Effect of correcting for gestational age at birth on population prevalence of early childhood undernutrition

Early childhood growth is an important indicator of a child’s future health and developmental potential [1, 2]. Childhood undernutrition—defined by weight and/or length/height more than 2 standard deviations below the standard population median for age and sex—has been associated with an increased risk of mortality [1, 3], poor cognitive development [4], lower school achievement [5], lower economic productivity [6], and a greater risk of adverse cardiometabolic outcomes [7‐9] later in life. Undernutrition represents a major public health burden in low- and middle-income countries [10, 11]. Preterm birth, defined as births < 37 weeks gestational age (GA), has been consistently identified as a major risk factor for undernutrition in early life [12‐15]. However, population-based estimates of undernutrition attributable to preterm-birth may be biased if they do not account for the GA of preterm-born children in the application of growth standards.

The World Health Organization Child Growth Standards (WHO-GS) are the most widely accepted international normative standards for evaluating postnatal growth among children less than 5 years of age [16, 17]. These standards are not directly applicable to children born preterm using postnatal age because they are based on a reference population of term-born children (births between 37 and 42 weeks GA) [18]. In clinical practice, guidelines commonly recommend correcting the postnatal age of children born preterm for the number of weeks that birth occurs prior to term gestation (40 weeks) to generate a ‘GA-corrected age’ (CA), which is then used to apply the WHO-GS up to 24 or 36 months of postnatal age [19, 20]. This strategy along with the recent publication of the GA-specific norms, INTERGROWTH-21st newborn size standards (IG-NS) [21] and INTERGROWTH-21st very preterm size at birth references (IG-VPBR) [22], provide new methods to account for GA in the standardization of anthropometric measures at birth and postnatally among children born across range of GA in population-based studies.

The importance of accounting for GA in the timescale for evaluating neonatal outcomes is well-recognized in perinatal epidemiology [23], and the implications of disregarding GA at birth in the application of growth standards/reference have been previously established in clinical settings [24, 25]. Yet, in population-based epidemiologic studies of undernutrition in LMICS, in which children born across a wide range of GA are typically included, CA is rarely used in the application of growth standards to preterm-born children [26]. Ignoring GA at birth penalizes children with shortened gestational duration and conflates those who are small but well-nourished given their GA at birth with children who have biologically meaningful deficits in nutritional status.

To quantify the implications of not accounting for GA at birth in the application of child growth standards for population-based estimates of undernutrition in LMICs, we compared the use of a ‘CA’ strategy, which applied GA-specific standards at birth in conjunction with WHO-GS using CA for preterm-born children postnatally, versus postnatal age in the application of WHO-GS for all infants (as is done conventionally), to estimate mean nutritional indices and indicators (prevalence of stunting and underweight) in a population-based cohort of term- and preterm-born children less than 5 years of age in Brazil. Additionally, we aimed to estimate the proportion of stunting and underweight attributable to preterm birth in the first 5 years of life using CA versus postnatal age strategy.

Of 4287 births in the cohort, data from 4107 children were eligible for inclusion in this study. Children born preterm (n = 482; 11.7%) had a mean ± SD GA of 34.4 ± 2.5 weeks (median (interquartile range): 35 (2.0) weeks). Of the 482 children born preterm, 19.2% (n = 92) were born at ≤ 33^6/7 weeks and 81.3% (n = 391) were born between 34^0/7 and 36^6/7 weeks GA. Mean ± SD GA at birth among term-born children was 39.2 ± 1.5 weeks.

As expected, children born preterm were on average shorter and lighter than term-born children at birth and at each subsequent follow-up visit up to 48-months when anthropometric indices were assessed using WHO-GS with postnatal age. However, the magnitude of the mean differences in LAZ and WAZ among term- versus preterm-born children estimated using postnatal age attenuated with increasing age (Tables 1, 2).

Using IG-VPBR and IG-NS at birth (i.e., CA vs postnatal age) shifted the distributions of LAZ and WAZ higher and resulted in significantly higher mean z scores compared to WHO-GS. The effect was greatest among children born preterm as expected, but was also evident among term newborns (see Additional file 1: Figure S2; Tables 1, 2). Compared to estimates derived using WHO-GS at birth, distributions of LAZ and WAZ based on IG-VPBR and IG-NS had lower variance, shorter left-tails, and were more normal appearing (i.e., skewness and kurtosis values were closer to a normal distribution; data not shown). At follow-up visits, children born preterm continued to have higher mean LAZ and WAZ using CA compared to postnatal age; however, the magnitude of the effect attenuated over time (Tables 1, 2) and the distributions of LAZ and WAZ based on CA versus postnatal age were essentially overlapping by the 12 month visit (see Additional file 1: Figure S2).

At birth, using CA (vs postnatal age) attenuated the overall prevalence of stunting (8.3 vs 11.6%) and underweight (3.7 vs 6.9%) (Tables 3, 4). Similarly, at the 3-month visit, using CA versus postnatal age reduced the overall prevalence of stunting (5.1 vs 7.8%) and underweight (6.6 vs 9.3%). Although estimates varied slightly, inferences remained unchanged when CA was used for all children (including term-born children) at postnatal follow-up visits (see Additional file 1: Tables S1 and S2) or when IG-PPFS with postmenstrual age were used for preterm-born children at the 3-month visit (see Additional file 1: Tables S3 and S4). The effect of using CA for children born preterm was attenuated by the 12 month visit.

Compared to term-born infants, children born preterm had higher estimated mean LAZ/WAZ and attenuated odds of stunting/underweight at birth and 3 months when using CA, which was opposite to the inferences based on postnatal age. From 12 to 48 months, children born preterm were at higher risks of stunting and underweight relative to term-born children when using both CA and postnatal age; however, ORs and %PARs were consistently overestimated using postnatal age (Tables 3, 4). The apparent patterns of change in ORs/%PARs with age were substantially influenced by the choice of age scale: using postnatal age, the odds ratio for stunting and underweight among preterm-born versus term-born children decreased from birth up to at least the 24-month visit, whereas using CA the odds ratios increased from birth up to at least the 24-month visit (Fig. 2). Similarly, using postnatal age, the %PAR of stunting and underweight due to preterm birth decreased from birth up to the 48-month visit, but an opposite (increasing) trend was observed when CA was used to estimate the %PAR due to preterm birth (Tables 3, 4). Of note, the magnitude of the estimates and the trends over time for stunting and underweight among children born preterm derived using CA were more closely aligned with trends in wasting, a measure of undernutrition that is estimated independently of age (Fig. 2; see Additional file 1: Table S5).

In a cross-sectional random sample of children from birth to 48 months of age, using GA-specific norms at birth alone and WHO-GS using postnatal age for all children after birth substantially attenuated the mean LAZ/WAZ and the prevalence of undernutrition at the population-level compared to using WHO-GS at birth and postnatal visits. In comparison, using CA for preterm-born children instead of postnatal age for all children when applying the WHO-GS postnatally did not greatly influence the overall estimates of the mean LAZ/WAZ or prevalence of undernutrition; however, use of postnatal age overestimated the relative odds of stunting and underweight in preterm-born versus term-born children considerably (Table 5).

Using postnatal age in the application of growth standards for all children in a birth cohort in Brazil overestimated the population-average estimates of undernutrition and inflated the associations between preterm birth and undernutrition. Incorporating information about gestational duration by using CA altered indices and indicators of child nutritional status at the population-level and changed inferences with respect to the relative odds of undernutrition among children born preterm in the newborn and early infancy period. The odds of stunting and underweight due to preterm birth were diminished using CA versus postnatal age at birth; however, as expected, the effect of using CA compared to postnatal age on population-average estimates of child nutritional status progressively attenuated overtime. Beyond the 12-month follow-up, the effect of using CA on population-average cross-sectional estimates appeared to be minimal. Although the inferences at the 12- and 24-month follow-up in this study are consistent with previously reported findings from this cohort [15], the magnitude of the associations between preterm birth and stunting and underweight were attenuated when CA compared to postnatal age was used. Importantly, in contrast to estimates derived using postnatal age, we observed an increasing postnatal association between preterm birth and stunting and underweight during the first 2 years of life when CA was used. This suggests that preterm-born children in this cohort were not more likely than term infants to be undernourished at birth, but had a higher likelihood of becoming stunted and underweight beyond the early postnatal period.

As expected, using postnatal age for children born preterm in the application of WHO-GS leads to systematic bias in the measurement of nutritional status during infancy as anthropometric measures are standardized based on expected size for a given age from birth (i.e. postnatal age) as opposed to age from conception (estimated by the postmenstrual age scale). The extent to which anthropometric z scores are misclassified—and therefore the effect of misclassification on measures of association where child nutritional status is either the exposure or outcome—depends on the incidence of preterm birth in the source population as well as the distribution of GA among children born preterm. In LMICs where a large proportion of children are born preterm—estimated to range from ~ 10 to 18% [34]—the misclassification of infant nutritional status is non-negligible. Given the importance of gestational duration in early infancy, the effect of using CA on relative measures of associations is apparent even in the context of a cross-sectional survey in which anthropometric data are pooled from a sample of infants and young children at various ages. These findings demonstrate that population-based epidemiologic studies in LMICs [12‐15] may overestimate the prevalence of undernutrition and the risk of undernutrition among preterm-born relative to term-born children, if GA is unavailable or unused in the application of growth standards.

Although the importance of using CA in analyses of anthropometric data is most apparent in early infancy (i.e., birth and 3 months), there are no known disadvantages of using CA up to 24 months of age in epidemiologic analyses, as is commonly done in clinical practice. In fact, using a GA-specific age scale in the application of growth standards confers several analytical advantages, including: (1) lower frequency of “biologically-implausible” outliers (i.e. LAZ < −6 or > 6 SD; WAZ < −6 or > 5 SD) [32], which may in part be due to incorrect assignment of z scores to preterm-born children, (2) reduced variance in the distribution of z scores leading to greater precision in the standard errors, and (3) improved normality of the z score distributions during early infancy.

An important limitation of this study was the use of a derived GA variable based on LMP and Dubowitz methods. These methods for GA assessment are less precise in comparison to first or second trimester ultrasound assessment. To minimize the risk of measurement error, we excluded all observations with implausible birthweight and GA combinations, which may have resulted in some selection bias by excluding observations at extreme ends of the birthweight-for-GA distribution. However, given that available ultrasound measurements in this cohort were prone to selection bias for preterm births, the systematically derived GA estimate provided a more reliable measure of GA for this study. Importantly, inferences regarding the effect of using CA compared to postnatal age for preterm-born children in the application of growth standards are robust against GA estimation methods. The effect of using CA for children born preterm in the application of WHO-GS in this study was likely a conservative estimate given the lower prevalence of undernutrition in Pelotas, Brazil, compared to countries in Sub-Saharan Africa and South Asia [10]. In addition, due to the repeated serial cross-sectional design of our analysis, we did not analyze individual-level longitudinal patterns of growth among preterm-born children using different age-scales in the application of growth standards. Furthermore, although we used the best available standards at each follow-up visit (e.g., IG-NS, IG-PPFS, and WHO-GS), it was difficult to fully disentangle the implications of using different standards (INTERGROWTH-21st vs WHO-GS) from GA-correction effect alone for comparisons at birth. However, the effects of using CA for preterm-born children (or using CA for all children in sensitivity analyses) at the 3-month and later visits were assessed using only WHO-GS, and therefore demonstrated that observed differences in CA versus postnatal age estimates were not an artifact of differences in standards but a true effect of accounting for GA in the standardization of anthropometric indices and indicators. Nonetheless, further research is needed to fully explore the implications of combining multiple different standards for longitudinal assessment of child growth trajectories.

In summary, this study suggests that the choice of age scale used to standardize anthropometric measures has substantial and meaningful implications for population-level epidemiologic inferences related to patterns of growth faltering and risk factors that contribute to undernutrition in early life. In LMICs with concurrently high prevalence of undernutrition and high incidence of preterm births, the discrepancies between postnatal age versus CA estimates of undernutrition at the population-level would be expected to be greater than the effect observed in this study, and may potentially misguide public health intervention priorities to reduce the population burden of undernutrition Future epidemiologic studies should therefore collect high-quality information about GA whenever possible and incorporate the GA information in the analyses of anthropometric data, at least at birth and during infancy. Longitudinal studies that account for heterogeneity in GA at birth in the application of growth standards are needed to better understand postnatal growth trajectories among children born preterm, the contribution of preterm birth to undernutrition in early life, and the implications for health outcomes in later life.