Journal of the National Cancer Institute Advance Access originally published online on April 29, 2008
JNCI Journal of the National Cancer Institute 2008 100(9):607-610; doi:10.1093/jnci/djn132
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Published by Oxford University Press 2008.
EDITORIALS |
Estimating the Cost of Cancer Care in the United States: A Work Very Much in Progress
Affiliation of author: Department of Health Policy and Management, Rollins School of Public Health and Winship Cancer Institute, Emory University, Atlanta, GA
Correspondence to: Joseph Lipscomb, PhD, Department of Health Policy and Management, Rollins School of Public Health, Emory University, Room 642, 1518 Clifton Road North East, Atlanta, GA 30322 (e-mail: jlipsco{at}sph.emory.edu).
There are many important uses for conceptually sound, empirically strong estimates of cancer care costs. Such uses include assessment of the aggregate economic burden of cancer at the national, state, or local level; economic evaluation of cancer interventions by such approaches as cost-effectiveness analysis; and delineation of the cost of specific cancer interventions at the individual patient level to support informed patient and physician decision making (1).
There are also many viable ways to produce such cancer cost estimates (2). Analysts must decide on the scope of the analysis (eg, national or patient level); its cancer focus (eg, one or several tumor sites or all cancers); data sources (eg, cancer registries, insurance claims, patient surveys, or medical records); perspective regarding who bears the cost burden (eg, the patient, third-party payer, or society at large); cost flow time frame (eg, prevalence costs in a specific year for all cancer patients alive in that year, incidence costs for all patients newly diagnosed with cancer in that year and followed over time, or both); and various methodological issues (eg, how to arrive at the net cancer-attributable cost of cancer care).
As a consequence, expert analysts investigating the same general issue often use different methods and do not always arrive at the same conclusion. For example, cancer care has an enormous cost burden at the national level, but exactly how much? The National Cancer Institute (NCI) estimated that the total direct medical costs associated with cancer diagnosis and treatment in the United States in 2004 was $72.1 billion (3). This analysis relied substantially on cancer registry data from the Surveillance, Epidemiology, and End Results (SEER) program and patient enrollment and claims data from the Medicare program. In contrast, Thorpe et al. (4) used different data sources (primarily individual self-reports of cancer diagnoses and health-care utilization extracted from federally sponsored national surveys) and reported findings for the year 2000 that suggest (after adjusting for inflation) that US cancer care expenditures for 2004 were about $46 billion.
This example is but one of many in cancer care costing in which different experts have chosen different data sources and methods to evaluate a specific issue and have ended up with different results. In fact, such variations in approach to cost analysis are not necessarily a bad thing. Different arenas of application (eg, national economic burden estimates vs cost-effectiveness analysis) require somewhat different costing approaches. Moreover, as yet there is no consensus on the best way to proceed, whatever the arena of application.
Still, it is well to ask whether there are specific approaches to cancer care cost analysis with a sufficiently strong conceptual and empirical base that (although yet incomplete and imperfect) indicate an appropriate approach for the development of mutually consistent cost estimates for a variety of purposes. In this regard, the important paper by Yabroff et al. (5) appearing in this issue of the Journal is of particular interest. It is the latest in a series of contributions from the NCI's Health Services and Economics Branch and Statistical Research and Applications Branch. The authors use SEER–Medicare data and a strategic set of modeling assumptions to produce estimates of the net incidence cost of cancer in the United States for the projected number of individuals aged 65 years or older who were diagnosed in 2004 with any one of the 18 most prevalent cancers (or with a cancer falling into a residual, all-other category). This study appears to be the first to integrate up-to-date estimates of cost of care and survival to derive long-term (in this case, 5 years) national-level estimates of incidence cost for a broad array of the most common types of cancers (albeit for the elderly only).
The Yabroff et al. (5) analyses proceed in the following several steps.
- Their analyses include data from all cancer patients in SEER registry areas diagnosed during the period of 1973–2002 who were aged 65 years or older (and thus eligible for Medicare) for some portion of the interval from 1999 through 2003—the study's observation period for computation of cancer care costs—and who also met certain inclusion criteria. Because complete Medicare encounter and insurance claims data are available only for enrollees who have fee-for-service coverage that includes both Medicare Part A and Part B, the authors had to exclude approximately 20% of potentially eligible patients who were in Medicare-managed care during the study period of 1999–2003 or who did not have both Part A and Part B coverage at any time during this period. Moreover, for patients who had fee-for-service coverage and who were covered by both Part A and Part B for only a portion of the time during the study period, their months in managed care and/or without Part B coverage were excluded from the analytical dataset.
- To set the stage for computing the net, cancer-attributable cost of the medical care received by these cancer patients, each patient was matched with up to five control subjects who were drawn from a 5% random sample of Medicare beneficiaries who, during the study period of 1999–2003, were aged 65 years or older and had no SEER-recorded cancer diagnosis. Control subjects were matched to cancer patients on sex, 5-year age interval (eg, 65–69 or 70–74 years), and SEER registry.
- Costs were estimated by the phase-of-care approach, as previously described in cancer costing studies; see, for example, Brown et al. (6) and Taplin et al. (7). Basically, the care for each patient is divided into an initial phase (the first 12 months after diagnosis), a last-year-of-life phase (the final 12 months before death), and the continuing phase (all remaining months between the initial and last phases). The patient's total Medicare-covered costs, which are based not on charges but on what Medicare pays providers, are then distributed month by month across these phases, according to a set of decision rules. To take one arbitrary example, if a cancer patient was diagnosed on January 1, 1999, and died on December 31, 2003, the first 12 months of Medicare costs would be assigned to the initial phase, the final 12 months of costs to the last-year-of-life phase, and the remaining 36 months (from 2000 through 2002) of costs to the continuing phase.
- Next, each control subject is assigned a pseudodiagnosis date identical to that of the corresponding matched cancer patient. The control subject's observed total Medicare costs over the relevant portion of 1999–2003 are assigned month by month to the continuing phase or to the last-year-of-life phase according to a set of decision rules. For each cancer site and each phase of care, a mean cost per month is computed by dividing total observed costs by the total observed months (pooling across all relevant cancer patients). Likewise for the control subjects who are matched to the cancer patients for a particular cancer site, a mean cost per month is computed for their continuing-phase months and also for their last-year-of-life months.
- The net cost of cancer care (for a given cancer site and phase of care) is the difference between the applicable mean values for the case patients and control subjects. The decisions rules for computing net cost in the last-year-of-life phase distinguish further between whether the cancer patient dies from cancer (so that the relevant costs for control subjects would be computed on the basis of their continuing-phase months) or with cancer but from other causes (so that the relevant costs for control subjects would be based on their last-year-of-life months). Although this case–control approach for computing cancer-attributable costs has been refined over the years (6–8), Yabroff et al. (5) demonstrate how it can be readily applied to estimate the attributable costs for multiple cancer sites concurrently.
- To model the incidence costs for cancer patients from the selected year of diagnosis into the future, patient-level survival estimates are required. To obtain these estimates for each cancer site studied, the authors used the SEER database. They identified all patients who were aged 65 years or older and who were diagnosed with cancer sometime during the period of 1998–2004 and obtained the survival status of each patient for each month during this period. These data were pooled (by cancer site) to estimate the probability of death for each month after diagnosis, given survival up to that month, over the first 5 years after diagnosis.
- For each cancer site, these phase-specific mean cost estimates and monthly mortality probabilities were applied to the total number of cancer patients aged 65 years or older and projected nationwide for 2004 to estimate the aggregate net cost of care over the first 5 years after diagnosis. For each cancer type, the nationwide incidence count is derived by inflating the total number of cancer patients in 2004 reported across all 17 SEER registries by a factor of (1/0.26) because these registries now cover approximately 26% of the US population. Aggregate 5-year costs were reported as costs discounted to present value (the standard and correct approach) and also as undiscounted costs to promote transparency. All costs are stated in year 2004 dollars (with adjustments having been made for intertemporal price inflation effects and geographic variations in input prices).
From this framework, Yabroff et al. (5) present a host of interesting findings. Aggregate per-patient net cost of care for elderly cancer patients varied by phase of care (statistically significantly higher for initial and last-year-of-life phases than for the continuing phase); by stage at diagnosis (generally higher for patients with late-stage cancer); and by cancer site, with a rank ordering of sites that was generally different between men and women and also different by phase of care. That said, among the 18 cancer types examined, brain and other nervous system cancers were universally the most costly per patient for men in each phase of care and in total for the 5-year cost projection period; for women, brain and other nervous system cancers were the most expensive for the initial and last-year-of-life phase, but ovarian cancer was the most expensive overall. Melanoma of the skin had the lowest net cost per patient for men and women across phases and overall.
When all projected 2004 cancer patients were considered, the most costly cancers in aggregate (in discounted dollars) among women were lung ($2 billion), colorectal ($1.6 billion), and breast ($1.4 billion) cancers. Among men, the most costly cancers were prostate ($2.3 billion), lung ($2.2 billion), and colorectal ($1.5) cancers. For elderly men and women combined, the most costly cancers in the nation were lung and colorectal cancers. Few of these individual findings are startling; yet, taken together, they provide the scientifically strongest picture yet, of the incidence costs of cancer in aggregate and by tumor type for the elderly in the United States.
Moreover, the analytical framework—that is, the conceptual and empirical base—that supports the calculations reported in Yabroff et al. (5) can be adapted to obtain cancer cost estimates for the various arenas of application above. For example,
- To derive the cancer-attributable cost of care associated with incidence cancers from a lifetime, rather than 5-year, perspective for those aged 65 years or older would require the development and application of monthly mortality probabilities (more generally, survival curve analyses) that extend from the date of diagnosis through many years into the future—conceptually, a natural extension of the analyses discussed above.
- By use of the techniques applied by Yabroff et al. (9), the cancer-attributable cost of care associated with prevalence cases of cancer among the US elderly in any given year can be readily derived from SEER–Medicare data by use of the phase-of-care framework as described in Yabroff et al. (5). Projection of the prevalence costs of colorectal cancer over time, as described (9), indicates the feasibility and value of developing long-term (in that analysis, through the year 2020) estimates of cancer site–specific mortality and also the importance of sensitivity analyses to gauge the impact on projected costs of variations in cancer incidence, survival rates, and the real (inflation-adjusted) cost of cancer care over time.
- SEER–Medicare data are already being used in cost-effectiveness analyses of cancer interventions. A phase-of-care costing model may be adopted, as in a recent cost-effectiveness analysis of DNA stool testing to screen for colorectal cancer (10), or other assumptions may be imposed concerning the distribution of the flow of medical care costs over time, such as in a recent study (11) of the cost-effectiveness of switching postmenopausal patients with early-stage breast cancer from tamoxifen to a particular aromatase inhibitor. Either way, the costing framework in Yabroff et al. (5) would naturally facilitate the estimation of intervention-specific, patient-specific costs over time—precisely what cost-effectiveness analyses in cancer usually require.
- One can argue that basically the same type of intervention-specific projections of medical care costs required for cost-effectiveness analyses would yield data that could provide information for patient–provider discussions regarding the possible costs of various types of cancer care or treatment. Although there is little evidence that cancer cost projections have been put to this use, doing so would constitute an additional potential payoff from fully exploiting the analytical capability of detailed costing frameworks such as the one used by Yabroff et al. (5).
Finally, what would it take to make this strong, NCI-fostered costing framework stronger still? Work that is underway should continue, and new efforts should be considered to expand the empirical base for the monthly phase-of-care cost and survival estimates as follows: For the elderly, medical cost estimates are required for those enrolled in Medicare-managed care and those who are Medicare fee-for-service but not enrolled in Part B. For all of these individuals, cost estimates are needed for nursing home care, over-the-counter drugs, and other services not covered at all by Medicare. For the nonelderly (those younger than 65 years), direct estimates of the costs associated with cancer-related and non–cancer-related medical care should be obtained, so that it is no longer necessary to infer these costs by extrapolation from Medicare-based costs for the elderly. Yabroff et al. (5) clearly recognize this issue, although addressing it was far beyond the scope of their analyses. Nonelderly individuals may be covered by private health insurance indemnity plans, some form of managed care, Medicaid, or the Veterans Affairs health system, or they may be uninsured (and thus their data would not be in a claims database). There has apparently been no successful effort to date to bring this diverse set of public and private payers together (at, eg, the state level) to develop a near-population–based data system that could track the costs of care for nonelderly patients. Yabroff et al. (5) aptly point to the NCI-sponsored Health Maintenance Organization (HMO) Cancer Research Network as a potential source of cost observations for those under age 65, as well as those aged 65 years or older who are enrolled in Medicare-managed care [see also Ritzwoller et al. (12)]. Howard et al. (13) used the MedStat/Marketscan data from private insurance claims to estimate cancer-related costs for the nonelderly. In addition, nationally representative surveys that seek self-reports of cancer-related resource use from self-identified cancer patients can fill important data gaps and otherwise complement what can be learned through linked registry–claims data sources. Important sources of survey data that have already been applied successfully to cancer care cost determination include the federally sponsored National Medical Care Expenditure Survey and, in particular, the Medical Expenditure Panel Survey (4).
For all cancer patients (elderly and nonelderly), the direct nonmedical costs associated with cancer care should be estimated. In fact, Yabroff et al. (14) recently calculated the opportunity cost of the patient's time involved in cancer treatment (often termed patient time costs) and evaluated these costs for various cancer sites on a per-patient basis and a national basis. Much work remains, however, in estimating the cost associated with informal caregiving and other resource commitments not priced out in the marketplace. Likewise, for all cancer patients, accounting accurately for what is traditionally termed the indirect cost burden associated with cancer is an ongoing challenge. Included in such costs is the value of the lost productivity (from a societal perspective) or lost earnings (from a patient perspective) due to the excess morbidity or premature mortality attributable to cancer. For the year 2004, the indirect costs associated with cancer nationwide were estimated to be $118 billion (3)—an enormous figure that will no doubt be subject to ongoing appraisal as the NCI and others turn their sights on this comparatively understudied component of the total cancer burden.
REFERENCES
1. Lipscomb J, Donaldson M, Hiatt RA. Cancer outcomes research and the arenas of application. J Natl Cancer Inst Monogr (2004) 33(1):1–7.
2. Yabroff KR, Warren JL, Brown ML. Costs of cancer care in the USA: a descriptive review. Nat Clin Pract Oncol (2007) 4(11):643–656.[CrossRef][Web of Science][Medline]
3. National Cancer Institute. Cancer trends progress report—2007 update. http://progressreport.cancer.gov. Accessed March 21, 2008.
4. Thorpe KE, Florence CS, Joski P. Which medical conditions account for the rise in health care spending? Health Aff. (Web exclusive) (2004) W4-437–W4-445. http://content.healthaffairs.org/cgi/reprint/hlthaff.w4.437v1.pdf. Accessed March 23, 2008.
5. Yabroff KR, Lamont EB, Mariotto A, et al. Cost of care for elderly cancer patients in the United States. J Natl Cancer Inst (2008) 100(9):630–641.
6. Brown ML, Riley GF, Schussler N, Etzioni RD. Estimating health care costs related to cancer treatment from SEER-Medicare data. Med Care (2002) 40(8, suppl):104–117.
7. Taplin SH, Barlow W, Urban N, et al. Stage, age, comorbidity, and direct costs of colon, prostate, and breast cancer care. J Natl Cancer Inst (1995) 87(6):417–426.
8. Baker MS, Kessler LG, Urban N, et al. Estimating the treatment costs of breast and lung cancer. Med Care (1991) 29(1):40–49.[CrossRef][Web of Science][Medline]
9. Yabroff KR, Mariotto AB, Feuer E, Brown ML. Projections of the costs associated with colorectal cancer in the United States, 2000–2020 [published online ahead of print October 2, 2007]. Health Econ. www.interscience.wiley.com http://www3.interscience.wiley.com/cgi-bin/fulltext/116325138/PDFSTART. doi:10.1002/hec.1307.
10. Zauber AG, Lansdorp-Vogelaar I, Wilschut J, Knudsen AB, van Ballegooijen M, Kuntz KM. Report to the Agency for Healthcare Research and Quality and the Centers for Medicare & Medicaid Services from the Cancer Intervention and Surveillance Modeling Network (CISNET) for MISCAN and SimCRC Models. In: Cost-Effectiveness of DNA Stool Testing to Screen for Colorectal Cancer (2007) Rockville, MD: Agency for Healthcare Research and Quality. http://www.cms.hhs.gov/determinationprocess/downloads/id52TA.pdf. Accessed on March 22, 2008.
11. Thompson D, Taylor DCA, Montoya EL, Winer EP, Jones SE, Weinstein MC. Cost-effectiveness of switching to exemestane after 2 to 3 years of therapy with tamoxifen in postmenopausal women with early-stage breast cancer. Value Health (2007) 10(5):367–376.[CrossRef][Web of Science][Medline]
12. Ritzwoller DP, Goodman MJ, Maciosek MV, et al. Creating standard cost measures across integrated health care delivery systems. J Natl Cancer Inst Monogr (2005) 35(2):80–87.
13. Howard DH, Molinari N-A, Thorpe KE. National estimates of medical costs incurred by non-elderly cancer patients. Cancer (2004) 100:883–891.[CrossRef][Web of Science][Medline]
14. Yabroff KR, Davis WW, Lamont EB, et al. Patient time costs associated with cancer care. J Natl Cancer Inst (2007) 99(1):14–23.
Related Articles in JNCI
CiteULike 
Connotea 
Del.icio.us What's this?
J Natl Cancer Inst 2008 100: 630-641.
J Natl Cancer Inst 2008 100: 603.
This article has been cited by other articles:
|
|
 |
|
  B. Handy The Clinical Utility of Tumor Markers Lab Med, February 1, 2009; 40(2): 99 - 103. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||