Response to: Reassessing the causal role of early life adiposity in breast cancer

We thank Schooling et al. for their interest in our recent article¹ which aims to identify the independent roles of adult adiposity and childhood adiposity in female breast cancer (BC). In their comments, Schooling et al. pointed out that the inverse association of childhood adiposity with BC could be explained by selection bias arising from only selecting survivors of childhood adiposity or BC. Actually, such survival bias is likely to be present in all epidemiological analyses, particularly considering a disease outcome in an elderly population.² In a simulation study,³ Burgess et al. assessed the impact of selection bias in the context of Mendelian randomization (MR), and demonstrated that selection bias can significantly influence causal effect estimates only when selection into the study is strongly influenced by risk factors. The inverse probability weighting approach has been developed to adjust for this, but it cannot easily be implemented as it requires estimates of the probability of selection.

Although we acknowledge that our study is inevitably influenced by survival bias to some extent, we think such bias is not likely to have a big impact. The reasons are as follows. First, participants’ selection into the Breast Cancer Association Consortium (BCAC) study is not likely influenced by risk factors [here referring to body mass index (BMI)], as most BCAC studies recruited all cases without selection in the sample hospital during a certain time.⁴ Second, our analyses are based on BC cases from the population-based or hospital-based studies which included women aged 18–79 years,⁴ indicating that not only older and middle-aged women but also young women were enrolled. In addition, an undeniable fact was that a very small number of women diagnosed with BC are younger than 35⁵ as the disease mainly occurs in middle-aged and older women, with the highest incidence rates observed in the 65 to 69 age group⁵ which is close to the mean age of BCAC participants (57 years old). Last but not least, the mortality attributable to BC among young women was far less than that of middle-aged and older women in the UK in 2000 (around the recruitment year of BCAC), at only 1.4, 5.9 and 12.1 per 100 000 persons for women aged 25–29, 30–34 and 35–39 years, respectively.⁶ Therefore, the selection bias due to such a small proportion of missing deaths is likely to be very weak.

Moreover, in their letter, Schooling et al. used participant reports of BC in mother and siblings to surrogate the participant’s liability to lifetime BC, and found that childhood BMI was unrelated to BC.⁷ They further substantiated the existence of survival bias by identifying the smallest magnitude of association for liability to BC from the BCAC with mothers’ lifespan. This preliminary investigation of survival bias is quite inspiring but the challenge remains, as it was not easy to confidently assess the impact of survival bias on the obesity-BC association in MR analyses with limited data. Consequently, to confirm this impact, rigorous study designs with additional data are recommended for future studies and compelling evidence from simulation studies are also warranted.

Author contributions

Y.H., J.X.and X.J. drafted the letter; S.B., H.Z. and P. K. provided expert advice; Y.L., X. Z, X.W., C.X., J.L., N.W. and L.Z. edited the letter.

Funding

This work was supported by the grants from the National Natural Science Foundation of China (No. 81874282).

Conflict of interest

None declared.