TO THE EDITOR: We perused with interest the article titled, “Risk of Breast Cancer in Patients with Thyroid Cancer Receiving or Not Receiving 131I Treatment: A Nationwide Population-Based Cohort Study” (1). The article reported a nationwide population-based cohort study showing an increased risk of breast cancer among thyroid cancer patients who have received 131I therapy, with an adjusted hazard ratio of 1.34 (95% confidence interval [CI], 1.06–1.69). The association was stronger after a follow-up of more than 5 y, with an adjusted hazard ratio of 1.81 (95% CI, 1.27–2.57). For patients with thyroid cancer but without 131I therapy, the increment in risk was not statistically significant, with adjusted hazard ratios of 1.26 (95% CI, 0.90–1.76) and 1.28 (95% CI, 0.72–2.26) for the overall group and the subgroup having more than 5 y of follow-up, respectively. The result was compatible with the existing knowledge of radiation-induced carcinogenesis, a process that takes several years to decades (2).
However, this study also showed that cumulative 131I dose (>4.44 GBq vs. ≤4.44 GBq) is not associated with an increased risk. The authors hypothesized that the main reason for carcinogenesis is not radiation exposure but an increase in sodium iodide symporter expression before 131I treatment. Nevertheless, it seems that “immortal time bias” (3–6) could explain the lack of correlation between radiation dose and breast cancer risk. Because the occurrence of breast cancer is the primary endpoint and also the end of the observation period, as stated by the authors, the effect of further 131I treatment for residual or recurrent thyroid cancer after breast cancer diagnosis would be ignored. Consequently, the patients who “survive” longer have a greater chance of receiving more courses of 131I treatment and, thus, a higher cumulative dose. This will bias the estimated hazard ratio of treatment effect toward zero, that is, a false protective effect. Indeed, if the bias is corrected, a high cumulative 131I dose might actually be associated with an increased breast cancer incidence.
Another problem is that the differences in baseline characteristics and survival between patients with and without 131I may also be correlated to risk of breast cancer. In addition to multivariable regression analysis, as already done in this study, propensity score analysis and competing-risk analysis may further clarify this issue.
Immortal time bias is common in observational cohort studies. A review of the literature by van Walraven et al. demonstrated that, in leading medical journals, more than 40% of clinical studies using survival analyses with a time-dependent factor were susceptible to immortal time bias (7). Fortunately, the bias could be removed by time-dependent analysis, such as Cox regression with cumulative 131I dose as a time-dependent covariate (3,5). We hope that a reanalysis of the original data can be performed to clarify this issue.
Footnotes
Published online May 5, 2016.
- © 2016 by the Society of Nuclear Medicine and Molecular Imaging, Inc.