The association between pre-pregnancy body mass index and perinatal death and the role of gestational age at delivery

Introduction The pathophysiology behind the association between obesity and perinatal death is not fully understood but may be in part due to higher rates of pregnancy complications at earlier gestation amongst obese women. We aimed to quantify the proportion of perinatal deaths amongst obese and overweight women mediated by gestational age at stillbirth or live birth. Methods The study included all singleton births at ≥20 weeks’ gestation in British Columbia, 2004–2017, and excluded pregnancy terminations. The proportion of the association between BMI and perinatal death mediated by gestational age at delivery (in weeks) was estimated using natural effect models, with adjustment for potential confounders. Sensitivity analyses for unmeasured confounding and women missing BMI were conducted. Results Of 392,820 included women, 20.6% were overweight and 12.8% obese. Women with higher BMI had a lower gestational age at delivery. Perinatal mortality was 0.5% (1834 pregnancies); and was elevated in overweight (adjusted odds ratio [AOR] = 1.22, 95% confidence interval [CI] 1.08–1.37) and obese women (AOR = 1.55, 95% CI 1.36–1.77). Mediation analysis showed that 63.1% of the association between obesity and perinatal death was mediated by gestational age at delivery (natural indirect effect AOR = 1.32, 95% CI 1.23–1.42, natural direct effect AOR = 1.18, 95% CI 1.05–1.32). Similar, but smaller effects were seen when comparing overweight women vs. women with a normal BMI. Estimated effects were not affected by adjustment for additional risk factors for perinatal death or sensitivity analyses for missing data. Conclusion Obese pregnancies have a higher risk of perinatal death in part mediated by a lower gestational age at delivery.


No confounder of M and Y is effected by A
In our text, A=pre-pregnancy BMI, M = gestational age at delivery, C is the set of confounders listed in the main text, and Y = perinatal death. Note also that our outcome is very rare (< 1%) and therefore the above odds ratios approximate the risk ratios.
Mediation analysis in presence of exposure induced mediator outcome confounding As discussed in the main text, it is likely that in our example (as in many other applied situations) there is likely to be some confounder, L of gestational age and perinatal death that is also caused by pre-pregnancy BMI (e.g. pre-pregnancy diabetes) and therefore assumption 4 above may be violated.
In this case, Vanderweele & Vansteelandt (3)(4)(5) have shown that we can still estimate 'interventional' versions of the direct and indirect effects listed above.
In our case, thee above natural direct and indirect effects correspond for each individual to setting the gestational age at delivery of those with high BMI to values of what it would be if they were normal BMI. On the other hand, the interventional versions of these effects correspond to the direct and indirect effects if those with high BMI had there gestational age randomly drawn from the distribution of gestational ages of those with normal BMI. That is, we now have effects based on population rather than individual counterfactuals. In our case, corresponding to shifting the gestational age at delivery distribution of obese women to match the gestational age delivery distribution of those of normal BMI.
Formally, assuming the above notation and letting M ∼ f a|c , (i.e. the distribution of M amongst those with A = a, conditional on C = c) and G a|c be a random draw from f , then (3)(4) Interventional direct effect : OR IDE a,G a * |c |C = odds{Y a,G a * |c = 1|C} odds{Y a * ,G a * |c = 1|C} and Interventional indirect effect : OR IIE a,G a * |c |C = odds{Y a,G a|c = 1|C} odds{Y a,G a * |c = 1|C} These effects can be identified and unbiased assuming that L is adjusted for. Our second set of adjusted odds ratios in the main text correspond to these effects where L = {pre-pregnancy diabetes, congenital anomolies}. Our analyses were broadly conistent after adjustment for L, but as mentioned, there are additional potential variables in L that could effect our results.

Estimation of effects with multiple versions of the treatment
When studying exposures like obesity, there is much debate on how results should be interrupted given that reducing obesity may be done many ways (i.e. that the treatment is not well defined) and therefore the idea of a 'causal effect' of obesity (or similar exposures) cannot be estimated (6)(7)(8)(9)(10)(11). In cases such as this, we can still estimate the effect of no one being obese by changing the determinants of obesity to reflect the determinants of those non-obese in the population (see 3.2 in 11). Of course this does not correspond to a specific intervention on obesity, but the estimate of this 'effect' is function of the population distribution of obesity (ref). By using a potential outcomes framework in this study we have tried to be clear about the assumptions required for our analyses, and the corresponding limitations. Exposures like obesity will continue to be studied in clinical and epidemiological studies regardless of these issues, and we hope by being explicit about the limitations the results can be interpreted appropriately. Further, in this example, our focus was on explaining pathways in the general population of obese women and generating potential explanations.
Although no 'fix-all' intervention exists for extending pregnancies, a hypothetical intervention here is plausible. By beginning the messy process of untangling such pathways we can begin to further refine questions, and eventually tailor appropriate obstetric responses.