Breast cancer is the most common cancer and second leading cause of cancer mortality in the United States and Texas [
Relatively few studies have been conducted to examine how racial and ethnic status affects the geographic distribution of breast cancer mortality, and no study has been conducted to investigate this at the county level in the state of Texas. Identification of geographic patterns of breast cancer mortality based on racial and ethnic status could provide impetus to conduct further investigations and target health resources for prevention and treatment in specific geographic areas. The Spatial Scan Statistic method developed by Kulldorff [
A previous study published by Zhan and Lin in 2003 [
The case definition for this study was a death due to malignant neoplasm of breast cancer (C50) as listed by the International Classification of Diseases, 10th Revision (ICD-10) in the female population of Texas for the years 2000 to 2008. The breast cancer mortality data at the county level for the years 2000 to 2008 were obtained from the Texas Cancer Registry, Cancer Epidemiology and Surveillance Branch, Texas Department of State Health Services. The Texas Cancer Registry classifies cancer mortality data according to the Surveillance, Epidemiology, and End Results (SEER) “Cause of Death Recode”, as given by the SEER Cause of Death Recode 1969+ (9/17/2004) (
In all, 22,820 breast cancer deaths occurred in Texas between the years 2000 to 2008, of which 15,234 (66.8%) were attributed to NHW, 3,503 (15.4%) to NHB, and 3,770 (16.5%) to Hispanics. At-risk population data at the county level for Texas covering the same time period were obtained from the US Census Bureau. Geographic coordinate data (i.e., county centroids) that represent the locations of the 254 Texas county polygons were specified by the 2000 US Bureau of the Census.
We used Spatial Scan Statistics to examine the presence of breast cancer clusters. We ran Purely Spatial Analyses using 4 different models (Discrete Poisson with and without covariates, Bernoulli and Multinomial models). For the Discrete Poisson model, we assumed that the number of deaths in each county was Poisson distributed and ran the model with and without adjustment of racial/ethnic status of the population. For the Bernoulli model, we classified breast cancer deaths among NHW as controls and breast cancer deaths among other racial/ethnic groups as cases, thereby creating 3 comparisons (NHB versus NHW, Hispanics versus NHW and “other races” versus NHW). For the Multinomial model, we classified breast cancer deaths into 4 racial/ethnic categories (NHW, NHB, Hispanics, and other races). The level of statistical significance used for this study was 0.05. We used the SaTScan v9.1.1 released on March 9, 2011 to run the Spatial Scan Statistics. The Spatial Scan Statistics detect high-risk areas of cases by gradually scanning a window across space and noting the number of observed and expected observations inside the window. The window sizes are varied continuously up to a prespecified maximum size. The most likely cluster (the cluster least likely to be due to chance) is assigned to a window with the maximum likelihood. We used the scanning window in the shape of a circle and specified the maximum window size as one that included 50% of the at-risk population throughout our analyses. We also report secondary clusters that cause a rejection of the null hypothesis (i.e., the log likelihood ratio of secondary clusters in the real data is higher than that of the most likely cluster in the simulated data sets) and do not overlap the most likely cluster. MapInfo Professional version 8.0 was used to create the thematic map based on detected counties by Spatial Scan Statistics.
This study did not need to be approved by the Committee for the Protection of Human Subjects because it used aggregated data on county-level breast cancer mortality in Texas.
Results of spatial analysis using the Discrete Poisson model without any covariate adjustment suggests that there were five statistically significant clusters of counties with a high rate of female breast cancer mortality in Texas for the years 2000 to 2008 (Table
List of breast cancer mortality clusters using Discrete Poisson model at county level in Texas for the years 2000 to 2008.
Cluster | Observed cases | Expected Cases | RR | LLR | |
---|---|---|---|---|---|
(a) Without covariate adjustment | |||||
Most likely cluster | 3388 | 2754.18 | 1.27 | 78.04 | <0.001 |
Secondary cluster 1 | 496 | 319.35 | 1.57 | 42.43 | <0.001 |
Secondary cluster 3 | 525 | 382.84 | 1.38 | 24.08 | <0.001 |
Secondary cluster 4 | 1992 | 1775.27 | 1.13 | 13.84 | <0.001 |
Secondary cluster 5 | 119 | 71.95 | 1.66 | 12.88 | <0.001 |
Not significant cluster | 52 | 33.23 | 1.57 | 4.52 | 0.654 |
(b) With adjustment of race/ethnic status of population | |||||
Most likely cluster | 4550 | 3843.34 | 1.23 | 74.66 | <0.001 |
Secondary cluster 1 | 476 | 363.25 | 1.32 | 16.21 | <0.001 |
Not significant cluster | 119 | 85.39 | 1.40 | 5.91 | 0.26 |
Not significant cluster | 1209 | 1111.12 | 1.09 | 4.41 | 0.68 |
Not significant cluster | 713 | 640.85 | 1.12 | 4.04 | 0.80 |
Not significant cluster | 413 | 370.87 | 1.12 | 2.35 | 1.00 |
Abbreviations: RR, Relative risk; LLR, Log-likelihood ratio.
Breast cancer mortality clusters at the county level in Texas for the years 2000 to 2008 using Discrete Poisson model without covariate adjustment (a), Discrete Poisson model with race/ethnicity as covariate (b), Multinomial model (c), Bernoulli model—non-Hispanic blacks (NHBs) versus non-Hispanic whites (NHWs) (d), Bernoulli model—Hispanics (H) versus NHW (e) and Bernoulli model: other races (O) versus NHW (f).
Results of spatial analysis using the Multinomial model suggest that there were three statistically significant clusters of counties where the distribution of risk for female breast cancer mortality based on race/ethnicity is statistically significantly different from the remaining regions of Texas (Table
List of breast cancer mortality clusters using Multinomial model at county level in Texas for the years 2000 to 2008.
Cluster | Observed cases | Expected cases | RR | LLR | |
---|---|---|---|---|---|
Most likely cluster | 1437, 164, 1659, 32 | 2202.32, 505.41, | 0.62, 0.29, 4.77, 0.65 | 1327.21 | 0.001 |
Secondary cluster 1 | 7586, 2706, 887, 193 | 7607.78, 1745.89, | 0.99, 3.43, 0.31, 1.47 | 1093.82 | 0.001 |
Secondary cluster 2 | 4348, 437, 419, 76 | 3532.28, 810.61, | 1.32, 0.47, 0.42, 1.01 | 418.51 | 0.001 |
Abbreviations: RR: relative risk; LLR: log-likelihood ratio; NHW: non-Hispanic whites; NHB: non-Hispanic blacks; H: Hispanics; O: other races.
Table
List of breast cancer mortality clusters for non-Hispanic blacks, Hispanics, and other races using Bernoulli model compared to non-Hispanic whites at county level in Texas for the years 2000 to 2008.
Cluster | Observed cases | Expected Cases | RR | LLR | |
---|---|---|---|---|---|
(a) Non-Hispanic blacks versus non-Hispanic whites | |||||
Most likely cluster | 2531 | 1744.27 | 2.63 | 447.78 | <0.001 |
(b) Hispanics versus non-Hispanic whites | |||||
Most likely cluster | 1659 | 607.22 | 4.13 | 1122.37 | <0.001 |
Secondary cluster | 510 | 148.27 | 3.83 | 430.95 | <0.001 |
(c) Other races versus non-Hispanic whites | |||||
Most likely cluster | 115 | 53.66 | 2.77 | 34.75 | <0.001 |
Secondary cluster | 101 | 68.04 | 1.70 | 9.36 | 0.012 |
Not significant cluster | 10 | 3.90 | 2.61 | 3.47 | 0.89 |
Abbreviations: RR: relative risk; LLR: log-likelihood ratio.
We found several statistically significant clusters for female breast cancer mortality at the county level in Texas through Purely Spatial Analyses. Five significant clusters were found through the Discrete Poisson model without any covariate adjustment, while the same model after adjusting for racial/ethnic status detected two significant clusters with different geographic distributions. The Multinomial model detected three significant clusters with different distributions of risk based on racial/ethnic status. The Bernoulli model found one significant cluster for NHB versus NHW, while two significant clusters were detected for Hispanics versus NHW and another two for “other races” versus NHW.
Zhan and Lin (2003) conducted spatial cluster analysis using the Discrete Poisson model without any covariate adjustment on female breast cancer mortality data for the years 1990 to 1997 and found two significant clusters [
We found three significant clusters using Multinomial analysis where risk for breast cancer mortality differed based on racial/ethnic status. The cluster in the northeast area of Texas had significantly highest risk for NHB and lowest risk of Hispanics, while the cluster in the southern area of Texas had significantly highest risk for Hispanics and lowest risk for NHB. It is important to note that risk is not substantially different among racial/ethnic groups in the northern Texas cluster. Findings from the Bernoulli model analysis support the results of the Multinomial analysis, indicating that breast cancer mortality is heterogeneously distributed based on racial/ethnic status.
NHB and Hispanics are more likely to be diagnosed with later stages of breast cancer [
We identified one cluster in northeast Texas that had higher risk of breast cancer mortality for NHB compared to NHW though the Bernoulli model analysis. The location of this cluster was the same as that found by the Multinomial analysis; however, the risk was slightly lower. This could be due to fact that we included other races in the Multinomial model analysis and breast cancer incidence and mortality have been observed to be lower among this group compared to NHW, NHB, and Hispanics. For example, cancer incidence and mortality per 100,000 were 51.7 and 8.6, respectively, among Asian/Pacific Islanders for the years 2001 to 2005, while these same rates were 125.4 and 24.3 among NHW, 116.2 and 35.6 among NHB, and 84.6 and 17.2 among Hispanics [
Using Bernouli model analysis, we identified two clusters, one in the western part of the state and another along the border with Mexico, where Hispanics had significantly higher risk of breast cancer mortality compared to NHW. The most likely cluster was the same as that found by the Multinomial analysis; however, the risk was slightly lower. For “other races,” two significant clusters were detected, with each covering four counties around the Houston and Dallas metropolitan areas, respectively. The cluster around Houston had higher risk compared to the Dallas area. This might be related to the fact that the proportion of total population for other races in Houston was higher compared to Dallas.
This type of cluster analysis at the county level can provide useful information to policy makers for the following reasons. The Department of State Health Services divides all Texas counties into Health Service Regions (HSRs), identified numerically from 1 to 11, to provide comprehensive public health services to the citizens of Texas through 8 regional public health offices (Figure Approximately half of Texas counties did not have accredited permanent mammography facilities in 2008 as reported in the Texas Cancer Facts & Figures, 2008 [ Results of our spatial analyses at the county level provide useful information to guide future spatial analyses at finer scales in Texas. Moreover, they are useful jumping-off points to conduct subcluster analyses at the county level or finer scales for a particular population group. For example, the census block or tract level analysis of female breast cancer mortality among NHB could be conducted in the HSRs 4, 5, 6, and 7 covering secondary cluster 2 identified through the Multinomial model (Figure
Map of Health Service Regions of the Department of State Health Services in Texas.
Map of accredited on-site or mobile mammography services by county.
There are several limitations to the present study. We used breast cancer mortality data aggregated at the county level, which could affect the sensitivity of cluster detection. The geographic distribution of total population and number of breast cancer mortality cases at the county level in Texas is heterogeneous, with some counties having much lower mortality than other counties. This could affect study power so that some potential clusters might be missed. Lack of significance for some secondary clusters might be due to this reason or it might be due to the fact that the test is conservative, that is, we compared secondary clusters with the most likely cluster from the simulated datasets. Results may not be comparative to other studies using data aggregated at different geographic scales. Also, we included only race/ethnicity as a confounder in our analyses, yet there are many other known or hypothesized risk factors for breast cancer that we did not analyze, such as age at diagnosis [
Results of our analyses indicate that breast cancer mortality at the county level in Texas is distributed heterogeneously based on racial/ethnic status. The evidence suggests that highest rates of female breast cancer mortality have shifted over time from southeastern areas towards northern and eastern areas of the state. In Texas, NHB had highest risk for breast cancer mortality in the northeastern region and lowest risk in the southern region, while Hispanics had highest risk in the southern region along the border with Mexico and lowest risk in the northeastern region. These findings, along with continuing trends toward urbanization, growing numbers of Hispanic residents, and increasing levels of poverty for many minorities, provide challenges and opportunities for Texas policy makers and health advocates. More research is needed to make informed decisions about effective and efficient distribution of health care resources to reduce breast cancer disparities for Texas residents.
This work is based on Arvind Bambhroliya’s independent research completed at the School of Public Health at the University of Texas Health Science Center at Houston under the supervision of Dr. Ken Sexton. Cancer mortality data were provided by the Texas Cancer Registry, Cancer Epidemiology and Surveillance Branch, Texas Department of State Health Services, Austin,