Seton Healthcare Family, a health care system serving about 1.9 million people in central Texas, was looking for a way to reduce readmissions for patients with congestive heart failure (CHF), which the company says is costly -- to the tune of an estimated $35 billion in the United States. Seton Healthcare says 500,000 new cases are diagnosed every year and more than half of CHF patients need to be readmitted within six months after treatment.
"Hospitals are penalized [by Medicare] for having too many patients who need to be readmitted," says chief strategy officer Travis Froehlich. To predict which patients are most at risk for readmission, Seton Healthcare uses ICPA to identify known risk factors like smoking, so health professionals can focus their efforts on keeping patients at home more efficiently.
To accomplish this, Seton is using the ICPA for Healthcare tool, which lets officials mine unstructured data using natural language processing and search technologies. The company says more than 80% of healthcare data is unstructured and consists of physician notes, registration forms, discharge summaries, echocardiograms and other medical documents.
When combined with structured data, all this can paint a more accurate picture of trends, patterns and deviations, allowing clinicians to make better treatment decisions. She says this data was found in the history and physical part of the medical record in a narrative section.
The information gained from the unstructured data, however, is only as good as the user-created linguistic models using ICPA, Seton says. The accuracy of the linguistic and predictive models themselves depends on how well the model has been optimized by the user.
In terms of ROI, "The cost and angst to get all this information we need in structured data fields is pretty tremendous," says Froehlich. "This has the potential to reduce the need to provide a data field for every possible piece of data and of making everyone type everything into every single field ... which drives people up a wall. Intuitively we believe there is a cost benefit."
Seton Healthcare senior epidemiologist Christine Jesser concurs. "Highly skilled clinicians are spending inordinate amounts of time entering data into structured fields and this can reduce the time and effort it takes." While clinicians were already aware that smoking is an important consideration when looking at someone with CHF, she says the tool came back with some results they didn't expect for predicting the probability of a person's readmission to the hospital.
"The interesting things we found were their living status and were they in assisted living situations, and whether they had drug and alcohol abuse,'' Jesser says. "Those were some social factors that were only found in unstructured data that emerged as important predictors in the model."