Association of caesarean delivery with offspring health outcomes in full-cohort versus sibling-comparison studies: a comparative meta-analysis and simulation study

Background Full-cohort and sibling-comparison designs have yielded inconsistent results about the impacts of caesarean delivery on offspring health outcomes, with the effect estimates from the latter being more likely directed towards the null value. We hypothesized that the seemingly conservative results obtained from the sibling-comparison design might be attributed to inadequate adjustment for non-shared confounders between siblings, particularly maternal age at delivery. Methods A systematic review and meta-analysis was first conducted. PubMed, Embase, and the Web of Science were searched from database inception to April 6, 2022. Included studies (1) examined the association of caesarean delivery, whether elective or emergency, with offspring health outcomes; (2) simultaneously conducted full-cohort and sibling-comparison analyses; and (3) reported adjusted effect estimates with 95% confidence intervals (95% CIs). No language restrictions were applied. Data were extracted by 2 reviewers independently. Three-level meta-analytic models were used to calculate the pooled odds ratios (ORs) and 95% CIs for caesarean versus vaginal delivery on multiple offspring health outcomes separately for full-cohort and sibling-comparison designs. Subgroup analyses were performed based on the method of adjustment for maternal age at delivery. A simulation study was then conducted. The simulated datasets were generated with some key parameters derived from the meta-analysis. Results Eighteen studies involving 21,854,828 individuals were included. The outcomes assessed included mental and behavioral disorders; endocrine, nutritional and metabolic diseases; asthma; cardiorespiratory fitness; and multiple sclerosis. The overall pooled OR for estimates from the full-cohort design was 1.14 (95% CI: 1.11 to 1.17), higher than that for estimates from the sibling-comparison design (OR = 1.08; 95% CI: 1.02 to 1.14). Stratified analyses showed that estimates from the sibling-comparison design varied considerably across studies using different methods to adjust for maternal age at delivery in multivariate analyses, while those from the full-cohort design were rather stable: in studies that did not adjust maternal age at delivery, the pooled OR of full-cohort vs. sibling-comparison design was 1.10 (95% CI: 0.99 to 1.22) vs. 1.06 (95% CI: 0.85 to 1.31), in studies adjusting it as a categorical variable, 1.15 (95% CI: 1.11 to 1.19) vs. 1.07 (95% CI: 1.00 to 1.15), and in studies adjusting it as a continuous variable, 1.12 (95% CI: 1.05 to 1.19) vs. 1.12 (95% CI: 0.98 to 1.29). The severe underestimation bias related to the inadequate adjustment of maternal age at delivery in sibling-comparison analyses was fully replicated in the simulation study. Conclusions Sibling-comparison analyses may underestimate the association of caesarean delivery with multiple offspring health outcomes due to inadequate adjustment of non-shared confounders, such as maternal age at delivery. Thus, we should be cautious when interpreting the seemingly conservative results of sibling-comparison analyses in delivery-related studies. Supplementary Information The online version contains supplementary material available at 10.1186/s12916-023-03030-2


Background
Caesarean delivery plays a crucial role in tackling medical conditions, such as abnormal placentation, dystocia, fetal distress, and previous caesarean delivery [1].Over the past 5 decades, the global caesarean delivery rate has increased from 5% in 1970 to 21.1% in 2018 [2], exceeding the level of 15% endorsed by WHO [3].The growing popularity of caesarean delivery has caused widespread concern about its potential negative impacts on maternal and offspring health [4].Population-based cohort studies from different settings suggest an association of caesarean delivery with multiple health outcomes in offspring, such as obesity, asthma, type 1 diabetes, and attention deficit hyperactivity disorder (ADHD) [5][6][7][8][9], but whether these findings reveal causation has remained much debated primarily due to potential biases from uncontrollable confounders.More recently, studies have attempted to sidestep such confounding effects by using a sibling-comparison design, which could presumably adjust for unmeasured confounding factors shared by siblings (e.g., cultural background, parental characteristics, and child-rearing practices) and thus may generate more reliable results in some contexts [10,11].In most studies that simultaneously used these two designs, the sibling-comparison analyses did generate less significant results with respect to the impacts of caesarean delivery on offspring health outcomes, enhancing the speculation that the associations observed in full-cohort analyses were likely due to uncontrolled or residual confounding [12][13][14][15][16].However, whether sibling-comparison analyses are more reliable than full-cohort analyses in this specific context remains largely unknown.
Mathematically, effect estimates from studies with sibling-comparison versus unpaired full-cohort design may be more biased due to the confounding of non-shared factors among siblings [17].Maternal age at delivery may be an important non-shared confounder in delivery-related studies using a sibling-comparison design.Specifically, in these studies, only sibling pairs that differ in delivery mode will be informative on the estimated associations.Given that caesarean delivery after a previous vaginal birth is more frequent than vaginal birth after a previous caesarean (VBAC) [18][19][20], the artificial selection of siblings with different delivery modes would lead to a systematic upwards bias in the maternal age for caesarean-born compared to vaginally-born siblings, as compared with a full-cohort design.In the meanwhile, higher maternal age might be associated with lower risks of adverse health outcomes of offspring, as older mothers generally have higher socioeconomic status and better parenting experience [21].This indicates that maternal age at delivery, as a confounding factor, may counteractively reduce the negative impacts of caesarean delivery on offspring health outcomes.Therefore, we raised the hypothesis that sibling-comparison studies, compared with full-cohort studies, would be more likely to underestimate the true association of caesarean delivery with offspring health outcomes due to inadequate adjustment for maternal age at delivery.
In this study, we first performed a systematic review and comparative meta-analysis for studies using both full-cohort and sibling-comparison designs to investigate the association between all caesarean delivery, including both elective and emergency caesarean delivery, and offspring health outcomes, with a particular focus on the impacts of different handling methods of adjustment for maternal age at delivery in multivariate regression models.We then conducted a simulation study to explore whether the results of the meta-analysis could be replicated mathematically.

Methods
This systematic review and meta-analysis was conducted and reported according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines [22].

Search strategy and eligibility
We initially searched PubMed, Embase, and the Web of Science on November 4, 2020, and updated the search on April 6, 2022.We combined terms related to "caesarean delivery", "cohort study", and "siblings comparison design" without restrictions on language and health outcomes.Full details of the search strategy are provided in Additional file 1.We also checked the reference lists of relevant reviews for additional studies.After importing studies searched from databases into Endnote and excluding duplicate records, two authors (HY and XW or ZG) browsed titles and abstracts to initially determine potential eligible studies and then scanned full text to assess for final inclusion.Studies were included if they met all criteria: (1) they were historical or prospective cohort studies that simultaneously conducted full-cohort and sibling-comparison analyses; (2) they examined the association of caesarean delivery compared with vaginal delivery with offspring health outcomes; and (3) they reported relative risk (RR), odds ratio (OR), or hazard ratio (HR) with 95% confidence interval (CI).All searches and screening were independently conducted by two authors (HY and XW or ZG), and a third author resolved disagreements by discussion and adjudication.

Data extraction and quality assessment
Two authors (HY and XW or ZG) independently extracted the following information from each study using a predetermined form: (1) first author and year of publication; (2) characteristics of the study, including study design, study location, study period, characteristics of the participants, sample size, groups of exposure, and outcome measures; and (3) effect estimates from both full-cohort and sibling-comparison analyses, including the number of participants, calculated effect size (e.g., OR, RR or HR [95% CI]), and details of adjustment for confounders.Whenever possible, we extracted the effect estimates that were most fully adjusted in the studies; if adjusted estimates were not available, unadjusted ones were extracted.If a study classified caesarean delivery into elective caesarean delivery and emergency caesarean delivery, we extracted all information on effect estimates.When needed, we contacted the original author for clarification.
Two reviewers (XW and ZG or HY) independently assessed the quality of the included studies according to the Newcastle-Ottawa Scale, which was developed to assess the risk of bias in observational studies including cohort studies [23].Study group selection (4 stars), comparability between groups (2 stars), and outcome measure (3 stars) are considered in the scale for cohort study, with the maximum being 9 stars.We defined ≥ 7 stars as high quality, 4-6 as medium quality, and ≤ 3 as low quality.Two reviewers (XW and ZG or HY) independently extracted data and assessed the quality of the included studies, and any discrepancies were resolved by discussion with a third investigator.

Data synthesis and statistical analysis
The primary analysis was to estimate the overall pooled ORs with the 95% CIs for caesarean delivery versus vaginal delivery on offspring health outcomes derived from full-cohort and sibling-comparison analyses separately.All adjusted effect sizes, including those for either elective or emergency caesarean delivery, were taken into account, implying that multiple effect sizes from the same studies may be included.Therefore, three-level meta-analytic models were used to pool the estimates to account for the dependence within studies, and the restricted maximum likelihood estimations were used to obtain the parameters [24].Moreover, a comparative analysis was carried out to evaluate the justification for using threelevel models, as opposed to ordinary two-level models.
Since adverse offspring health outcomes were rare [25,26], we regarded HR and RR as approximate ORs [27].Statistical heterogeneity was assessed using the I 2 and Q statistic, and the sources of heterogeneity were explored by conducting subgroup analyses according to the type of caesarean delivery (elective caesarean delivery or emergency caesarean delivery), type of outcomes, method of adjustment for maternal age at delivery (without adjustment, adjusting as a categorical variable, or adjusting as a continuous variable).In the subgroup analysis concerning the type of caesarean delivery, two-level randomeffects models based on the generic invariance method were used to pool the results as only one effect size in each study was included.To assess the robustness of the results, sensitivity analyses were made by serially excluding each study.Funnel plots and Begg's rank correlation test were used to assess potential publication bias [28].
In the simulation study, we created a hypothetical cohort of over a million mother-child pairs with varying maternal ages at delivery based on the results of the meta-analysis (e.g., the overall pooled ORs of caesarean delivery on offspring health outcomes) and those from the literature (e.g., the prevalence of caesarean delivery).In this simulated cohort, approximately 20% of the children were siblings, while the remaining ones were independent observations.With the assumption that increasing maternal age at delivery is associated with a higher chance of caesarean delivery as well as a lower risk of adverse health outcomes of offspring [21,29], the mode of delivery and the health outcome of each child were simulated.We performed both full-cohort and sibling-comparison analyses and compared the estimated effects at different levels of sibling similarity (i.e., correlation of maternal age at delivery among siblings) and for different methods of adjustment for maternal age at delivery (i.e., without adjustment, adjusting by 10-year age categories, adjusting by 5-year age categories, or adjusting as a continuous variable).Each scenario was simulated 100 times, after which the median and interquartile range over the 100 estimates were calculated.The simulations only focused on maternal age at delivery as the confounding factor, without considering any other potential confounders.Full details of the simulation study are provided in Additional file 2 [2,21,[29][30][31].
Statistical analyses were performed using R software (version 4.2.2), and statistical tests were two-sided with a significance level of 0.05.

Associations between caesarean delivery and offspring health outcomes
The three-level meta-analytic models revealed that caesarean delivery compared to vaginal delivery was significantly associated with increased risk of adverse offspring health outcomes.The pooling of effect estimates based on full-cohort analyses generated a summary OR of 1.14 (95% CI: 1.11 to 1.17), with 62.0% of the total variation attributed to between-study heterogeneity (level-3 I 2 = 62.0%;Q(df ) = 113.0(30);P < 0.01) (Fig. 2).Meanwhile, the pooled OR was significantly lower for estimates based on sibling-comparison analyses (P < 0.01), with a value of 1.08 (95% CI: 1.02 to 1.14) and 57.6% of the total variation attributed to between-study heterogeneity (I 2 = 57.6%;Q(df ) = 72.3(30); P < 0.01) (Fig. 2).The comparison between the three-level models and the twolevel models showed that the former provided better fits (Additional file 3: Table S2).
The discrepancies in the results between full-cohort and sibling-comparison analyses, as anticipated, appeared to vary with methods of adjustment for maternal age at delivery.Regarding the estimates that did not adjust for maternal age at delivery, the pooled OR based on fullcohort analyses was 1.10 (95% CI: 0.99 to 1.22), while that based on sibling-comparison analyses was 1.06 (95% CI: 0.85 to 1.31).In the estimates that adjusted for maternal age at delivery as a categorical variable, the pooled ORs Fig. 2 Caesarean delivery compared with vaginal delivery on offspring health outcomes of full-cohort and sibling-comparison analyses were 1.15 (95% CI: 1.11 to 1.19) and 1.07 (95% CI: 1.00 to 1.15), respectively.Notably, among the remaining estimates that adjusted for maternal age at delivery as a continuous variable, the pooled ORs based on full-cohort and siblingcomparison analyses were 1.12 (95% CI: 1.05 to 1.19) and 1.12 (95% CI: 0.98 to 1.29), respectively.

Sensitivity analyses and assessment of publication bias
In the primary leave-1-out analyses, omitting any study did not significantly change the estimated effect size (Additional file 3: Table S3).The funnel plots suggested an absence of publication bias, whether based on fullcohort or sibling-comparison analyses (Additional file 4: Figure S1), and the Begg's rank correlation test also did not indicate significant publication bias of the included studies (Additional file 3: Table S4).

Simulations
We simulated scenarios where insufficient adjustment for maternal age at delivery may lead to discrepancies between the results of full-cohort and sibling-comparison analyses.The distributions of the estimates derived from the two designs are shown in Fig. 3.
When siblings were less similar regarding maternal age at delivery (i.e., the correlation of maternal age at delivery between siblings was equal to 0.3), the difference between the estimates from the two designs increased as the adjustment became more insufficient.Specifically, when we adjusted maternal age at delivery as a continuous variable, the results from both designs were approximately equal to the true effect, while the estimates derived from full-cohort analyses were more concentrated.
When we adjusted for maternal age at delivery as a categorical variable, the estimates from full-cohort analyses were still relatively close to the actual effect, while those from sibling-comparison analyses were far from the true value.As the similarity of maternal age at delivery increased, the difference between the results of the two designs decreased.For example, when we did not adjust for maternal age at delivery, the difference in the median of the estimates from the two designs changed from 0.25 to 0.05 as the correlation of maternal age at delivery between siblings changed from 0.3 to 0.9.In addition, we also found that regardless of whether conditional logistic regression or the between-within model was used in sibling-comparison analyses, the results of the simulation study were robust (Additional file 2).

Principal findings
To our knowledge, this study is the first to synthesize and comprehensively investigate the associations of caesarean delivery with offspring health outcomes generated by full-cohort and sibling-comparison analyses.Given the high rate and potential adverse impacts of caesarean delivery, clarification of the seemingly contradictory evidence from these two types of analyses is of clinical and public health significance.As anticipated, the pooled OR of caesarean delivery with offspring health outcomes derived from sibling-comparison analyses was more conservative than that derived from full-cohort analyses.This phenomenon was more pronounced in the subgroup of studies that did not adjust for maternal age at delivery or adjusted for it as a categorical covariate.Previous research has pointed out mathematically that the estimates from sibling-comparison design may be more biased when siblings are less similar regarding nonshared confounders [17].In this study, we considered maternal age at delivery to be a main non-shared confounder for the following reasons.First, a vaginal delivery after previous caesarean is less frequent than a caesarean delivery after previous vaginal birth [18][19][20], so in sibling-comparison studies, children delivered by caesarean delivery were more likely to be born to older mothers.Therefore, the difference in maternal age at delivery between the two delivery modes in sibling-comparison studies is inherently larger than that in full-cohort studies.Meanwhile, maternal age at delivery is closely related to the health and well-being of offspring [43], since it relates to biological, social, economic, and behavioral factors that may affect a child's development [44][45][46].Older mothers generally have higher socioeconomic status and better parenting experience [47].Thus, increasing maternal age might be associated with a lower risk of adverse health outcomes of offspring, which may in turn reduce the negative impacts of caesarean delivery on offspring health outcomes [48].In addition, similar to many other continuous covariates, maternal age at delivery was often adjusted categorically in multivariate regression models.Adjustment for continuous confounders as categorical variables may inevitably result in residual confounding [49], and given the design nature, a sibling-comparison design compared to a full-cohort design would be particularly susceptible to such confounding [17].Therefore, the effect estimates generated by sibling-comparison studies may be more likely to underestimate the underlying relationship between caesarean delivery and offspring health outcomes.The simulation study further supported our hypothesis as well as findings from the meta-analysis.Simulated results demonstrated that the estimates from the fullcohort analyses were more concentrated, more accurate, and less affected by the inadequate adjustment of maternal age at delivery.In contrast, the estimates from the sibling-comparison analyses were dispersed and more susceptible to the influence of residual confounding.Notably, consistent with the findings in the meta-analysis, when we insufficiently adjusted for maternal age at delivery, the estimates of full-cohort analyses were always closer to the true value we set.Although fully adjusting confounders is far more complex than we simulated, we believe that the results of ordinary cohort studies with large sample sizes would be more accurate and robust than those of sibling-comparison studies, especially when the adjustment for non-shared confounders such as maternal age at delivery is inadequate.
Interestingly, we noticed that the effect of caesarean delivery on endocrine, nutritional and metabolic diseases, especially obesity or overweight, appeared to be overestimated, but not underestimated, in siblingcomparison analyses.A previous study found that when maternal age was greater than 30 years, it was associated with a higher risk of offspring being overweight or obese [50].This may be due to the high prevalence of obesity among older women [51,52], which may in turn negatively impact the development of the offspring's metabolic system and ultimately result in metabolic diseases in offspring [53,54].Therefore, contrary to previous scenarios, older maternal age at delivery was positively associated with the outcome at this time, so sibling-comparison analyses compared to full-cohort analyses would be more likely to overestimate the effect size when the adjustment for maternal age at delivery was inadequate.

Limitations of the study
This study has several limitations.First, multiple types of health outcomes, with potentially high heterogeneity, were included in the analyses.Although the subgroup analysis concerning different types of health outcomes was performed, the number of studies in some subgroups was limited.However, this meta-analysis did not focus on the effects of caesarean delivery on offspring health outcomes but rather on comparing the estimates of the effects from different designs.Second, the effect estimates of the included studies were inconsistent, including ORs, RRs, and HRs.We regarded both HRs and RRs as ORs to obtain a relatively conservative estimate.Third, due to the limited number of studies available, only maternal age at delivery was used as a proxy for similar inverse confounders.Future studies should investigate additional confounders to obtain a more comprehensive understanding of the associations.Fourth, the models we used in the simulation study may not perfectly reflect real-world scenarios.For instance, maternal age at delivery was considered as the only confounder, and the association of maternal age at delivery with offspring health outcomes was simply assumed to be linear.However, since the aim of the simulation study is to illustrate how inverse confounders such as maternal age at delivery may lead to the underestimation of sibling-comparison analyses, the discrepancy between the models and reality may not affect the results.Fifth, most included studies used data from Swedish or Danish national registers and might fail to be well-represented worldwide.Reassuringly, the results of these studies were proven to be consistent with those from other settings [55][56][57].

Conclusions
The results of our meta-analysis and simulation study indicated that sibling-comparison analyses may underestimate the association of caesarean delivery with multiple offspring health outcomes due to inadequate adjustment of non-shared confounders such as maternal age at delivery.In contrast, full-cohort analyses provide more reliable estimates of this association.Therefore, it is advisable to future delivery-related studies to give priority to the large-sample cohort design.If using the sibling-comparison design, it is essential to carefully consider the impact of non-shared confounders and be cautious about the interpretation of the results.
• fast, convenient online submission • thorough peer review by experienced researchers in your field • rapid publication on acceptance • support for research data, including large and complex data types • gold Open Access which fosters wider collaboration and increased citations maximum visibility for your research: over 100M website views per year

•
At BMC, research is always in progress.

Learn more biomedcentral.com/submissions
Ready to submit your research Ready to submit your research ?Choose BMC and benefit from: ? Choose BMC and benefit from:

Fig. 1
Fig. 1 Flow diagram for study identification and selection

Fig. 3
Fig.3Distributions of estimates in the simulation study.The black dashed line indicates the "true effect" of caesarean delivery on offspring health that we set according to the results of our meta-analysis."Cor" represents the correlation of maternal age at delivery between siblings

Table 1
Characteristics of the included studies

Table 1 (
continued) Abbreviations: PCS Prospective cohort study, ADHD Attention deficit hyperactivity disorder, ASD Autism spectrum disorder, ASR Acute stress response, BMI Body mass index, CPAP Continuous positive airway pressure, OCD Obsessive-compulsive disorder, PTSD Posttraumatic stress disorder, SLE Systemic lupus erythematosus