Predicting and Preventing Hospital Readmissions with IBM SPSS
Hospital readmissions are estimated to cost over $41 billion annually. Readmissions, specifically when patients are readmitted within 30 days of discharge, are quality metrics that hospitals watch closely. Healthcare organizations seek to understand what conditions are bringing patients back to the hospital. Many readmissions could be prevented if hospitals can identify the factors that indicate readmission in the patient population.
The healthcare industry is facing financial pressure to reduce readmissions. The CMS (Centers for Medicare & Medicaid Services) has named thousands of hospitals that will lose up to 2% of their Medicare reimbursement as part of the Affordable Care Act program that aims to curb hospital readmission rates. In FY 2015 these penalties will rise up to 3% of Medicare reimbursements. These penalties are applied to all Medicare payments to the Hospitals. It is likely that other payers will follow suit. This means that hospitals must reduce readmissions to maximize reimbursement.
It is estimated that nearly one in every five Medicare patients returns to the hospital within 30 days of discharge (Rau, 2012). The national rate of hospital readmission is approximately 19 percent, but the rate of readmission varies throughout the country. This has large implications for hospitals and health systems, as readmissions are costly and often result in poor outcomes for patients.
Ironside Advanced Analytics and IBM SPSS
New capabilities enable hospitals to target patients at risk for readmissions. Electronic medical records allow predictive models to be built using real-time information for real-time scoring. Additionally, dashboards and visualizations provide this real-time information in mobile-friendly formats, which ensures that interaction between clinicians and patients can occur throughout the hospitalization and aftercare.
Multiple factors from a wide variety of sources may prove to be significant indicators for readmission. It is essential that both structured and unstructured data be available for analysis. IBM SPSS Modeler uses advanced data preparation techniques that make it possible to evaluate many potential indicators from a wide variety of sources. IBM SPSS Text Analytics, using Natural Language Processing, extracts new insights from unstructured text.
Patient populations vary widely, and no single patient model fits all populations. IBM SPSS Modeler is a powerful predictive analytics platform designed to bring predictive intelligence to decisions made by organizations. It provides a range of advanced algorithms, data manipulation, and automated preparation and modeling techniques to build predictive models that can be implemented to improve outcomes.
Hospital readmission models created with SPSS uncover hidden patterns in the data. Indicators of readmissions are recognized and patients at risk are identified. Patient risk scores and factors provide actionable information to clinicians and administrators at the point of impact. IBM Cognos Business Intelligence provides the interactive data visualizations to these readmission indicators and highlights the appropriate actions.
Predicting Hospital Readmissions for Congestive Heart Failure
Ironside, an IBM Business Partner headquartered in Lexington, Massachusetts, used IBM Analytics to predict Congestive Heart Failure (CHF) for a hospital in New York. The Ironside Advanced Analytics team created a predictive model with IBM SPSS Modeler based on information about 352 CHF-related admissions over the span of 200 days.
The types of data incorporated into the model were derived from multiple datasets and included:
- Demographic characteristics of the patient
- Characteristics related to the hospitalization, such as the admission source, admission type, and others.
- After care: Discharge disposition and Discharge type
- Chief complaint upon admission
- Length of hospitalization
- Medical procedures during hospitalization
- Medications administered during hospitalization
Using IBM SPSS, the Ironside Advanced Analytics team created a model to predict the probability that a hospitalized CHF patient will be readmitted within 30 days of discharge. The CHF model is a strong predictor of readmissions. The CHF model predicted that 43 patients would be readmitted, and 84% of those predicted patients were actually readmitted (36 out of 43). The CHF model also predicted that 309 patients would not be readmitted, and 92% of those predicted patients were not readmitted (284 out of 309).
The CHF model identified 15 significant predictors of CHF hospital readmissions. These significant predictors included discharge disposition, marital status, chief complaint, and seven “Groups of Medications and Procedures” based on a factor analysis of the different medicines and procedures administered to the patient.
IBM Cognos dashboards containing advanced visualizations were used to provide both a historical and future view of hospital readmissions to administrators. Clinicians were presented with patient scoring, including readmissions indicators sent to the clinicians’ mobile devices.
Summary and Next Steps
As hospitals endeavor to reduce readmissions and improve patient outcomes, Ironside’s Data Science & Advanced Analytics team can help identify those patients who are at risk of being readmitted. The leading indicators of hospital readmissions can then be brought to those hospital decision makers who are at the point of impact, providing a crucial set of markers that lead to effective preventative measures and lower costs. Contact us to learn more about how Ironside can put powerful intelligence and decision making information for those afflicted with chronic illness or disease, in the hands of providers and clinicians at your organization.