Authors' reply: Epidemiologic Study of NDMA Contaminated Valsartan Use And Risk of Cancer. Reassuring or Inconclusive?
Dear Drs. Etminan and Mansournia.
Thank you for taking interest in our study and for your considerations regarding exposure definition, follow-up time, and statistical precision.
We analysed whether ever-use of NDMA contaminated valsartan was associated with increased cancer risk compared to use of non-contaminated valsartan. This does indeed compare to an intention-to-treat like approach. Exposed follow-up time began after the first filled prescription for contaminated valsartan (applying a 1-year lag-time). The follow-up time as NDMA ever-exposed was on average 3.5 years (11 920/3450) and the median follow-up time was 3.8 years. Importantly, the ever-user analyses do not introduce exposure misclassification. By definition, a patient is considered an ever-user after the first NDMA contaminated prescription fill regardless of whether the patient uses non-contaminated valsartan later.
We fully agree that relying solely on ever-use is insufficient in this as well as other drug-cancer association studies.1 Hence, we carried out dose-response analyses for cumulated dose of contaminated valsartan as well. For the dose-response analyses, the follow-up time for each stratum of cumulated dose cannot be calculated by dividing person years for each stratum with the total number of ever-exposed, because not all the 3450 ever-exposed patients reached the highest dose category. Although this was not supplied in the article, the median follow-up time of individuals was 3.1 years after having entered the highest dose category (excluding the 1-year lag-time we applied), among those that entered this category. Importantly, however, we fully agree that our results only pertain to short-term risk given the limited follow-up time and latency of most cancers. Accordingly, we emphasized that long-term risk is unknown and should be addressed in future studies in the manuscript. Similarly, we have already, as suggested, highlighted the upper bounds of the confidence intervals in the discussion section of our manuscript.
Lastly, Drs. Etminan and Mansournia suggest that death should be considered as a competing event. There were 425 deaths in our sample, corresponding to a rate of 0.036 per 1000 person years. We have performed a competing risk analysis based on Fine and Gray’s proportional subhazards model for our main estimate, using all-cause death as the competing event. This yielded subhazard ratios that were virtually identical to the hazard ratios reported in our paper: The estimate for ever exposure to NDMA was changed from 1.09 (0.85 to 1.41) to 1.12 (0.88 to 1.43), while the estimates for the three strata in the dose response analysis was changed from 1.15, 0.99, and 1.11 to 1.14, 1.00, and 1.17, respectively. As such, we conclude that there is no reason to suspect bias from informative censoring by death.
1. Pottegård A, Friis S, Stürmer T, Hallas J, Bahmanyar S. Considerations for Pharmacoepidemiological Studies of Drug-Cancer Associations. Basic Clin Pharmacol Toxicol 2018;122:451-9. doi:10.1111/bcpt.12946. pmid:29265740
Competing interests: No competing interests