Neonatal nutrition and early childhood body composition in infants born extremely preterm

Background & aims: Previous studies have observed changes in fat and fat-free mass among preterm infants when compared to term-born infants. However, these studies have mainly focused on moderate or very preterm infants, with a scope limited to the ﬁ rst few years of life. We aimed to compare body composition in extremely preterm infants to term-born infants in early childhood. Additionally, we investigated whether early neonatal nutrition was associated with the distribution of fat-and fat-free mass in later life. Methods: The study used dual-energy x-ray absorptiometry to evaluate the body composition of 52 children aged 6 e 9-years, of whom 35 were born extremely preterm and 17 were born at term and was analyzed using multivariate linear regression. Nutritional intakes of ﬂ uids, energy, and macronutrients during the ﬁ rst eight postnatal weeks for 26 extremely preterm infants were investigated in relation to body composition at age 6 e 9 years using Bayesian regression analysis and Gradient Boosting Machine. Results: Children born extremely preterm had smaller head circumference (con ﬁ dence interval (cid:1) 8.7 to (cid:1) 1.7), shorter height (con ﬁ dence interval (cid:1) 2.7 to (cid:1) 0.6), higher waist to height ratio (con ﬁ dence interval 0.01 e 0.05) and lower fat-free mass (con ﬁ dence interval (cid:1) 3.9 to (cid:1) 0.49), compared to children born at full-term. Children born extremely preterm had a differing response to amount of ﬂ uid and macro-nutrient intake for both fat mass index and fat-free mass index. A bimodal response showed high intake of ﬂ uid and macronutrients as associated with high fat mass index for some children, whereas others demonstrated an inverse association, suggesting analysis on cohort-level as problematic. Conclusions: Childhood body composition differs between extremely preterm infants and term-born infants. Extremely preterm infants display differing responses in their body composition to varying levels of ﬂ uids and macronutrient intake during the neonatal period. © 2024 The Author(s). Published by Elsevier Ltd on behalf of European Society for Clinical Nutrition and Metabolism. This is an open access article under the CC BY license (http://creativecommons.org/licenses/ by/4.0/).


Introduction
Extremely preterm infants, defined as those born before 28 weeks gestational age, are a vulnerable patient group at high risk for neonatal morbidity and growth failure while admitted to the neonatal intensive care unit (NICU) [1e4].Later in life, it has also been demonstrated that preterm infants are shorter and weigh less compared to their term-born counterparts [5].Asides from their smaller size, preterm infants also differ in body composition compared to term-born infants [5e8].Body composition, defined as the distribution of muscles, bone, water and fat and fat-free mass, has been established as an important indicator of children's health [9].A growing body of evidence indicates a connection between fat mass, fat-free mass, and long-term outcomes [9e13].Previous research has revealed that increased fat-free mass during hospitalization is linked to improved neurological and motor outcomes at one year of corrected age in very low birthweight infants [10].Additionally, both fat mass and fat-free mass are associated with cerebellar volume at term [11].Further, early-life body composition has shown correlations with cardiovascular and metabolic health in adulthood [12e14].A longitudinal study from Australia followed extremely preterm infants from 2 to 22 years of age, found that extremely preterm infants compared to term infants had higher body mass index (BMI) which was associated with poorer cardiometabolic health including higher visceral fat volume, and increased levels of triglycerides and systolic blood pressure [15].The cause behind altered body composition among preterm infants is not fully understood and probably multifactorial.However, one potential risk factor is believed to be rapid catch-up growth which can partly be mitigated by nutritional management during nutritional period [16].While the implications of nutritional management during the neonatal period is fairly well understood, less is known about the impact of neonatal nutrition during childhood [17].Feeding breastmilk is positively associated with more beneficial body composition in term infants [18].However, to meet the high nutritional requirements in infants born extremely preterm, breastmilk often needs to be enriched with human milk fortifiers [16,19e21].The effect of this nutritional diet on long-term health outcomes and body composition is not fully elucidated.
Previous studies have mainly included very low birth weight infants (birth weight <1500 g) and limited their body composition assessments to the first few months of life [22e26].Consequently, body composition at long-term follow up in extremely low birth weight infants (birth weight of around <1000 g) have not been as intensively studied [5].The primary aim of this study was to compare body composition in extremely preterm infants to term infants in childhood, which we hypothesized would differ.We also wanted to investigate if early neonatal nutrition is associated with the distribution of fat-and fat-free mass later in life and if all infants responded the same to early neonatal nutrition or if there are individual variations due to other factors.

Study design, setting and population
We conducted a retrospective cohort study combining data from two independent, partially overlapping studies on the same population treated at Karolinska University Hospital, hereafter referred to as cohort A and cohort B. Fig. 1 shows a flow-chart of included and excluded children from both cohorts.

Cohort A
Cohort A was comprised of 52 children, of whom 35 were born extremely preterm.The control group consisted of 17 healthy termborn children.Children in cohort A were born between 2009 and 2011 and included in a study investigating renal function in relation to nephrocalcinosis which has been described previously [27].In the original study, all patients born extremely preterm before 28 weeks of gestational age at the Karolinska University Hospital, without major congenital malformations, were considered eligible for inclusion.All children were assessed for body composition at the age of 6e9 years using Dual Energy X-ray Absorptiometry (DEXA), and matched for sex and age.

Cohort B
Cohort B included extremely preterm infants born in Stockholm between 2004 and 2011 from a study collecting daily nutritional data during the first eight postnatal weeks [28].A total of 26 extremely preterm infants from cohort A had nutritional data available.However, data on early nutritional intake for the control group was not available.

Data collection of anthropometric measures and body composition
Data collection for anthropometric measurements and body composition took place in 2016 and 2017 at Karolinska University Hospital in Solna, Sweden.DEXA, that uses a very small amount of ionizing radiation, was used for assessment of body composition and is widely regarded as the gold standard for assessing body composition due to its minimal radiation dose [29].GE Lunar Prodigy DXA system (GE Healthcare, Chicago, IL, USA) 2006, was used for all children and the DEXA assessment was conducted by a radiologic technologist.From the DEXA assessment, fat mass index (FMI) and fat-free mass index (FFMI) analysis was calculated as follows: Anthropometric measurements, including weight, height, head circumference, and waist circumference, were taken by a trained study nurse.Waist to height ratio was calculated by dividing the waist measurement with individuals' height.For growth during the neonatal period, the Fenton Z-score was used as a reference for the growth curve [30].Karlberg was used as the reference growth curve for the anthropometric data at follow-up as this is the reference growth curve used in the Swedish population of children [31].Anthropometric measures, including parental weight, and children's physical activity levels, were self-reported by parents.Data collection was interview-based and conducted by a study nurse during the study follow-up visit.

Data collection of early nutrition
This study included a comprehensive daily nutritional assessment of 26 extremely preterm infants.This assessment encompassed all fluids administered to the infants, such as parental nutrition, enteral nutrition, medications, and supplements, throughout the first eight postnatal weeks.The assessments included mothers' own breastmilk or and donor breastmilk analyzed for macronutrient content.The nutritional data was methodically gathered and evaluated using the Nutrium nutritional calculation software (nutrium.se).

Statistical analysis
Statistical analyses were made using STATA (StataCorp.2021.Stata Statistical Software: Release 17. College Station, TX: StataCorp LLC) and R R Core Team (2023).R: A language and environment for statistical computing.R Foundation for Statistical Computing, Vienna, Austria, for frequentist analyses.The level of statistical significance was defined as p < 0.05 for frequentist analysis.Comparison of means and medians between extremely preterm infants and term-born infants were made by using independent sample Ttest and nonparametric test including ManneWhitney U-test and the Fishers exact test.ShapiroeWilks was used to test for normal distribution.We used bootstrap with 200 samples when performing nonparametric tests.Differences of body composition between groups were made using multivariate linear regression.We adjusted for age at follow-up, maternal-and paternal BMI, and physical activity (above or below 3 h per week).Two variables among the term infants exhibited missing data, which where imputed in the analysis.For head circumference, we used the mean value of available data for imputation.For APGAR score 5 and 10, we used the most common number which was 10.These methods were deemed appropriate as the term infants was a healthy control group, and it was anticipated that these variables would exhibit minimal variation within this population.
To investigate potential variations in early neonatal nutrition responses, an ad hoc analysis was performed to determine which nutritional variables and other predictors were most influential in determining childhood body composition.As a starting point, frequentist linear regression analyses using a simplified model with total nutrient intake during weeks 1e8, as well as other relevant factors, were used as predictors and either FMI or FFMI as the response.The following variables were included in the model: incidence of necrotizing enterocolitis (NEC) and surgical NEC, sepsis, sex, gestational age, birth weight, mechanical ventilation, age at follow-up, maternal and paternal BMI and the nutritional values for total fluid, energy, protein, carbohydrates, and fat intake during the first eight weeks of life, using scaled and centered data for the predictor variables.To achieve a deeper understanding of the distributions among children related to nutrient intake and FMI and FFMI, Bayesian regression analysis using the same model and data was performed, using the settings: adapt_delta ¼ 0.95, max_treedepth ¼ 13 and iter ¼ 10000.Bayesian regression offered a complement to the frequentist statistics enabling robust analysis of complex data incorporating various types of predictors, including categorical variables, interactions, and nonlinear relationships.
As the nutritional management of fluid and macronutrient intakes can change considerably during the first eight postnatal weeks, the macronutrient and fluid intake were also divided into weekly subcategories and analyzed individually using a Gradient Boosting Machine (GBM) [32], which is effective when analyzing complex, high-dimensional data with unclear interaction.In addition to weekly nutrient data, expanded data in the form of inflammation associated predictors including C-reactive protein (CRP) and use of antibiotics during the first eight weeks of life.were included as predictors in the GBM analysis.However, use of antibiotics was excluded from the final analysis since no interaction was observed.For CRP, the highest registered value every week was used.Analysis of bimodality of GBM output was performed with Hartigan's dip test and adjusted by Bonferroni correction for multiple testing.Analysis of difference in proportion of patients showing a positive/negative correlation to sickness was done with Fisher's exact test with Benjamini-Hochberg procedure correction for multiple testing.The incidence of NEC, gastrointestinal surgery, late-onset sepsis, mechanical ventilation, and weekly CRP levels as proxies for the severity of illness.Infants were then divided into two groups: those with high severity of illness and those with low severity of illness.

Body composition in extremely preterm and term-born infants
Body composition and anthropometric measurements were analyzed in 52 children.Infants and maternal baseline characteristics are presented in Table 1.There were similar numbers of boys and girls in both groups, with slightly more boys in the control group.As expected, there were significantly more multiple births and caesarean sections in the group of infants born extremely preterm than in the control group.Birth weight Z-score did not differ between the groups and the majority of infants had a birth weight average for gestational age.
Anthropometric data including measurements of weight, height, head circumference and waist circumference along with body composition measurements are shown in Table 2. Children in the control group were six months older at follow-up on average.Children in the extreme prematurity group exhibited a lower weight, shorter stature, and a smaller head circumference.However, the differences in weight were no longer evident when adjusting for age at follow-up, whereas the differences in height and head circumference remained significant.Body composition is presented in both fat and fat-free kg, percentage, and FMI.In the crude analysis, extremely preterm infants had significantly higher total body fat and significantly lower fat-free mass.However, the significant difference in total body fat did not remain after adjustment for confounders, whereas the difference in fat-free mass remained significant.Waist to height ratio was significantly higher in the extremely preterm group in both crude and adjusted analysis compared to the control group.There was no difference in BMI, fat mass (kg), or FMI between children born extremely preterm and term born infants.Adjustments were made for age at follow-up, maternal and paternal BMI, and physical activity as defined above or below 3 h per week.

Association between neonatal nutrition and body composition during childhood
A total of 26 infants from the extremely preterm infants in cohort A had available nutritional data during the first eight postnatal weeks.The mean intake of total fluids (mL/kg/day), energy (kcal/kg/day), fat (g/kg/day) and carbohydrates (g/kg/day) was consistent with nutritional guidelines of that time whereas mean protein (g/kg/day) intake was lower than recommendations from the European Society for Paediatric Gastroenterology Hepatology and Nutrition (ESPGHAN) guidelines [21].Nutritional intake was also consistent with previously reported mean intakes for the entire cohort B by Westin et al. [28].
There was no significant association between any macronutrients or fluid intake during the first eight postnatal weeks and body composition during childhood when analyzed with traditional frequentist regression (Supplementary Table 2).However, a differing response, demonstrated by the bimodal shape of the posterior probability distributions was observed for both FMI and FFMI when analyzed with Bayesian linear regression (Supplementary Table 3).As shown in Fig. 2, this was observed for energy and all the macronutrients, including, protein, carbohydrates, fat, fluids and for other predictors such as sex and parental BMI in relation to both FMI and FFMI.The differing response implies that children reacted differently to macronutrient intake.For instance, for some children a high energy intake was associated with higher FMI, whereas others demonstrated an inverse association.

Factors associated with a differing response of neonatal nutritional intake for FMI and FFMI
Fig. 3 depicts a SHAP plot with nutritional variables dived into weekly subcategories and other predictors and how these affected FMI and FFMI according to the model.The most influential predictors/features on model output are presented on the top of the figure with declining order.The figure also demonstrates if the variable has a bimodal distribution and the distribution along the x-axis for each datapoint indicating a positive or negative influence on model outcome for each individual patient.The most influential nutritional predictors for FMI were intake of fluid at seven weeks of life, fat intake during the first weeks of life and energy during the fifth week of life.Two other strong predictors were birth weight and age at follow-up.The most influential nutritional predictors for FFMI were intake of carbohydrates at week one, fluid intake at week four and protein intake at week eight.CRP during the first two weeks of life was also an important predictor for FFMI.

Associations with differing response for FMI and FFMI and severity of illness during neonatal period
We hypothesized that severity of illness during the first weeks of life could have an important influence on difference in body composition and observed bimodal shape.Infants were then divided into two groups, those with high severity of illness and those with low severity of illness.The mean FMI was higher for the high severity of illness group (4.9 compared to 4.1); however, the difference was not statically significant (CI -2.1 to 0.55).The mean FFMI was also higher for the high severity of illness group (11.0 compared to 10.5), but the difference not statistically significant (CI -3.7 to 1.06).Variables with bimodal shape in Fig. 3 with were tested for effect on FMI and FFMI and we found a significant difference in response between the two groups shown for many nutritional variables such as intake of fat during the first week of life but also for some other predictors such as maternal and paternal BMI.All variables with bimodal shape and test of significance are presented in supporting information (Table 4).

Discussion
In this retrospective cohort study, we observed differences in the body composition of extremely preterm infants compared to term-born infants during childhood at ages 6e9 years of age.Extremely preterm infants had lower fat-free mass, higher waist to height ratio, shorter height and had smaller head circumference.We also found associations with neonatal nutrition during the first eight postnatal weeks and body composition during childhood.There were no statistically significant results indicating how early nutritional intake was associated with body composition on the study population as a whole.However, a differing response on the impact of early nutrition on body composition in children born extremely preterm was observed.This suggests that children exhibit varying responses to levels of macronutrients and total fluid intake resulting in non-significant findings in traditional frequentist regression due to differing responses.When results are not significant in frequentist statistics, it can be because they average out positive and negative responses, masking the underlying distribution.Bayesian regression, however, analyzes the full distribution of responses, revealing details such as bimodality that frequentist methods, which focus on means and standard deviations, can overlook.For example, in our study, some children with a high energy intake had high fat mass during childhood at age 6e9 years old, whereas for others the inverse association was observed.
Fig. 2. Violin-plots showing the posterior probability distributions of Bayesian linear regressions of mean macronutrients intake during the first eight postnatal weeks as well as other predictors and the effect on fat mass index (FMI) and fat-free mass index (FFMI) during childhood.The red brackets show the 0.05 to 0.95 credible interval.The Y-axis has been scaled using a generalized logarithmic transformation to improve readability.Previous research on anthropometric measurements and body composition in preterm infants has mainly included infants born at or after gestational week 28 [3,22e26,33,34].In these studies, preterm infants have been found to have an altered body composition compared to term-born infants.Our study stands out as one of a few studies focusing solely on extremely preterm infants.In addition, previous studies have mainly focused on differences in body composition after hospital discharge around full-term age and during the first year of life, whereas we report on body composition during childhood at age 6e9 years.A systematic review published in 2020 examined the growth and body composition of children born extremely preterm from childhood to adulthood, finding that extremely preterm infants had lower weight, height, and head circumference compared to term-born infants [5].Only five out of 17 studies reported on body composition.Although there was evidence of aberrant body composition, the results were inconclusive.Two of the studies included children of a similar age to those in our present study [35,36].The first was a cohort study including 93 extremely low birth weight infants and 87 controls followed up at around eleven years of age in Belgium and similar to our results found that extremely low birth weight infants had shorter height and had a smaller head circumference than term-born infants [35].Additionally, the group of preterm infants weighed less and had a higher fat mass, as observed in our study's crude analysis.However, the study by Raaijmakers et al., used a different technique (Bodystat QuadScan 4000) when assessing body composition.The study by Kwinta P et al., was a cross-sectional study in Poland describing the body composition of 81 extremely low birth weight infants [36].Comparable to our study, they found that these infants had shorter height, had smaller head circumference and lower FFM than termborn infants at seven years of age.

Association between neonatal nutrition and body composition during childhood
Previous studies have reported on nutritional intake during the neonatal period and its relation to body composition.However, these studies have often been restricted to the first years of life and did not find any negative associations on body composition in relation to macronutrient intake levels [20,22,37e44].Less is known about the long-term effect of high macronutrient intake on body composition.To the best of our knowledge, only one prior study has examined the relationship between nutritional intake during the neonatal period and body composition in childhood.The study by Stutte S et al., found that early protein intake was linked to higher levels of fat mass, abdominal fat, and hip fat mass at 9.5 years of age [45].The study did not identify any other statistically significant associations of nutritional intake.In our study, we observed that children born extremely preterm with severe neonatal morbidity tended to react similarly to amounts of fluids and macronutrient intake in relation to body composition, but in a differing manner forming two distinct populations.Additionally, our findings indicate that macronutrients were associated with body composition during some, but not all, of the first eight weeks of life.For example, carbohydrates intake during week one and week three were associated with FMI, but not week two.In addition, energy and carbohydrates seemed to be more associated with FMI whereas protein and fluid intake was more associated with FFMI, however this does not hold true for all investigated nutritional weeks.Given the available data, we refrain from attempting to elucidate the underlying mechanisms driving these results, as there remains a possibility that they are random findings.These findings are noteworthy, but it is important to exercise caution when interpreting them due to the limited sample size.Further investigation is required to understand the optimal nutritional treatment during the neonatal period for extremely preterm infants with and without severe neonatal morbidity.

Strengths and limitations
A strength with this study is the availability of individual data during the neonatal period and follow-up, as well as parental data, which allows for adjustment of several confounding factors.The confounders adjusted for in regression models were considered the most important but we cannot exclude the possibility that results are explained by residual confounding factors.Possible confounders that we identified, but for which we did not have complete data, included socio-economic status and children's genetic propensity for obesity and other genetics related to body composition.Although this was partially controlled for by adjusting for parental BMI and parental ethnicity as a proxy for socioeconomic status, as there is evidence that immigrant status influence socioeconomic status [46].The nutritional data from the neonatal period was comprehensive, including all fluid, energy, and macronutrient intakes during the first eight postnatal weeks.Previous research on nutritional intake during early life has mainly focused on whether the infant was given breastmilk or formula [25,37,40,47].However, in our study, we analyzed individual data on nutritional intakes from mothers' own breastmilk and donor breastmilk to determine the macronutrient content.This enabled a more precise analysis the association between neonatal nutritional intake and body composition.The study nurse obtained anthropometric measurements of the children.Additionally, body composition was measured using DEXA, which is considered the gold standard [29].
We acknowledge several limitations, including recall bias from self-reported variables, such as parents' weight and children's physical activity.However, outcome and exposure measures, including data on body composition and nutritional intake, were not self-reported thereby not subject to biased handling.Data from the neonatal period also did not rely on parents' self-reporting, as it was derived from medical journals.Other factors that could have implications on body composition are prevalence of pubertal stage, primarily for children at 9 years of age, which we did not have information on.In addition, we also lacked data on children's current and childhood diet and information about prevalence and duration of breastfeeding for the preterm group.One limitation was the relatively small sample size, which likely impacted the statistical power and limited the ability to detect small but clinically significant associations.This study may be subject to selection bias as it relied on parents who volunteered to participate, which may limit the representativeness of the study population.Also, we wish to highlight the elevated incidence of NEC at 26% in our cohort.It is important to consider these limitations when interpreting the study results.

Clinical implications of the study results
Extremely preterm infants are a vulnerable patient group during the neonatal period and a potential risk group for morbidity later in life such as higher risk for cardiovascular diseases and metabolic syndrome [12e14].Although adequate neonatal nutrition is crucial for promoting growth and ultimately organ development, our results indicate that the response to nutritional intake in terms of body composition largely differs between individuals as indicated by our reported bimodal distribution.This highlights the need for more individualized nutritional therapy.Nevertheless, the optimal nutritional approach cannot be determined based on the results in this study alone but needs to be confirmed by studies with larger sample size.

Conclusions
This study observed differences in body composition, including smaller head circumference, shorter height, and lower fat-free mass, among children born extremely preterm compared to healthy term-born infants during early childhood.This study did not identify any significant evidence on how early nutrition affects long-term body composition.However, it was observed that children born extremely preterm exhibit differential responses to varying amounts of fluid and macronutrient intake in terms of body composition at the age of 6e9 years.Given the relatively small sample size in our study, it is not possible to draw clear conclusions about the clinical implications of nutritional management based on these results.Future research should investigate these findings using a longitudinal study design with a larger population and collect data on nutritional intake during childhood.

Fig. 1 .
Fig. 1.Flow-chart of included and excluded children from cohort A and cohort B.

Fig. 3 .
Fig. 3. SHAP plots showing influence of predictors/features on fat mass index (FMI) and fat-free mass index (FFMI) during childhood including weekly nutritional intake of fluids and macronutrients during the first eight weeks of life.A gradient boosting machine (GBM) analysis was used to incorporate the larger number of parameters (listed on left side of figure).Distance along x axis indicate influence of predictor while color of each datapoint indicate initial predictor value.Predictors marked with red color indicate a statistically significant bimodal distribution of GBM output.Predictors/Features with SHAP-value equal to zero have been removed from plots.

Table 1
Demographics/characteristics of extremely preterm infants and control group of term infants in cohort A at birth and during neonatal period.
[10]ssing data on eight infants in control group.Mean value of observed data was used for imputation on missing data.bMissingdatafor APGAR score 5 and 10 for seven infants.The most common numbers[10]were imputed which was deemed appropriate given that the control group were healthy full-term infants.cIndependent sample T-test was used for parametric variables and Mann Whitney U-test for non-parametric variables.

Table 2
Differences in growth with anthropometric measurements and body composition data of extremely preterm and term born infants during follow-up.a a Data shown as mean, standard deviation (SD), crude and adjusted risk ratio (RR) and 95% confidence interval (CI).b Differences were tested using multivariate linear regression and adjusted for age at follow-up, maternal body mass index (BMI), paternal BMI and physical activity.c Adjusted for only age at follow-up.d Abbreviations: standard deviation score (SDS), body mass index (BMI).