Avalere assessed the accuracy of the prediction model used for case-mix adjustment for each of the 17 cancer types in the radiation oncology model. For prostate cancer, actual Medicare FFS payments are substantially higher, on average, than estimated by the prediction model: 41% higher for professional component (PC) payments and 32% higher for technical component (TC) payments (Figure 1).

Background

On July 10, the Centers for Medicare & Medicaid Services (CMS) proposed a mandatory alternative payment model for radiation oncology that would test the effects of making prospective, episode-based payments for radiation therapy (RT) services over a 90-day episode of care for 17 cancer types. Participants would include physician group practices (PGPs), freestanding RT centers, and hospital outpatient departments (HOPDs). The model would replace Medicare fee-for-service (FFS) payments with capitated payments via a PC and a TC. The model would run for 5 years and begin either January 1 or April 1, 2020.

The capitated payments are determined by national base rates for each cancer type (trended over time), a case-mix adjustment for patient risk factors (e.g., cancer type, age, sex), and a historical adjustment to account for any remaining differences between these combined amounts (base rate, case mix adjustment) and a participant’s actual per episode FFS payments during the 2015–2017 base period. National base rates reflect average Medicare FFS payments per episode during the base period. The case-mix adjustment uses prediction model coefficients to predict the influence of patient risk factors on FFS payments per episode. The historical adjustment, which can be positive or negative, accounts for the remaining difference between case-mix adjusted base rates and a participant’s actual per episode FFS payments during the base period. Participants with historically higher per-episode FFS payments compared to national base rates, even after adjusting for patient risk, have a portion of their base period per-episode FFS payments identified as inefficient.

Findings from the Analysis

Overall, CMS’ prediction model accurately predicts episode costs for most cancer types. For prostate cancer, however, actual Medicare FFS payments are higher, on average, than estimated by the prediction model. The main factor driving the inaccurate prediction of episode costs for prostate cancer is CMS’ decision to only use episodes in which radiation is delivered in a hospital outpatient setting. Almost half of prostate cancer episodes have radiation treatment in a freestanding center; these episodes are excluded when estimating the effect of patient risk factors. If the effects of patient risk factors are estimated using all episodes, regardless of the care setting for radiation delivery, then the accuracy of the prediction model improves, most substantially for prostate cancer episodes.

Because CMS’ prediction model is based solely on episodes that occur in the hospital outpatient setting, CMS will identify a high portion of prostate cancer care as inefficient costs. Participants’ inefficient historical costs are weaned out of their capitated payments over time, reduced by 10% in performance year 1 to 30% in performance year 5. Accordingly, under CMS’ current method, a greater number of participants treating prostate cancer outside a hospital outpatient setting will be identified as inefficient and receive lower capitated payments over time, compared to an alternative method that estimates patient risk factors for radiation treatment based on episodes occurring in either a hospital outpatient setting or freestanding center.

Note: If the coefficient is closer to 1, then CMS is, on average, more accurately predicting episode payments.

Methodology

Avalere performed this analysis using a combination of its access to 100% of Medicare FFS claims, under a CMS research data use agreement, and the publicly available radiation oncology episode file published by CMS alongside the preliminary rule for the radiation oncology model.  Using episodes in which radiation was delivered in a hospital outpatient setting, Avalere replicated the ordinary least squares (OLS) regressions described in the preliminary rule to estimate the coefficients for the patient risk factors included in prediction model during the 2015-2017 base period. Avalere ran separate OLS regressions for both the professional component and technical component.  In addition, Avalere reran the OLS regressions using all episodes.

Funding for this research was provided by American Society for Radiation Oncology.

To receive Avalere updates, connect with us.