Association between mode of delivery and body mass index at 4-5 years in White British and Pakistani children: the Born in Bradford birth cohort

There is limited published research on the direct association of mode of delivery and child BMI, at the age of children starting school (4–5 years). This association has not been investigated in UK South Asian mothers compared to WB mothers.

The aim of this study was to determine if there is any association between mode of delivery (CS and vaginal delivery) and BMI at 4–5 years of age, in the Born in Bradford (BiB) cohort, and if this differs between ethnicities (WB and Pakistani).

Born in Bradford (BiB) is a longitudinal multi-ethnic birth cohort study. BiB aims to investigate parent and child wellbeing by examining physiological, environmental and genetic factors in the City of Bradford [31]. Bradford is situated in the north of England; it is ethnically diverse and has high levels of socio-economic deprivation. BiB recruitment occurred from September 2007 to December 2010. Women who attended the Bradford Royal Infirmary at 26–28 weeks gestation for a routine glucose-tolerance test, which is offered to all women booked to give birth in Bradford, were invited to join the study and 87% of those approached agreed to participate. It is at this point that women were weighed, and their height measured. Weight at first antenatal clinic assessment (median 12 weeks’ gestation) was extracted from the antenatal records and this weight, together with height measured at recruitment, was used to calculate the woman’s early pregnancy BMI (kg/m²).

The BiB population is broadly representative of the maternal population in Bradford [31]. Twelve thousand four hundred fifty-three pregnant women gave consent to be involved with the study. One woman could contribute more than a single pregnancy, resulting in a total of 13,776 pregnancies being reported, which gave rise to 13,858 children. A baseline questionnaire was conducted at recruitment. The mother self-reported most variables, including ethnicity, socioeconomic indicators, alcohol-related and smoking habits. In this analysis, the mother’s self-reported ethnicity was used to define their child’s ethnicity. School nurse teams took anthropometric measurements of children in reception class (aged 4–5) as part of the National Child Measurement Programme (NCMP). Where anthropometric measurements were missing from NCMP data, it was possible to obtain some of these measurements from Primary Care and Child Health Records at this age. Of the non-missing anthropometric data, 82.2% originated from NCMP, 10.2% from Primary Care, 7.6% from Child Health Records. A subgroup of mothers was followed up for data on breastfeeding, at 6, 12, 18, 24 and 36 months. This subgroup was part of BiB1000, a nested cohort of the BiB prospective birth cohort [32].

From the total 13,858 children enrolled in BiB, 7448 were excluded due to not meeting the inclusion criteria of having data on mode of delivery, BMI at 4–5 years, singleton birth and being WB or Pakistani. As approximately a third of the children enrolled in BiB are missing outcome data (BMI at 4–5 years), a characteristics table comparing the final study population (n = 6410) and those with missing BMI at 4–5 years (n = 4154) is presented in the supplementary material. There is no suggestion of selection bias at this stage.

Mothers self-reporting an ethnicity other than WB or Pakistani were excluded as they were a very heterogeneous ethnic group (429 = Indian, 288 = White Other, 253 = Bangladeshi, 226 = Black, 108 = Mixed-White and Black, 61 = Mixed-White and South Asian, 309 = other). This left a sample size of 6410 children (Fig. 1). The original BiB cohort had a similar distribution of child BMI and mode of delivery to the study population.

Exposed participants were children who were delivered by elective or emergency CS. Unexposed participants were children who were delivered vaginally; including normal, forceps and ventouse extraction deliveries. Mode of delivery was recorded by a midwife or paediatrician within the first 6 h of life. Paper forms with handwritten notes were entered into the routine eClipse electronic maternity record as neonatal data.

BMI values of children aged 4–5 years, recorded as part of the NCMP by school nurse teams [33] or obtained from Primary Care or Child Health Records, were transformed to a standardised measure (z-scores). The z-scores were calculated using the LMS method. This is prepared via an Excel spreadsheet, which can be obtained online for free [34]. The LMS growth application includes access to a 1990 UK (UK90) reference population. Using this reference, each individual is assigned a z-score which adjusts for age, sex and the BMI distribution for skewness [35]. The UK90 reference group is recommended for population monitoring and clinical assessment in children aged 4 years and over [36]. It serves as an anchor for comparison; it is used by the NCMP and has been used for other BiB studies [37]. Children with BMI z-scores above the 85th percentile were classified as overweight, and those above the 95th percentile as obese [38].

Sample size calculations were conducted in OpenEpi [39]. This study had a power of 99%, determined from a post-hoc power calculation using the parameters from this study (vaginal to CS ratio of 3.71 and prevalence of childhood overweight or obesity in vaginal births at 14.8%), and the odds ratio from a previous study (odds ratio of 2.10 (95% CI 1.36–3.23) of obesity in children aged 7 years, by CS compared to vaginal birth) [10].

The forward selection approach was used to create regression models. Potential confounders were added individually according to their effect size. The covariate was retained in the model if there was an appreciable (10%) difference in effect size of mode of delivery on z-score.

The final multivariable linear regression model assessed the association between mode of delivery and BMI z-score. Preliminary analysis confirmed the assumptions of the regression were met; z-scores were normally distributed and lacked collinearity. BMI z-scores were also categorised and logistic regression models performed to obtain odd ratios for being overweight and obese in children delivered by CS. All the study population (n = 6410) contributed to the unadjusted regression modeling. Complete case analysis was used for the adjusted regression modeling (n = 6115).

Potential effect modification was judged by stratifying the final model by ethnicity to observe the separate association of mode of delivery on z-score in WB and Pakistani children. A formal test for effect modification was also conducted; a likelihood ratio test compared the final model with a model which also included an interaction term between mode of delivery and ethnicity.

The number and proportion of patients missing data on descriptive variables was described. Complete case analysis was used for the multivariate analysis. No imputation was performed.

Approximately 85% of the study population had missing data on breastfeeding (n = 5439 missing). Due to the large proportion of missingness, breastfeeding was not assessed in this study. Maternal parity (n = 273 missing) and maternal BMI (n = 275 missing) were missing for about 4% of the study population.

Tables 1 and 2 summarise the baseline characteristics of the study population stratified by mode of delivery and ethnicity, also visualised in Fig. 2. In this study, 21.3% (n = 1361) of babies were delivered by CS. There were more Pakistani mothers (54.6%, n = 3502) than WB mothers (45.4%, n = 2908). Amongst Pakistani mothers, 19.8% had CS, whereas the CS prevalence among WB mothers was higher at 23.0%. Most children were underweight/ healthy weight (84.7%), and fewer were overweight (10.0%) or obese (5.2%) (Table 2). The mean BMI z-score was slightly higher in CS deliveries (0.32) than vaginal deliveries (0.22) at age 4–5 years. Furthermore, the mean BMI z-score was higher in WB children (0.43) than Pakistani children (0.08).

Almost half the mothers in this study population were underweight or healthy weight (49.2%). Obesity was more prevalent in women who gave birth by CS (30.3%), compared to mothers giving birth vaginally (18.9%). WB mothers had a slightly higher BMI (+ 1.30 kg/m²) than Pakistani women. Leading to a higher prevalence of obesity in WB mothers (25.0%) than Pakistani women (18.2%).

Mothers who had CS often had achieved a higher level of education than mothers with vaginal deliveries (26.7% of CS mothers, 21.0% of vaginal mothers). More Pakistani women achieved higher than A-level qualifications compared to WB women (24.6 vs 19.3%).

Most mothers were not currently employed (67.7%). Current employment was more common in women who had CS births, compared to those with vaginal births (49.9 and 40.2%, respectively) and also more common in WB women (66.2%), compared to Pakistani women (22.4%). Additionally, a higher proportion of mothers having vaginal deliveries (44.5%) received benefits than those having CS deliveries (36.5%). Receiving benefits was more common in Pakistani mothers (47.8%) than WB mothers (36.7%).

The mean age of mothers who had a CS was 28.8 years old, which was 1.7 years older than those who gave birth vaginally. Also, mothers having vaginal deliveries were more likely to be multiparous (62.8%) compared to mothers having CS deliveries (54.8%). The difference in mean gestation period between CS and vaginal deliveries was minimal (4 days difference). A larger proportion of preterm births was experienced by mothers who had a CS (8.6%), compared to mothers who gave birth vaginally (4.1%).

Only 10 out of 3490 Pakistani mothers (0.3%) drank alcohol during pregnancy or 3 months before, whereas 70.7% of WB mothers reported alcohol consumption. Further to this, mothers having CS were marginally more likely to have drunk alcohol during pregnancy or 3 months before (36.4% of CS mothers and 31.2% of vaginal mothers). More WB women reportedly smoked during pregnancy than Pakistani women (33.1% of WB women, 3.0% of Pakistani women).

Gestational diabetes was more prevalent in mothers having CS (10.8%) compared to those having vaginal deliveries (6.8%). Additionally, prevalence was higher in Pakistani mothers than WB mothers (10.0 and 4.9% respectively).

Children born by CS had a mean birthweight 28.3 g lower than vaginal births. Irrespective of mode of delivery, children with WB mothers had a higher mean birthweight than those with Pakistani mothers (3357.6 g and 3138.9 g, respectively).

The unadjusted linear regression calculated the predicted difference in z-score between mode of delivery, the z-score being higher with CS (n = 6410, difference = 0.103; 95% CI = 0.035–0.170) (Fig. 3). The adjusted model calculated the predicted mean difference in z-score between mode of delivery, controlling for all factors which met the confounding criteria (ethnicity, maternal BMI (continuous), maternal job status and maternal drinking of alcohol during pregnancy or 3 months before), there was no difference in child BMI z-score (n = 6115, difference = 0.005; 95% CI = −0.062–0.072) (Fig. 3).

The proportion of variance in BMI z-scores explained by the mode of delivery was 7.49% for the adjusted model (adjusted R² value). The F ratio (71.76) shows how much variability the model can explain relative to how much it cannot explain. The standard error (0.034) did not differ between unadjusted and adjusted models, suggesting an absence of collinearity.

The adjusted logistic regression models with the outcome of overweight and obesity obtained odds ratios and confidence intervals of no strong statistical support for a difference in odds between mode of delivery (Fig. 4).

Stratified analysis suggests the adjusted association of mode of delivery on BMI z-score is similar, irrespective of ethnicity (Fig. 3). Additionally, there was no evidence of effect modification from the likelihood ratio test (p = 0.97). When a test for effect modification was performed on the categorised z-scores, similar results were obtained. There was weak evidence of effect modification by ethnicity on the association between mode of delivery and overweight (p = 0.14), and no evidence for effect modification by ethnicity on the association between mode of delivery and obesity (p = 0.79).