Intergenerational determinants of offspring size at birth: a life course and graphical analysis using the Aberdeen Children of the 1950s Study (ACONF)

Abstract

Background: Size at birth has taken on renewed significance due to its now well-established association with many health and health-related outcomes in both the immediate perinatal period and across the entire life course. Optimizing fetal growth to improve both neonatal survival and population health is the focus of much research and policy development, although most efforts have concentrated on either the period of pregnancy itself or the period immediately preceding it.

Methods: Intergenerational data linked to the Aberdeen Children of the 1950s (ACONF) study were used to examine the influence of grandparental and parental life course biological and social variables on the distribution of offspring size at birth. Guided stepwise multivariable methods and a graphical approach were used to assess the relative importance of these temporally ordered and highly correlated life course measures.

Results: Both distal and proximal grandparental and parental life course biological and social factors predicted offspring size at birth. Inequalities in size at birth, according to adult maternal socioeconomic indicators, were found to be largely generated by the continuity of the social environment across generations, and the inequalities in maternal early life growth were predicted by the adult grandparental social environment during the mother’s early life. Mother’s own size at birth predicted her offspring’s intrauterine growth, independent of her adult biological and social characteristics.

Conclusions: A mother’s childhood social environment and her early growth are both important predictors of her offspring’s size at birth. Population strategies aimed at optimizing size at birth require broader social and intergenerational considerations, in addition to focusing on the health of mothers in the immediate pregnancy period.

Background

Historically, size at birth has been an important marker of both individual and population health.¹ Recently, size at birth has taken on renewed significance. Smaller size at birth in particular has been associated with an increased risk for several chronic diseases some decades later, in adulthood, in addition to an increased chance of developing risk factors for these diseases.^2,3 Despite these life course associations, however, much current research designed to optimize pregnancy outcomes limits the context of determinants of size at birth to concurrently measured adult parental or pregnancy-specific characteristics. This focus ignores known intergenerational associations in size at birth and the possibility that other more temporally distal periods in a mother’s life course may influence her capacity to sustain and adequately nourish her own offspring in adulthood. Determinants of offspring size that focus on pregnancy and the immediate perinatal period are undoubtedly of clinical importance when the aim is to understand risk factors for poor intrauterine growth and optimize individual pregnancy care, however, this time-limited focus may be problematic when the intent is to understand where population-level interventions might be best directed to optimize intrauterine growth.⁴

The idea that reproductive success in adulthood is influenced by a mother’s childhood social environment and her life course growth, rather than just her attained adult characteristics, is not new. This relationship was suggested over 70 years ago when Kermack et al.⁵ postulated that the health of adults was largely determined by their health as children, and that the health of infants was in turn dependent on the health of their mothers. Studies by Baird and his colleagues, which considered the perinatal outcomes of infants born in Aberdeen between 1948 and 1972 in relation to the childhood social environments of their mothers, reiterated this association.^6–11 Until recently^12–14 there has been a paucity of datasets including reliable and accurate measures of birthweight and gestational age for two or more generations across the normal population range,^15–18 and fewer still that also include measures of the social environment across generations.^19,20 Even with the emerging availability of such datasets, the analytical approach required to consider the cumulative influence of proximal and distal biological and social factors remains challenging.²¹ For example, in analyses that combine data from five cohort studies in low- and middle-income countries (from the COHORTS group), conditional height and weight measures have been utilized to overcome collinearity issues encountered when repeated measures of height and weight are used in models trying to disentangle the relative contribution of different periods of growth for both adult and reproductive outcomes.^32–35 These analyses have demonstrated that each period of growth contributes to intergenerational outcomes, often with early growth in the first 2 years of life being of enduring importance for next generation birthweight as well as for later health outcomes, such as high blood pressure.

The hypothesis that reproductive outcomes are influenced by earlier periods of a mother’s development was assessed here using a unique intergenerational dataset containing extensive parental, perinatal and developmental data for 3932 mothers born between 1950 and 1955 in Aberdeen and their 7928 offspring born between 1967 and 2001 throughout Scotland. This population-based dataset from a high-income population has considerable advantages over other intergenerational datasets, which either: focus on selected subgroups of mothers and infants (e.g. low birthweight or small for gestational age mothers and/or infants); contain incomplete information about all the potential second-generation deliveries to first-generation females; or have limited socioeconomic variables across generations.^15–20 We also introduce a graphical analytical approach, a ‘temporal map’, as an adjunct to more traditional multivariable regression methods to illustrate pictorially how both distal and proximal life course and intergenerational social and biological variables act together over time to predict second-generation offspring size at birth.

Methods

Study population—Aberdeen intergenerational dataset

The Aberdeen Children of the 1950s Study (ACONF), which collected data on almost 15 000 Aberdeen schoolchildren in the early 1960s, provided the early-life maternal and grandparental data for the intergenerational analyses. It has been described in detail elsewhere.^23,24 Extension from the ACONF dataset to an intergenerational dataset, with information on adult maternal characteristics and second-generation births, relied on linking the first generation females, all those born in Aberdeen between 1950 and 1955 in ACONF, to pregnancy and delivery records for the period when the females were of reproductive age (that is 12–50 years). These second-generation linkages, undertaken in 2000 and 2001, were sought for the 4997 of the 5866 (85.2%) females from the ACONF who had complete early-life information about their own intrauterine growth and development. Information regarding second-generation (offspring) deliveries was independently obtained from two sources. First, probabilistic linkage was used to link females to the routinely collected Scottish maternity discharge information (SMR2) between 1969 and 1999. Second, a pre-existing internal exact linkage system was used to match females to delivery information contained in the Aberdeen Maternity and Neonatal Databank (AMND) between 1967 and 2001. The resulting collated intergenerational dataset consisted of 7014 viable second-generation deliveries (that is live-born singletons, weighing at least 500 g at delivery with a gestational age of at least 24 completed weeks)²⁶ linked to 3485 of the 4997 (70%) first-generation singleton females. This corresponded to a mean of 2.01 second-generation deliveries per first-generation woman linked. This linkage rate of 70% was almost certainly an underestimate of the true rate of reproduction which, given age-specific fertility rates for females born between 1950 and 1955 in Scotland, should have been closer to 85% if all the first-generation females had survived to reproductive age. However, the average number of live-born children per woman was appropriate.²⁵ Sensitivity analyses were used to consider the potential for bias in the results that the deficit may have created, particularly given that reproductive histories for first-generation females who may have migrated outside Scotland were unavailable. These analyses confirmed that the results obtained from the available data were robust to repeated reclassification of women from non-linked to linked status (details presented fully elsewhere^28,29).

Intergenerational data

This dataset has some unique characteristics that make it ideal for considering determinants of second-generation offspring size at birth; in particular, it has detailed information on grandparental characteristics and perinatal information around the time of the mother’s own fetal development, as well as during the mother’s own pregnancies in adult life. Specifically, the intergenerational dataset includes variables relating to three generations: G1 refers to the 3485 first-generation females, born 1950 to 1955 in Aberdeen; G2 refers to the 7080 offspring of the G1 females, born 1967 to 2001 throughout Scotland; and G0 refers to the parents of the G1 females, that is the grandparents of the G2 offspring (Figure 1). The G1 population is made up of all adult survivors of Aberdeen births between 1950 and 1955 and their reproductive events that occurred in Scotland during their adult life.

Perinatal data collected for both G1 and G2 deliveries included: pregnancy complications (gestational hypertension, pre-eclampsia, antepartum haemorrhage and gestational diabetes); infant sex; birthweight (measured in grams); and gestational age in completed weeks (according to a combination of date of last menstrual period, clinical obstetric assessment and ultrasound assessment for the later G2 deliveries); maternal and grandmaternal height; parental age at pregnancy; maternal parity; and parental and grandparental occupation at the time of each pregnancy. G1 females’ childhood height, weight and age (in months) at school entry were taken from the ACONF, as were data on family size, structure and socioeconomic indices. Information on smoking in pregnancy was available for G1 females linked to G2 deliveries through the AMND system, but these data were not collected routinely on the SMR2 forms until after 1996; therefore, information on smoking was only available for 3602 (51%) of the 7014 second-generation pregnancies. There was no evidence to suggest that the delivery records retrieved from the AMND were systematically different from those found in SMR2, once year of delivery was accounted for.²⁸

Size at birth in G1 and G2 refers to standard deviation (SD) scores for sex-specific birthweight according to completed week of gestational age at delivery. The SD scores were internally standardized for both G1 and G2 infants (that is within the population-based sample). Internal standardization was used because complete records of birthweight and gestation at delivery were not available for births before 1967, when the SMR2 system was initiated, and because G1 represented a population group of females born in Aberdeen between 1950 and 1955 who remained in Aberdeen and survived to school entry. One SD difference in fetal growth was equivalent to approximately 560 g difference in birthweight for a female infant born at 40 completed weeks of gestation in both generations.

Measures of size were also available in childhood and adult life for G1 females; however, these measures of size, including size at birth, were highly correlated repeated measures. Therefore, two conditional measures of change in size (weight and/or height conditional on the closest earlier measure) were calculated: between birth and childhood, and between childhood and adult life. The conditional growth measures were adjusted for age at measurement and also internally standardized. Importantly, these predicted conditional growth measures from birth to adulthood are uncorrelated with each other to allow them to be entered together into a regression model to determine how each period of maternal growth is associated with G2 offspring size at birth.

Paternal occupation was available for both G0 and G1 and was coded according to the Registrar General’s classification of occupations for the appropriate time period. G0 and G1 maternal occupation and education were also available but less complete than the paternal information; hence paternal occupational measures were used as proxy measures for the socioeconomic environment across the two generations.

Statistical methods

Regression models were used to assess the marginal and conditional contribution of each G0 and G1 life course variable on G2 offspring size at birth for all 6369 intergenerational pairs of G1 mothers and their G2 infants. Analyses were then repeated on the restricted set of 3602 intergenerational pairs of G1 mothers and G2 infants for whom data on maternal smoking in pregnancy was available in addition to complete life course measures. A temporally guided stepwise regression approach was then utilized, with G2 size at birth (SD score) as the outcome throughout, as described below. Paternal social class and maternal smoking were treated as categorical explanatory variables, and maternal pre-eclampsia as a binary variable (yes or no). All other explanatory variables were treated as continuous. Likelihood ratio tests were used to check that there was no evidence of departure from linearity for these continuous variables or of statistical interaction.

The variables were entered in four temporal groupings beginning with the most distal influences (G0) through the most proximal (adult G1). This method allowed for consideration of whether each predicted G2 size at birth independently or whether their influences had potentially been mediated by more proximal measures of later maternal development or status. Robust standard errors were calculated to account for the repeated G0 and G1 information in consecutive G2 deliveries to the same G1 mother. Since multiple statistical testing was used, the significance of each variable was assessed according to crude associations and biological plausibility, in addition to the level of statistical significance.

Using a ‘temporal map’

A graphical display of the ordered, standardized (where appropriate) regression coefficients was plotted to make explicit the impact of adding successively more proximal variables into the mutually adjusted regression model, where the outcome variable was G2 size at birth (as an SD score) throughout. This is referred to as a ‘temporal map’ because it allowed the predictive capacity of each temporally ordered explanatory variable on G2 size at birth to be tracked over time. As each group of temporally grouped explanatory variables is added to the model, the temporal map plots the cumulative effect of adding variables more proximal to the G2 size at birth (dependent variable) on the regression coefficients of the more temporally distal variables. This provides an explicit illustration of change in the relative importance of each life course and intergenerational variable over time, and how some temporally distal variables predict offspring size at birth independently of more proximal variables or how their predictive capacity is mediated by the more temporally proximal variables. A simplified temporal map (limiting coefficients displayed) is provided here to illustrate the salient intergenerational relationships with coefficients drawn from the full life course and intergenerational model.

Results

Intergenerational data

The crude and mutually adjusted regression coefficients for all life course variables remained largely unchanged between the analysis of the greater number of intergenerational pairs of mothers and offspring (n = 6369) and the subset with maternal smoking information available in pregnancy (3602). Only the results for the latter pairs, with G1 smoking information, are provided here. The 3602 intergenerational pairs of G1 mothers and their G2 offspring include 1858 G1 mothers with an average of 1.93 G2 offspring each. The life course and intergenerational variables and their univariate associations with G2 mean birthweight and size at birth SD scores are detailed in Table 1.

Intergenerational associations in offspring size at birth

The variation in G2 size at birth according to G0 adult social class (a proxy measure of the G1 mother’s early-life social environment) was of the same order and magnitude as the variation according to G1 adult social class. In each case, the average difference in size at birth between G2 offspring born to those in the two most advantaged social groups (I & II) compared with the most disadvantaged (Other in G0 and IV & V in G1) groups were 0.38 and 0.37 SD, respectively, for G0 and G1 extreme social class categories, with a gradient between these two extremes (Table 1).

G2 offspring size at birth was also strongly associated with each mutually exclusive period of G1 maternal conditional growth. On average there was a 0.68 SD difference in size at birth between G2 infants born to G1 mothers in the highest compared with the lowest quintile of their own intrauterine growth, with a graded effect on size at birth between these extremes: an average 0.38 SD between G2 infants born to G1 mothers in the highest compared with the lowest quintile of conditional childhood growth (that is the rate of growth in early childhood independent of size at birth), with a graded effect between these extremes; and an average 0.43 SD difference between G2 infants born to G1 mothers in the highest compared to the lowest quintile of conditional height change between childhood and adulthood (that is the rate of height change between childhood and adulthood, independent of the early childhood growth), also with a graded effect between these extremes (Table 1).

Adult (G1) parental associations with offspring (G2) size at birth

G2 size at birth was associated with parental adult characteristics in the expected directions. In particular, offspring size at birth was positively associated with G1 attained maternal adult height (0.04 SD increase for every 1 cm increase in maternal adult height), maternal age at pregnancy (0.03 SD increase for every 5-year increase in maternal age, with no significant evidence of a tailing off of this graded effect in the oldest females) and maternal parity (0.15 SD increase for each subsequent delivery). G2 size at birth was significantly negatively associated with maternal smoking in pregnancy (0.18 SD decrease for every 10 cigarettes smoked per day during pregnancy) and maternal pre-eclampsia during pregnancy (0.14 SD decrease). There was also a positive gradient in G2 size at birth according to increasing G1 paternal adult socioeconomic status measured at the time of the G1 pregnancy (Table 2). This gradient was not fully explained by differences in other maternal adult characteristics, including smoking in pregnancy (Table 2).

Stepwise analyses of intergenerational and life course determinants of G2 offspring size at birth

(i) Step 1: G0 adult characteristics

The mutually adjusted associations between G0 characteristics and G2 size at birth were largely consistent in terms of relative magnitude and direction with the associations between corresponding G1 adult measures and G2 size at birth (seen in Table 2). Exceptions to this intergenerational consistency were apparent for the negative association between G0 maternal parity and G2 size at birth and the lack of a significant association between G0 pre-eclampsia and G2 size at birth in the mutually adjusted model. The graded effect of G0 social class on G2 size at birth was attenuated after mutual adjustment for all other G0 characteristics, but it remained an important predictor (Table 3, step 1).

(ii) Step 2: early G1 maternal growth

Every 1 SD increase in G1 maternal size at birth was associated with an average increase of 0.23 SD in G2 size at birth when adjusted for G0 and G1 adult determinants of offspring size at birth, namely maternal age, adult height, parity, hypertension in pregnancy, maternal smoking in pregnancy and paternal social class. Every 1 SD increase in conditional growth to childhood was significantly associated with a further increase of 0.11 SD in G2 size at birth (Table 3, step 2). The influence of G0 parental social class (proxy for G1 early social environment) was substantially reduced after adjusting for G1 early growth (intrauterine and conditional childhood). In contrast, G0 parity was negatively associated with, and G0 age at delivery and G0 maternal height positively associated with, G2 size at birth in the mutually adjusted model (Table 3, step 2). However, the positive association of G0 maternal height was attenuated once both G1 size at birth and conditional growth in childhood were added to the model.

(iii) Step 3: G1 adult characteristics

Conditional maternal growth between childhood and adulthood was positively associated with G2 size at birth (Table 3, column 1) after adjustment for G0 characteristics and earlier measures of G1 maternal growth (size at birth and conditional childhood growth). In this model, G0 maternal adult height became non-significantly associated with G2 size at birth, after the addition of G1 adult height to the model (Table 3, step 3). A significant, positive association remained between grandmaternal (G0) and maternal (G1) age at delivery and G2 size at birth. By contrast, grandmaternal (G0) parity and maternal parity (G1) at delivery, although still significant, were associated in opposite directions with G2 size at birth, a finding that is consistent with reported effects of intergenerational parity on childhood anthropometry.²⁰ Maternal adult (G1 paternal) social class continued to have a significant graded association with G2 size at birth, with infants born to parents in more disadvantaged social classes being significantly smaller at birth than those born to more advantaged parents even after accounting for the earlier variables.

(iv) Step 4: G1 maternal pregnancy-specific characteristics

G1 maternal pre-eclampsia in pregnancy was not a significant predictor of G2 size at birth after allowing for the influence of all earlier G0 and G1 measures. G1 maternal smoking in pregnancy remained an important predictor of reduced G2 size at birth, with the magnitude of the effect of smoking 10 cigarettes per day in pregnancy being similar to the loss of the intergenerational predictive effect of 1 SD of a mother’s own size at birth. In the final mutually adjusted model, each of the largely independent conditional measures of maternal life course growth remained significant predictors of G2 size at birth, together with contemporary adult measures of G1 parity, age at delivery, adult parental social status and maternal smoking in pregnancy (Table 3, step 4).

Temporal map

The graphical approach (Figure 2) illustrates how more temporally distal G0 and early G1 variables, although not necessarily remaining significant in the final multivariable regression model, are associated with G2 size at birth via their associations with the more proximal G1 social and biological variables that remained significant in the final regression model (Table 3, step 4). For simplicity only a selection of the life course variables are shown on the temporal map provided in this paper (full variable temporal map not shown). The coefficients shown on the temporal map in Figure 2 are for the extreme categories of G0 and G1 adult social class—that is social class I & II (most advantaged) compared with IV and V and life course maternal growth (maternal size at birth, and conditional childhood and conditional child-to-adult growth). The temporal map displays the regression coefficients for a step change in each of the life course and intergenerational variables with respect to G2 (offspring) size at birth (SD score), and illustrates how these coefficients change over time as the more proximal intergenerational variables are entered into the model. The numbering on the horizontal axis refers to the temporal sequence in which the variables were entered into the model (from most distal to most proximal, singly or grouped temporally and corresponding to the variables used in Steps 1 through 4 in Table 3). The coefficients are shown on the map when they remain significant in the model, after which time they are removed (e.g. G0 social class at time point 9). All size and conditional growth measures were entered as standardized variables (SD scores) so that their relative importance for predicting G2 size at birth could be directly compared.

Figure 2 demonstrates that the inequalities seen in G2 size at birth (SD score) according to grandparental (G0) social class diminished as statistically independent measures of G1 conditional maternal life course growth were sequentially entered into the regression model. By contrast, the conditional measures of G1 maternal growth between her size at birth and her adult life remained important predictors of G2 size at birth independently of more proximal predictors. The intergenerational association of G1 fetal growth on G2 fetal growth was stable over time and was not attenuated by the addition of more temporally proximal G1 adult or pregnancy-specific characteristics to the final model.

Discussion

Strategies aimed at improving population health by optimizing size at birth have largely focused their efforts on the health of mothers in the immediate pre-pregnancy and pregnancy time periods. In addition, there has also been a focus on ‘at-risk’ populations such as those mothers likely to have low birthweight infants. Whereas it has long been theoretically acknowledged that reproductive outcomes are influenced by earlier periods of a mother’s development,^5–7 there have been few datasets able to test this explicitly or to consider the importance of these early-life factors in relation to traditionally investigated adult characteristics.^31–34 Recently the associations of different patterns of maternal growth have been considered for birthweight SD scores (not adjusted for gestational age in weeks) in five cohorts from low- and middle-income countries where low birthweight and stunting are common, but these associations have been far less common in more homogeneous populations from high-income countries such as the UK. Even where such datasets are available, there remain significant analytical challenges given the highly correlated nature of repeated life course measures.²¹ The discrete periods across a woman’s life course that influence her offspring’s size at birth are not independent. They represent snapshots of the continuum of life course development between a mother’s own intrauterine life and her adult reproductive life. Utilizing the data from each of these periods in a life course approach requires not only life course data but also analyses that take account of the temporal dimension of the data and the collinearity of repeated measures over time. Using the Aberdeen Children of the 1950s intergenerational dataset, which provided extensive biological and social life course data across three generations, and a graphical method which made explicit the temporal relationship between highly correlated life course variables, it was clear that offspring size at birth was predicted by biological and social characteristics acting together across time and across generations.

Using this life course and intergenerational approach illustrates that contemporaneous socioeconomic inequalities seen in offspring size at birth are strongly determined by the continuity of social environments across generations and by the inequalities in maternal growth throughout her life course from before birth predicted by her early childhood social environment in particular. Using conditional and statistically independent standardized explanatory variables to capture change in maternal growth over different time periods allowed an assessment of the importance of each period of maternal life course growth for her own offspring’s size at birth. Overall, a mother’s own intrauterine growth was an important intergenerational predictor of her offspring’s fetal growth. Her conditional growth in early childhood and between childhood and adult reproductive age also contributed independently to her capacity to grow her own offspring, but to a lesser extent than that of her own fetal growth. These findings compare favourably to similar analyses using conditional growth measures from the COHORTs group who have used pooled data from five cohorts from low- and middle-income countries to demonstrate that weight and height gain in early life are important predictors of attained adult maternal height as well as of offspring birthweight.^34,35

In the Aberdeen intergenerational dataset, offspring size at birth was associated with maternal and paternal adult characteristics measured contemporaneously with the index pregnancy in ways consistent with previous studies,³⁰ and these adult variables were also noted to be reflections of the cumulative influences of more temporally distal intergenerational variables across the life course, shaped in particular by the early-life social environment. This finding is similar to that of a Swedish study which has recently shown that the social environment shared across generations is a significant contributor to intergenerational correlations in size at birth. Even maternal smoking in pregnancy, a variable with a more immediate influence on offspring size at birth, can be viewed as an indicator of socioeconomic background.²⁰

One of the strengths of the Aberdeen intergenerational dataset was that it represented a complete cohort of first-generation females linked to their offspring records over a time period capturing their entire reproductive years (mothers aged 12–50 years). Measures of size at birth were available from perinatal records across the full population range of gestational age and birthweight for both generations, unlike many other intergenerational studies that have either been restricted to term deliveries or have had limited information about gestation at delivery in one or both generations and have therefore been limited to birthweight as an intergenerational explanatory or outcome variable.^33–37 Additionally, a diverse set of both social and biological life course measures was available for the mother’s generation.

However, the dataset had limited information on the life course development of the fathers of the second-generation offspring. (They were not necessarily part of the original Children of the 1950s cohort.) Therefore, the analyses were limited to a focus on maternal characteristics. This precludes a consideration of the influence of paternal life course growth on offspring’s size at birth, although for offspring size at birth, rather than life course growth, it is generally acknowledged that maternal size is the more influential determinant.^18,37–39 A further limitation was the absence of maternal data between late childhood and adulthood and lack of prospectively collected measures on the females through early adult life. Nevertheless, the Aberdeen population represents a stable population, with 90% of the original ACONF cohort residing in the same geographical area throughout their reproductive lives. This, combined with high rates of follow-up using routine data linkage in the Grampian region, means that despite the lack of interim information, retrospective data on this time period tended to be available for this population. One further limitation of retrospective data collection’s dependence on routine linkage was that information on maternal smoking status in pregnancy was limited to half the adult females. However, there is no indication that the associations between life course variables and second generation offspring size at birth differed between the group with smoking data available and those for whom this information was missing.

The population analyses applied using this cohort can help direct research attention to the periods in the life course of greatest influence for specific health outcomes, reproductive or otherwise. However, a limitation of population data that deals with gross measures is that it cannot capture the detailed mechanisms that inevitably mediate all growth, from intrauterine development to adulthood. There remain important questions to be answered regarding the more specific genetic and non-genetic determinants of offspring size at birth, that inevitably underlie these life course and intergenerational associations.

Conclusion

Offspring size at birth is the result of a complex interplay of life course biological and social variables acting together over several generations, with enduring influences of the maternal early-life social environment as well as the more immediate perinatal environment. Whereas interventions aimed at individual mothers in pregnancy, such as reducing rates of maternal smoking, and nutritional supplementation in undernourished women, have been shown to be beneficial for offspring growth, at a population level interventions required to optimize fetal growth across the whole range of birth size may require longer-term life course strategies, which recognize the importance of both biological and social environmental influences over time and across generations.³¹ The analyses carried out using the unique Aberdeen intergenerational cohort suggest that strategies aimed at optimizing offspring size at birth and, therefore, potentially infant and adult health, require a focus on growing healthy mothers in one generation (and probably fathers too) in order to grow healthy children in the next.

Conflict of interest: None declared.

References