Fax: (410) 706-5394
Out-of-pocket health care expenditure burden for Medicare beneficiaries with cancer
Article first published online: 7 DEC 2012
Copyright © 2012 American Cancer Society
Volume 119, Issue 6, pages 1257–1265, 15 March 2013
How to Cite
Davidoff, A. J., Erten, M., Shaffer, T., Shoemaker, J. S., Zuckerman, I. H., Pandya, N., Tai, M.-H., Ke, X. and Stuart, B. (2013), Out-of-pocket health care expenditure burden for Medicare beneficiaries with cancer. Cancer, 119: 1257–1265. doi: 10.1002/cncr.27848
- Issue published online: 4 MAR 2013
- Article first published online: 7 DEC 2012
- Manuscript Accepted: 23 AUG 2012
- Manuscript Revised: 1 AUG 2012
- Manuscript Received: 31 MAY 2012
- out-of-pocket burden
BACKGROUND: There is increasing concern regarding the financial burden of care on cancer patients and their families. Medicare beneficiaries often have extensive comorbidities and limited financial resources, and may face substantial cost sharing even with supplemental coverage. In the current study, the authors examined out-of-pocket (OOP) spending and burden relative to income for Medicare beneficiaries with cancer.
METHODS: This retrospective, observational study pooled data for 1997 through 2007 from the Medicare Current Beneficiary Survey linked to Medicare claims. Medicare beneficiaries with newly diagnosed cancer were selected using claims-based diagnoses. Generalized linear models were used to estimate OOP spending. Logistic regression models identified factors associated with a high OOP burden, defined as spending > 20% of one's income during the cancer diagnosis and subsequent year.
RESULTS: The cohort included 1868 beneficiaries with and 10,047 without cancer. Compared with the noncancer cohort, cancer patients were older, had more comorbidities, and were more likely to lack supplemental coverage. The mean OOP spending for cancer patients was $4727. Cancer patients faced an adjusted $976 (P < .01) incremental OOP spending. Greater than one-quarter (28%) of beneficiaries with cancer experienced a high OOP burden compared with 16% of beneficiaries without cancer (P < .001). Supplemental insurance and higher income were found to be protective against a high OOP burden, whereas assets, comorbidity, and receipt of cancer-directed radiation and antineoplastic therapy were associated with a higher OOP burden.
CONCLUSIONS: Medicare beneficiaries with cancer face a higher OOP burden than their counterparts without cancer; some of the higher burden was explained by the higher comorbidity burden and lack of supplemental insurance noted among these patients. Financial pressures may discourage some elderly patients from pursuing treatment. Cancer 2013. © 2012 American Cancer Society.
Medical care spending on cancer has increased dramatically in recent years.1 As a result, the cost of care per patient has grown substantially,2 and there is a growing concern that the potential out-of-pocket (OOP) cost may discourage treatment, and that the realized OOP cost creates substantial financial hardships for cancer patients and their families.3,4 The issue of OOP burden for the Medicare beneficiary population is more compelling and complex than for younger adults. Cancer is a disease associated with aging, with a reported incidence among older adults that is nearly 10 times the rate in those aged < 65 years.5,6 The Medicare program provides basic coverage for cancer care, but has substantial cost sharing for covered services. For example, Medicare Part B covers treatments administered in physician offices, but only covers 80% of reimbursable costs. In fact, the Medicare program covered less than one-half of beneficiary total health care spending in 2006.7 This was true even after implementation of the Part D prescription drug benefit. The problem of OOP spending is exacerbated by the lower income and assets available for older compared with younger adults.8
Most Medicare beneficiaries have supplemental insurance, which limits OOP spending requirements. Estimates from 2008 indicate that approximately 50% of beneficiaries had employer-sponsored insurance (ESI) or other private insurance; 15% had Medicaid; and 24% were enrolled in Medicare Advantage plans, leaving 10% of Medicare beneficiaries with only Medicare fee-for-service.7 Before the implementation of Medicare Part D, approximately one-third of beneficiaries lacked prescription drug coverage at some point during the year.9 This dropped to 10% with the implementation of Medicare Part D.7 Despite the high prevalence of supplemental insurance, the exposure to OOP burden can still be significant. Estimates from 2006 suggest that beneficiaries paid 25% of medical care costs OOP. The median OOP spending on health care was reportedly 5.4% of income.7 Recent studies have indicated that Medicare Part D implementation reduced OOP spending on prescription medications, but the change only affected a percentage of the total OOP exposure.10,11
Although several studies have focused on OOP spending in the Medicare beneficiary population12-17 and have examined access to care among cancer survivors broadly defined,18,19 to the best of our knowledge only 1 study to date has examined OOP spending for beneficiaries with cancer, but this relied on self-reported cancer diagnosis and treatment.20 Furthermore, to our knowledge there is no information regarding OOP spending among Medicare beneficiaries with newly diagnosed cancer. These patients and their physicians are in the difficult process of making key treatment decisions, and there is little information concerning OOP spending associated with treatment in the context of the beneficiary's overall health care use and financial resources. We begin to fill this knowledge gap by examining OOP spending and burden relative to income for Medicare beneficiaries with cancer.
MATERIALS AND METHODS
Medicare Current Beneficiary Survey (MCBS) data were pooled from 1997 through 2007. Pooling increased the analytic sample size and permitted testing for trends in OOP burden over time, including 2 years after the implementation of Medicare Part D. The MCBS is nationally representative of Medicare beneficiaries. Approximately 4500 new beneficiaries are sampled each year, with the beneficiary remaining in the survey for up to 4 years, subject to death or loss to follow-up. We used data from the MCBS Income and Assets supplement, which is administered in the spring of each year and elicits detailed responses regarding assets and income sources and amounts from the prior calendar year.
The MCBS collects data regarding a wide array of demographic, insurance, health status, and access to care measures. Health care use and spending data, including prescription medication events (PME), are collected using Explanation of Benefit forms, provider bills, and prescription medication containers. Expenditure data distinguish between payments by Medicare, public and private insurers, and OOP payments. Survey data are linked to detailed Medicare Parts A and B claims.
We constructed 4-year cohorts of Medicare beneficiaries who were diagnosed with cancer at any time during their tenure in the survey. To ensure completeness of the Medicare claims and PME files, we excluded beneficiaries who resided in a long-term care facility, those not continuously enrolled in both Medicare Parts A and B, or, alternatively, those enrolled in a Medicare health maintenance organization during the study period. We selected International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) codes for qualifying cancer diagnoses (140-172, 174-208, 225, 227.3, and 227.4); beneficiaries were required to have at least 1 inpatient or 2 outpatient or medical provider claims with a qualifying cancer diagnosis. Outpatient or provider claims had to be at least 30 days, but no more than 13 months, apart. An index date was assigned based on the first qualifying cancer claim. Each 4-year cohort identified with cancer was matched with a comparison cohort of beneficiaries without cancer. Index years were assigned for the noncancer group to match the distribution for cancer beneficiaries. We identified the subset of “newly diagnosed” cases using a prior 12-month period during which the beneficiaries did not qualify as having cancer based on the previously described definition.
The cancer site was identified based on the ICD-9-CM codes that qualified the respondent as a beneficiary with cancer. When ICD-9-CM codes indicated involvement of > 1 site, we applied a hierarchy, assigning beneficiaries to the cancer site that is more commonly the source of metastases in a secondary site.21 For example, if the beneficiary had diagnoses for cancer in the breast and bone, we assigned the beneficiary to the breast cancer group. We created indicators for lung, colorectal, breast, and prostate cancer, grouping beneficiaries with all other cancer sites and without a known primary tumor.
Measurement of Key Variables
Dependent measures included total OOP spending and a binary indicator for OOP spending that was > 20% of income. OOP spending for health care services included direct payments to providers as well as unpaid liabilities. Spending was measured for the index and subsequent calendar year for a maximum of 24 months. A 24-month spending window was chosen because of a limitation of the PME file, which does not capture specific dates of service within the year. We allowed the 2-year period to capture drug expenditures associated with initial treatment, particularly for those patients diagnosed later in their index year. Income was measured as reported in the MCBS Income and Assets supplement. Underreporting of income has been observed in the MCBS relative to the Current Population Survey, particularly among married individuals.22 To address this limitation, we generated income adjustment factors specific to respondent sex and marital status based on comparison between the MCBS and the Current Population Survey.23 For married persons, we attributed one-half of reported income to the respondent. Annual income was prorated for beneficiaries who died during the year. All spending and income amounts were inflated to constant 2007 dollars, using the Consumer Price Index. After adjustment, we divided the OOP spending by income, and created indicators for a ratio > 20%.
We used measures of beneficiary demographics, health status, and health behaviors reported during the index year. Supplemental medical and prescription drug coverage were measured using a combination of administrative enrollment indicators (Medicare and Medicaid), supplemented by self-reported enrollment in up to 5 private plans. We generated mutually exclusive categories that combined medical and drug coverage: 1) ESI with prescription coverage; 2) other private medical and prescription coverage; 3) public medical and prescription coverage; 4) any medical without prescription coverage; 5) prescription-only coverage; and 6) neither medical nor prescription coverage. Categories were designed to span the period before and after the implementation of Medicare Part D, and therefore did not specify the source of the prescription drug coverage. Hence, a Medicare beneficiary who was dual-reported enrolled in Medicaid and receiving prescription coverage through Medicaid in 2005 would be included in the same category as a Medicaid dual-enrolled individual who received Part D coverage in 2006 or later. Health status measures included self-reported functional status limitations and counts of hierarchical coexisting conditions (HCCs). HCCs are claims-based indicators developed as part of the risk adjustment system used by the Centers for Medicare and Medicaid Services to establish capitated payment rates for Medicare Advantage plans.
Major categories of cancer treatment (surgery, radiation, and antineoplastic therapy) were identified based on procedure codes in Medicare claims and specific therapeutic drug classes in the PME files. We identified the Healthcare Common Procedure Coding System (HCPCS) codes for the Medicare Part B treatments (surgery, radiation, and antineoplastic therapy), and pharmacologic drug classes from the First Databank (available at www.fdbhealth.com) dictionary for prescription antineoplastics. We created indicators for beneficiary receipt of each type of cancer-directed treatment during the index or subsequent year.
Bivariate analyses described OOP spending and OOP burden relative to income, comparing beneficiaries with cancer with the comparison cohort. Multivariate models estimated within the cancer cohort examined the association between beneficiary characteristics and OOP spending level and high OOP burden. Characteristics included supplemental insurance; assets; income as a percentage of the federal poverty level (%FPL); demographics; attitudes toward care seeking; and health status, including primary cancer site and type of cancer-directed therapy. Models controlled for months from January to cancer index month, months observed after the cancer index month, and an indicator if the beneficiary died during the observation period. The effect of Medicare Part D implementation on OOP spending was assessed by including the index year in multivariate models, using 2004 through 2005 as a reference period.
OOP spending models were estimated using a generalized linear model with a γ distribution and a log link. Marginal effects were computed to reflect the magnitude of the effects of a 1-unit change in the independent variable. Models indicating an OOP burden > 20% of income were estimated using logistic regression analysis. Marginal probabilities reflect the change in the outcome when there is a unit change in each independent measure. Bivariate analyses were weighted using cross-sectional sampling weights; all statistical tests were adjusted for the complex survey design of the MCBS. Analyses were performed using SAS version 9.2 (SAS Institute Inc, Cary, NC) and Stata 12 (StataCorp, College Station, Tex) statistical software. The study was approved by the University of Maryland Institutional Review Board.
The sample included 1868 beneficiaries who were newly diagnosed with cancer, and 10,047 in the comparison group without cancer. Beneficiaries with cancer were older, more likely to be men (51.2% vs 44.0%) and married (56.4% vs 52.0%), lacked supplemental medical coverage, and had greater assets compared with the comparison group. Among beneficiaries with cancer, 16.8% had breast cancer, 11.9% had lung cancer, 21.5% had prostate cancer, and 14.8% had colorectal cancer (Table 1).
|Characteristic||With Cancer||Without Cancer|
|No.||1868 %||10,047 %|
|Other, metastatic unknown||35.0||—|
|Private, ESI with Rx||36.5||35.4|
|Private, other with Rx||7.9||9.3|
|Public with Rx||12.5||17.6|
|Income as % FPLa|
|Aged ≥65 y and former SSDI||7.1||7.0|
|Functional status limitationsb|
|Attitudes regarding care seeking|
|“Usually go to doctor as soon as you feel bad”||39.8||34.6|
Cumulative 2-year spending was $41,561 for Medicare beneficiaries with cancer and $17,630 for the comparison group (Table 2). Beneficiaries with cancer paid $4727 OOP; the comparison group paid $3209. With adjustment for study characteristics, beneficiaries with cancer faced an incremental $976 OOP spending over those without cancer (data not shown). The mean OOP spending was 13.4% of the mean income for beneficiaries with cancer compared with 8.5% for those without cancer. At the individual level, the distribution of OOP burden included extremely large values associated with small denominators that are bounded at 0, so that the mean burden is difficult to assess. The median value for OOP/income was 10.0% for beneficiaries with cancer compared with 6.1% for those without cancer. Large OOP burdens were much more common among beneficiaries with cancer; approximately one-half (49.9%) spent at least 10% of their income OOP, and 27.6% spent 20% of their income compared with approximately one-third (34.3%) and 15.8%, respectively, among the comparison group.
|With Cancer||Without Cancer|
|Total health care spending||$41,561||$1078||$17,630||$310||<.001|
|By payment source|
|Percentage with OOP/income|
Figure 1 illustrates OOP spending by selected characteristics among Medicare beneficiaries with cancer. OOP spending varied by age, educational attainment, income relative to FPL, assets, HCC counts, limitations in functional status, attitudes regarding care seeking, and cancer sites and treatment.
Table 3 presents the adjusted effects of individual characteristics on OOP spending levels for beneficiaries with cancer. Beneficiaries with ESI or public medical coverage experienced lower OOP spending ($1075 and $2439, respectively) compared with those without any supplemental coverage. Beneficiaries with income between 100% and 200% of the FPL experienced $1127 higher spending compared with those living below the FPL. With respect to demographic characteristics, OOP spending declined with increasing age and educational attainment, but effects associated with demographic characteristics including sex, race, or marital status were not found. Beneficiaries who sought care “as soon as [they] feel bad” faced $1283 higher OOP spending. The number of chronic health conditions was found to be positively associated with OOP spending; compared with those with ≤ 1 chronic health conditions, beneficiaries with ≥ 9 conditions faced an OOP spending increment of $3275. Receipt of radiation and antineoplastic therapy were associated with higher OOP spending ($1526 and $1470, respectively; P < .01). OOP spending levels ranged between $1048 and $819 lower between 1997 and 2003. Spending did not differ for 2006 through 2007 (after implementation of Medicare Part D) compared with the reference period of 2004 to 2005. In sensitivity analyses, we limited the dependent variable to OOP spending associated with prescription medications, but again failed to find effects associated with Medicare Part D implementation on spending for the cancer cohort (data not shown).
|Total OOP Spending (Marginal Effects)||OOP/Income >20% (Marginal Probability)|
|Cancer site (reference: breast)|
|Supplemental insurance (reference: none)|
|Private, ESI with Rx||−1075a||−.120a|
|Private, other with Rx||−1011b||−.058c|
|Public with Rx||−2439a||−.136a|
|Income as % FPL (reference: ≤100%)|
|Assets (reference: ≤$2500)|
|Age (reference: 65–69 y)|
|Sex (reference: man)|
|Race (reference: white, non-Hispanic)|
|Marital status (reference: unmarried)|
|Education (reference: no HS)|
|HCC count (reference: 0–1)|
|Functional status limitations (reference: 0–1)|
|Attitude regarding care seeking|
|“Usually go to doctor as soon as you feel bad”||1283a||.048b|
|Year (reference group: 2004–2005)|
Characteristics associated with very high OOP burdens relative to income revealed different patterns (Table 3). Beneficiaries with ESI or public insurance with prescription coverage or medical coverage only were less likely to meet this threshold (12.0, 13.6, and 5.9 percentage points less likely, respectively). Increasing asset levels were associated with a higher probability, whereas income relative to poverty > 200% was associated with a lower probability of meeting the 20% threshold. Increasing comorbidity and functional status limitations were associated with higher probabilities of very high OOP burden, with increases of 12.7 percentage and 22.4 percentage points noted among beneficiaries with 5 to 8 and ≥ 9 HCCs, respectively. Receiving radiation or antineoplastic therapy was found to be associated with increased probabilities of a very high OOP burden. There were no significant time trends noted.
The results of the current study indicate that OOP spending and the burden of OOP spending relative to income is substantially higher in Medicare beneficiaries with newly diagnosed cancer compared with beneficiaries without cancer. Unadjusted OOP spending was found to be 47% higher for beneficiaries with cancer compared with those without cancer, and the probability of experiencing a high OOP burden was 75% higher. The gap between the unadjusted and adjusted effects of having cancer suggests that beneficiary characteristics explain some of the difference. However, the presence of these differences does not mitigate the financial burden experienced by Medicare beneficiaries faced with cancer.
We found that OOP spending is associated with increasing income and comorbidity burden as well as receipt of outpatient cancer-directed therapies, such as radiation and antineoplastic therapy. It is especially interesting that we observed that both ESI and public medical insurance with prescription drug coverage were associated with reduced OOP spending levels, and that even supplemental medical insurance without prescription coverage appeared to protect beneficiaries against a high OOP burden. This finding suggests that a large percentage of OOP spending for beneficiaries with cancer is associated with treatments covered through the Medicare medical (Parts A and B) benefits.
The results of the current analysis fill an important gap in understanding OOP spending for Medicare beneficiaries with cancer. Several recent studies have used Surveillance, Epidemiology, and End Results tumor registry data, linked to Medicare claims, to examine spending for beneficiaries with cancer,2,24 and the percentage of spending not covered by Medicare.25 However, as is clear from the current analysis, supplemental insurance can reduce the net beneficiary liability substantially. To get a complete picture of the financial burden for Medicare beneficiaries with cancer, it is necessary to understand not only what Medicare does not pay, but what percentage remains uncovered by supplemental insurance. This analysis also fills an important gap by providing information concerning the role of income and assets. Researchers have noted that cancer patients report spending OOP exceeding their income.13,14 Higher assets were associated with higher OOP burden relative to income, whereas higher income levels were associated with a lower OOP burden. This pattern suggests that beneficiaries may be spending down assets to pay for their care, and that considering income alone as the denominator for a burden measure, without controls for assets, may be misleading. Ideally, we would be able to track asset changes in response to OOP spending, and potentially use a composite burden measure that includes both income flow and liquid assets that are readily available for spending; however, the data from the current study did not support this type of measure.
The current analysis also provides insight into the importance of patient attitudes with regard to spending. The results suggest that patients who readily seek out physician care are likely to experience higher OOP spending, and face a higher OOP burden. Although this relation may appear to be intuitive, it has been difficult to quantify the influence of patient attitudes and preferences in determining treatment and outcomes. These results confirm the important role of patient attitudes concerning care seeking, and the significance of future research to incorporate this information into population-based analyses of cancer treatment.
Our estimates of OOP burden for the Medicare population are substantially higher than the burden for nonelderly persons with cancer. A recent study3 found that 13.4% of cancer patients aged 18 years to 64 years had a high OOP burden, also defined as OOP spending exceeding 20% of income. Our estimate of 27.6% is more than twice that percentage. Some of the difference in these estimates may be explained by the higher comorbidity noted among Medicare beneficiaries. However, the denominator for the OOP burden measure, annual income, drops substantially with retirement and Medicare enrollment so that a large percentage of the dramatic upward shift in OOP burden is likely explained on that basis.
It is important to note that this is not a structural analysis, in that many factors such as socioeconomic and underlying health status, as well as insurance availability and premiums, are likely to affect the decision to enroll in supplemental insurance. In turn, these choices will further affect both total and OOP spending.15 The current analysis reports on the net effects (the association between beneficiary characteristics, treatment, and high OOP spending) but does not disentangle the mechanisms through which OOP spending occurs. Although they may affect decisions related to supplemental insurance enrollment, we did not include OOP premiums for Medicare Part B or supplemental insurance in our estimate of OOP spending. Our rationale was that premiums paid are independent of actual total spending. To include them could dilute real differences in OOP spending related to beneficiary characteristics.
The current study is subject to limitations commonly associated with the use of survey and claims data. Our approaches toward identifying cancer patients and assigning cancer sites using claims data are subject to potential misclassification, which may reduce estimated differences between beneficiaries with and without cancer.26 The MCBS gleans information regarding payments by supplemental insurers and the remaining OOP burden from provider bills and insurer Explanation of Benefit forms. Because these data are assembled by the respondent, there is the potential for incomplete reporting. However, the MCBS has been in the field for nearly 20 years and remains a principal source of information concerning OOP spending for Medicare beneficiaries. Our ability to calculate a measure of OOP burden for married respondents was limited because of the lack of information regarding OOP health care spending for the spouse. To overcome this limitation, we attributed one-half of the married couple's income to the MCBS respondent and calculated an individual burden measure. Our results were sensitive to this approach. If we assumed instead that 100% of a couple's income was available to the MCBS respondent, then the percentage of beneficiaries with a very high OOP burden was reduced by approximately one-third. Because the true allocation of a couple's resources cannot be determined from the data, our preferred approach assumes that, on average, the MCBS respondent draws on approximately one-half of the resources.
In conclusion, Medicare beneficiaries with cancer face a higher OOP burden than their counterparts without cancer, and only a percentage of the difference was explained by a higher comorbidity burden. The financial pressures of the OOP burden associated with cancer treatment may discourage some Medicare beneficiaries with cancer from pursuing further treatment. The results of the current study suggest an increase in OOP spending over time. Given the increased use of expensive biologics and oral cancer therapies over the past decade, future research should extend this analysis to include more recent time periods.
Funding for this research was provided by a grant from the American Cancer Society (RSGI-1 0-1 09-0 1-CPHPS). Additional support was provided by GlaxoSmithKline.
CONFLICT OF INTEREST DISCLOSURES
The authors made no disclosures.
- 4The USA Today/Kaiser Family Foundation/Harvard School of Public Health. National Survey of Households Affected by Cancer. http://www.kff.org/kaiserpolls/upload/7591.pdf. Accessed September 25, 2012.
- 5Ries LAG, Melbert D, Krapcho M, et al, eds. SEER Cancer Statistics Review, 1975-2005. Bethesda, MD: National Cancer Institute; Year. seer.cancer.gov/csr/1975_2005/. Based on November 2007 SEER data submission, posted to the SEER Web site, 2008. Access date.
- 6American Cancer Society. ACS Cancer Facts and Figures Supplemental Data 2008. http://www.cancer.org/acs/groups/content/@nho/documents/document/2008cafffinalsecuredpdf.pdf Accessed September 25, 2012.
- 7Kaiser Family Foundation. Medicare Chartbook. 4th ed. Menlo Park, CA: The Henry J. Kaiser Family Foundation; 2012.www.kff.org/medicare/upload/8103.pdf. Accessed May 28, 2010..
- 8The Burden of Out-of-Pocket Health Spending Among Older Versus Younger Adults: Analysis from the Consumer Expenditure Survey, 1998-2003. Menlo Park, CA: The Henry J. Kaiser Family Foundation; Year. www.kff.org/medicare/7686.cfm. Accessed October 12, 2008., , , .
- 9The Henry J. Kaiser Family Foundation. Medicare Chartbook. 3rd ed. Menlo Park, CA: The Henry J. Kaiser Family Foundation; 2005. www.kff.org/medicare/upload/Medicare-Chart-Book-3rd-Edition-Summer-2005-Report.pdf. Accessed May 24, 2012.
- 21Derivation of malignancy status from ICD-9 codes. AMIA Annu Symp Proc. 2003: 1050., , , et al.
- 23Eligibility and take-up of the Medicare part D low income subsidy. Inquiry In press., , , , .