- Machine learning has a lot of promise, but it's still far from being used in routine clinical practice.
- Ochsner Health System in Louisiana has been piloting a new technology tool for predicting a specific kind of medical emergency, before it's too late.
- It aims to put this tool to work today.
Have you ever seen a medical TV show where a group of doctors will rush over to a "coding" patient whose heart has stopped while dramatic music plays?
This race against the clock happens quite often in real life. And many times, the resuscitation attempt isn't enough.
So one hospital group, Ochsner Health System in New Orleans, has been working with Epic, the company that makes its medical record software, and Microsoft's cloud service Azure, to figure out whether technology can pinpoint which patients are likely to code, well before a cardiac or respiratory arrest.
Ochsner's team has been piloting the tool in the past three months, looking at whether various signals from a patient are likely to result in the kind of rapid deterioration that results in a code.
These data points are things like dropping blood pressure, a rising heart rate, a changing value in the blood work, or some combination of the above, as well as demographic factors like age and the patient's medical history.
That's more data than a human doctor could juggle on her own, said Dr. Richard Milani, Ochsner's chief clinical transformation officer. And the existing tools, he said, are no more sophisticated than the tools that are used to figure out credit scores, which have a high rate of false negatives and positives.
That means some flagged patients are perfectly fine, while others slip through the cracks.
"It's all these very nuanced little things that when added together are bigger than the sum of their parts," he told CNBC in a phone interview.
It can take time for companies to collect and analyze all of this data. But that's starting to change as more hospitals form alliances with technology companies, which include data-sharing agreements.
The three organizations plan to publish the results in a study and share them with the broader medical community.
The data from the pilot is still being assessed, said Milani, but he added that the number of codes, meaning emergencies that require immediate resuscitation of the patient, dropped by 44 percent outside of the intensive care unit.
The hope is that other hospitals will follow suit once the data gets published. One potential stumbling block, however, is training health providers to prevent codes using this kind of data, rather than waiting to respond to them.
Milani said the plan is to deploy the tool in all of its hospitals, with an ultimate goal of eliminating codes among floor patients. Ochsner's network has 30 hospitals and more than 80 health centers.
"This isn't just about an algorithm," said Seth Hain, Epic's director of analytics and machine learning. "It's really about the opportunity that all health organizations have to put machine learning into practice."