I-Corps: Software platform for predicting hospital patient re-admissions

Vanderbilt University, Nashville TN

Investigators

Abstract

The broader impact/commercial potential of this I-Corps project is the development of predictive modeling tools to help reduce preventable hospital readmissions. One in every eight patients discharged from acute care hospitals is readmitted within 30 days. Over a quarter of these readmissions could be avoided with appropriate and timely healthcare interventions. The Centers for Medicare and Medicaid Services estimate that they spent over $17 billion per year on avoidable readmissions in 2015. In addition, there are nearly 100 hospitals that are fined over $1M annually for having hospital readmission rates much higher than industry averages. Insurers, accountable care organizations, self-insured employers such as large hospital systems, and health plans seek to decrease preventable readmissions to provide high quality care while managing their medical loss ratios. The busiest hospitals, consistently operating near or over maximum bed capacity lose revenue from low acuity preventable readmissions that reduce the institution’s case-mix index. The goal for the proposed technology is to provide better care to patients while simultaneously lowering costs for hospitals and insurers alike. This I-Corps project is based on the development of a software platform that includes machine learning algorithms to predict hospital readmission risk and identify specific factors contributing most to that risk for individual patients. The proof-of-concept for this technology was built using data from approximately 80,000 patient encounters over two years at a major academic medical center. It outperformed widely used industry standards by approximately 40%. These algorithms incorporate both modifiable and unmodifiable risk factors including various social determinants of health and incorporate fairness criteria to ensure predictions don’t reinforce biases of societal structures. Since these contributing risk factors may vary widely from one population to the next, each healthcare system or insurer requires their own unique predictive model based on their data. The proposed next steps are to identify and prioritize customer needs for the application of this technology such as algorithm validation services for each customer’s patient population, electronic health record interoperability, and user interface design. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.