The NDBC was drawn from a population of 119,799 living in a 12 km2 area of south Delhi during 1969-72 [14,15]. 20,755 married women of reproductive age were recruited and followed regularly every other month to record menstrual dates. During recruitment, a social worker obtained information on maternal schooling and age, household structure including family income, number of family-members, ownership and type of residence, and sanitation and water supply facilities. Women who became pregnant were followed every two months initially and on alternate days from the 37th week of gestation to determine the pregnancy outcome. There were 9169 pregnancies, resulting in 8181 live births. Survival status and anthropometric measurements (length and weight) of these babies were recorded within 72 hrs of birth, at the ages of 3, 6, 9 and 12 months (±7 days) and every 6 months until 21 years by trained personnel.

Data analysis was performed using SPSS version 20.0. Student’s t-test and Chi square test were used to compare descriptive statistics between alive and dead cases. Associations of maternal age at birth with mortality were determined using Cox Proportional Hazard Model [16]. Maternal age was initially used in a continuous format and the quadratic term was used to assess the non-linear associations. Subsequently, it was divided into five groups (£19, 20-24, 25-29, 30-34 and ³35 years) with 20-24 years (maximum sample size) as the reference category.

The associations between maternal age and offspring mortality were evaluated in a stepwise manner. Crude analyses adjusted for the child’s sex, followed by adjustment for confounders, and later for additional mediators. We included only those potential confounders and mediators, which were significantly (P<0.05) different between children who survived and those who died. Confounders included for adjustment were socio-economic factors (maternal education, per capita annual household income and household wealth). Household wealth scores were derived from the 1st principal component [17] for the combination of type of housing and ownership, sanitation, water supply and crowding (number of people/room); a higher score related to better wealth. The potential mediators available and considered for additional adjustment were behavioural (place of delivery and breastfeeding status), and biological (birth weight and gestation). As breastfeeding status was relevant only for the post-perinatal deaths, it was not included. The final primary analyses models were: (i) Model 1- adjusted for sex, (ii) Model 2- adjusted for sex and socio-economic confounders (maternal education, household income and household wealth); and (iii) Model 3- adjusted for sex, socio-economic confounders and mediators (place of delivery, gestation and birth weight). A sensitivity analysis was also performed on Model 3 with additional adjustments for breastfeeding status (only for post-perinatal deaths). Linear and quadratic associations between maternal age and socio-economic confounders and mediators were also analyzed.

TABLE I Sex-wise Mortality Distribution 

All the socio-economic confounders (maternal education, household income and household wealth), and mediators (place of delivery, gestation and birthweight) had inverted U-shaped relationship with maternal age (P£0.001 for quadratic term) (Web Table II). Both younger and older age of mothers was associated with lower education, household income, wealth, birthweight and gestation, and less likely to deliver in the healthcare services. Maternal age was unrelated to breastfeeding status.

TABLE II Association Between Different Maternal Age-groups and Offspring Mortality

All deaths (stillbirths and mortality till five years of age): Table II depicts the risk of offspring mortality across the five maternal age groups. In comparison to mothers aged 20-24 years, younger (£19 years) and older (³35 years) maternal ages were associated with higher offspring mortality (stillbirth – 5 years) (HR:1.68; 95% CI 1.16, 2.43 and HR 1.48; 95% CI 1.01, 2.16, respectively). After adjustment for socio-economic confounders, this higher risk persisted for younger mothers (HR 1.51; 95% CI 1.03, 2.20) but not for older mothers (HR 0.99; 95% CI 0.66, 1.48). On further adjustment for mediators, offspring of both younger and older mothers had a higher risk of mortality (HR 2.14; 95% CI 1.25, 3.64 and HR 1.74; 95% CI 1.02, 2.97, respectively). In order to estimate the change in effect size of the association with additional confounder and mediator adjustments (which led to reductions in sample size), models 1 and 2 were run on the available sample for the fully adjusted model 3 (Fig. 1). The hazard ratios for both younger and older mothers were sequentially attenuated from the crude to the fully adjusted models. The mothers available for fully adjusted model 3 (after reduction in sample size) were comparatively educated and had marginally higher household income and wealth score.

Fig. 1 Hazard ratio for mortality across different age groups of maternal age at childbirth (a) all deaths till five years including stillbirths (Number of deaths/total sample: 156/4154); (b) Perinatal deaths (Number of deaths/total sample: 29/4154); (c) Post-perinatal deaths (Number of deaths/total sample: 127/3894). (Model 1: adjusted for sex; Model 2: further adjusted for socio-economic confounders and Model 3: further adjusted for mediators (type of delivery, gestation and birth weight). The bars represent 95% confidence interval for the hazard ratio and figures at the top of the bars are P value for significant age groups. Ref: Reference age group.

Post-perinatal or perinatal deaths: A similar pattern was found for post-perinatal mortality; the increased risk being statistically significant (P<0.05) for all three models in younger but not older mothers. The attenuation pattern was similar for perinatal deaths but the increased risk was not statistically significant. There were no instances for which the point estimate in one time interval was outside the 95% confidence for the other time interval, thereby suggesting that the effect sizes were similar or hazard was proportional in both perinatal and post-perinatal categories.

On sensitivity analyses (data not presented), the mortality risk for younger and older mothers remained similar after additional adjustments for birth-order (all three death categories) and breastfeeding (post-perinatal deaths).

In this prospective cohort study, offspring of young (<20 years) mothers had an increased risk of mortality from the perinatal period up to age five years, primarily after the early neonatal period. An apparently similar dis-advantage in older (>35 years) mothers was principally a reflection of their adverse socio-economic profile.

Persistence of a higher overall mortality risk in children of young mothers, despite adjustments for confounders and mediators, suggests a causal relationship. Similar effects were evident for post-perinatal deaths but not for perinatal mortality. This could either reflect a true biological difference or insufficient statistical power for the perinatal mortality component, which showed broadly similar associations (29-95 deaths in various models). The confounder-adjusted association for post-perinatal mortality was further attenuated after the introduction of mediators and, except breastfeeding, the other three biological and behavioural factors (place of delivery, gestation and birth weight) were significantly related to young maternal age. The increased risk appears to be partly operating through lower birth weight and gestation [6], and less utilization of health care services (home delivery). These factors; however, are of limited relevance for the stillbirth component of perinatal mortality as the event is likely to determine the birth weight, gestation and access to health care rather than the converse. In contrast, the increased overall mortality risk in older mothers was not evident after socio-economic adjustments. Older maternal age may thus not biologically predispose the offspring to higher mortality, and older mothers are also likely to be more experienced in child care practices. In a recent meta-analysis of five birth cohorts from LMIC (of which NDBC was one) children of older mothers had a higher risk of preterm birth, but had better nutritional status and schooling after similar confounder adjustment [6]. Older mothers available for the fully adjusted model 3 had higher education and wealth score, which along with a lower sample size could explain the observed statistically significant associations.

Earlier cross-sectional data, including pooled analyses from 118 demographic and health surveys conducted between 1990 and 2008 in 55 low and middle income countries (LMIC), also documented a higher risk of perinatal, neonatal, infant and under-five mortality in young mothers [8-12,18-23]. It is suggested that this risk may operate through both biological and social mechanisms. Some studies also documented a higher risk in older mothers or J or U shaped association, particularly for unadjusted models [18,24]. However, this evidence has important limitations: (a) Cross-sectional design and variation in context and time period; (b) Sub-optimal confounder adjustments; (c) Non-linear relationships have been rarely explored; and (d) Prospective data collection, to minimize bias, is mostly restricted to developed countries. Three population-based cohorts in Brazil (1982, 1993 and 2004) observed an increased risk of post-neonatal infant mortality (confounder adjusted OR 1.6; 95% CI 1.2, 2.1) in children of young (<20 years) mothers but not for stillbirths, perinatal deaths or neonatal mortality [13]. Further adjustment for mediating variables (place of delivery, gestation and birth weight) led to the disappearance of the excess of post-neonatal mortality. It was concluded that social and environmental factors may be more important than biological immaturity for this increased mortality. However, in our data, the increased risk for post-perinatal deaths persisted even after confounder and mediator adjustment, suggesting a causal relationship. These observed differences, among other factors, could relate to contextual variability, baseline mortality risk, social characteristics of young mothers, social and health care support systems and methodological differences (surveillance versus prospective cohort follow up, including or excluding mothers ³30 years and restricting outcomes to infant or under-five mortality). We thus hypothesize that young maternal age predisposes the offspring to higher post-perinatal mortality, which only partly operates through socio-economic deprivation and biological-behavioural mediators (lower birthweight and gestation, and poorer access to healthcare); the additional precise biological mechanisms need further exploration.