Multistage Carcinogenesis and the Incidence of Thyroid Cancer in the US by Sex, Race, Stage and Histology

Rafael Meza; Joanne T. Chang


BMC Public Health. 2015;15(789) 

In This Article


In this paper we investigated thyroid cancer incidence trends by sex, race, stage and histology in the US using multistage carcinogenesis and age-period-cohort models. Our analyses suggested that period rather than cohort is more significant in determining thyroid cancer rates, with the possible exception of distant cancers, and that the increase of thyroid cancer by calendar-year is consistent for all sex, race, and histology combinations analyzed. These results together with the observation from joinpoint analyses (appendix) showing that the largest recent increases have occurred primarily for localized cancers suggest that the rising incidence may be predominantly due to more intensive surveillance and improved diagnostics. Nonetheless, the consistent calendar-year increases by race, which have occurred in presence of significant disparities in health care access by socioeconomic status and race,[4,23,24] suggest that an environmental factor may be also at play.

Previous analyses have compared thyroid cancer incidence trends by sex, stage, and race using joinpoint and APC models.[2,3,16,25,40] Here we complement these with multistage modeling analyses, which allowed us to generate hypotheses about the biological mechanisms of thyroid cancer tumor initiation, promotion and malignant conversion. For instance, the multistage analyses show that the 3-fold higher thyroid cancer incidence in women versus men can be explained by 1.5-fold higher rates of initiation and promotion (premalignant tumor growth). These imply that women get 50 % more tumors and that those progress 50 % more rapidly to cancer than men's. The lower thyroid cancer incidence in blacks can be attributed in part to lower tumor promotion rates versus whites. Sub-analyses of thyroid cancer incidence by stage reveal gradients of tumor initiation (localized > regional > distant) and promotion (localized < regional < distant) that suggest heterogeneity in aggressiveness from tumor onset with clear implications for the overdiagnosis of slow growing tumors under active screening and surveillance.

The estimated TSCE hazards (age-specific incidence) display a pattern where the asymptote is reached in middle age (roughly flat after age 40). This appears to be in contrast with previous analyses of thyroid cancer based on age-period-cohort models that have found age-specific patterns by period that decrease after age 40, or by cohort that increase until age 70.[40] Our results are indeed consistent with these previous findings, once the TSCE hazards are multiplied by the corresponding estimated period and cohort effects as shown in Fig. 4 (and Additional file 1: Figures S7, S8, S9 and 10s). This suggests that a constant age-specific thyroid cancer risk from age 40–50 is consistent with the SEER data, and that the observed decreases in age-specific risk in different years (period) or increases in age-specific risk until late in life by cohort, might just be representing un-adjusted secular trends and not true age patterns in risk.

Analyses of thyroid cancer incidence by histological type (papillary vs follicular) show that interestingly estimated promotion rates do not seem to vary by histology, but that initiation of papillary cancers is about 6 and 1.5 times higher than that of follicular cancers in females and males, respectively. The higher initiation rates could be attributed to higher underlying mutation rates or to a higher number of susceptible cells. Interestingly, the estimated period effects do not vary much by histology, suggesting a common risk factor or an underlying cause behind the thyroid cancer increase with calendar year (potentially increased screening). However, the estimated birth-cohort effects do vary by histology, and show a decreasing trend for follicular cancers since the early 1900s possibly reflecting the reductions in risk due to iodization of salt in the US.[41]

Our study has several limitations. First, in common with other analyses of cancer registry data, we were unable to assess exposures to relevant risk factors, such as dietary patterns and environmental exposures, in the underlying population and cancer cases. Nonetheless, the SEER registry allowed us to analyze trends in thyroid cancer incidence since the 1970s by sex, race, stage and histology. Second, our multistage models are clearly a simplification of the underlying biology of thyroid cancer incidence and neglect the contribution of relevant risk factors. As mentioned above, this is largely due to the lack of risk factor data from the SEER cancer registry. However, the age-period-cohort approach allowed us to control for secular trends in the estimation of the biological parameters. Moreover, the estimated TSCE models provide better fits to the data than more complex models that allow for additional carcinogenesis stages (data not shown), and the differences in biological parameters provide plausible explanations to the observed differences in age-specific thyroid cancer risk. Third, age-period-cohort models have an inherent non-identifiability problem that makes it impossible to estimate uniquely the period and cohort effects.[42,43] However, replacing the age effects with the hazard of a multistage model resolves, at least in theory, the non-identifiability problem, allowing us to estimate uniquely the secular trends.[36] Finally, the smaller sample sizes for some demographic groups, like black males, or for specific histologies, like follicular cancer in men, preclude us from doing more detailed analyses.