Incidence of Differentiated Thyroid Cancer by Socioeconomic Status and Urban Residence: Canada 1991–2006

Introduction

Thyroid cancer is the seventh most common cancer affecting women, and the most common endocrine cancer in humans (1). The vast majority of thyroid cancer is classified as differentiated thyroid cancer (DTC), which is largely asymptomatic and has an extremely high cure rate when treated. The incidence of thyroid cancer in the United States in 2002 was estimated to be about 24,000 cases/year, with DTC accounting for about 97% of these cases (2). Risk factors for thyroid cancer include exposure to ionizing radiation, family history of thyroid cancer, genetic factors, history of thyroid disease, and iodine deficiency or excess (3). As these cancers are largely asymptomatic on presentation, they are often detected incidentally during imaging studies of the neck performed for other reasons.

The incidence of DTC is rising in developed countries and has been for several decades. In the United States, analysis of the National Cancer Institute's Surveillance, Epidemiology and End Results (SEER) program demonstrated an increase in the incidence of thyroid cancer from 3.6/100,000 in 1973 to 8.7/100,000 in 2002, a 2.4-fold increase (2). In Denmark, incidence increased from 1.43/100,000 in 1996 to 2.16/100,000 in 2008 (4). Similarly, in the Netherlands, the estimated annual percent change for papillary thyroid cancer (the most common variant of DTC) was +3.5% from 1989 to 2009 (5). Data from Ontario, the largest Canadian province, showed an increase in new cases of DTC from 403 in 1990 to 990 in 2001 (a 146% increase) (1).

To our knowledge, there has not been a Canadian study using nationwide Cancer Registry data to examine the relationship between thyroid cancer incidence, socioeconomic status (SES), and geographic location in Canada. The purpose of this study was to answer the following three questions. First, what was the magnitude of increased incidence of thyroid cancer in Canada from 1992 to 2007? Second, is there an association between SES and thyroid cancer incidence in Canada? Third, does the relationship between SES and the incidence of thyroid cancer vary by rural/urban status?

Materials and Methods

Two different sources were used to draw our data: the Canadian Cancer Registry (CCR) data file and the Canadian Census of Population from Statistics Canada. The CCR data file contains patient demographic and tumor-specific data on each tumor included in provincial and territorial cancer registries from 1992 to 2007 inclusive. It provides information on all cases of cancer diagnosed in individuals whose usual place of residence is Canada including all rural and urban areas of each province and territory (6). Thus, observed differences in the incidence of thyroid cancer between urban and rural areas reflect differences in the extent to which cancers are being diagnosed, or differences in the actual incidence of thyroid cancer, or both. Names and personal identification numbers were removed and replaced with unique identifiers in the version of the file released for this study. All analysis was within the University of New Brunswick Research Data Centre (NB-RDC), and all output was vetted for release using enhanced vetting methods required by Statistics Canada. Ethics approval is not required for research projects using data stored in the NB-RDC. To our knowledge, this is the first time CCR data have been available and used to study the relationship between thyroid cancer incidence and SES in Canada.

The long form of the Canadian Census of population, which occurs every five years in Canada, contains demographic and socioeconomic data on 20% of the Canadian population. For this study, data from the census years (CYs) 1991, 1996, 2001, and 2006 were available. The narrowest level of disaggregation at which census information is released by Statistics Canada is the dissemination area (DA), a small, relatively stable geographic unit composed of one or more adjacent dissemination blocks, including a population of 400–700 individuals. DAs cover the whole territory of Canada. For each DA and for each of the four CYs, information was collected on age for men and women, average household income, proportion of adult residents with at least a university degree, and other sociodemographic information such as the proportion of the population born outside of Canada and the proportion of the population that belongs to an ethnic minority group. Although average income in the DA might be influenced by large income outliers, median income is not available at the DA level for all of the CYs. Each DA is assigned to one of three categories of rural/urban status based on definitions provided by Statistics Canada: city, if the DA is located in a census metropolitan area (an area with a total population of at least 100,000 of which 50,000 or more must live in an urban core); town, if the DA is located in a census agglomeration (an area with a total population of between 10,000 and 100,000); and rural, if the DA is not located in either a CMA or CA (7). The number of DAs for which census information was available was 32,825 in 1991, 38,016 in 1996, 46,909 in 2001, and 52,443 in 2006. In order to control for wage and price differences across different regions, the income quintile (InQ) for each DA was defined relative to other DAs in the associated census division (CD), which Statistics Canada defines as a group of neighboring municipalities joined together for the purposes of regional planning and managing common services. The number of CDs in Canada was in the range of 288–290 across the four CYs, and changes occurred to only a few CD boundaries each CY. DAs within each CD were sorted by average income then assigned to one of five InQs. Although the DA of an individual in the CCR database is not reported, the postal code of residence is disclosed. PCCF+ is a sophisticated statistical tool provided by Statistics Canada that maps postal codes to DAs. Statistics Canada revises geographic boundaries and updates PCCF+ at each CY. Hence, in order to match postal code and DA information as closely as possible, cancer cases were assigned to CYs as follows: cases diagnosed in 1992–1995 were associated with socioeconomic characteristics according to the 1991 census; cases diagnosed in the 1996–2000 period were associated with data of the 1996 census; cases diagnosed in the 2001–2005 period were associated with data of the 2001 census; and cases diagnosed in the 2006–2007 interval were associated with data from the 2006 census. Census data could be linked to 96.9% of DAs for the CY 1991, 97.4% for the CY 1996, 99.0% for the CY 2001, and 99.2% in the CY 2006.

For each CY, the unit of observation for the analysis of incidence was the DA, and the key variable of interest was the number of cases of thyroid cancer diagnosed in adults over the age of 18 in each DA over a relevant period of time corresponding to the CY. At this level of geography, it is not feasible to disaggregate counts further by age and sex. In the regression analysis, controls were included for the age and sex composition of the adult population of each DA. The exposure variable was the adult population in the DA at the CY multiplied by the number of years in the corresponding time interval for that census (two, four, or five years). Negative binomial regression models were estimated where neighborhood socioeconomic status, as measured by the InQ of the DA, was captured by a 0/1 binary variable for each InQ, with the highest InQ specified as the baseline. To capture the changes in incidence over time, we defined indicator variables for each CY from 1996 to 2006 with 1991 as the reference year. In our first specification, differences in rural/urban status were captured by a binary variable for each type of region as defined above. In our second specification, we interacted InQ variables with indicators for residence in cities, towns, and rural areas to allow the relationship between socioeconomic status and incidence of thyroid cancer to vary by urban/rural residence. Both regressions included detailed controls for the age/sex composition of the DA. Since healthcare in Canada is administered at the provincial level, both regressions also included indicator variables for each province or territory of residence in order to capture regional unobserved effects important to the incidence and diagnosis of thyroid cancer.

Results

We present our baseline regression results in Table 1. Taking the 1991 census year (corresponding to the years 1992–1995) as the reference year in the regression, the incidence rate of thyroid cancer per 100,000 person years increased over time by 27% in 1996 (incidence rate ratio [IRR] 1.27, p=0.00 [CI 1.21–1.32]; Table 1), by 104% in 2001 (IRR 2.04, p=0.00 [CI 1.96, 2.13]), and by 156% in 2006 (IRR 2.56, p=0.00 [CI 2.43–2.70]) in Canada.

A significantly positive relationship was found between higher InQ and thyroid cancer incidence. The lowest InQ's thyroid cancer incidence rate was 78% (IRR 0.78, p=0.00 [CI 0.75–0.81]) of the highest InQ, whereas the middle InQ had an incidence rate of 88% of the highest InQ (IRR 0.88, p=0.00 [CI 0.84–0.91]; Table 1). There was a statistically significant monotonic increase in incidence by InQ from lowest to the highest.

The thyroid cancer incidence rate was found to vary as a function of the rural/urban status of the DA, with the incidence rate in towns and rural areas at least 25% lower than in cities (IRR 0.72, p=0.00 [CI 0.69–0.74] for towns; IRR 0.75, p=0.00 [CI 0.72–0.77] for rural areas; Table 1).

Table 2 reports results from a second regression that allowed for differences in the effects of InQ on thyroid cancer to vary between cities, towns, and rural areas. Since the highest InQ is the reference category, controls for town and rural residence reflect differences in incidence compared to city residents for the highest InQ. Compared to cities, the incidence rate for the highest InQ was 34% lower in towns (IRR 0.66, p=0.00 [CI 0.61–0.71]) and 39% lower in rural areas (IRR 0.61, p=0.00 [CI 0.56–0.66]). The effect of InQ on incidence in cities was pronounced and highly significant, with a 16% lower incidence rate for the middle InQ compared to the highest (IRR 0.84, p=0.00 [CI 0.81–0.91]) and a 28% lower incidence rate for the lowest InQ compared to the highest (IRR 0.72, p=0.00 [CI 0.75–0.81]). For towns, the effect of InQ on incidence rate was significant for the lower InQs only and was of smaller magnitude. The incidence rate was 13% lower for the lowest InQ compared to the highest (IRR 0.87, p=0.01 [CI 0.78–0.97]) but was not significantly lower for the middle InQ. For residents of rural areas, there was no significant difference in the incidence rate of thyroid cancer by InQ even for the lowest InQ (IRR 1.05, p=0.38 [CI 0.94–1.17]).

Discussion

Our study shows that the incidence of thyroid cancer in Canada increased dramatically from 1991 to 2006, which confirmed similar results from other jurisdictions (2 –5,8). We identified a strong association between SES (as measured by InQ) and thyroid cancer incidence in Canada, with increased incidence rates among individuals in higher InQs. We found that thyroid cancer incidence was significantly higher in cities than in towns or rural areas. In addition, when we allowed the effect of InQ on incidence to vary by rural/urban status, the effect of SES on incidence was greatest in cities, moderate in towns, and nonexistent in rural areas.

InQ is defined relative to the CD in which the dissemination area is located. CDs vary in their composition of rural and urban areas, and for those CDs that are composed of both rural and urban areas, there may be correlation between InQ and rural/urban status. However, in our data, the correlation coefficient between InQ and a categorical variable reflecting rural, town, and city status is only 0.034. Furthermore, in our expanded model with interactions, estimates of the effect of rural/urban status on incidence are allowed to vary within IQs. That we observe significant variation in rural/urban status within quintiles is one of the key findings of our paper.

An interesting result of our study is that even within the Canadian universal healthcare system, socioeconomic disparities in the diagnosis of thyroid cancer still exist. One theory is that this increase in thyroid cancer incidence can largely be attributed to increased detection, rather than a true increase in the incidence of the disease. Access to imaging studies of the neck, including ultrasound, computed tomography, and magnetic resonance imaging, has increased in developed countries over the last several decades. A review of administrative data in Ontario, Canada, showed an increase in the overall number of imaging tests of the neck per capita from 0.17 to 0.43 between 1993 and 2006 (8). These data also showed that women received more imaging tests than men, and that the difference in imaging frequency between men and women was increasing over time. These imaging studies are able to detect tumors that are not palpable, and may be smaller than clinically evident tumors. In Ontario, Kent et al. showed that a significantly higher number of small (<2 cm) impalpable tumors were resected in 2001 compared with 1990, while the incidence of tumors 2–4 cm in size remained stable (1). A similar finding was noted in the SEER data; between 1988 and 2002, 49% of the increase in thyroid cancer incidence consisted of tumors measuring 1 cm or smaller, and 87% consisted of tumors measuring 2 cm or smaller (2).

Access to diagnostic imaging (and thus the potential for detection of subclinical thyroid cancers) may be affected by various factors. In the United States, Morris et al. demonstrated that a variety of SES metrics, including medical coverage, income, ethnicity, and education, are correlated with thyroid cancer incidence, with increased incidence in higher socioeconomic groups (9). In Canada, healthcare is universally available, and thus the socioeconomic disparity in imaging access may be different from the United States. However, other factors may influence access to diagnostic imaging, such as urban or rural residence and proximity to a major healthcare center. Alternatively, it is possible that individuals of higher SES are more likely to be aware of the existence of screening tests, and more likely to receive screening tests, than individuals of lower SES. A similar relationship between SES and screening for oral cancer was identified by our group in a previous study (7). While the theory that access to imaging is likely the most important variable in thyroid cancer incidence, other possible explanations, including diet, pollutants, or exposure to toxins could also be considered.

There are important limitations to our study. We did not possess individual-level income data, and thus we used DA-level information on household income as a substitute, which is less precise than individual-level information. Furthermore, while household income reflects income from all sources including pensions and investment income, SES could be measured more accurately with individual-level information on other determinants of SES such as education level and occupation. Income alone may be biased in favor of current SES rather than long-run SES. Another limitation is that we have no direct evidence that imaging access is increased in cities compared to rural areas or towns. This may be an important consideration for further study.

In conclusion, our study demonstrates an increased incidence in thyroid cancer in Canada from 1991 to 2006. We also found that the incidence of thyroid cancer was higher in cities than in towns or rural areas. Higher SES was positively correlated with higher rates of thyroid cancer incidence overall, but subgroup analysis showed that this relationship was only observed in cities, rather than towns or rural areas.