Upcoding or Selection? Evidence from Medicare on Squishy Risk Adjustment

Diagnosis-based subsidies, also known as risk adjustment, are widely used in US health insurance
markets to deal with problems of adverse selection and cream-skimming. The widespread use of these
subsidies has generated broad policy, research, and popular interest in the idea of upcoding—the notion
that diagnosed medical conditions may reflect behaviors of health plans and providers to game the
payment system, rather than solely characteristics of patients. We introduce a model showing that
coding differences across health plans have important consequences for public finances and consumer
choices, whether or not such differences arise from gaming. We then develop and implement a novel
strategy for identifying coding differences across insurers in equilibrium in the presence of selection.
Empirically, we examine how coding intensity in Medicare differs between the traditional fee-for-service
option, in which coding incentives are weak, and Medicare Advantage, in which insurers receive diagnosis-based
subsidies. Our estimates imply that enrollees in private Medicare Advantage plans generate 6% to
16% higher diagnosis-based risk scores than the same enrollees would generate under fee-for-service
Medicare. Consistent with a principal-agent problem faced by insurers attempting to induce their providers
to upcode, we find that coding intensity increases with the level of vertical integration between insurers
and the physicians with whom they contract. Absent a coding inflation correction, our findings imply
excess public payments to Medicare Advantage plans of around $10 billion annually. This differential
subsidy also distorts consumers' choices toward private Medicare plans and away from fee-for-service
Medicare.