Modeling for Lower Hospital Readmissions
If you’re a traveler, being a frequent flyer is good thing that can bring some nice perks from the airline — free trips, seat upgrades, extra legroom. That’s not so if you’re someone seeking medical attention.
Hospitals will provide care for their frequent flyers — patients who return time and again after discharge — but you probably won’t find much fawning going on. These are the people who will send a hospital’s preventable readmissions rate into the penalty zone instituted by the Centers for Medicare & Medicaid Services. On the other hand, for better or worse, they do make for some good modeling.
The number of previous admissions is highly influential in determining somebody’s probability of readmission, Michael Hollenbeck, vice president of healthcare solutions at Predixion Software, told me recently. Frequent flyers and, conversely, infrequent flyers are naturally occurring clusters when hospitals and other healthcare industry participants “let the machine learning process itself help them to understand where the natural breaks in clusters are.”
With help from machine learning processes, a hospital might also find naturally occurring clusters around conditions of disease states, Hollenbeck said. This is especially the case when hospitals have specialties, such as oncology or diabetes treatment. “They might find in the beginning that maybe there’s an issue with a specific unit, or that there’s some type of HR isuse, and the modeling can help pick those out.”
In other words, one approach hospitals can take in determining who is likely to be readmitted is letting the machine learning algorithms find patient segments by running against the known readmission population. That’s the approach that Thrive HDS, which I mentioned yesterday, is taking now. “We’ve just sent the first cohort of data and are looking at those things as we speak,” said Curt Sellke, vice president of analytics at Thrive HDS, which recently formed to provide predictive analytics and other technology services to accountable care organizations and other healthcare providers.
Full Article