Unintended Consequences: Empirical Studies of Continuity of Care and Financial Incentive Gaming in Primary Care

Abstract

Strengthening primary care is a priority for health systems globally. Policies to modernize primary care can however undermine valuable aspects of incumbent delivery models. This dissertation uses data from primary care health records in England to study disruption of continuity of care and gaming responses to financial incentives as two important unintended consequences of recent primary care reforms. Chapter 1 employs a prospective cohort study design to examine the relationship between continuity of care with a primary care physician and the risks of death and emergency hospitalization for older, medically complex patients. We find better relational continuity predicts lower mortality and emergency hospitalization risk, suggesting payers should consider incentivising the provision of good doctor-patient continuity for older, medically complex patients. Chapter 2 uses multilevel models to explore longitudinal trends in continuity of care over a decade. We find continuity fell for all patient groups but fastest for older and medically more complex patients. We additionally find larger practices tend to provide poorer continuity of care, with substantial variation between providers in the rates of decline of continuity. These findings suggest consolidation of primary care providers into larger organizations may have negative consequences for continuity and point to the possibility of provider-level factors that may be targets for continuity improvement efforts. Chapter 3 proposes two novel methods for identifying gaming by clinicians exposed to pay-for-performance incentives. To identify gaming in the management of hypertension, we develop a method for estimating provider rounding behavior when recording blood pressure readings in medical records. We then look for evidence of differential changes in rounding behavior around payment thresholds following incentivization. We also exploit patterns in near threshold retesting for indicators based on blood test results for diabetes and cholesterol control to identify whether providers increase intensity of testing to achieve results within payment thresholds. We find evidence of changes in blood pressure rounding consistent with gaming behavior, but no such changes for blood-test-dependent targets. These findings suggest it is possible to identify gaming behavior in patient clinical care to characterize gaming tendency at the level of individual providers.