In this cohort study, it was found that undergoing a CS was not associated with an increased risk of overweight and obesity in children, and there was no difference between ethnic groups.
Mothers who had undergone a CS were generally of higher socio-economic status than those who had vaginal deliveries; CS women were more educated, more likely to be currently employed, more likely to have a stable housing situation and less likely to be receiving benefits. Also, mothers who experienced CS had baseline characteristics to suggest they had poorer health than mothers giving birth vaginally; CS mothers had a higher mean BMI, were more likely to drink alcohol during pregnancy or 3 months before and more prevalent gestational diabetes. The distribution of alcohol drinking varies vastly between ethnicities, this is most likely due to religious beliefs (30). This explains the very low prevalence of alcohol drinking and avoidance of smoking amongst Pakistani women.
The linear regression for the adjusted model offers no evidence for a difference in BMI z-score between children born via CS and vaginal deliveries. The low adjusted R2 value suggests there are other variables which have an influence on the primary association. The adjusted logistic regression models also suggest no evidence for children delivered by CS having different odds of being overweight or obese, compared to children of vaginal deliveries.
The stratified analysis and formal test for effect modification both indicate there is no evidence that the association between mode of delivery and children’s BMI z-scores varied by ethnicity.
As discussed in the introduction, previous studies have varied interpretations. Two leading systematic reviews suggest there is evidence that CS increases child BMI (7, 8). However, there were several studies which found no statistical association between mode of delivery and child BMI. The findings from this project are compatible with the later studies mentioned.
Two out of three studies conducted in the UK concluded there was no ‘statistical significant’ difference in risk of childhood overweight or obesity between modes of delivery, at 3 years old (38) and 5 years old (27).
Furthermore, maternal BMI explained most of the observed association in this study and was hence the main confounding factor. All previous studies looking at the association between mode of delivery and BMI, also adjusted for this factor. Additionally, none of the previous studies identified any confounders that had a biologically plausible reason to include them in the unadjusted model.
However, this study also differs with previous research. The other UK study found that CS increased the odds of being overweight or obese, at 7 years old (18). This was a study which used data from the Avon Longitudinal Study of Parents and Children (ALSPAC); participants were recruited from the Avon area if they were born in 1991–1992.
Several confounders (child gender, gestational factors and child feeding patterns) were adjusted for in the ALSPAC study but did not meet the confounding criteria to be adjusted for in this study. There were also inconsistencies with other factors adjusted for in this study compared to previous studies, such as not adjusting for antibiotics during pregnancy (19). Different factors could have met the confounding criteria in previous studies due to their population type, for example, by having a different BMI distribution as the children were leaner.
Adjusting for ethnicity was not seen in previous research in the UK. Due to the large proportion of Pakistani women in this BiB study, there was sufficient power to investigate differences between WB and Pakistani ethnic groups, whereas this would not be possible in studies like ALSPAC. As previous studies did not adjust for ethnicity, other variables could have acted as confounders. Overall, the differences in study design, study population and confounding adjustments could explain the inconsistent conclusions reached.
The large sample size used in this study allowed sufficient power to identify any meaningful differences in association between BMI z-scores of two different modes of delivery. Additionally, consistent statistical methodology with previous studies was used and there was minimal recruitment bias due to the BiB study having a high recruitment rate of 87%.
As CS and vaginal deliveries are very different procedures, in theory, there was no opportunity for this to be incorrectly recorded. Hence no information bias, in the form of differential misclassification, should have occurred. Furthermore, observer bias would not arise when recording the child’s BMI, as nurses taking anthropometric measurements at ages 4–5 were blinded to information regarding the child’s mode of delivery.
There is evidence to suggest BMI measurements systematically underestimate childhood adiposity (39). This has also been specifically investigated in South Asians with evidence to suggest that BMI additionally appears to underestimate adiposity in this ethnic group. Despite South Asians being generally smaller and lighter, they seem to have greater relative fatness compared to white European populations (40).
Most of the data on covariates were collected in the baseline questionnaire, completed by the mother. As data were self-reported, information bias in the form of differential misclassification could have occurred which would tend the results to overestimate or underestimate the true association. An example would be smoking as this is a likely factor to be underreported. Underreporting could underestimate the association between mode of delivery and smoking, which would have led to it not being adjusted for in the final analysis.
There may be residual confounding which is obscuring the true effect of mode of delivery on child BMI. The low adjusted R2 value implies other factors could have an influence on the association, therefore suggesting factors which were not included in the analysis explained some of the association. These could be factors such as amount of exercise, feeding pattern of the child or breastfeeding. Adding the effect of breastfeeding on childhood adiposity as a covariable in the multivariable analysis would have been desirable, but this was not possible due to the missing data on breastfeeding. Approximately 85% of the study population had missing data on breastfeeding due to the data being collected in a subgroup of women who participated in the BiB1000 as described in the study design section.
Additionally, the generalisability of the results is limited to the BiB population, however, similarities with other multi-ethnic UK cities are likely to be seen.