We present two papers motivated by rising concerns about healthcare quality and costs.
 
First, we introduce a new statistical decision-making algorithm to enable medical decision-makers to personalize treatment choices at the individual-level and thereby improve patient outcomes. We formulate this problem as a multi-armed bandit with high-dimensional covariates, and present a new efficient bandit algorithm based on the LASSO estimator. Our regret analysis establishes that our algorithm achieves near-optimal performance; the key step is proving a new oracle inequality that guarantees the convergence of the LASSO estimator despite the non-i.i.d. data induced by the bandit policy. We evaluate our algorithm on the real-world clinical problem of warfarin dosing. Our algorithm outperforms existing bandit methods as well as physicians to correctly dose a majority of patients.
 
Second, Medicare has initiated several pay-for-performance mechanisms to encourage medical decision-makers to adopt such data-driven methods for improving patient outcomes. However, these initiatives may be undermined if hospitals engage in upcoding, the practice of mis-reporting patient outcomes to increase reimbursement. We empirically estimate that over 11,000 hospital-acquired infections are upcoded a year, drawing into question the effectiveness and fairness of Medicare’s pay-for-performance initiatives. We make several policy recommendations based on our results.