This article describes research related to the design of a payment model for specialty oncology services for possible testing by the Center for Medicare and Medicaid Innovation at the Centers for Medicare & Medicaid Services (CMS). Cancer is a common and costly condition. Episode-based payment, which aims to create incentives for high-quality, low-cost care, has been identified as a promising alternative payment model for oncology care (Bach, Mirkin, and Luke, 2011; McClellan et al., 2013a). Episode-based payment systems can provide flexibility to health care providers to select among the most effective and efficient treatment alternatives, including activities that are not currently reimbursed under Medicare payment policies (Bach, Mirkin, and Luke, 2011). However, the model design also needs to ensure that high-quality care is delivered and that beneficial treatments are not withheld from patients.
CMS asked MITRE and RAND to conduct analyses to inform design decisions related to an episode-based oncology model for Medicare beneficiaries undergoing chemotherapy treatment for cancer. In particular, this report focuses on analyses of Medicare claims data related to the definition of the initiation of an episode of chemotherapy, patterns of spending during and surrounding episodes of chemotherapy, and attribution of episodes of chemotherapy to physician practices. Chemotherapy and its administration accounts for approximately 20 percent of Medicare spending on oncology care (McClellan et al., 2013a).
The claims data analyses in this report provide one source of information for consideration in payment model design. Other sources of information provide important complementary insights. Claims data are useful for understanding patterns of utilization, but they include limited information about the clinical context or appropriateness of care. The oncology clinical evidence and associated practice guidelines provide this type of complementary information. Several other reports from this project also provide complementary sources of information related to model design. A previous report summarized a comprehensive environmental scan of oncology care payment reform options (McClellan et al., 2013a). The Brookings Institution and MITRE convened a technical expert panel to discuss these reform options and provide input on how to best design payment and delivery reform models (McClellan et al., 2013b).
Based on evidence from the environmental scan and feedback from both the stakeholder interviews and the technical expert panel, CMS chose to move forward with developing an episode-based oncology model that incorporates care within oncology practices. The model would be designed for testing in the traditional Medicare fee-for-service (FFS) program (Parts A and B). The details of the model have yet to be determined. The analyses in this report are intended to support decisionmaking related to model design.
The study population included Medicare beneficiaries receiving chemotherapy treatment for cancer. The primary study sample was drawn from a 100-percent sample of national Medicare FFS claims files. A supplementary sample was drawn from Surveillance, Epidemiology, and End Results-Medicare (SEER-Medicare) data files, which include information on cancer stage and diagnosis date, in addition to claims information.
Beneficiaries initiating chemotherapy treatment in 2010 were identified using 2009 and 2010 chemotherapy drug claims. Chemotherapy drugs were classified into two categories: “likely chemotherapy,” including drugs that are nearly always used to treat cancer, and “possible chemotherapy,” which includes chemotherapies with other clinical indications—for example, treatment of autoimmune diseases. Chemotherapy and related drugs can be physician-administered or self-administered. Depending on the drug, formulation, and Medicare payment provisions, claims for these drugs can appear in Carrier, Outpatient, Part D, or Durable Medical Equipment (DME) claims, all of which were included in these analyses. In order to describe patterns of health spending, we classified all Medicare claims for services provided to the study population using procedure codes and other information on the claims.
Definition of an Episode of Chemotherapy
We used claims data analysis to focus on two key parameters in model design for episodes of care for patients receiving chemotherapy treatment for cancer: episode initiation and termination. The date of first diagnosis of cancer, the date of first chemotherapy, and the dates of visits with physicians related to oncology care could all potentially be used to identify the initiation of an episode of care. We used SEER-Medicare data to measure the time between the first diagnosis of cancer and initiation of chemotherapy for all patients in the sample initiating chemotherapy in 2003–2009. We found that the time between the primary cancer diagnosis and chemotherapy initiation varied widely across patients, ranging from one day to over seven years, with a median of 2.4 months. A substantial number (16.2 percent) of beneficiaries with a primary cancer diagnosis on a claim did not have a claim for an ambulatory care visit for a cancer diagnosis or a claim for a chemotherapy drug (specifically, claims for “possible” or “likely” chemotherapy drugs, either in the Carrier, Part D, Outpatient, or DME claim files), an unexpected finding. Almost all patients with chemotherapy drug claims also had physician visits for cancer diagnoses within 14 days.
In subsequent analyses, we used the first chemotherapy drug claim in 2010 (with no prior claims within six months) as the marker of the initiation of an episode of care. We examined the types of chemotherapy drugs that marked episode initiation. Chemotherapy initiation was mostly concentrated within several types of chemotherapy drugs, but those drugs varied by type of cancer. For most types of cancer, a substantial proportion of initiations were for drugs classified as “possible chemotherapy.” For example, for breast cancer, 18 percent of episodes initiated with tamoxifen citrate. For lymphoma, 47 percent of episodes initiated with rituximab. A chemotherapy payment model will need to include provisions to ensure that patients receiving “possible chemotherapy” drugs are actually cancer patients that are eligible for the model.
Considering the termination of an episode, the simplest approach would be a fixed period of time following initiation. However, several events that occur frequently during episodes could also potentially be considered as markers to signal the end of an episode. We examined several potential markers of this type, including a period of time without chemotherapy utilization (as indicated by claims for either chemotherapy drugs or administration), mortality, hospice, and loss of Medicare Part A and Part B enrollment (e.g., due to enrollment in Medicare Advantage).
We found that there is considerable variability in the length of time beneficiaries receive chemotherapy treatment. The length of uninterrupted chemotherapy treatment was greatest for people with breast cancer: The mean time from chemotherapy initiation to a two-month period with no chemotherapy treatment was five months (standard deviation [SD], five months). The length of uninterrupted chemotherapy treatment was shortest for people with pancreatic cancer: The mean time from chemotherapy initiation to a two-month period with no chemotherapy treatment was four months (standard deviation [SD], four months).
We found that repeating periods of chemotherapy were common. For example, among patients with chemotherapy treatment for breast cancer (excluding decedents), 72 percent had a gap in chemotherapy with duration g of at least one month that was followed by reinitiation of chemotherapy. These repeating periods could be part of a planned therapeutic strategy and considered as continuations of the same episode of care, or they could reflect new courses of treatment that could be considered new episodes of care. It is impossible to distinguish between these scenarios using claims data analysis. The payment model could be designed so that each repeating period is treated as a separate episode, or it could be designed to recognize longer periods of treatment, including active chemotherapy and gaps of several months, which would lead to substantially longer periods of eligibility for the model.
As expected, we found that mortality occurs frequently among Medicare beneficiaries receiving chemotherapy treatment. Mortality within 12 months from the first date of chemotherapy ranged from 7 percent of patients (breast cancer) to 62 percent of patients (pancreatic cancer). These findings indicate that many patients will die during an episode of chemotherapy that makes them eligible for the payment model. Hospice use was also common among Medicare beneficiaries with cancer; the rate of hospice participation ranged from 5 percent (breast and prostate cancer) to 49 percent (pancreatic cancer). Disenrollment from Medicare Part A or B in the time period following chemotherapy initiation was very rare; 2–4 percent of the study population disenrolled for at least one month.
We found that Medicare payments escalate sharply in the four months prior to chemotherapy initiation, which may reflect diagnosis and treatment planning, and then peak in the first month of chemotherapy. Monthly spending falls at varying rates in the first six months after chemotherapy initiation and is relatively flat between months eight and 18 after chemotherapy initiation. We found that 16 to 25 percent of total spending in the 24-month window around chemotherapy initiation that we examined occurs in the six months prior to chemotherapy, representing the work-up prior to initiation (i.e., laboratory, imaging, and Evaluation and Management [E&M] payments) and hospitalizations. Between 52 and 71 percent of total spending occurs through the first six months of chemotherapy, and between 77 and 90 percent of spending occurs within the first year of chemotherapy.
The average level of total monthly payments varied considerably across cancers, with the highest spending peak of $9,972 for lymphoma, and peaks of $3,109 for breast cancer and $2,135 for prostate cancer. Monthly Medicare spending for beneficiaries with cancer who receive chemotherapy was substantial. The service categories forming the components of cancer care were numerous, and their relative importance varied across cancer types. Chemotherapy drug and administration (10 to 31 percent of total spending) and inpatient care (25 to 45 percent of total spending), however, were nearly always important contributors to overall spending and may represent important opportunities for improving the efficiency and coordination of care.
We also found that for each type of cancer, there exists great variation in the total costs of care across patients, with the top quartile of patients incurring substantially higher costs than the bottom three quartiles. In principle, this variation could reflect variability in the clinical severity of patients. In a subanalysis, however, we examined quartiles of total spending within patients with each type of cancer and the stage of cancer at diagnosis. The variability across quartiles within a stage and site is comparable to the variation in spending across quartiles not accounting for stage at diagnosis. While there may be other clinical characteristics driving service utilization, this evidence may suggest that differences in treatment patterns independent of severity contribute to variation in utilization and spending.
Attribution of Episodes to Practices
Attribution methodologies are needed to associate patients undergoing episodes of chemotherapy treatment with physician practices that could participate in a payment model. We explored two alternative approaches for attributing chemotherapy episodes to practices: a rule that attributed the episode to the practice responsible for the plurality of cancer-related visits for evaluation and management services, which entails a retrospective approach, and a prospective attribution rule that attributed episodes to the practice responsible for the trigger chemotherapy claim (i.e., the claim that is used to identify the initiation of the chemotherapy treatment episode).
The two rules produced the same results (i.e., attributed the episode to the same practice) for 78.7 percent of episodes. The prospective rule successfully attributed a higher percentage of episodes (99.6 percent vs. 97.4 percent). We then examined the percentage of episodes attributed to clinicians who were most likely to be characterized as “chemotherapy providers” by virtue of submitting Carrier or DME claims for chemotherapy. Using the results of our clinician attribution (which was conducted in parallel with our practice-level attribution), we found that a total of 87.0 percent of episodes were attributed to clinicians who provide physician-administered chemotherapy using the prospective rule, as compared with 80.4 percent of these episodes attributed to clinicians under the plurality rule.
Finally, we measured the percentage of payments per episode made to the attributed practice across a range of different payment categories. We focused on the sample of episodes that produced discordant attribution results. We found that the prospective attribution rule was more likely to attribute the episode to the practice that was primarily responsible for chemotherapy spending. In two other payment categories (all outpatient and total payments), the practice attributed under the prospective rule was responsible for a higher proportion of payments.
Characteristics of Practices with Attributed Oncology Episodes
Using the results from the prospective attribution method, we derived summaries of the practices that were attributed at least one episode. We found that the distribution of episodes per practice is extremely skewed, with 62.1 percent of practices attributed only one or two episodes. A payment model that used a minimum annual episode volume of ten episodes would reduce the sample of practices participating in the program by 79 percent in our study population. However, because chemotherapy episodes are clustered primarily in high-volume practices, the ten-episode minimum entails a loss of only 9 percent of episodes from the analysis. Practices with at least ten attributed episodes had a mean of 3.0 affiliated oncologists and 6.5 total physicians with at least one attributed episode. The number of episodes attributed to each practice and the number of physicians and oncologists with attributed episodes per practice each increase as the overall practice volume minimum is raised. The low volume of episodes attributed to many practices will likely present a challenge to the measurement of performance for episodes of care.
We also examined mean payments, by type of cancer, for the subset of practices that were attributed 40 or more episodes (we selected a cutoff of 40 episodes for these analyses to provide a profile of the practices most likely to participate in the payment model). Mean 12-month episode costs ranged from $58,392 for episodes of leukemia to $19,591 for prostate cancer. Six-month episode costs followed similar patterns.
The results of this study provide one source of information for consideration in the design of an oncology payment model. The analyses in this article describe the initiation and termination of episodes of chemotherapy, spending patterns for patients initiating chemotherapy, and the results of claims-based methods for attributing chemotherapy patients to oncology practices. In future analyses, we will simulate the potential effects of oncology payment models and identify key design considerations.
Bach, P. B., J. N. Mirkin, and J. J. Luke, “Episode-Based Payment for Cancer Care: A Proposed Pilot for Medicare,” Health Affairs, Vol. 30, No. 3, 2011, pp. 500–509.
McClellan, M., Kavita Patel, John O'Shea, Jeffrey Nadel, Andrea Thoumi, and Judith Tobin, Specialty Payment Model Opportunities Assessment and Design: Environmental Scan for Oncology, Washington, D.C.: The Brookings Institution, 2013a. As of August 20, 2014:
McClellan, M., Kavita Patel, John O'Shea, Jeffrey Nadel, Andrea Thoumi, and Judith Tobin, Specialty Payment Model Opportunities Assessment and Design: Summary of the Technical Expert Panel for Oncology, Washington, D.C.: The Brookings Institution, 2013b. As of August 20, 2014:
The research addressed in this article was conducted in RAND Health, a division of the RAND Corporation, under a subcontract to MITRE.