a collaboration of five birth cohorts from LMICs, in which mothers were recruited before or during pregnancy, and the children followed up to adulthood.26 In this prospective study, the cohorts include the 1982 Pelotas Cohort (Brazil); the Institute of Nutrition of Central America and Panama Nutrition Trial Cohort (Guatemala); the New Delhi Cohort (India); the Cebu Longitudinal Health and Nutrition Survey (Philippines); and the Birth to Twenty Cohort (South Africa; appendix p 1).26 All studies were approved by appropriate institutional ethics committees. Informed verbal or written consent was obtained at recruitment from the mothers in the original birth cohort studies. Informed verbal or written consent was obtained at each round of follow-up, from a parent for childhood follow-ups and from the cohort member themselves for the adult follow-up. The mother’s age at the birth of the index child was calculated from data obtained at interview before or during pregnancy, or, for South African data, from birth notification forms. Birthweight was measured by researchers (Brazil, India, and Guatemala) or obtained from hospital records (the Philippines and South Africa). Gestational age was calculated from the last menstrual period date obtained by prospective surveillance (Guatemala and India), from the mother at recruitment (the Philippines), or from medical records (Brazil and South Africa). In the Philippines, gestational age was obtained for low birthweight babies by newborn clinical assessment.27 Low birthweight was defined as <2500 g, preterm birth as gestational age <37 weeks, and smallness-for-gestational-age as birthweight below the age-specific and sex-specific 10th percentile of a US reference population.28 In all sites, post-natal weight and height were measured longitudinally with standardised methods. Measurements at 2 years were available in all sites, and 4-year measurements in all except the Philippines, where the next available age (8·5 years) was used to define so-called mid-childhood size. Height-for-age and weight-for-height were converted into Z-scores (HAZ and WHZ, respectively) with the WHO growth reference. Stunting and wasting were defined as HAZ and WHZ below −2 SDs, respectively. Breastfeeding data were recorded prospectively with different methods in each cohort; for this analysis we used duration in months of any breastfeeding (as opposed to exclusive breastfeeding), which was available for all cohorts except India. Height, weight, and waist circumference were measured with standardised techniques. Fat and fat-free mass were measured with site-specific methods, as previously described.29 In Brazil, bioelectrical impedance was measured and results corrected based on a validation study that used isotopic methods; data are only available for men. In Guatemala, weight, height, and abdominal circumference were measured and entered into a hydrostatic-weighing validated equation. The Indian and Filipino cohorts used published equations that have been validated for use in Asian populations for estimation of body fat from skinfold measures. South Africa used dual X-ray absorptiometry (Hologic Delphi, Bedford, MA, USA). Fat mass was calculated as: % body fat × body weight, and fat-free mass as weight–fat mass. Overweight was defined as a body-mass index (BMI) of 25 kg/m2 or more and obesity as BMI 30 kg/m2 or more. Blood pressure was measured seated, after a 5–10 min rest, with appropriate cuff sizes and a variety of devices, as previously described.30 High blood pressure was defined as systolic blood pressure 130 mm Hg or greater or diastolic blood pressure 85 mm Hg or greater.31 Fasting glucose was measured in all sites except Brazil, where a random sample was collected and glucose values adjusted for time since the last meal.32 Impaired fasting glucose was defined as a fasting glucose concentration of 6·1 mmol/L or greater but less than 7·0 mmol/L and diabetes as a concentration of 7·0 mmol/L or greater.33 Pregnant women were excluded from all these analyses. Socioeconomic status was considered a potential confounding factor and was assessed with five variables: maternal schooling, marital status, wealth index, urban or rural residence, and ethnic origin. Wealth index was a score derived in each cohort based on type of housing and ownership of household assets (appendix p 2). The Brazilian, Indian, and South African cohorts are urban, and the Guatemalan cohort is rural; in the mixed Filipino cohort, a so-called urbanicity index was used.34 Brazil and South Africa had white, black, Asian, and other ethnic subgroups. Maternal height was potentially both a confounder of maternal age effects (due to secular trends in height) and a mediator (due to younger mothers not having attained final height). Maternal parity and breastfeeding duration were potential mediators (older mothers tend to have higher parity and younger mothers might breastfeed for a shorter time). Parity was coded as 1, 2, 3, or 4 or more. The number of offspring for whom outcomes were available diminished with age at follow-up; for example, n=17  903 (81%) for birthweight and n=10 376 (47%) for adult blood pressure. The maternal age distribution was similar in successive waves of follow-up (appendix p 3). For each outcome, we used the maximum sample with available data. To test the representativeness of our analysis sample, we compared maternal age in those included in the analysis (with data for any childhood or adult outcome of interest) and those not included, using t tests (appendix p 4). We used maternal age as a continuous variable where possible, but used categories (≤19 years, ≥35 years, and 5-year intervening bands) to check for linearity and for tables, figures, and odds ratio calculations. Percentage of body fat and wealth index were non-normally distributed and were Fisher-Yates transformed.35 We first analysed associations between maternal age and outcomes in each cohort with multiple linear regression for continuous outcomes and multiple logistic regression for dichotomous outcomes. We assessed non-linear associations with quadratic terms. We then produced pooled analyses, including main effects for each cohort and interaction terms for cohort and control variables. We tested for heterogeneity among cohorts with F tests, comparing sums of squares explained when effects were or were not allowed to vary across sites.35 We used a sequence of regression models: (1) adjusted for sex, and adult age (adult outcomes only); (2) further adjusted for socioeconomic variables; (3) further adjusted for maternal height; (4) further adjusted for breastfeeding duration; and (5) further adjusted for parity. Missing maternal wealth and schooling values were imputed with regression analysis of known values on other socioeconomic variables. Missing maternal height values were not imputed, and a dummy variable (0=not missing; 1=missing) to represent missing value was included in regression models. The Guatemala cohort is based on a randomised controlled trial of a protein and energy supplement for pregnant women and children;26 it comprises children living in the trial villages who were born or were younger than 7 years of age at any time between 1969 and 1977. We tested for interactions between maternal age and intervention group in this cohort, but noted no consistent evidence of interactions. In Guatemala, the 2344 participants came from 768 families; in India, the 5395 came from 5313 families; there were no siblings in the other cohorts. We used linear mixed modelling to assess whether siblings affected the associations of outcomes with maternal age, and found that they made little difference (appendix p 9). We therefore present our findings without adjustment for sibships. All analyses were done with SPSS version 21 and Stata version 12. The funders had no role in the study design or conduct; the management, analysis, or interpretation of the data; the preparation, review, or approval of the report; or the decision to submit the manuscript for publication. CO and CHDF had full access to all the data and take responsibility for the integrity of the data and the accuracy of the data analysis. CHDF had the responsibility to submit the report for publication.