Newborn weight change and childhood cardio-metabolic traits – a prospective cohort study

The participants of this study are part of the Generation XXI birth cohort [16], assembled between April 2005 and August 2006, after delivery, during the hospital stay, at the five public units providing obstetrical and neonatal care in the metropolitan area of Porto, Portugal. Follow-up assessments of the cohort have been undertaken when the children were aged 4 years (April 2009 – July 2011), and 7 (April 2012 – January 2014).

All newborns were routinely weighed at birth and, since November 2005, whenever possible, newborns additionally had a second weight measurement performed by a trained examiner during their hospital stay. Since November 2005, 5034 newborns were recruited to Generation XXI, of which 4449 were full-term singletons without known congenital anomalies. Of those 4449 children, a random sub-sample of 1806 newborns had the second neonatal weight measurement. This group of 1806 children were eligible for this study and of that 1806, 471 had missing information on exact time of measurement during hospital stay, 28 were measured after 96 h of life (the period of interest was the first 96 h of life) and 19 were considered outliers [1st/3rd quartile ±3 times the interquartile range corresponding to those with weight loss higher than 0.50% of birthweight per hour (n = 15) and those with weigh gain higher than 0.19% of birthweight per hour (n = 4)]. Of the remaining 1288, 312 children had complete data on all key variables and are the participants included in this study. Figure 2 shows the study flow chart of participants. Similar characteristics were found between participants and eligible non-participants (Additional file 1).

The second weight measurement of the newborn was performed by trained examiners, during the hospital stay, but independently of routine procedures. Newborns were weighted after the questionnaire to the mother, without clothes or diaper. The same digital scales were used (seca®) to weight all newborns to the nearest 1 g, after waiting for the scale to stabilize. The date and time of measurement were registered. The measurement time varied from 6.3 to 96.0 h of life, mean of 45.3 (SD 19.4) hours. We calculated NWC using the formula:

NWC(%) = ((estimated minimum weight – birthweight) / birthweight) × 100.

where estimated minimum weight was the predicted weight at 52.3 h of life - the mean nadir time of lowest weight in the first 96 h in European infants [13]. This was estimated for each child using a cubic regression model as described in the statistical analysis section.

Information on family and personal history of disease, socio-demographic characteristics, maternal pre-pregnancy anthropometric parameters, and intra-uterine exposures were collected during a face-to-face interview conducted during the hospital stay by trained interviewers. These interviews took place 24 to 72 h after delivery. Data on delivery and newborn characteristics, including birthweight and gestational age, were abstracted from clinical records by the same interviewers.

At 4 and 7 years of age, trained researchers performed anthropometric and blood pressure measurements and obtained a fasting blood sample, according to standard procedures. Waist circumference measurements were taken with an inextensible tape measure to the nearest 0.1 cm, at the umbilicus level, with the child in a standing position, the abdomen relaxed, arms at the sides and feet positioned together [17]. Blood pressure was measured with an electronic sphygmomanometer (Omron®), with the child conformably sitting in a chair, with the cuff on the non-dominant arm, 2–3 cm above the elbow (without clothes compressing the arm). Two measurements of systolic (SBP) and diastolic (DPB) blood pressure, separated by at least 5 min, were taken after 10-min rest. If the difference between them was less than 5 mmHg for SBP or DBP, the mean was calculated. If the difference was larger than 5 mmHg a third measurement was taken and the mean of the 2 closest values was used [18]. After an overnight fast, a venous blood sample was collected before 11 a.m., according to standard procedures, after applying a topical analgesic cream (EMLA cream). This blood sample was centrifuged at 3500 rpm for 10 min and then the supernatant (serum) was stored at − 80 °C. Glucose was measured using UV enzymatic assay (hexokinase method), total and high density lipoprotein-cholesterol (for posterior calculation of LDL-C using Friedewald equation) [19], and triglycerides (TG) using an enzymatic colorimetric assay, in the Clinical Pathology Department of Centro Hospitalar São João, Porto, Portugal.

Our outcomes were: glucose, LDL-C, TG, waist circumference, SBP and DBP. In order to be able to compare magnitudes of association between different cardio-metabolic traits, we generated age- and sex-specific z-scores. For fasting glucose, LDL-C, TG and waist circumference this was done using the age- (in 6-months categories) and sex-specific means and standard deviations (SD) from the whole Generation XXI cohort. For SBP and DBP, we generated age-, sex- and height-specific z-scores using the means and standard deviations recommended by the American Academy of Pediatrics, in order to generate measures of BP that are independent of height (a major contribution to BP variation in children) [18]. High levels of the outcomes were considered when above the 90th percentile.

All the phases of the study complied with the Ethical Principles for Medical Research Involving Human Subjects expressed in the Declaration of Helsinki. The study was approved by the University of Porto Medical School/ Centro Hospitalar São João ethics committee and all parents or legal representative signed an informed consent according Helsinki.

STROBE checklist for the present manuscript can be found in Additional file 2.

The estimated nadir time was 52.3 h of life, thus, weights used to calculate NWC were birthweight and the weight estimated at 52.3 h of life. A cubic model with random intercept and slope by subject [weight(t)~3241.442 + (−9.378) × t + 0.119 × t2 + (−0.0004) × t3 + b0i + b1i × t] was used to estimate the weight according to the newborn’s age represented as t in the formula (this analysis was performed using R version 2.14.1) [13].

Proportions were compared using the chi-square test, and means were compared using student’s t-test or ANOVA (analysis performed using SPSS version 21.0). Pearson correlations were also computed. Crude and adjusted linear regression coefficients (β) and 95% confidence intervals (95% CI) were computed using path analysis. Full information maximum likelihood estimation was used to handle missing values, assuming missing at random [20]. We conducted path analysis based on the theoretical model depicted in Fig. 1 and tested the fit of the model with potential confounders. The final model included NWC, maternal education, maternal pre-pregnancy BMI, gestational age, and birthweight as explanatory variables. Path analysis was performed with Mplus software (version 7); 95% confidence intervals were calculated by bootstrapping. The fit of the models was assessed using different indexes: the Comparative Fit Index (CFI) [21], the Tucker–Lewis Index (TLI) [22], and the Root Mean Square Error of Approximation (RMSEA) [23]. A good model fit is indicated by a CFI and TLI values ≥0.90 and values of RMSEA close to 0.

A large proportion of the main cohort participants were not included in our analyses. However, distributions of maternal and neonatal characteristics between participants and eligible non-participants were similar. Our finding that NWC during the first 96 h was not associated with childhood cardio-metabolic traits might be due to lack of statistical power, but the point estimates were all close to the null and the 95%CI were narrow, suggesting that we had adequate power to obtain precise estimates and rule out an important association.

As weight measurements were not taken at regular periods for each newborn, it is possible that the precise nadir of NWC was not detected. Nevertheless, a systematic review [12] found a mean NWC ranged from − 5.7% to − 6.6% and a nadir around the 2nd and 3rd days following birth. These results support our methodology as we found a mean NWC of − 6.7% occurring at 52.3 h of life [13], suggesting that we correctly estimate the actual nadir in our sample.

We used path analysis to explore associations. This is an extension of regression analysis, which allows for simultaneous estimation of the interrelations between variables. We used it here because it is a useful method for comparing the magnitudes of effects between variables with complex interrelations or to test the plausibility of mediation effects [36].

Our exposure – NWC – measures the growth occurring within 96 h, which is a very narrow period. It is known that the measurement error on weight change is higher and the capacity to identify associations between growth periods and outcomes reduced, when the measurements are closer in time and the weight change is smaller [37]. Since measurement error in NWC is unlikely to be related to later child cardio-metabolic outcomes (the midwives and trained researchers who measured birth and later weight would have no way of knowing the future cardio-metabolic traits of the newborns they were assessing), the expectation would be for this random error to attenuate the observed associations, however the bias would also depend on relationships of the error to other explanatory variables in the model. We doubt it would fully explain the consistent null results we find across all traits.

Although cardio-metabolic traits in early childhood track throughout childhood (as we showed) and into adulthood, clinically abnormal values of cardio-metabolic risk factors, such as fasting glucose or SBP, are unusual in childhood [38], and it is possible that the medium/long-term effects of NWC on cardio-metabolic health are not yet fully evident at the ages we have assessed. Future research conducted as this population ages will enable this possibility to be explored.