Hospital inpatient care makes up nearly a third of all health care expenses in the U.S. — the better part of $400 billion per year. Obviously, the length of a patient’s stay — on average, 4.5 days — has a big impact on this cost. That length of stay is also an indication of a hospital’s efficiency and efficacy — how quickly and how effectively it treats patients. Those factors express the quality of care and the amount of cost.
Note that if a hospital can treat patients efficiently, it can treat more patients. Thus, an accurate prediction of “length of stay” becomes an important metric to improve care and reduce cost not only for the patient, but also for the community that hospital serves.
Win win.
So how do we get to an accurate prediction of length of stay?
Auburn researcher Pankush Kalgotra took a look at this complex, wicked problem and created a solution that significantly improves the accuracy of length-of-stay predictions.
When a patient is admitted to the hospital, the current intake process typically considers only a small portion of patient information — immediate symptoms and medical history. But length of stay is driven by many other factors – a patient’s individual characteristics, treatment plans and disease interactions.
Kalgotra and his research partner, Ramesh Sharda at Oklahoma State University, devised a two-step process. First, they looked at historical data — 20 million patient records — and compiled a database of 15,000 diseases that are often found together — the co-morbidities. Then, they used the patients’ symptoms at the time of admission to find the probable co-morbidities that are not diagnosed but may show up during the hospital stay. This information can augment treatment and measures can be taken before the expected disease occurs.
The patient benefits, the stay is shortened, the hospital can better manage critical resources and insurers can plan for and accommodate the real-world costs of each patient admission.
Kalgotra and Sharda published their findings in the Journal of Management Information Systems.
Can Big Data Improve Care?
