We determined, using county-level data from the NCI, that the annual age-adjusted incidence rates of breast and prostate cancer in the U.S. between 2000/2001 and 2004/2005 were correlated at the county level (Table 1 and Figure 4). In general, counties with a high incidence of breast cancer also had a high incidence of prostate cancer, and vice versa. The correlation coefficient between these two cancers was greater than the correlation coefficient between these cancers and other cancers that are not hormonally regulated (Table 1), suggesting that risk factors for both breast and prostate cancers either cluster together spatially or the two cancers share common risk factors.
The correlation between these cancers increased from 0.332 to 0.552 when we used a geographically-weighted regression model, which accounted for data within a 200 km radius. This sudden increase in the correlation coefficient suggests similar risk factors for these cancers at a geographical area greater than the county level. These results also suggest our county level correlation is unlikely to be due to the county's cancer detection and reporting system.
The parameter estimates calculated for each county in our geographically-weighted regression model indicated over 76% of the counties had a significant positive association between breast and prostate cancer. This relationship varied across the U.S. and was strongest in the eastern area of the Midwest and adjacent areas of the Southeast and Southern U.S (Figure 4). The areas, where the standardized parameter estimates were the highest, were often where the hot and cold clusters for breast and prostate cancers overlapped (Figures 2, 3, and 4). There were only a few areas where the parameter estimates suggested a negative correlation between breast and prostate cancer, and these values were not statistically significant (i.e. blue areas in Figure 4). For the most part, our data suggested the rates for both of these cancers were positively correlated. This study identifies counties, as well as larger geographic areas within the U.S. where this correlation is strongest and weakest, which is useful for further research into potential factors driving the incidence of these cancers.
Both cancers also had a distinct north-south distribution (Figures 2 and 3), with the exception of the area known as "cancer alley" in the states of Louisiana and Mississippi . In general, areas with higher than expected incidence of cancer (hot clusters) were located in the northern states and areas with lower than expected incidence of cancer (cold clusters) were in the southern states. This trend has also been reported by Schwartz and Hanchette  for prostate cancer mortality rates in the U.S. A U.S.-wide spatial analysis has not been reported for breast cancer; however, there have been reports of higher occurrence of breast cancer mortality in the northeastern U.S. than the southeastern part of the country [20, 21].
There are several possible explanations for the north-south pattern of breast and prostate cancers. One explanation proposed by several researchers is the low exposure to ultraviolet radiation (UV) in the northern states, especially during the winter months [19, 22], which is believed to result in lower vitamin D levels . There are several independent researchers who have experimentally documented the beneficial effects of vitamin D on differentiation and proliferation for cell types with vitamin D receptors such as prostate and breast cells [23, 24]. There are also several epidemiological studies that have examined UV exposure as a modifying factor for breast and prostate cancers and found a protective effect [19, 24].
Another risk factor that may contribute to the clustering of cancer in the North may be low temperature, which almost always confounds UV exposure. That is, areas with a high UV index generally have high temperature and those with a low UV index have lower temperature. Temperature has a significant effect on ecological processes. Experiments have demonstrated that the biodegradation of certain organic compounds, including endocrine disruptors and chelation of heavy metals, is temperature-dependant and slower at colder temperatures [25, 26]. It is also documented that semi-volatile organic chemicals (i.e. PCBs) precipitate out of the atmosphere more efficiently at cold temperatures and during snow events [27–32]. There may, therefore, be an interaction between precipitation, temperature, and atmospheric pollution, and exposure to endocrine disruptors, which have been associated with an increase in risk of both breast and prostate cancer [6, 33, 34], may be greater at higher altitudes and latitudes. This phenomenon would occur on a global scale and may explain the higher incidence of cancers at higher latitudes that have been reported in numerous countries .
There are also other differences in the distribution of risk and protective factors across the U.S. that may partially explain the north-south distribution of cancer observed in this study. For example, cultural differences that increase or decrease the risk of cancer (i.e. behavior and diets) may be unevenly distributed between the northern and southern U.S. It is also possible that the rate of other diseases, such as cardiovascular disease, is higher in the southern U.S.  thereby resulting in premature mortality and lower incidence of cancer in these areas. Because this study was an ecological study and data were obtained at the county level, we could not adjust for differences in individual risk factors. However, we were able to adjust for age and race by using Caucasians only and age-adjusted rates in our analyses. So it is unlikely that age and race played a significant role in the distribution pattern observed.
Ethnicity may have contributed to the distribution pattern as we could not obtain data on Caucasians that were not of Hispanic origin. Because individuals of Hispanic origin have lower risks of breast and prostate cancers , and the distribution of individuals that are of Hispanic origin is not even throughout the continental U.S., this factor may have contributed to the north-south distribution pattern. However, it is unlikely that this factor was the only reason for the north-south distribution because other researchers have noted a similar pattern in other countries .
Socioeconomic status is a known risk factor for many cancers and their outcome. To minimize the effect of this variable on our outcome of interest we used incidence data instead of mortality data. Although socioeconomic status is associated with the detection of cancer, it is most likely less dependent on the availability of adequate health care than mortality rates, which is strongly influenced by the treatment received by the patient. We also corrected for this variable in our GWR model by including the county's average annual unemployment rate between 2001 and 2004. Despite this, it is still possible that this parameter biased the findings of the correlation and influenced the cluster analyses. However, the fact that other types of cancers were not as strongly correlated with breast and prostate cancers at the county level (Table 1) suggests breast and prostate cancers are correlated (i.e. counties with high incidence rates of breast cancer also tend to have a high incidence rate of prostate cancer and vice versa) regardless of the effect of socioeconomic status.
One other possible bias in this study was the disparity in the size of the counties within the continental U.S. In general, the counties in the east were much smaller than those in the west, which may have affected our cluster analyses. Because the predominant pattern observed was north to south, and this pattern was consistent using different distances to measure clusters (data not shown), we felt that the east to west variation in the size of individual counties most likely did not affect our overall conclusions. Further, the north-south disease pattern observed in this study is consistent with other research that has found a relationship between latitude and breast and prostate cancers in other areas of the world .
There were a few inconsistencies in the spatial distribution and correlation between breast and prostate cancers. For example, there was a small cluster of counties in the south known as "cancer alley" that had a high incidence of prostate cancer, but did not have higher than expected breast cancer rates. Similarly, there were a few clusters of counties with a high incidence of breast cancer in the southeast that did not coincide with elevated prostate cancer (Figure 2 and 3). The variation in the parameter estimates from our GWR analysis also suggests the relationship between these cancers varies and, therefore they may not be completely homologous. If we had refined our case definition and only included specific types of breast or prostate cancers that are more likely to be analogous (i.e. similar cell types and responsive to specific types of steroid hormones), the distribution may have overlapped better. Further, the aggregation of data at the county level renders it impossible to analyze information at a smaller spatial scale. The reason we used county level data is because it was age-adjusted, averaged over several years, and readily available for the entire U.S.
There are multiple factors that may act synergistically on prostate and breast cell types, while others may act antagonistically on these tissues [4, 6], which may account for some of the inconsistencies in the distribution of the two types of cancers. Risk factors for these cancers may also not be equally distributed within the male and female populations in a county. Despite the differences in the distribution of these cancers, the distinct north-south spatial pattern and the positive correlation between the cancers warrants further investigation to identify the factors driving these patterns. A model that includes variables such as socioeconomic status, incidence of other diseases, temperature, precipitation, pollution and UV indices, and controls for ethnicity would provide insight into the epidemiology of breast and prostate cancers. The findings of this study add to the growing evidence in the literature that prostate and breast cancers have similar risk factors and patho-physiological mechanisms.