TO THE EDITOR: We have meticulously and enthusiastically read the paper by Umlauft et al. in the January issue of The Journal of Nuclear Medicine (1). The authors purposed to examine the risk of developing diabetes mellitus after radiopeptide therapy for neuroendocrine tumors and the effect that the development of diabetes would have on all‐cause mortality. They concluded that there does not seem to be an increased risk for development of diabetes and that development of diabetes does not seem to increase mortality. This study made a considerable contribution to this area, but some methodologic issues must be considered to avoid misinterpretations.
The authors incorporated all types of mortality into one category and then examined the relationship of diabetes mellitus with this combined-outcome category. Although the power of statistical testing is improved using the combined outcome, the homogeneity of the relationships of diabetes mellitus with cause-specific mortalities must be considered a main assumption. This assumption might have been violated in the study of Umlauft et al. since the strength of the relationship of diabetes mellitus with all-cause mortality may differ from the strengths of its relationships with cause-specific mortalities (2). More sophisticated statistical methods have been newly presented to efficiently assess the relationship of exposure with multiple outcomes such as cause-specific mortalities (3).
Second, the authors used the Cox proportional-hazards model to compare the hazard of diabetes mellitus among 3 treatment modalities. However, the proportional-hazards assumption—one of the most important assumptions in the Cox model—has been violated in their study, as shown in their Figure 3C (1). Hence, variants of this model, such as the stratified or extended Cox regression model, must be applied to avoid any misleading findings (4).
Footnotes
Published online Feb. 16, 2017.
- © 2017 by the Society of Nuclear Medicine and Molecular Imaging.