Leveraging Data Science to Understand and Predict Hospital Readmissions in Diabetes Patients

Authors

  • Katherine Brown
  • Joseph Bivens
  • Scott VandePolder
  • William Eberle

Abstract

The fundamental goal in healthcare is simple: improve patient health. An important facet to this is reducing the number of hospital admissions for a patient. Reaching this goal becomes incredibly challenging when a patient is diagnosed with a chronic disease, such as diabetes. The ability to predict complications leading to hospital readmission before the situation worsens is highly important to hospitals across the world.

Reducing readmission for any patient demographic provides enormous benefits to all parties involved. Patients benefit from the more comprehensive and complete care. Also, by limiting a patient’s stay in a hospital, patients are less exposed to dangerous diseases such as staph infections. Hospitals benefit as more resources can be allocated to other patients.

Identification of such complication trends requires processing large-scale patient and treatment data. Our project attempts this analysis with the University of California, Irvine data repository on hospital readmissions of diabetic patients. We define a hospital readmission in terms of the number of days since the original admission. Patients were either not readmitted at all, “NO”, admitted within thirty days, “<30”, or admitted after thirty days have passed, “>30”.

We investigate this data using R statistical computing software, performing analysis to discover and quantify relationships among the available data. Our goal is to develop predictive models that identify patients at an increased risk of readmission. When incorporated into a utility for healthcare providers, these predictive diagnostics could help patients receive the targeted care they need, lowering their chances of readmission.

Downloads

Published

2017-05-17

Issue

Section

Computer Science