Disparities Despite Coverage: Gaps in Colorectal Cancer Screening Among Medicare Beneficiaries

Background  Despite its effectiveness in reducing mortality, colorectal cancer (CRC) screening rates are low, especially among low-income and minority groups; however, physician recommendation can increase screening rates.

Methods  We performed a multilevel analysis of the Medicare Current Beneficiary Survey data linked to Medicare claims and the Area Resource File to identify determinants of racial and socioeconomic disparities in CRC screening among 9985 Medicare Parts A and B beneficiaries with a usual physician. Recent CRC screening was defined as receipt of either a home fecal occult blood test, flexible sigmoidoscopy, or colonoscopy at recommended intervals.

Results  Unadjusted rates of screening were 48% for white and 39% for black beneficiaries (P<.001). Racial differences in CRC screening receipt were eliminated after adjustment for socioeconomic status as measured by income and education. Socioeconomic status disparities decreased but remained significant after adjustment for personal and health system factors. Awareness of CRC (adjusted odds ratio, 2.76; 95% confidence interval, 2.29-3.33) and having a primary care generalist (vs another specialist) as one's usual physician (adjusted odds ratio, 1.31; 95% confidence interval, 1.12-1.53) were associated with higher odds of screening, controlling for other factors. The odds of screening were also higher among those whose usual physician was rated more highly on information-giving skills.

Conclusions  Racial differences in CRC screening rates among Medicare beneficiaries with a usual physician are explained by differences in socioeconomic status. Beneficiaries with a primary care generalist as their usual physician had higher rates of CRC screening receipt. Increased efforts to make Medicare beneficiaries aware of the benefits of CRC screening may capitalize on the associations found in this study between CRC knowledge, physician information giving, and timely screening.

Despite a wealth of evidence about its effectiveness in reducing colorectal cancer (CRC) mortality, CRC screening rates are far below ideal levels.^1,2 Rates are particularly low among black and low-income persons who disproportionately experience late-stage diagnosis and excess CRC mortality.^2-5 Age-adjusted CRC mortality rates are 28 per 100 000 among blacks vs 21 per 100 000 among whites.²

We were interested in understanding, first, whether racial differences in CRC screening rates persisted in a group with near universal insurance coverage, that is, Medicare Parts A and B beneficiaries, 93% of whom have a usual physician.^6-11 We also wanted to investigate which health system features, in addition to having a usual source of care,^7-11 facilitate screening and help reduce racial and socioeconomic disparities in CRC screening rates.

The Institute of Medicine outlined health care system features that might present additional hurdles to receiving care among lower-income and minority persons.¹² Features that may have particular relevance to CRC screening include the characteristics of one’s usual physician (specialty, communication, and quality), geographic availability, ability to move through system bureaucracies, reimbursement structures, tiers of care delivery within Medicare due to differing levels of supplementary coverage, and access to colonoscopy.

Medicare beneficiaries are an ideal group in which to analyze such health system factors in relation to CRC screening for different groups. First, this is a racially diverse age group for whom CRC screening is recommended.¹ Second, their primary insurance coverage and access to a usual health care provider helps to decrease the socioeconomic confound in race-based rates of screening. Finally, efforts to improve equity in screening through better understanding of system factors have special relevance to Medicare because 75% of Medicare expenditures are on behalf of persons with annual incomes at or below $25 000.¹³

Using the year 2000 Medicare Current Beneficiary Survey (MCBS),¹⁴ linked claims, and the Area Resource File,¹⁵ we examined CRC screening within this health system context. We aimed to (1) quantify the size of any racial differences in the receipt of CRC screening among beneficiaries and the extent to which racial differences were confounded by socioeconomic status (SES); (2) determine which features of the health care system, in addition to having a usual health care provider, were associated with higher screening rates; and (3) ascertain whether these features differed for socioeconomic and racial groups of beneficiaries.

The MCBS, including the Access to Care components,¹⁴ is an annual, nationally representative in-person survey conducted by the Centers for Medicare and Medicaid Services (CMS). It captures data on beneficiaries’ socioeconomic and demographic characteristics, health status, health care utilization, supplementary insurance, access to services, satisfaction, perceptions of health care provider quality, and usual source of care. The year 2000 survey is the most recent MCBS asking about CRC screening. Beneficiaries are encouraged to have their medical records available at the interview. Survey information is augmented with data from Medicare claims. When linked with the Area Resource File,¹⁵ county data on resources can also be examined.

The sample of 15 339 respondents from the year 2000 MCBS Access to Care survey was drawn by CMS from its Medicare enrollment file. The first stage of sampling was the selection of 107 primary sampling units, consisting of groups of counties chosen to represent the nation. Within primary sampling units, the sample was further restricted to addresses within certain subareas corresponding to ZIP codes. Beneficiaries residing in these areas were selected by systematic random sampling within age strata. The response rate for the MCBS, during which the Access to Care items were asked, was about 84%.¹⁴

For this study, racial groups other than “black” or “white” could not be included owing to small sample sizes. Persons who were institutionalized, who were younger than 65 years, or who had end-stage renal disease were excluded. In the remaining sample of persons with Medicare Parts A and B coverage, 95% had a usual place of care and 93% (of the total) reported having a specific physician. There were 11 154 respondents who constituted this latter group. After excluding from this group persons with a prior diagnosis of CRC or with gastrointestinal symptoms or with large amounts of missing data, we were left with 9985 respondents on whom we could analyze CRC screening measures. This sample yielded adequate statistical power to analyze our outcomes in multivariate models.¹⁶

The dependent variable was a dichotomous measure of receipt of a recent CRC screening test. A beneficiary was considered “recently screened” if either a home fecal occult blood test had been performed in the past 12 months, a flexible sigmoidoscopy was received in the past 5 years, or a colonoscopy was performed in the past 5 years. Even though screening colonoscopy is recommended every 10 years,¹ the wording of MCBS response options on colonoscopy end at “5 or more years ago.” The MCBS did not ask about barium enema.

We relied on self-report of CRC screening from the MCBS rather than from claims because claims do not distinguish well between screening and diagnostic tests because health care providers have been slow to use the new screening codes.¹⁷ In addition, claims do not capture tests paid for by non-Medicare sources or colonoscopy received during the entire age-eligible period (outside of the index year for which claims are linked to the MCBS). Screening rates from the MCBS are consistent with those found in other national data.⁹

The main independent variables of interest were beneficiary race and SES (income and education) and aspects of the health care financing (supplementary insurance) and delivery system (usual physician’s specialty, perceived quality of care from usual physician, availability of specialists, and health maintenance organization [HMO] status). Supplementary insurance status was categorized as “none,” “Medicaid” (dually eligible), or “private” (employer-sponsored or self-purchased private insurance). Persons receiving assistance to pay for Part B, co-payments, and deductibles (Qualified Medicare Beneficiary and Specified Low-Income Medicare Beneficiary) were included in the Medicaid group.

“Primary care generalists” included physicians practicing family practice, internal medicine, geriatrics, or general practice as their primary specialty. Specialty type of one’s usual physician was determined primarily from beneficiary response to the MCBS on “usual physician” type and secondarily from the unique health care provider identification number from claims when the survey response on physician type was absent. Only in 2 categories did we use the claims to override the survey reported data: 1 case (3% of respondents) was for a beneficiary who reported general practitioner/family practitioner/internal medicine and the claims suggested that the usual physician ought to have been a common medical subspecialist. The other case was for those missing the survey response on usual physician’s specialty (13% of respondents); therefore, we identified from claims the physician with whom a beneficiary had the most physician office visits. Claims that were referrals from generalists to specialists or emergency department visits were eliminated from this group.

Three scales on medical care quality delivered by one’s usual physician were assessed: technical skills (4 items, Cronbach α = .89), interpersonal manner (4 items, Cronbach α = 0.79), and information giving (4 items, Cronbach α = 0.84). The grouping of items yielding these dimensions and scores were based on the MCBS factor analysis by Lee and Kasper.¹⁸ We also included, as a separate covariate, the MCBS item that asked about satisfaction with “availability of specialists when the respondent thinks [she or he] needs it.”

Additional beneficiary level covariates included age, sex, marital status, health status, attitudes toward health care, rural vs urban residence, and awareness of CRC. Five items on health care–seeking behavior were included. These were whether the beneficiary (1) “Worries about health more than others your age”; (2) “Would do almost anything to avoid going to doctor”; (3) “When sick, would try to keep sickness to self”; (4) “Usually goes to doctor as soon as you feel bad”; and (5) “Had health problem you think doctor should see but didn’t.”

Because more frequent visits provide greater opportunity for screening, we also controlled for frequency of physician outpatient visits from claims (excluding emergency department visits) in the past 12 months. Because HMO beneficiaries lack claims, they were assigned to a third “don’t know” group for this controlling variable. Logistic regression models were run both with and without the HMO subgroup (18% of the sample) to assess whether the estimates changed.

The following 7 different county level variables from the Area Resource File were examined: HMO penetration, presence of large hospitals, number of physicians per elderly patient, number of gastroenterologists per elderly patient, percentage of population in poverty, median family income, and per capita income.

The beneficiary was the unit of observation. Univariate, bivariate, and stratified analyses were performed, including assessment for interaction. Logistic regression models were then built. First, race (with age and sex) was entered into the model. Then other beneficiary characteristics were added, followed by the health system characteristics. A hierarchical model was then constructed to assess whether there was any significant clustering of MCBS respondents among physicians. We found no significant random effects¹⁹ or clustering at the physician level; this was not surprising because more than 75% of the physicians from the linked claims saw exactly 1 MCBS patient. Finally, a hierarchical model was created with each of the county level variables (one at a time) as the second level. The random effects at the county level were not significant, and estimates of the main effects did not change.¹⁹ Thus, final models are presented as 1-level models. In addition, none of the county level variables were associated with screening use in the multivariate models. The final model had a c statistic of 0.734, suggesting good model fit.²⁰ Sampling weights accounting for the multistage sample design and nonresponse were used to obtain national estimates with SAS (version 9.0; SAS Institute Inc, Cary, NC) callable SUDAAN (version 9.0.0; Research Triangle Institute, Research Triangle Park, NC).

Characteristics of the sample are presented in Table 1. Almost half of these beneficiaries had total annual incomes of $20 000 or less. In terms of supplementary insurance, 24% had none and 8% had Medicaid.

For comparison purposes, MCBS respondents with Medicare Part B coverage who lacked a usual source of care had a recent CRC screening rate of only 21%. In the focus sample for this study (those with a usual physician and Medicare Parts A and B), the recent CRC screening rates were higher but still far from their recommended levels: white vs black beneficiaries’ rates were 48% vs 39%, respectively (P<.001) (Table 2). Absolute differences in recent screening rates were much greater by income and education than by race (24% difference from lowest to highest income group; 25% difference from lowest to highest education groups; 9% difference between whites and blacks). Dually eligible beneficiaries had even lower rates of screening than did those with no supplemental coverage (30% vs 45%, respectively, P<.001). When stratified by income and secondary insurance, CRC screening rates also did not differ by race (Figure). No significant interactions were found between race and either beneficiary or health system characteristics with respect to screening receipt.

Initially, analyses were carried out separately for the fecal occult blood test outcome and the endoscopy outcome; however, subgroup comparisons were consistent with those for the summary variable on receipt of either timely fecal occult blood test or endoscopy. Table 3 presents the adjusted odds of receipt of a recent CRC screening test before and after adding beneficiary and health system factors to the model. Racial differences in receipt of CRC screening disappeared with adjustment for patient SES (model 2). Beneficiaries who had heard about CRC in the past had significantly higher rates of timely screening (model 4).

At the health system level (model 5), several modifiable features of care were associated with CRC screening. Controlling for other factors, patients whose usual physician was a primary care generalist rather than another type of specialist had significantly higher odds of CRC screening. Patients whose usual physician’s information-giving skills were rated more highly had higher odds of CRC screening. Availability of specialty care and being in an HMO were associated with timely screening. (The model was also run excluding the HMO subgroup, and the estimates did not change significantly.) While they helped reduce socioeconomic disparities in screening, none of the health system covariates eliminated them.

Despite having Medicare Parts A and B coverage and a usual physician, the rates of recent CRC screening differed greatly for Medicare beneficiaries with lower vs higher SES. Racial differences in CRC screening receipt were fully explained by SES. It is important to note that racial differences in CRC screening rates still exist. However, the observed racial differences are explained by lower SES in black compared with white beneficiaries rather than by race itself. Higher educational status and awareness of CRC were each significantly associated with screening receipt. Health care system features associated with higher odds of CRC screening included having private supplementary insurance, having a primary care generalist rather than another type of specialist as one’s usual physician, having a usual physician rated more highly on information-giving skills, perception of availability of specialist care when needed, and HMO enrollment. Each of these factors had similar benefits for screening receipt among whites and blacks.

Medicare claims studies from the 1990s found that CRC screening rates were significantly lower for black than for white beneficiaries despite controlling for county-level SES.^22-24 An analysis of National Health Interview Survey (NHIS) data²⁵ found that among men younger than 65 years, but not among women, observed black-white differences in CRC screening rates were fully explained by SES, insurance, and presence of a usual source of care. However, the sample of black NHIS respondents older than 65 years was too small to draw conclusions about factors associated with racial differences in CRC screening for Medicare beneficiaries.²⁵ To these prior data, our analyses add the finding that among Medicare beneficiaries with a usual health care provider, SES eliminates racial differences in receipt of a recent CRC screening test. This finding on CRC screening can be added to recent findings that racial differences in late-stage CRC diagnosis were explained by SES.²⁶

It is possible that older studies did not find an elimination of racial differences in CRC screening with adjustment for SES^22-24 because only county-level data on average SES was available. Individual level data on income and education allows for more specific control of confounding by SES.

The impact of SES on CRC screening receipt may be partly due to competing demands for limited income. Small increases in out-of-pocket costs reduce patient interest in screening.²⁷ Differences in preventive care-seeking and knowledge of CRC screening associated with educational status are also likely. Limited colonoscopy resources in low-income areas are another likely barrier. Qualitative data suggest that health care providers are reluctant to order even the less costly fecal occult blood test if colonoscopy referrals are not available for follow-up of abnormal fecal occult blood test results.^28,29

In 1998, Medicare started to cover some costs of CRC screening. Comparisons of Behavioral Risk Factor Surveillance System data before and after 1998 show an increase in CRC screening among low-income (<$25 000) beneficiaries.^30,31 In our study, dual eligibles and those with no secondary coverage had significantly lower CRC screening rates compared with their counterparts with private secondary insurance. Combined with earlier findings on out-of-pocket costs,²⁷ our findings suggest that basic Medicare Parts A and B coverage, which still requires the beneficiary to pay a substantial deductible and co-payment for colonoscopy, may be inadequate to overcome SES differentials in screening receipt.

Beneficiaries with a primary care generalist as their usual physician had higher odds of timely CRC screening, adjusting for health status and other potential confounders. A substantial proportion of beneficiaries (15%-20%) see only specialists for their care and identify a specialist as their usual health care provider.³² A study comparing men in 2 Medicare-managed care plans found that men in the plan that required beneficiaries to choose a primary care generalist as a usual health care provider had no socioeconomic differentials in use of CRC screening, whereas men in the plan not requiring a primary care physician had persistent socioeconomic differentials in screening use.³³

In our analyses, beneficiaries whose regular physician was rated more highly on information-giving skills had higher CRC screening rates. This aspect of patient-physician communication was also associated with the receipt of timely breast and cervical cancer screening³⁴ and with CRC screening in other studies.^35,36 This association seems to be especially strong among low-income persons.^34,35

Higher rates of CRC screening among HMO enrollees is consistent with HMOs’ high prioritization of preventive service delivery and use of system interventions such as reminder systems.³⁷ Unlike breast and cervical cancer screening, CRC screening was not a Health Employer Data Information Set (HEDIS) reporting requirement at the time of the 2000 MCBS survey, but it has been added to the 2004 HEDIS.³⁸

First, while we excluded from our sample those with gastrointestinal symptoms, we cannot ascertain whether some of the reported use was for diagnostic rather than for screening reasons.³⁹ Second, it is possible that patients who see a primary care physician as their usual health care provider may be more prevention-oriented compared with those who see another type of specialist as their usual health care provider. Third, area data on colonoscopy resources might have explained more of the socioeconomic disparities in screening rates.^28,40 Fourth, CRC screening is necessary but not sufficient to improve CRC mortality rates. Access to timely treatment is also important.^41,42 Finally, it is possible that later adoption of newer screening technologies, such as colonoscopy, by vulnerable groups creates a short-term disparity, whereas if one waited a decade to measure rates, differences might diminish. The most recent national data indicate that large differentials in unadjusted CRC screening rates persist, with absolute differences between whites and blacks and between low- and high-income groups from 10% to 20% in some states.⁴³ Only future data will answer this concern definitively.

In conclusion, racial differences in CRC screening rates among Medicare beneficiaries with a usual physician can be explained by SES. However, disturbing socioeconomic disparities persist in receipt of timely CRC screening among beneficiaries. Beneficiaries with a primary care generalist as their usual source of care have higher rates of CRC screening receipt. Increased efforts to make Medicare beneficiaries aware of CRC screening could capitalize on the associations found in this study between education, knowledge about CRC, and timely screening receipt.

Correspondence: Ann S. O’Malley, MD, MPH, Center for Studying Health System Change, 600 Maryland Ave SW, Suite 550, Washington, DC 20024-2512 (aomalley@hschange.org).

Accepted for Publication: June 06, 2005.

Financial Disclosure: None.

Funding/Support: This study was funded by grants NCI-KO7 CA 91848 (Dr O’Malley) and KO5 CA96940 (Dr Mandelblatt) from the National Cancer Institute, National Institutes of Health, Bethesda, Md.

Disclaimer: Dr O’Malley had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Additional Information: The MCBS data and claims are covered under the Data Use Agreement (No. 13459) with the Centers for Medicare and Medicaid, Baltimore, Md.

Acknowledgment: We thank Christopher Hogan, PhD, for excellent technical management of the claims data files and creation of the merged data sets for analysis. We also thank the reviewers for their comments.