To encourage hospitals to improve quality of care for older adults, Medicare penalizes hospitals with higher than expected 30-day readmissions under the Hospital Readmissions Reduction Program (HRRP). In 2017, HRRP penalized 80% of hospitals an average of $205,000 each for a total of $526 million and 89% of hospitals serving large proportions of low-income patients, compared to 80% of all hospitals, had reduced payments under HRRP. Critics of the HRRP contend that the penalties are unfair because, rather than only reflecting low-quality care, they also reflect unmeasured social and medical factors. Such factors can result in greater penalties for safety-net hospitals caring for socially vulnerable patients, such as minorities and low-income patients.
However, research has been unable to convincingly distinguish whether hospitals are penalized because they have lower quality, or because their patients are unobservably sicker. We will address this problem using big clinical data from Michigan Surgical Quality Collaborative on over 500,000 surgical episodes between 2008 and 2018. These data contain rich information about patient disposition and risk, information that is far more detailed than that contained in health care claims data (used for risk-adjustment under the HRRP). We will use deep-learning machine learning strategies to develop a “gold standard” risk-prediction model using health care claims, and claims enriched with the the MSQC data. We will then evaluate the effects of social risk on within- and between-hospital differences in risk-adjusted hospital readmission rates.
This analysis will yield disruptive insights about whether social factors are proxy for unobserved clinical risk or for hospitals with poorer clinical quality. The project will combine unique perspectives about surgical outcomes (Justin Dimick), social risk (Geoff Hoffman), and financial incentives (Andrew Ryan) to improve clinical care and health care policy.