Predictive Analytics Used to Help a Large Hospital Group Reduce Readmission Rates
For a large Hospital Group hoping to reduce readmission rates by 3% annually, predictive analytics is helping doctors pinpoint patients with high readmission risk. Hospital staff then administer additional medical care to these patients and thereby reduce readmission rates.
The Hospital Group generates an individualized prediction of a patient’s readmission rate at the time of diagnosis. Using the derived predictions from the analysis, the Hospital Group reaps the following annual savings:
• Reduces 6,000 occurrences of patient readmission.
• Avoids $4 million in potential Medicare penalties.
• Saves approximately $72 million in medical service costs.
• Utilizes resources more efficiently by providing extra care to high-risk patients.
• Improves hospital rating based on lower readmission rate and increased patient satisfaction.
Recent changes in federal legislation have made hospitals restructure the way they manage patients to save money and avoid government penalties. Section 3025 of the Affordable Care Act added section 1886(q) to the Social Security Act, which took effect October 1, 2012. It established the Hospital Readmissions Reduction Program, which requires the Centers for Medicare & Medicaid Services (CMS)—a federal agency whose mission is to ensure effective healthcare coverage and to promote quality care for Americans—to reduce payments to hospitals with excess readmissions.
Consequently, hospitals have been seeking ways to reduce readmission rates across the board. Doing so would not only reduce unnecessary costs, it would help hospitals avoid CMS-levied payment penalties.
The Hospital Readmissions Program Accuracy and Accountability Act1 requires CMS to account for patient socioeconomic status when calculating risk-adjusted readmission penalties. Holding all other factors constant, socioeconomic conditions—such as poverty, low literacy, limited English proficiency, minimal social support, poor living conditions, and limited community resources—likely have direct and significant impacts on avoidable hospital readmissions. Adjusting for these factors would improve accountability and quality of care.
Per capita healthcare costs in the US are the highest in the world and have trended upward for decades. Reducing the number of unnecessary readmissions by even a few percent could create huge savings.
A readmission is defined as a hospitalization that occurs approximately 30 days after a previous hospital stay. Readmissions are often the result of a patient’s initial problem not being resolved. They can also be caused by a patient’s mismanagement of the original condition, misunderstanding how to manage the condition, or lack of access to additional medical services or medications.2
According to a recent study, a patient’s socioeconomic conditions can have direct and significant impacts on avoidable extended hospital stays.3 Adjusting for these factors improves accountability and quality of care.
Socioeconomic data along with electronic medical records (EMR) provide a patient’s history and living standard for the model. This data is collected in an enterprise data hub (EDH). The EDH facilitates data loading, cleansing, and association or linking between different datasets, such as EMR and socioeconomic data, for every patient.
Using the “Random Forests” algorithm, Intel helped the Hospital Group build models based on this linked dataset. These models predict a readmission risk score during the admission process for each patient based on his/her EMR and socioeconomic data. The Hospital Group assigns each patient who is admitted a risk score, thus creating a high-risk bin, which consists of the top 5% of patients by risk score. With an accurate prediction, hospital administrators can suggest a special care plan for patients identified as high risk.
By focusing on patients in the high-risk bin, the Hospital Group can target patients with a higher likelihood of readmission for additional care during their first visit, and thus reduce the readmission rate of these patients. As a side benefit, this frees up resources they could use to help an additional 300 to 500% more patients.
The requirements necessary for the data storage solution include analysis of a patient’s readmission prediction, scope of the patient’s medical condition, accuracy of the prediction, growing population data, easy ingestion of diverse data, fault tolerance, low cost, and security of the data.
Working with Intel, the Hospital Group selected several data sources for the predictive model, including EMR from a relational model and additional socioeconomic data such as housing prices and availability of healthcare within the immediate area of each hospital.
By using predictive analytics with Cloudera, the Hospital Group takes advantage of more unconventional data sources to produce more accurate readmission predictions. Cloudera has the power to ingest unrelated, unstructured, and semi-structured data sources, which the Hospital Group uses to enrich existing medical data.
The readmission predictive models Intel helped create for the Hospital Group are very successful at identifying patients who are a high risk for readmission after an initial hospital stay. With these more accurate predictions, hospital administrators can now suggest special care plans for high-risk patients.
2. Silow-Carroll, Sharon; Edwards, Jennifer N.; and Lashbrook, Aimee; Health management associates. “Reducing Hospital Readmissions: Lessons from Top-Performing Hospitals” synthesis report. April 2011.