Upcoding of active cancer diagnosis in Medicare Advantage and traditional Medicare.

Abstract

229 Background: Medicare Advantage (MA) plans have been known to engage in more intensive diagnostic coding practices to maximize risk-adjusted capitated payments. Regulators have deemed cancer a high-risk diagnosis susceptible to miscoding, as the challenges in distinguishing current vs. historical malignancies may allow MA plans to “upcode” patients in remission as active cancer cases, thereby increasing plan profits. Despite this concern, no peer-reviewed studies have examined cancer coding behaviors in a representative sample of MA enrollees. Methods: Using 2015-2019 SEER-linked Medicare data, we compared longitudinal trajectories of diagnosis coding between MA and traditional Medicare (TM), encompassing the year of diagnosis (Y) and the fourth year after diagnosis (Y+4). We included aged Medicare beneficiaries diagnosed with early-stage colorectal cancer, a cancer with relatively high survival and long remission periods that may be conducive to upcoding. In each beneficiary-year, we assessed the probability of having any active diagnosis and a diagnosis supported by claims for cancer-directed surgical, systemic, or radiation treatments within a 6-month window. We estimated linear probability models, controlling for age at diagnosis, cancer type (colon vs. rectum), cancer stage, race and ethnicity, dual enrollment in Medicaid, Yost index, and year of diagnosis, and county and calendar year fixed effects. Given concerns about missing MA encounter data, we restricted to MA plans with more complete records (as recommended in the literature). Results: We included 119,284 beneficiary-year observations (TM = 67.1%, MA = 32.9%). From Y to Y+4, the rate of any active diagnosis decreased from 94.9% to 57.6% in MA and 90.1% to 58.6% in TM. However, the rate of supported coding was substantially lower: at Y+4, only 5.2% and 9.3% of beneficiaries in MA and TM, respectively, had evidence of treatment. When averaged across all periods, MA beneficiaries had a 1.97 percentage point (pp; 95%CI:1.35, 2.59; P < 0.001) greater probability of active diagnosis coding, but a 3.59 pp (95%CI:-4.13, -3.05; P < 0.001) lower probability of supported coding (representing a 9% difference relative to the mean in TM, 38.9%). These differences tended to be concentrated in earlier survivorship periods. Among claims with unsupported diagnosis coding, evaluation and management visits were the most common procedure codes (64% in MA and 54% in TM). Conclusions: Higher levels of active diagnosis coding, combined with lower levels of observed treatment for cancer-directed care in MA, suggest excess payments to MA plans for patients with a history of colorectal cancer. Reforms to MA risk adjustment that account for the provision of cancer care services, as well as efforts to improve the accuracy of coding (in both TM and MA), are needed to calibrate MA payments for cancer care delivery.