PREDICTING HOSPITAL READMISSION SIN THE MEDICARE POPULATION
Avoidable hospital readmissions cost taxpayers billions of dollars each year. The Medicare Payment Advisory Commission has estimated that almost $12 billion is spent annually by Medicare on potentially preventable readmissions within 30 days of a patient’s discharge from a hospital [1]. The Medicare program has begun to apply financial penalties to hospitals that have excessive risk-adjusted readmission rates. There is much interest in the health policy and medical communities in the ability to accurately predict which patients are at high risk of being readmitted. Not only are there strong financial reasons to avoid readmissions, readmission to the hospital can be a sign of poor clinical care and can indicate a worsening of a patient’s condition [2]. If doctors and nurses were aware of which patients were at highest risk, they could focus their efforts on these patients and could improve coordination of care with post-acute providers and family physicians. There has been some interest in this problem in the machine learning community as well. The Heritage Health Competition was a predictive modeling competition with the objective of predicting hospital readmissions, with a $3 million cash prize. However, the dataset used for that competition was highly de-identified and thus was missing much of the key information useful for predictions. It also had a low number of patients who were generally healthy1 . In this paper, I will apply machine learning methods to a dataset of Medicare claims to predict which patients are at a high risk of being readmitted to the hospital. I will then compare my results to the performance of risk adjustment models currently used by the Medicare program to predict readmissions.
Research Paper Link: Download Paper