Cesarean birth is not associated with early childhood body mass index

Exposure

Birth information was obtained from the Perinatal Statistics Collection and included all births in SA, 1999–2005. Information is collected by midwives and reported to the SA DHA. Women with vertex presentation, who previously had a cesarean and then had an elective cesarean or a spontaneous vaginal birth were eligible. All other births, including women who intended vaginal delivery but then needed emergency Cesarean section, were excluded. Exposures were chosen to reduce confounding by indication.

Outcome

Anthropometrical data (2003–2012) was provided by the Women's and Children's Hospital Network, SA DHA, from community-based well-child health checks. Children aged 3–6 years were included because this age group represents peak attendance for a preschool health check. Height (shoes removed) and weight (light clothing) were collected by nurses. Children's BMI was converted to age- and sex-specific BMI z-scores (BMIz) using the World Health Organization's Child Growth Standards 3, 4. Height-for-age z-scores (HFAz) and weight-for-age z-scores (WFAz) were also calculated. Z-scores were generated by the zanthro programme in STATA (13.0, TX, USA).

Confounding

The following potential confounders were identified a priori; maternal age (years), antenatal care (hospital, obstetrician, general practitioner, other), antenatal visits (≤7, 8–12, ≥13), medical conditions during pregnancy (asthma as a marker of chronic disease, diabetes, hypertension), smoking in pregnancy (yes/no), gestational age, birthweight for gestational age z-scores 5, mother had a partner (yes/no), maternal ethnicity (Caucasian, Aboriginal or Torres Strait Islander, Asian, Other), maternal occupation 6, neighbourhood-level indicators of socioeconomic disadvantage 7 and remote residence 8.

Statistical analyses

The distribution of confounders across the exposure groups were compared by t-tests and χ² tests. Associations were analysed using augmented inverse probability of treatment weighting (aipw) 9, 10. This method generates the probability of having a cesarean given an individual's confounders and weights a regression by this probability to calculate the average treatment effect (ATE). Weighting helps to better balance confounders. The ATE can be interpreted as the average effect for the population of cesarean compared to vaginal birth on anthropometry. Sensitivity analyses excluding women with diabetes or hypertension were undertaken because of associations with planned cesareans and childhood anthropometry.

Of 16 375 women who previously had a cesarean, 61% birthed by elective cesarean (9925/16 375) and 11% had spontaneous vaginal birth (1842/16 375). The remaining 28% had emergency cesarean or instrument-assisted deliveries (4608/16 375) and were excluded. Successful linkage to anthropometrical data included 4099 children (3909 with BMIz measurements, 4817 WFAz and 3993 HFAz).

Sample characteristics (Table 1) show that although the exposure groups intended to minimize confounding by indication, imbalances remained. For example, women who had cesarean births were older and had higher proportions of diabetes and antenatal care by an obstetrician than those who had vaginal births.

In unadjusted analyses, HFAz was similar but BMIz and WFAz were higher among children born by cesarean compared to vaginal births (Table 2). The apiw analysis suggested no differences in BMIz, WFAz and HFAz according to mode of delivery. Sensitivity analyses excluding women with diabetes and hypertension were consistent with these main findings (BMIz ATE = 0.05, 95%CI −0.46, 0.54, p = 0.856, n = 3180; WFAz ATE = 0.48, 95%CI −0.32, 1.29, p = 0.240, n = 3909; HFAz ATE = 0.38 95%CI −0.38, 1.15, p = 0.329, n = 3251).

In our analyses, we found little evidence of an association between cesarean birth and the BMI of 3 to 6-year-old children. Previously, cesarean birth has been associated with 18% higher odds of obesity (95%CI 1.09, 1.27) 11, although there is evidence of publication bias 11. Other large population-based studies have reported null 12 and positive 13 associations between non-medically indicated cesareans and adiposity. It is difficult to judge whether differing analytical approaches, samples or definitions of cesareans (e.g. non-medically indicated nullipara, low-risk multiparous) have contributed to the mixed findings. Whether an association is manifested at younger or older ages is conflicting 14, 15, and our dataset has too few children at older ages to examine this issue.

We attempted to reduce confounding by accounting for many variables associated with both cesarean and obesity. This was important because baseline characteristics demonstrated imbalances. A potential limitation is that confounding by women with high risk pregnancies (e.g. type 1 diabetes) may remain, as these women often plan cesarean births and their children may be at higher risks of obesity. However, excluding women with diabetes or hypertension supported the main analyses. Another limitation is we were unable to adjust for other known confounders including maternal pre-pregnancy BMI 16, pregnancy weight gain or physical activity, as these data were not available. We hypothesize that adjustment for these confounders is likely to attenuate the association between cesarean and BMI even further.

A critique of this work may be that it is not ‘representative’ of typical cesareans, but representativeness was not our goal. Our intent was to examine whether cesarean section causes obesity by attempting to address both confounding and confounding by indication with more rigorous epidemiological methods than past studies. There is no doubt that the gut microbiome of children differs between cesarean and vaginal births; however, this study casts doubt over cesarean-induced microbiome differences on anthropometry of young children.