Introduction

Preterm birth is one of the leading causes of perinatal mortality and morbidity1, and some evidence suggest associations with long-term health and social outcomes2,3.

Exposures during fetal life may influence postnatal growth and cardio-metabolic health4,5. Much literature has demonstrated that birth weight associates with postnatal body mass index (BMI) with positive linear J and U-shaped associations reported6,7,8. Birth weight is strongly related to gestational age at birth (GA)9, but the extent to which the association between birth weight and later BMI reflects differences in GA is unclear10.

Some studies have investigated the associations between GA and BMI and, respectively, height3,11,12,13,14,15,16,17,18,19,20 and reported positive associations in childhood13,21, while findings are mixed in adulthood20,22. However, wide variations in study design, mode of data collection, and use of covariates in the studies making it difficult to evaluate the evidence23. Some studies assessed gestational duration as a dichotomous variable reporting all preterm children in same exposure group3,24, while others compared growth of extremely preterm with term, which may hide important differences by degree of preterm3. We identified only two previous publications that reported associations with BMI and height for subgroups of preterm and term born children13,25.

In addition, majority of studies have examined the association without repeat measurements of BMI or height; hence, evidence on the longitudinal association between gestational age and BMI and height is limited. A life-course approach that enables tracking of BMI and height during infancy, childhood and in adolescence by where individuals have reached their maximum height and a stabilized BMI may add to the evidence26.

The aim of this study was to examine longitudinal associations between length of gestation at birth and trajectories of BMI and height, respectively, from birth through adolescence. We explored sex-specific associations due to natural differences in growth27.

Methods

Study design

This is a longitudinal study using data from the Danish National Birth Cohort (DNBC), which includes information on 96,822 live-born children and their mothers28. Baseline information was planned to be collected at 12 and 30 weeks gestation, and follow-ups with information on height and weight were planned to be collected at 18 months and 7, 11 and 18 years. Information is linked with nation-wide registry data at Statistics Denmark29,30,31,32 as described below. Details of the DNBC are given elsewhere28.

Study population

Eligible for this study was any live-born singleton in the DNBC with a gestational age of 23–43 completed weeks at birth (n = 92,615) (Fig. 1). We excluded individuals who emigrated or died (n = 1826) during follow-up, and excluded individuals with less than two growth measurements after birth and without information on potential confounders: maternal education (n = 173), maternal pre-pregnancy BMI (n = 3894) and/or household income (n = 286). Before defining two specific populations for analysis, individuals with missing information on birth length (n = 374) or birth length < 30 cm or > 80 cm (n = 685), and with less than two measurements of height after birth (n = 22,749) were excluded.

Figure 1
figure 1

Flow chart of eligibility and inclusion in the study population in the study.

Full size image

In the sample for height, we excluded three additional individuals with height values ± 5 standard deviations from the mean following recommendations from Vidmar et al. leaving 62,625 individuals for analysis33.

In the sample for BMI, we excluded an additional 620 individuals without information on birth weight or with implausible combined values of birth weight and GA (n = 26), and individuals with less than two measurements of BMI (n = 594). Lastly, 39 individuals with BMI values ± 5 standard deviations from the mean was excluded, which left 61,969 individuals for analysis33.

Gestational age

Information on GA in days was obtained from the Danish Medical Birth Register29. The GA was reported to the register by the midwife at birth, based on results of ultra sound scans (at the time not part of recommended care, however made on almost 80% of women around week 18), anamnestic information about LMP and cycle length and regularity, and the clinical judgement of the child. GA was converted to completed weeks and categorized into seven groups: extremely preterm (23–27 weeks), very preterm (28–31 weeks), moderately preterm (32–33 weeks), late preterm (34–36 weeks), early term (37–38 weeks), term (39–41 weeks), and post term (42–43 weeks).

Body mass index and height

Information on birth weight and birth length was derived from the Danish Medical Birth Register29. Mothers were interviewed when the child was around 18 months and asked to report the weight and height registered in the ‘Child’s book’ at the preventive child examinations in general practice at scheduled ages of 5 and 12 months (measured at 1–22 months). These were used in the analyses together with height and weight reported by mothers at questionnaire-based follow-ups, scheduled at ages of 7 (measured at 5–7 years) and 11 years (measured at 10–14 years)28. At the 18 year follow-up (in practice at 17–19 years), height and weight were self-reported by the adolescent.

For each child, standardized scores (z-scores) of BMI and length at birth, and BMI and height after birth to age 19 years (228 months) were calculated separately for boys and girls. Standardization was done internally using 1-month categories.

Confounders

Confounders were selected a priori based on previous evidence2,34,35,36. Information on maternal height (cm), maternal pre-pregnancy BMI (kg/m2), and maternal smoking during pregnancy (yes/no) was taken from questionnaires at 12 and 30 weeks of gestation. Information on maternal age at delivery (continuous in years); gestational diabetes, ICD-10: O24 (yes/no); gestational hypertension, ICD-10: O13 (yes/no); and pre-eclampsia, ICD-10: O14 (yes/no) was derived from Danish Medical Birth Registry. Maternal education was obtained from the Danish Population’s Education Register and operationalised as highest ongoing or completed education at child’s birth according to international classification standards37: low [ISCED-2011: 0–2], medium [ISCED-2011: 3–4], high [ISCED-2011: 5–8]). Household income was based on disposable household income extracted from the Income Statistics Register. The variable was divided by an equivalence factor according to household size (available at http://www.oecd.org) and recoded into internal quantiles per year. Birth weight was perceived as an intermediate variable on the causal pathway, hence not included in the models38.

Statistical analysis

Linear-mixed effects models were used to estimate the association between GA in categories and BMI and height z-scores, respectively, from birth through age 19 years (228 months) while adjusting for confounders39. Analyses were performed by sex due to natural differences in growth trajectories27.

Linear splines are a commonly used type of regression spline for repeated measurements of non-linear growth trajectories40. Therefore, to best approximate the relationship between GA and the standardized growth measurements, we included linear splines in our models with four internal knots (5, 12, 85 and 134 months) and two boundary knots (at birth and 228 months). The knots were chosen a priori based on amount of available data40.

The linear mixed-effects model with linear splines (LME) accommodated for repeated measurements from the same child, without imposing any structure on the correlations among the time points using an unstructured covariance matrix. Also, in the LME models we accounted for clustering of children with same mothers. Age was set to 0 months at birth and included as a continuous variable, and the actual age in months at measurements of BMI and height was included in the models.

Furthermore, nearly one-third of individuals was excluded from the eligible sample (32%) (S1 Table), thus by adding inverse probability weights (IPWs) to the LME models we sought to remedy bias from this selection. The IPWs were based on variables that predicted selection into our analysis sample: sex, maternal age, maternal education, maternal smoking during pregnancy, maternal pre-pregnancy BMI, household income, parity, and caesarean section41.

The fitted LME models were used to predict sex-specific mean BMI and height z-scores by categories of GA with 95% confidence intervals (CIs) at six ages corresponding to the data collections: birth, 5 months, 1 year, 7 years, 11 years, and 18 years. For prediction, we used the following: mean maternal age at delivery (31.2 years), mean maternal pre-pregnancy BMI (23.5 kg/m2), mean maternal height (1.69 m), maternal education (high), household income (4th quartile), maternal smoking during pregnancy (no), gestational diabetes (no), gestational hypertension (no), and preeclampsia (no).

We performed sensitivity analyses (S3–S6 Tables), including a model only adjusting for sociodemographic variables, and models without adjustment of covariates, IPW and clustering.

Statistical analyses were carried out using the statistical software R version 4.042 and packages for ‘nlme’ and ‘splines’43.

Ethics approval

All methods were carried out in accordance with relevant guidelines and regulations. The DNBC data collection has been approved by the Regional Scientific Ethical Committee for the Municipalities of Copenhagen and Frederiksberg, and the Danish Data Protection Agency. Informed consent for study participation was obtained from the mother upon enrolment, and confirmed by the adolescent at age 18 years. Approval of the study was obtained from the Danish Data Protection Agency through the joint notification of The Faculty of Health and Medical Sciences at The University of Copenhagen (SUND-2017-09) and the study has furthermore been approved by the DNBC Steering Committee.

Results

Descriptive statistics

Baseline characteristics differed across categories of GA in the analysis sample (Table 1). All categories of preterm infants, relative to term, were more likely to have mothers with low educational level, who smoked during pregnancy, were diagnosed with preeclampsia or gestational hypertension, and delivered by caesarean section. Mean age at timing of measurements for height and weight was similar across categories of GA at all data collections, whereas mean BMI (kg/m2) was similar for preterm and term individuals after age 5 months, whilst height was consistently lower for preterm individuals throughout childhood and adolescence (S2 Table).

Table 1 Selected characteristics of the study population from children born in the Danish National Birth Cohort (1996–2003), n = 62,625.
Full size table

Trajectories of predicted BMI z-score by categories of GA

For extremely, very, moderately and late preterm children, the mean BMI z-score increased noticeably in the first year of life, with catch-up growth (gain in z-score of > 0.67) in preterm infants between birth and 5 months (Table 2, S1 Fig.)44. Accordingly, mean differences in BMI z-score between preterm and term infants attenuated in the first 12 months, though BMI remained lower in preterm than term.

Table 2 Predicted mean BMI z-score from birth to age 18 years by category of gestational age, n = 61,969 (estimated from LME-model).
Full size table

By age 7 preterm children had a slightly lower BMI z-score than term children with largest mean differences for very preterm boys (− 0.37, 95% CI: − 0.80 to 0.06) and extremely preterm girls (− 0.68, 95% CI: − 2.04 to 0.68). During adolescence the mean BMI z-score was similar across categories of GA, with the largest mean differences relative to term children observed for extremely preterm girls at both 11 years (− 0.42, 95% CI: − 1.94 to 1.10) and 18 years (− 0.28, 95% CI: − 1.49 to 0.92). Results from the sensitivity analyses with sociodemographic variables were only slightly different with estimates away from the null, while sensitivity analyses without IPW and clustering, respectively, were consistent with the main findings (S3, S4 Tables).

Trajectories of predicted height z-score by categories of GA

Extremely, very, moderately and late preterm children experienced catch-up in height (gain in z-score of > 0.67) within the first year of life (Table 3, S2 Fig.). By age 1 year, preterm children remained shorter than term children with the largest mean differences observed in extremely preterm boys (− 1.22, 95% CI: − 2.41 to 0.03) and girls (− 1.56, 95% CI: − 2.58 to 0.54) relative to term counterparts.

Table 3 Predicted mean height z-scores from birth to age 18 years by category of gestational age, n = 62,625 (estimated from LME-model).
Full size table

The mean difference in height z-score between preterm and term attenuated between age 1 year and 7 years, yet extremely, very and moderately preterm remained relatively shorter than term. By age 11 and through age 18 years individuals born preterm remained slightly shorter than term with mean differences being greatest in extremely preterm girls (− 1.00, 95% CI: − 2.08 to 0.08). Results from the sensitivity analyses were overall similar to the main finding, though the analysis with adjustment for sociodemographic variables only scarcely changed the estimates away from the null (S5, S6 Tables).

Discussion

This study investigated trajectories of predicted BMI and height z-scores across categories of GA from birth through 18 years. In the first year of life, BMI and height were lower in preterm than in term infants. Mean difference in BMI and height between preterm and term attenuated during childhood, and continued to decrease towards zero for BMI by adolescence. At 18 years of age, individuals born preterm remained only slightly shorter than children born at term.

This study agrees with two recent publications from United Kingdom (UK) (n = 475) and Australia (n = 478) showing that a mean difference in BMI between extremely preterm and term is largest in infancy, whilst decreasing during childhood through adolescence21,45. The two studies reported lower mean difference in BMI at age 6 years (− 0.98, 95% CI: − 1.23 to − 0.73) and 8 years (− 0.42, 95% CI: − 0.67 to − 0.18), respectively, which corresponds the magnitude of association for our results in girls at age 7 years (− 0.71, 95% CI: − 2.10 to 0.67). Similar to our findings, the studies from UK and Australia found no difference in BMI for extremely preterm and term adolescents at age 18 and 19 years, respectively.

Very and moderately preterm boys and girls remained lighter than term after birth through 7 years in our study. This is in line with findings from a recent cohort study from Brazil (n = 3036) reporting lower mean BMI in both boys and girls (born ≤ 33 weeks) aged 6 years25. Also, we found that late preterm had similar mean BMI as term already following the first 5 months of life through adolescence, as reported previously in a Chinese study on 7 169 children aged 14 years, and at age 18 years in the study from Brazil46.

The studies from Australia, UK, Brazil and China did not adjust for potential confounders such as maternal pre-pregnancy BMI and pre-eclampsia. While the results are in accordance with our findings, we consider it a strength of our study that the estimates are adjusted for key confounders and sex-specific across seven categories of GA contrasting previous publications.

We further found evidence to support that GA is positively associated with height in infancy with results indicating a linear association. This is in line with findings from a British study on 18 818 singletons aged 3 and 5 years, despite the authors reporting greater magnitudes of association for each GA-category (23–31 weeks, 32–33 weeks, 34–36 weeks, 37–38 weeks), respectively3. The analyses, however, were not adjusted for maternal height, gestational diabetes, gestational hypertension or preeclampsia.

For extremely preterm children, our study suggests that the height remain shorter after birth through adolescence, although the mean differences attenuate with age. These findings agree with results from Australia and UK where extremely preterm was shorter than term peers through infancy, childhood and adolescence compared to term peers21,45. The study from UK reported a lower mean height z-score in 315 extremely preterm aged 6 years (− 0.95, 95% CI: − 1.16 to − 0.73), 11 years (− 0.71, 95% CI: − 0.92 to − 0.50), and 19 years (− 0.81, 95% CI: − 1.14 to − 0.47), respectively, compared with term peers, which corresponds to the likes of the study from Australia.

Our findings support that gestational duration affects height in early life, and that this pattern persists and includes the age where individuals reach their maximum height. This contrast previous genetically informed studies challenging that exposures during fetal life is associated with postnatal growth47.

A strength of this study is the large sample size, and the ability to assess associations of seven categories of GA with BMI and height at birth and at scheduled ages 5 months, 1, 7, 11 and 18 years. Also, we were able to include maternal anthropometrics and measures of gestational hypertension and diabetes and preeclampsia.

A key limitation of our study is missing data on one-third of participants due to loss-to-follow up or lack of reporting child height and weight, or maternal pre-pregnancy BMI. To account for potential biases due to missing data, in the main analyses we accounted for missing data using IPWs and found that results were broadly consistent with results obtained in sensitivity analyses without IPWs (S3–S6 Tables).

Lack of growth measurements between age 2 to 5 and 13 to 16 years is also a limitation, which means we are not able to make any inference during two periods of dynamic change in size, namely adiposity rebound and puberty.

We relied on parental reporting of child weight and height from 7 to 11 years, and self-report at 18 years. Parental reporting at child height and weight may be prone to systematic error of under- or over-reporting of BMI, but given characteristics of the data collections we do not assume this would affect our overall estimates markedly48. Self-reporting of current BMI is a reasonably valid tool49, hence we neither considered this to bias the results. Our study uses data from a healthy population50 and there is a slightly overrepresentation of socially advantaged families in the study population (S1 Table). However, given the particular magnitudes we do not consider this to affect the validity of internal comparisons within the cohort, while earlier studies have also demonstrated that estimates obtained from the DNBC have been almost identical to those from completely unselected populations50. Also, the sensitivity analyses without IPW were consistent with the main analyses. DNBC is almost entirely comprised of women of Danish ethnic origin, thus replication of our findings in more diverse populations might be useful.

Importantly, to understand the causal mechanisms between the associations of GA and BMI and height, respectively, intermediate factors such as birth weight and breast feeding would be relevant to include for investigate in future studies.

Clinical implications

For preterm infants, the largest mean differences in BMI and height relative to term infants appear in the first years of life. However, despite magnitudes of potential clinical relevance in infancy (mean difference > 0.5 z-score)51, findings from our main and sensitivity analyses are reassuring for the growth of preterm individuals with the majority of children reaching similar height and BMI of term peers by end of adolescence Importantly, for a clinical setting these findings should be further considered in combination with maternal characteristics (e.g., socioeconomic position, gestational diabetes and hypertension, preeclampsia, smoking during pregnancy, and comorbidities).

Conclusion

The lower BMI in preterm infants relative to term infants equalizes during childhood, such that by adolescence there is no clear difference. Height is strongly positively associated with GA in early childhood, whilst by end of adolescence preterm individuals remain only slightly shorter than term peers.