Fax: (916) 734-7946
Geographic variation of racial/ethnic disparities in colorectal cancer testing among medicare enrollees
Article first published online: 10 JAN 2011
Copyright © 2011 American Cancer Society
Volume 117, Issue 8, pages 1755–1763, 15 April 2011
How to Cite
Semrad, T. J., Tancredi, D. J., Baldwin, L.-M., Green, P. and Fenton, J. J. (2011), Geographic variation of racial/ethnic disparities in colorectal cancer testing among medicare enrollees. Cancer, 117: 1755–1763. doi: 10.1002/cncr.25668
- Issue published online: 6 APR 2011
- Article first published online: 10 JAN 2011
- Manuscript Accepted: 18 AUG 2010
- Manuscript Revised: 29 JUL 2010
- Manuscript Received: 24 MAY 2010
- cancer screening;
- colorectal cancer;
- geography of health;
The Medicare population has documented racial/ethnic disparities in colorectal cancer (CRC) screening, but it is unknown whether these disparities differ across geographic regions.
Among Medicare enrollees within 8 US states, we ascertained up-to-date CRC screening on December 31, 2003 (fecal occult blood testing in the prior year or sigmoidoscopy or colonoscopy in the prior 5 years). Logistic regression models tested for regional variation in up-to-date status among white versus different nonwhite populations (blacks, Asian/Pacific Islanders [APIs], Hispanics). We estimated regression-adjusted region-specific prevalence of up-to-date status by race/ethnicity and compared adjusted white versus nonwhite up-to-date prevalence across regions by using generalized least squares regression.
White versus nonwhite up-to-date status varied significantly across regions for blacks (P = .01) and APIs (P < .001) but not Hispanics (P = .62). Whereas the white versus black differences in proportion up-to-date were greatest in Atlanta (Georgia), rural Georgia, and the San Francisco Bay Area of California (range, 10%-16% differences, blacks<whites); there were no significant white versus black differences in Connecticut, Seattle (Washington) or Iowa. Whereas APIs had significantly lower up-to-date prevalence than whites in Michigan and the California regions of San Francisco, Los Angeles, and San Jose (range, 4%-15% differences, APIs<whites), APIs in Hawaii had higher up-to-date status than whites (52% vs 38% P < .001). White versus Hispanic differences were substantial but homogeneous across regions (range, 8%-16% differences, Hispanics<whites). In contrast to nonwhites, there was little geographic variation in up-to-date status among whites.
Significant geographic variation in up-to-date status among black and API Medicare enrollees is associated with heterogeneous racial/ethnic disparities for these groups across US regions. Cancer 2011. © 2011 American Cancer Society.
In 2009, colorectal cancer (CRC) accounted for 10% of new cancer cases and 9% of all cancer deaths in the United States.1 Screening by either fecal occult blood testing or flexible sigmoidoscopy has been demonstrated to reduce the incidence and mortality from CRC,2-3 and compelling observational data suggest a similar benefit from colonoscopy screening.4 Professional and public health societies now strongly recommend CRC screening for all adults older than age 50 years.5-6 Although Medicare has covered CRC screening since 1998, only half of Medicare enrollees have been screened according to guidelines.7-8
In addition to suboptimal overall uptake of CRC screening within the Medicare population, there is evidence of considerable persistent disparity in CRC screening rates among racial and ethnic groups.7, 9-12 Potential explanations for racial/ethnic disparities in screening use include differences in socioeconomics,13-15 access to care,13-14, 16 and cultural factors,15, 17-18 and the relative contribution of these factors to the development of disparity may differ across racial/ethnic groups.19 Previous work suggests geographic variation in CRC screening among the aggregate Medicare population.20-21 However, it is not known whether racial/ethnic differences in CRC screening also vary across geographic regions.
Geographic variation in healthcare disparities may develop due to regional differences in healthcare quality among whites and nonwhites.22 National survey data indicate CRC screening rates are significantly higher in regions where a greater proportion of primary care physicians routinely recommend CRC screening.23 Meanwhile, blacks disproportionately receive primary care from physicians with less training and fewer resources.24 Thus, regional variation in provider quality may lead to different CRC screening rates within racial/ethnic groups. Wide variation in white versus black disparities has been documented for certain preventive procedures among Medicare enrollees (eg, mammography, eye examination for diabetes)25; but to our knowledge, regional variation in white versus minority disparities in CRC screening have not been previously described.
In this study, we determined whether racial/ethnic differences in CRC testing varied across geographic regions in the Medicare population. We further assessed whether regional variation in disparities was attributable to varying use of CRC screening among whites, nonwhites, or both.
MATERIALS AND METHODS
Setting and Subjects
We identified subjects from the 2003 random 5% sample of Medicare enrollees living in US Surveillance, Epidemiology, and End Results (SEER) regions. SEER is a federally sponsored collaboration of regional cancer registries. We included data from 11 SEER registries in 8 states (California, Connecticut, Georgia, Hawaii, Iowa, Michigan, New Mexico, Washington) representing approximately 14% of the US population. The population of Medicare enrollees within these regions is comparable to the general Medicare population in terms of age, sex, poverty, and education but is slightly more urban and has a greater proportion of minorities and foreign-born residents.26
Patients were eligible for inclusion when they were aged 69 to 79 years as of January 1, 2003 and were continuously enrolled in fee-for-service Medicare (Parts A and B) for the 5 preceding years. We required 5 years of fee-for-service enrollment so that we could ascertain patient CRC screening up-to-date status on the basis of a continuous claim history for 5 years up to December 31, 2003. Whereas the 5% sample included patients with prior cancer diagnoses, we excluded patients who had claims for a diagnosis that was an indication for surveillance colonoscopy, including any prior diagnosis of CRC (according to SEER data), or any diagnosis of inflammatory bowel disease or colonic polyps on the prior 5 years of claims.
Identification of CRC Screening Tests
We identified CRC screening tests from the American Medical Association's Current Procedural Terminology (CPT), Healthcare Common Procedural Coding System (HCPCS), and the World Health Organization's International Classification of Disease, Ninth Revision, Clinical Modification (ICD-9-CM) procedure codes on claims from physician encounters, hospital outpatient encounters, and inpatient admissions. Fecal occult blood test codes included CPT code 82270 and HCPCS code G0107. Sigmoidoscopies were identified by CPT codes 45300, 45305, 45308-9, 45315, 45317, 45320, 45330, 45331, 45333, 45334, 45338, 45339; HCPCS code G0104, and ICD-9-CM codes 45.24, 48.23, 48.24. Colonoscopies were identified by CPT codes 45378, 45380, 45382, 45383, 45384, 45385; HCPCS codes G0105, G0121, and ICD-9-CM codes 45.23, 45.25, 45.42, 45.43, and 48.36. We excluded double-contrast barium enemas as there were very few claims for barium enema during the study period.
For the primary analysis, we defined patients as up-to-date with CRC testing on December 31, 2003 when 1 or more claims were submitted for either 1) fecal occult blood testing in 2003 or 2) sigmoidoscopy or colonoscopy from 1999 to 2003 (a 5-year observation period). In secondary analyses, we constructed a similar outcome but with a 10-year screening interval for colonoscopy, as suggested by some guidelines.5 For the subset of enrollees with 10 years of continuous enrollment from 1994-2003, we defined patients as up-to-date on December 31, 2003 under any of the following conditions 1) an FOBT claim was submitted in 2003, 2) a sigmoidoscopy claim was submitted in 1999-2003, or 3) a colonoscopy claim was submitted in 1994-2003. No attempt was made to delineate whether tests were performed for screening or diagnostic purposes because of misclassification concerns.27 We assumed that both screening and diagnostic tests fulfilled the criteria for screening tests, and both were used in calculating up-to-date status.
Race and Ethnicity
We categorized patient race/ethnicity as non-Hispanic white, black, Asian or Pacific Islander (API), or Hispanic by using the Medicare Enrollment Database race classification. Although this classification system is highly accurate for white and black categorization,28 the sensitivity for Hispanic and API classification is low. Nevertheless, the positive predictive value for Hispanic and API classification is sufficient to make meaningful comparisons between the up-to-date statuses within these groups and the non-Hispanic white group.29 Because we intended to make comparisons of up-to-date status among defined racial/ethnic groups, we excluded patients classified as “Other” or who had missing classification data. Furthermore, we excluded patients identified as Native American because their sample sizes were too small.
Although several registries joined the SEER program in 2000, we limited our samples to longstanding SEER regions because of the possibility that CRC diagnoses may not have been ascertained before 2000. Patients were included from 11 regions, including Atlanta, Georgia; rural Georgia (10 counties in rural Georgia); 3 California registries (San Francisco-Oakland, San Jose-Monterey, and Los Angeles County); Seattle-Puget Sound, Washington; Detroit, Michigan; and 4 statewide registries encompassing Connecticut, Hawaii, Iowa, and New Mexico. We excluded the Utah registry because of small samples of nonwhite patients.
We identified a range of patient-level covariates to allow adjustment for sociodemographic, medical, and environmental factors that could affect regional differences in CRC testing. We classified patients by age on December 31, 2003 (aged 70-74 years vs 75-79 years), sex, the rurality of patients' counties of residence based on the US Department of Agriculture Rural/Urban Continuum codes (metropolitan county; nonmetropolitan but adjacent to metropolitan county; nonmetropolitan, nonadjacent to metropolitan county, population ≥20,000; nonmetropolitan, nonadjacent to metropolitan county, population <20,000). We defined income level as the median income of elderly (aged ≥65 years) householders within the same residence ZIP code as reported in the 2000 US Census. We also calculated the Charlson Comorbidity Index for each patient from inpatient and outpatient Medicare claims during 2002.30
We used descriptive analyses to characterize the regional samples by race/ethnicity. Adjusted racial/ethnic differences in up-to-date status and regional heterogeneity in these differences were then assessed by using multivariate regression methods via generalized linear models. Because not all ethnic groups were well represented in each region and to provide customized covariate adjustments for each specific racial/ethnic comparison, we fit separate sets of regression models for Asians, African Americans, and Hispanics, with each set comparing 1 of these groups to whites. For each group, we first assessed geographic heterogeneity in adjusted up-to-date rate differences between white and nonwhite race/ethnicities, using a logistic regression model with up-to-date status as the dependent variable and main effects specified for the race/ethnicity indicator and region, an interaction term between race/ethnicity and region, and covariates (ie, age, sex, comorbidity, household income, and rural vs urban residence). For this model, we used the Wald test for the nonwhite race/ethnicity by region interaction term as an overall test for significant geographic variation in nonwhite versus white differences in up-to-date status.
Next we fit a generalized linear model with an identity link and binomial variance function (ie, a generalized least square regression) to estimate adjusted nonwhite versus white differences in up-to-date status prevalence for each racial/ethnic group within each region, while adjusting for covariates.31 The fitted model was also used to obtain regression-adjusted, up-to-date prevalence rates for each of the compared racial/ethnic groups by the method of conditional prediction to statistically control for covariates.32
In all models for each nonwhite subgroup, we excluded registries when there were <50 enrollees in that subgroup residing within the registry. To assess the sensitivity of results to potential underascertainment of CRC tests, we performed similar analyses by using a smaller cohort with longer-term Medicare enrollment, using the secondary definition of up-to-date status based on a 10-year observation period for colonoscopy. Because results were generally similar for the primary and secondary definitions of up-to-date, we present results using the primary definition.
Hypothesis tests were 2-tailed with a level of significance of α = .05. We used SAS (Version 9.1; SAS Institute, Cary, NC) to perform the analyses. Because some have argued that not all differences in racial/ethnic in health measures constitute disparities,33 we refer to statistically significant adjusted differences of at least 5 percentage points as “disparities,” and otherwise use the term “difference”.
A total of 53,990 Medicare enrollees in 11 geographically dispersed registry locations were available for analysis (Table 1). There was considerable variation across registries in the representation of nonwhite racial/ethnic subgroups. Black enrollees were substantially represented (≥50 individuals) in all registries except Hawaii and New Mexico. API enrollees had considerable representation in Hawaii, Detroit, Michigan, and the western registries (ie, California and Seattle). Substantial samples of Hispanics were present in the California registries, New Mexico, and Connecticut.
|Row % (No.)||Row % (No.)||Row % (No.)||Row % (No.)||No. (Column %)|
|Atlanta, GA||81 (2898)||19 (672)||<1 (19)||<1 (13)||3602 (7)|
|Connecticut||94 (6516)||4 (294)||<1 (35)||1 (87)||6932 (13)|
|Hawaii||36 (492)||1 (16)||62 (845)||1 (17)||1370 (3)|
|Iowa||99 (8258)||1 (96)||<1 (23)||<1 (10)||8387 (16)|
|Los Angeles, CA||72 (6257)||8 (704)||12 (1012)||9 (793)||8766 (16)|
|Detroit, MI||82 (7597)||17 (1604)||<1 (56)||<1 (43)||9300 (17)|
|New Mexico||88 (2845)||1 (42)||<1 (11)||11 (350)||3248 (6)|
|Rural Georgia||70 (230)||30 (97)||0 (0)||0 (0)||327 (<1)|
|San Francisco, CA||71 (2736)||8 (307)||18 (683)||3 (117)||3843 (7)|
|San Jose, CA||80 (2115)||2 (58)||11 (283)||7 (179)||2635 (5)|
|Seattle, CA||95 (5293)||2 (126)||3 (140)||<1 (21)||5580 (10)|
|Total||84 (45,237)||7 (4016)||6 (3107)||3 (1630)||53,990 (100)|
Within individual logistic regression models that tested whether white versus nonwhite disparities differed by region, interaction terms were significant for blacks (P = .01) and for APIs (P < .001) but not for Hispanics (P = .62). Across regions with substantial black populations, white versus black differences in up-to-date CRC screening rates ranged from −0.5 percentage points in Iowa to 16.0 percentage points in San Jose (Fig. 1), and the estimated up-to-date prevalence for blacks ranged from 29% (San Jose, Calif) to 44% (Iowa). In these same regions, the estimated up-to-date prevalence for whites ranged from 43% (Los Angeles, Calif) to 49% (Atlanta, Ga).
We also observed variation in the up-to-date status among APIs (Fig. 2). The within-region adjusted prevalence of APIs that were up-to-date for CRC screening ranged from 32% to 52%, and API versus white adjusted up-to-date prevalence differences varied widely across regions (−13.9 to 15.0 percentage points). In contrast to the adjusted API versus white differences seen in Los Angeles, San Francisco, San Jose, and Michigan (range, 4-15 percentage points), API enrollees had significantly higher up-to-date rates than whites in Hawaii (adjusted prevalence difference, 13.9 percentage points).
Substantial white versus Hispanic disparities were present in all regions with sizable Hispanic populations (Fig. 3). However, as predicted by the overall model, the estimated disparity between Hispanic and white up-to-date status was similar across regions (range, 8-16%), although the Connecticut difference did not meet statistical significance (Fig. 3).
Overall, the estimated proportion of whites up-to-date was generally consistent across regions (Figs. 1-3). In most regions, the proportion of whites up-to-date ranged from 43% to 49%, although slightly lower estimates were observed for whites in Hawaii (38%) and New Mexico (39%). Thus, regional nonwhite versus white differences largely derived from regional variation in up-to-date rates among nonwhites.
In Hawaii, the proportion of up-to-date APIs (52%) was much greater than elsewhere, and the proportion of whites up-to-date (38%) was lower than in the mainland registries. We further explored the Hawaii findings by evaluating the unadjusted up-to-date status for the API and white populations in Hawaii by specific CRC tests (ie, FOBT vs sigmoidoscopy and/or colonoscopy). APIs had higher up-to-date rates both for FOBT (28% among APIs vs 16% among whites) and for sigmoidoscopy and/or colonoscopy (38% among APIs vs 31% among whites).
Racial/ethnic differences in CRC testing rates differ substantially by geographic location. Except for relatively consistent disparities between whites and Hispanics, differences in CRC testing rates between whites and nonwhites were substantial in some regions and minimal or nonexistent in others. In addition, regional differences in up-to-date status were primarily driven by variation in the up-to-date status of the minority group, whereas whites had similar rates of up-to-date status across most regions.
Because all subjects had insurance coverage for CRC screening by virtue of Medicare enrollment, several factors may have disproportionately impaired CRC screening delivery among minority patients in some regions but not others. First, it is plausible that nonwhites in some regions may be segregated within primary care practices and health systems that may be less likely to deliver CRC screening.13-14, 16, 24, 34 Second, nonwhite Medicare enrollees in some regions may have differentially lower access to primary care, which has been associated with increased CRC screening delivery.11 Third, nonwhites in some regions may have greater or lesser access to gastrointestinal specialists, whose numbers are lower in counties with a higher proportion of black residents but higher in counties with a higher proportions of API residents.35
A possible explanation for variation in local CRC testing rates among APIs is regional differences in ethnicity and culture, which may influence attitudes toward preventive care and CRC screening.17-18 For example, Chinese immigrants in Seattle identify the maintenance of energy (qi) and spirit (jing shen), exercise, and diet moderation rather than FOBT or endoscopy as means of CRC prevention.36 Meanwhile, Japanese-Americans comprise the largest ethnic subgroup within the API population residing in Hawaii and have relatively high self-reported FOBT use.37 In our study, APIs in Hawaii had greater up-to-date status in 2003 than all other racial/ethnic subgroups, including mainland whites, and Hawaiian APIs exhibited high rates of both FOBT and sigmoidoscopy and/or colonoscopy. With greater acculturation and more favorable socioeconomic status, Japanese-Americans may be favorably predisposed to CRC testing compared with other API subgroups. Meanwhile, research is needed to confirm and illuminate the reasons for comparatively low rates of CRC testing among white Medicare enrollees residing in Hawaii as compared with mainland whites.
There was a consistent 8-16 percentage point disparity in Hispanic versus white up-to-date testing rates in California and New Mexico. Colorectal cancer screening rates among Hispanics depend strongly on acculturation,19, 38-39 which could not be measured in this study. As most Hispanics in California and New Mexico are of Mexican heritage,40 it is possible that the level of acculturation among Hispanic Medicare enrollees may be similar in these regions. In Connecticut, where most Hispanics may be of Puerto Rican or Dominican descent,40 the adjusted Hispanic versus white difference (8 percentage points) was similar to California and New Mexico but did not meet statistical significance, likely due to the relatively small sample of Hispanics in this region. Nevertheless, other research suggests considerable differences in healthcare disparities across different Latino subgroups.41
Several additional limitations warrant discussion. Variation in SES within nonwhite ethnic groups could impart regional heterogeneity in CRC testing rates.13-15 Our models controlled for median income at the neighborhood level and urban versus rural residence, but these contextual measures could have allowed residual confounding by SES.42 Although the included registries represent a large fraction of the US Medicare population, sample sizes of nonwhites were small in some regions, and white versus nonwhite disparities may differ in other US regions. In addition, our primary definition of up-to-date status did not account for colonoscopies performed within the prior 10 years, but a sensitivity analysis within a subgroup with 10 years of claims data yielded similar results. Our analysis also could not differentiate CRC whether testing was being performed for screening or for diagnostic purposes.27 Furthermore, the accuracy of FOBT claims are uncertain,43 and regional variation in FOBT claim submission by providers caring for white versus nonwhite patients could impact estimated differences in CRC screening rates. Moreover, SEER regions are of variable size and, unlike hospital referral regions, do not represent single healthcare markets.44 Substantial variation in disparities is conceivable even within these regions. Finally, our study design did not permit direct assessment of regional CRC screening rates on cancer outcomes; however, expansion of the coverage of CRC screening by Medicare was associated with an increased proportion of CRC patients being diagnosed at an early stage.45
Although the overall rate of up-to-date screening remains suboptimal, we have demonstrated significant geographic variation in white versus nonwhite CRC testing rate differences for black and API, but not Hispanic Medicare enrollees. Furthermore, our results suggest that regional disparities largely derive from variation in up-to-date status among nonwhites, suggesting that screening rates may vary as much within racial/ethnic groups as between them. Research is needed to elucidate the local and regional barriers that impair CRC screening among nonwhite populations in some US locations but not others.
CONFLICT OF INTEREST DISCLOSURES
This study was supported by an American Cancer Society Mentored Research Scholars Grant to Dr. Fenton (MRSGT-05-214-01-CPPB).
- 5Screening and surveillance for the early detection of colorectal cancer and adenomatous polyps, 2008: a joint guideline from the American Cancer Society, the US Multi-Society Task Force on Colorectal Cancer, and the American College of Radiology. CA Cancer J Clin. 2008; 58: 130-160., , , et al.
- 13Predictors of colorectal cancer screening participation in the United States. Am J Gastroenterol. 2003; 98: 2082-2091., ,Direct Link:
- 20Geographic variation among Medicare beneficiaries in the use of colorectal carcinoma screening procedures. Am J Gastroenterol. 2004; 99: 1544-1550.,Direct Link:
- 22Institute of Medicine (U.S.). Committee on Understanding and Eliminating Racial and Ethnic Disparities in Health Care. Unequal Treatment: Confronting Racial and Ethnic Disparities in Health Care. Washington, DC: National Academy Press; 2002., , ,
- 23Association of regional variation in primary care physicians' colorectal cancer screening recommendations with individual use of colorectal cancer screening. Prev Chronic Dis. 2007; 4: A90., , , et al.
- 26Overview of the SEER-Medicare data: content, research applications, and generalizability to the United States elderly population. Med Care. 2002; 40(8 suppl):IV- 3-18., , , ,
- 37Hawaii Behavioral Risk Factors Surveillance Report 1994-2000, Hawaii Progress Toward Healthy People 2000. Honolulu, HI: Hawaii Department of Health, Community Health Division; 2002., , .
- 40US Census Bureau. Tables PCT1: Total population racial or ethnic grouping by Hispanic or Latino, Mexican, Puerto Rican, Cuban, other Hispanic or Latino. US Census Bureau website: http://factfinder.census.gov. Accessed May 7, 2010.