A study published in the Journal of the American Medical Association has investigated between vaginal birth, planned and unscheduled cesarean delivery, and subsequent health problems or early death of the child, reporting differences in risk of a range of outcomes between the three groups. Roundup comments accompanied this analysis.
Title, Date of Publication & Journal
‘Planned caesarean delivery at term and adverse outcomes in childhood health’
Tuesday 1 December 2015
Study’s main claims – and are they supported by the data
The paper’s stated primary aim is to investigate the impact of caesarean birth, be that planned or not, versus vaginal birth on the offspring developing asthma – assessed as the offspring requiring a hospital admission due to asthma at any time during the follow-up period. There are multiple secondary outcomes investigated in the paper, on varying subsets of the original dataset.
The authors state “In this national cohort of 321 287 offspring, planned caesarean delivery in a first pregnancy was associated with a small increase in risk of offspring asthma and death in childhood when compared with vaginal birth but not when compared with un-scheduled caesarean delivery.” The authors give a detailed discussion section listing many of the standard reasons to be cautious about over interpreting this result. In particular I congratulate the authors on their frank statement that some of the results are likely statistical artefacts: “The increased risk of type 1 diabetes following planned vs unscheduled caesarean delivery was unexpected…. In the absence of a clinical explanation, the borderline statistical significance of the findings support that these may be attributable to type I error.” (Type 1 error: In any statistical test there is a chance we reach the wrong conclusion, that is one of the unavoidable costs of making population-level statements based on a sample; a statistical type I error is to declare a significant result when in reality we should not, merely as a result of the random sample we happen to have obtained).
On the primary outcome, there may be some evidence from this observational study of a difference in risk of an asthma related hospital admission (which is not the same as saying the offspring develops asthma). However the findings are far from conclusive evidence on worse outcomes (across primary and secondary secondary, including all-cause death up to the age of 21) for offspring born via caesarean.
Any reports on this article should take care to be clear on the difference between a ‘planned caesarean’, ‘unplanned caesarean’, ‘scheduled caesarean’, ‘unscheduled caesarean’, and a ‘caesarean’. In many situations, a caesarean is not an elective procedure, but rather a life-saving procedure.
Although the authors highlight this issue, it should be stressed that the classification of scheduled or unscheduled caesarean is not without problem, not least there are numerous medical reasons for a caesarean that are not conveyed in such a crude classification. This potential mixture of reasons for caesarean is perhaps indicated by the paper’s many findings of no difference between the scheduled and unscheduled groups.
Given the small proportions, relative to the vaginal birth group, and the finding that even after adjustment the caesarean groups were similar; it seems odd not to have pooled them into a caesarean versus vaginal comparison. Instead, the authors only consider unscheduled versus scheduled, and scheduled versus vaginal. There is a third comparison of unscheduled versus vaginal that is nowhere in the paper, something that is curious. (Aside: the paper shows some results where there is no statistical evidence for a difference between an outcome for planned and unscheduled, but there is evidence of a difference between planned and vaginal; from this we cannot conclude there is a difference between unscheduled and vaginal – this is the missing third comparison).
There is little acknowledgement of the issues of mis-classification in hospital records, that there can be multiple reasons for hospital admission. Having an asthma related hospital admission, the primary outcome analysed across the whole cohort and all follow-up, is not equivalent to the offspring having asthma (a chronic condition not simply captured by a single hospital admission).
As the authors themselves state: “Further investigation is needed to understand whether the observed associations are causal”. This study is an observational historical cohort, and we should not draw any strong conclusions.
The number and variety of secondary outcomes is slightly excessive. Although the authors are clear in the manner these analyses were pre-defined, in keeping with best analysis practice, the different subsets under analysis (in the two different comparisons and across different variables) slightly confuses the sample size being used for each specific question (across most of the tables the sample size is changing on a row-by-row basis).
The cohort period, 1993-2007 with follow-up until 2015, covers a significant period with changing public health policies. Specifically the introduction of the smoking ban in 2006 in Scotland. Although the authors state they assessed the proportional hazards assumption underling Cox proportional hazards, there may be some question as to whether the changes, specifically in second hand smoke, may have a confounding affect. If, as the authors state, caesareans have become more common in recent times, then caesarean births are more likely in the more recent period conflated with the changes in smoking policy. Whether this will lessen or exacerbate the difference is unclear. The authors are, perhaps indirectly, accounting for this by including the mother’s smoking status.
On a statistical note, the authors use two methods for group comparisons of continuous variables: standard t-tests and Mann-Whitney tests. The latter is appropriate with non-normality or highly skewed data. It’s not clear why all the comparisons are not done using the Mann-Whitney test (as the less restrictive test).
In Table 1 we might be seeing a statistical fallacy – that with a big enough sample size, everything can seem significant. In Table 1, the maternal age of planned versus unscheduled caesarean is 29 and 29, with very similar inter-quartile ranges (25-33 and 24-32); and yet the p-value comes out as highly significant that these are different. The issue of a statistically significant difference versus a clinically significant difference may be coming into play. Again, considering the year of delivery, it’s hard to conclude these are meaningfully different despite being statistically different.
The paper considers a very large historical cohort. The authors should be congratulated on clearly defining their primary and secondary analyses upfront, to reassure the reader that there is minimal data-dredging (testing every possible outcome until one shows a significant difference).
This is a significant undertaking, and very valuable to the wider literature. Being able to link multiple databases, to follow offspring and assess their health outcomes over such a long follow-up period.
There a many unknowns on the long term effects of a caesarean, with multiple possible biological and clinical explanations for differing outcomes. This study has attempted to account for multiple factors (by adjusting the analysis for potential confounders). It adds to the overall evidence-base, and highlights the difficulty in addressing these types of public health issues.
Any specific expertise relevant to studied paper (beyond statistical)?
Before The Headlines is a service provided to the SMC by volunteer statisticians: members of the Royal Statistical Society (RSS), Statisticians in the Pharmaceutical Industry (PSI) and experienced statisticians in academia and research. A list of contributors, including affiliations, is available here.