The role of residual confounding from drug exposures on the association between sugary drinks and increased risk of cancer.
I read with interest the recently published work by Chazelas et al, which reported an increased risk of cancers among a prospective cohort in France. The authors alluded to the fact that their findings might have been plagued by residual confounding. They explained the potential for residual confounding using proxies such as body mass index, which was adjusted for in their analysis. However, they did not adjust for (or acknowledge) participants’ medication exposures (current or past), which have a huge potential of introducing residual confounding in their study, resulting in biased point estimates resulting in the observed increased risk of cancers.
Complete adjustment for confounders requires detailed information on lifestyle factors, comorbidities and medication information which are often incomplete in observational studies using electronic data sources, introducing residual confounding. The authors stated that “a wide range of confounding factors were included in the analyses and many sensitivity analyses”. However, this did not include current or prior medication exposures. Description of the NutriNet-Santé cohort, suggest that medication information was collected from participants.
Meta-analyses of randomised controlled trials and observational studies have linked medication exposures to increased risk of various cancers. Exposure to insulin has been linked to all cancers, colorectal and pancreatic cancers. Furthermore, the use of thiazolidinediones (banned in 2011 in Europe as such participants could have been exposed prior to the ban) and tumor necrotic factor inhibitors have also been associated with cancer risk. While adjustment for proxies such as BMI in place of adiposity might not fully account for residual confounding. The degree of residual confounding from medication exposures is likely to be more substantial, resulting in an increased risk of cancer observed.
Simulation analysis suggests that the validity of epidemiological findings depends largely on controlling for residual confounders, which are known to greatly bias observed estimates and subsequently lead to erroneous interpretation of study findings. Therefore, it is important that the observed increased risk of cancers be interpreted in the light of residual confounding resulting from the participants current or previous medication exposures, which was unaccounted for, in their study.
 Chazelas E, Srour B, Desmetz E, Kesse-Guyot E, Julia C, Deschamps V, et al. Sugary drink consumption and risk of cancer: results from NutriNet-Santé prospective cohort. BMJ 2019;366. doi:10.1136/bmj.l2408.
 EMA-ENCePP. Guide on Methodological Standards in Pharmacoepidemiology (Revision 7). 2018. doi:10.1371/journal.pone.0028435.Andrews.
 Hercberg S, Castetbon K, Czernichow S, Malon A, Mejean C, Kesse E, et al. The nutrinet-santé study: A web-based prospective study on the relationship between nutrition and health and determinants of dietary patterns and nutritional status. BMC Public Health 2010. doi:10.1186/1471-2458-10-242.
 Ioannidis JPA, Zhou Y, Chang CQ, Schully SD, Khoury MJ, Freedman AN. Potential increased risk of cancer from commonly used medications: An umbrella review of meta-analyses. Ann Oncol 2014. doi:10.1093/annonc/mdt372.
Competing interests: No competing interests