1 INTRODUCTION

Caesarean-born children may show differences in their gastrointestinal tract microbiome composition, their oral microbiota composition1, 2 and immune and inflammatory responses3-5 from vaginal-born ones.3, 4 These may be the mechanisms triggering higher future risk of adverse health outcomes such as obesity,6 asthma,7 gestational diabetes8 and cardiovascular diseases.9 In this respect, it has also been postulated that caesarean delivery may influence the establishment of caries,10-12 since it may interfere in salivary immunoglobulin A (IgA) levels13 and result in changes in neonate oral bacterial colonization,1, 2 although no convincing evidence supporting the long-term effect of delivery type on oral microbiome perturbations is available14 at this time.

Socio-behavioural determinants may be the rationale for caesarean-born children having less caries. Caesarean section rates are higher among highly educated women,15, 16 also have children with lower rates of caries.17 Caesarean section rates have nearly doubled since the turn of the century, soaring from 12% of all births in 2000 to 21% in 2015.15 In populations where the caesarean section rate is higher than the 15% accepted for medical purposes according to the World Health Organization, socioeconomic factors are the main reasons behind the option for this delivery type.15, 16

Few studies have explored the relationship between delivery type and early childhood caries (ECC),10-12, 18 showing disparate findings. In a study conducted in Thailand, caesarean-born children had a lower prevalence of caries than those born by vaginal delivery.10 Three other studies using multiple logistic regression analysis11, 12, 18 conducted in India, Denmark, and Sweden showed no association between caesarean section and ECC.

Previous studies of the association between caesarean section and ECC have used multivariate regression models adjusting for confounders of exposure and outcomes,10, 11, 18 household income,11, 12, 18 maternal weight,12 child's gender10, 12 and child birth weight.10, 12, 18 However, marginal structural models (MSM) are a helpful tool to control confounder bias based on counterfactual logic. The tool balances the confounders across the exposure, thus providing a better approach for estimating causal effects.19

Based on the premise that caesarean section newborns show differences in their microbiota colonization and immune response from vaginal-born ones, this study hypothesized that caesarean-born children would be at a higher risk of ECC. Thus, we proposed to analyse the effect of caesarean delivery and ECC, estimating the association in multivariate regression models and MSM.

2 METHODS 2.1 Study design

This was an observational study conducted on a historical cohort of children and their mothers in São Luís, Maranhão, Brazil, to assess the risk factors associated with ECC.

The city of São Luís has seven health districts, 53 public preschool units or communities, comprising 2740 enrolled students, according to the list of public daycare centres provided by the Municipal Department of Education. Mother-child dyads were selected for inclusion in the study by two-stage cluster sampling: initially, we selected the districts and then the preschool units, resulting in the selection of five districts and 15 schools. ECC was defined as the presence of one or more decayed, missing or filled tooth surfaces in any primary tooth in children aged 71 months or younger.20 Thus, all children aged 24 to 71 months were eligible for the study in these school units. An estimated sample of 633 individuals in a 2:1 ratio (422 vaginal deliveries and 211 caesarean deliveries) would have 90% power to identify a difference of 13% in the prevalence of caries between the two groups. The exposed and control groups were considered to have 59% and 72% prevalence of caries, respectively, in a two-tailed test at a 5% level of significance. An additional 10% was added to the sample size to cater for losses, and the study collected data from 697 children and their respective mothers.

2.2 Data collection procedures

Data were collected at the daycare centres from August 2013 to March 2015. The mothers answered a structured questionnaire, including mother and child socioeconomic, demographic and health data.

The mother reported maternal hypertension based on a medical diagnosis. Child birth weight data were retrieved from the Child Health Handbook's records. The waist circumference measurement was used in the mother's anthropometric evaluation, which was gauged using a seamless Sanny® tape, with a 1 mm precision at the midpoint between the last costal arch and the iliac crest.21

The clinical oral examination for caries lesions was performed under natural light, after tooth brushing, using the decayed, missing and filled primary teeth index (dmft) for children.22 Data were gathered by a previously calibrated single examiner (ELC) with an intra-examiner agreement kappa of 0.86.

2.3 Theoretical model based on causal diagrams of directed acyclic graphs (DAGs)

The proposed theoretical model was constructed to analyse the association between caesarean section and ECC. Maternal income and maternal schooling were the antecedent determinants, exerting their effect on the exposure caesarean section,23, 24 on the outcome ECC17, 25 and on the model's covariates maternal obesity,26 maternal hypertension27 and child birth weight.28 Maternal obesity,29 child birth weight10, 12, 18 and maternal hypertension30 explained the caesarean section. Maternal obesity26 and child birth weight12 were associated with ECC. We also assumed that maternal hypertension could be associated with ECC, since meta-analysis has shown that hypertension is positively associated with excessive sugar consumption30, 31 (Figure 1).

Theoretical model based on DAGs of the association between delivery type and ECC [Colour figure can be viewed at wileyonlinelibrary.com] 2.4 Study variables

The exposure variable was the delivery type (0-vaginal; 1-caesarean section). The outcome was the number of teeth with caries experience (dmft), analysed as a discrete variable. The monthly household income was measured in Brazilian minimum wages in the 2013-2014 period, using the American dollar as comparative reference (USD 333.00) and categorized as ≤1 minimum wage; >1 up to ≤2 minimum wages and >2 minimum wages. Maternal schooling was categorized into years of study as <8, 8 to 11 and >11 years. Maternal waist circumference was categorized as waist circumference <80 cm (no metabolic risk); waist circumference ≥ 80 and less than 88 cm (high metabolic risk) and waist circumference ≥ 88 cm (very high metabolic risk).29 Child birth weight was categorized as low weight <2500 g and adequate weight ≥ 2,500 g.32 Waist circumference, child birth weight and maternal hypertension would be confounding maternal variables in the association between delivery type and ECC.

2.5 Statistical analysis

The theoretical model was depicted in DAGs for identification, using the DAGitty software, version 3.0. DAGs were elaborated to guide the analysis of the effect of caesarean section on ECC, controlling for confounding factors most appropriately and avoiding unnecessary adjustments.33

The bivariate Poisson regression analysis was initially used to estimate the association between caesarean section and ECC. Then, a multivariate analysis was performed, adjusted for the minimal set of variables according to the DAG rules. Poisson regression estimated the means ratio (MR) and the 95% confidence interval (95% CI) at a significance level of 0.05.

A causal inference analysis was also used to estimate the association between delivery type and ECC. The MSM estimation process was assessed in two steps.34 First, by calculating the weight to assign to each covariate, reflecting which are under-represented or over-represented in the sample, and then balancing the confounders across the exposure (caesarean section) using inverse probability of treatment weighting (IPTW). In the second one, the MSM, including exposure and outcome, was estimated by Poisson regressions and weighted by the inverse probability of caesarean section. The Poisson regression with endogenous treatment effects (etpoisson) was used by margins to estimate the average treatment effect (ATE), the average difference between the treatment and control of potential outcomes. The mean ratio (MR) coefficient was also estimated by Poisson regression analysis, weighted by the inverse probability of treatment. The treatment-effects commands (teffects IPW) were employed to verify the exchangeability (balance the covariates). All analyses used STATA software version 15.0.

A descriptive statistical analysis was performed using the relative and absolute frequencies of categorical variables and mean and standard deviation for continuous variables.

The Ethics Committee of the Federal University of Maranhão approved this study under opinion protocol 23115012534/2008-41. All mothers participating in the study signed an informed consent form.

3 RESULTS

This study used data from 679 mothers and children from São Luis, Maranhão, Brazil. Most of the sample consisted of children born by vaginal delivery weighing ≥2500 g. Socioeconomic, health and anthropometric data of the mothers are summarized in Table 1.

TABLE 1. Descriptive characteristics of mother-child dyad of the study, historical cohort, São Luís, Maranhão, Brazil (n = 697) Variables Without ECC With ECC TOTAL N % N % N % Delivery type Vaginal 318 67.2 143 69,8 461 67.89 Caesarean section 155 32.8 62 30.2 217 31.96 Missing - - - - 1 0.15 Household income ≤1 MW 313 67.2 150 75.7 463 68.19 >1 ≤ 2 MW 105 22.5 32 16.2 137 20.18 >2 MW 48 10.3 16 8.1 64 9.43 Missing - - - - 15 2.20 Maternal schooling ≤8 years 280 59.1 135 65.8 415 61.12 9-11 years 219 27.2 51 24.9 180 26.51 ≥11 years 65 13.7 19 9.3 84 12.37 Missing - - - - Child's age 1 8 1.7 1 0.5 9 1.33 2 72 15.2 24 11.8 96 14.14 3 145 30.6 52 25.4 197 29.01 4 200 42.2 103 50.2 303 44.62 5 47 9.9 22 10.7 69 10.16 6 2 0.4 3 1.5 5 0.74 Maternal waist circumference (cm) <80 213 44.9 86 41.9 299 44.04 80-88 124 26.2 52 25.4 176 25.92 >88 137 28.9 67 32.7 204 30.04 Maternal hypertension Yes 49 11.2 15 7.9 64 9.43 No 387 88.8 173 92.1 560 82.47 Missing - - - - 55 8.10 Child birth weight <2500 g 39 9.5 17 10.2 56 8.25 ≥2500 g 373 90.5 150 89.8 523 77.02 Missing - - - - 100

14.73

DAG rules specify that all the following variables should be included in the minimally sufficient adjustment set for confounders to estimate the total effect of the association between caesarean section and ECC (nonmediated effect), applying the backdoor criterion: socioeconomic determinants, maternal hypertension, maternal obesity and birth weight.

In the bivariate model, caesarean delivery was protective against higher numbers of decayed teeth (MR 0.81; 95% CI 0.70-0.94; P = .005), and this association remained significant in a multivariate model adjusting for confounders (MR 0.78; 95% CI 0.67-0.91; P = .002). However, the caesarean delivery did not affect the number of decayed teeth (ATE = −0.35; P = .107) in MSM weighted by the inverse probability of exposure (Table 2).

TABLE 2. Statistical models of Poisson regression (bivariate and multivariate) and causal inference model of the association between caesarean section and ECC Analysis models Estimated CI (95%) P-value Bivariate Model 0.81 (MR) 0.70 to 0.94 0.005 Multivariate Modela 0.78 (MR) 0.67 to 0.91 0.002 Marginal Structural Modelb −0.35 (ATE) −0.78 to 0.07 0.107 0.99 (MR) 0.72 to 1.38 0.995 95% CI—confidence interval; significant P-value < 0.05. Abbreviations: ATE, Average treatment effects; MR, Means Ratio. a Adjusted for income, schooling, waist circumference, hypertension and birth weight. b Weighted by the inverse probability of exposure, considering income, schooling, waist circumference, hypertension and birth weight to caesarean section explanation.

We also tested the multivariate model, including all significant Poisson regression interactions, to address why the marginal estimate diverged from the multivariate estimate. However, the difference between the multivariate and the marginal estimate was maintained (data not shown in tables).

4 DISCUSSION

In this study, caesarean delivery was protective for greater ECC severity using regression models (bivariate and multivariate). However, caesarean section was not associated with caries severity in the estimation carried out with MSM.

The seemingly protective effect of caesarean section against ECC severity shown in the bivariate and multivariate regression models in our study was supported by findings in a study in Thailand, where vaginally delivered children had more ECC.10 Caesarean section rates in our study and the Thai study10 are high, exceeding the 15% acceptable by WHO.34 Socioeconomic factors23, 24 frequently determine high caesarean section rates. Meanwhile, higher socioeconomic status is a protective for ECC.17 Thus, socioeconomic factors may underlie the spurious association between caesarean section protective for ECC, as observed in bivariate regression models. In this study, the association of caesarean section was protective for ECC in the multivariate model. However, this association was not maintained when we estimated the association with MSM. In this respect, MSM is a more appropriate tool for inferring causality in observational studies than multivariate regression adjustment because MSM studies balanced the confounders across the exposure using IPTW.35 Thus, it is vital to employ an estimator that considers predictive exposure models using propensity scores and subsequent estimation by MSM in populations with high caesarean section rates as in this study.

When estimated by MSM, our data showed no association between delivery type and ECC severity (number of decayed teeth). These data corroborate the findings of three studies in Denmark,11 India12 and Sweden,18 which found no association between caesarean section and ECC. These studies were conducted in countries where caesarean section rates are close to those recommended by the WHO for medical indication.36, 37 By contrast, the use of caesarean sections in Brazil is associated with high socioeconomic status.38

Although those studies have shown no association between caesarean section and caries experience in children,11, 18 one study of this nature emphasized that children born by caesarean section had a higher mean dmft than those delivered vaginally.12 By contrast, another study showed that the caesarean section decreased the risk of developing caries after three years of age.18 Thus, besides the caesarean section's socioeconomic determinants, residual confounders may have been better controlled in our MSM analysis, since we observed no tendency in the association between caesarean section and caries severity.

This was the first study to use MSM to infer the association between caesarean section and ECC. The conventional standard multivariate regression estimates the effect of an intervention conditional on all confounders. Thus, unpredictably, the presence of several confounders may have biased the direction of the effect estimation.39 MSMs have advantages over conventional regression approaches for estimating the causal effect, such as focusing on the research question. Furthermore, MSM replicated the findings of a randomized clinical trial using observational data,39, 40 simulating that the entire study sample had been randomized (caesarean section versus vaginal delivery) using the counterfactual concept.

Ideally, to investigate the association between caesarean section and ECC, it would be necessary to randomize caesarean section to the covariate's distributions in such a way that has no influence on the relationship between the exposure and outcome. Therefore, that randomization was simulated using IPW as estimator and generated a new pseudo-population. This approach may avoid self-selection bias by treatment.39 The IPTW estimator was used to weight the data, simulating the treatment randomization to caesarean section, balancing groups and imitating a randomization procedure.

The substantial increase in caesarean section rates, especially in developing countries, is of concern because a strong association with the future risk of noncommunicable diseases has been suggested.41-44 Two main explanations link caesarean sections and adverse health outcomes: these are the microbiome1, 2, 4 and immune response3, 4, 45 theories. A vaginal delivery would be the ideal trigger to awaken the immune system as the newborns adapt to their ‘new habitat’.4 At the time of the caesarean section, the lack of exposure to faecal bacteria and the maternal vagina results in a different neonatal intestinal microflora and a deregulated immune system.46

A limitation of this study is its historical cohort design, with a risk of recall bias. However, we believe that mothers easily remember the delivery type exposure in the elapsed time of 3-4 years. Furthermore, we could not confirm whether the assumption ‘no unmeasured confounders’ was satisfied for using MSM, since this is an untestable assumption. Thus, based on the background knowledge, we proposed the set of measurable covariates depicted in the DAG to minimize the risk of inaccurate measurement of confounders. All other assumptions of exchangeability, positivity and well-defined exposures were satisfied, allowing the MSM to be used for the causal analysis of observational data.40

The use of the MSM enhanced our ability to make causal inferences. Accordingly, the caesarean section did not remain associated with higher ECC severity in the models in closer alignment with estimating causality. Higher socioeconomic status underlies caesarean section rates, and it is a determinant of children having lower rates of caries. This should be better controlled in approaches for estimating the causal effect of observational data.

ACKNOWLEDGMENTS

The authors are grateful to FAPEMA (Maranhão State Research and Scientific and Technological Development Foundation) and CNPq (National Council for Scientific and Technological Development) for their financial support.

CONFLICT OF INTEREST

All the authors declare that they have no conflict of interests.

AUTHOR CONTRIBUTIONS

Lorena LCL, Elizabeth LC, Sarah PM contributed to the acquisition of data, analysis and data interpretation and drafted the manuscript. Cayara MC, Rubenice AS and Fabian CF contributed to the study design, data interpretation and critically reviewed the manuscript. Cecilia CCR contributed to the conception, data analysis and data interpretation and critically reviewed the manuscript. All authors have read and approved the final article.

REFERENCES

1Li H, Wang J, Wu L, et al. The impacts of delivery mode on infant’s oral microflora. Sci Rep. 2018; 8: 11938. 2Lif Holgerson P, Harnevik L, Hernell O, Tanner ACR, Johansson I. Mode of birth delivery affects oral microbiota in infants. J Dent Res. 2011; 90(10): 1183- 1188. 3Malamitsi-Puchner A, Protonotariou E, Boutsikou T, Makrakis E, Sarandakou A, Creatsas G. The influence of the mode of delivery on circulating cytokine concentrations in the perinatal period. Early Hum Dev. 2005; 81(4): 387- 392. 4Hyde MJ, Mostyn A, Modi N, Kemp PR. The health implications of birth by Caesarean section. Biol Rev. 2012; 87(1): 229- 243. 5Sandall J, Tribe RM, Avery L, et al. Short-term and long-term effects of caesarean section on the health of women and children. Lancet. 2018; 392(10155): 1349- 1357. 6Tun HM, Bridgman SL, Chari R, et al. Roles of birth mode and infant gut microbiota in intergenerational transmission of overweight and obesity from mother to offspring. JAMA Pediatr. 2018; 172(4): 368- 377. 7Chu S, Chen Q, Chen Y, Bao Y, Wu M, Zhang J. Cesarean section without medical in