Use Case: Predicting Hospital Readmission Risk with DataRobot

This post was originally part of the DataRobot Community. Visit now to browse discussions and ask questions about the DataRobot AI Platform, data science, and more.

In this article, we’ll explain how to use predictive models for better outcomes and lower cost-of-care.

What’s the problem?

The federal government imposes penalties (in the form of withholding payment) based on a hospital’s readmission rate within 30 days of discharge. In 2017, Medicare expected to withhold $528 million, about $108 million more than the previous year. These penalties would punish 2,597 hospitals–more than half of the nation’s facilities. According to a Kaiser Health News analysis, while this is roughly the same number as in 2015, the average penalty is increasing by 20%. The costs of readmission can be staggering.

The challenge and solution

How do you identify patients with a high readmittance risk so that you can take preventive measures prior to or after discharge? It is impractical and cost prohibitive to monitor every patient after discharge. You could, however, use the hospital’s existing data to build predictive models that help prioritize patients based on readmission risk. The overall benefits from a readmission risk model are better patient outcomes, reduction in cost-of-care, and penalty avoidance. Using automated machine learning to develop the model, content experts can frame the problem and data scientists can focus on application and deployment. In this way, you develop an adaptable model that addresses provider, patient, and business needs.

For example…

Marcy is a clinical analyst for a large healthcare network; member hospitals are seeing a growing number of 30-day readmissions. Marcy has been tasked with identifying patients who have a high risk of readmittance. Using the network’s electronic health record (EHR) data, Marcy will build a predictive model to proactively identify those at risk for hospital readmission. With DataRobot, her model can show her–and the care providers–the basis for the prediction.

Training data: https://s3.amazonaws.com/datarobot-use-case-datasets/10k_diabetes.csv

Prediction data: https://s3.amazonaws.com/datarobot-use-case-datasets/10kDiabetesScoring.csv

Hospital Readmission Data Dictionary: https://s3.amazonaws.com/datarobot-use-case-datasets/Hospital+Readmission+Data+Dictionary+(1).pdf

About the author

Linda Haviland

Community Manager

Meet Linda Haviland