Examination of Readmissions after Cardiac Surgery in Pennsylvania: Development of Risk Models with Clinical Relevance

University Of Pennsylvania, Philadelphia PA

Investigators

Abstract

The Centers for Medicare and Medicaid Services (CMS) initiated policies meant to improve this period of care in an effort to encourage health care providers to engage in better coordination of care. In 2012, CMS began to assess penalties from ?excess? rehospitalization for selected diagnoses, including pneumonia, heart failure, and acute myocardial infarction. The financial incentive spurred hospitals to identify patients at risk and to devise strategies to reduce readmissions. Readmissions decreased.2 Given the program's apparent success, the list of admission diagnoses was expanded to include total hip arthroplasty, total knee arthroplasty, and chronic obstructive pulmonary disease (COPD) in 2015, and readmission after coronary artery bypass grafting was added beginning in 2017. The largest, most reliable studies of rehospitalization after cardiac surgery describe a rate of thirty day readmission of 13-19%.3-5 281,000 cardiac surgeries were reported to the Society of Thoracic Surgeons Adult Cardiac Surgery Database last year.6 A conservative estimate of the number of readmissions after heart surgery is therefore approximately 40,000. Multiple publications have developed logistic regression models designed to identify patients who are at increased risk of readmission, but these models have had disappointing predictive value, with c-statistics that range from 0.6 to 0.65.3-5, 8 Using data from the Pennsylvania Health Care Cost Containment Council, we seek to improve upon these models. The most common causes of readmission after heart surgery are heart failure, arrhythmia and infection. A recent publication from a consortium of academic centers described that these three diagnoses represented about half of all readmissions within 30 days.7 Identifying patients who are at high risk for cause-specific readmissions would facilitate tailored interventions to prevent this unwanted outcome. We seek to develop models that predict cause- specific readmissions and to define high-risk populations in whom targeted strategies to prevent readmission can be efficiently implemented. In Aim 1, we will test the hypothesis that a diagnosis-specific multivariable analysis will yield models that are more predictive and can better define a high risk population. Development of cause-specific readmission models would be a step forward in confronting this problem. These tools would aid hospitals and healthcare systems in implementing strategies to improve rates of readmission.