Editor’s Note: After we published this article, we learned of a preprint study of the Epic Deterioration Index for COVID-19 patients that’s mentioned in this article. Be sure to read the whole thread to understand some of the nuances of this quickly moving effort.
Not long ago, we shared the details on how Stanford was using machine learning to predict how patients with COVID-19 were at risk of having their condition worsen. There are many more such stories emerging as the virus continues to wreak havoc within many hospitals.
According to STAT News, hundreds of hospitals are also using the Epic technology deployed by Stanford. While the Epic tool, known as the deterioration index, wasn’t designed with COVID-19 in mind, hospitals are wading in anyway.
The speed at which hospitals are adopting this technology is virtually unprecedented. As the SN story notes, hospital IT departments usually test such a tool extensively, refine AI capabilities to meet their needs then restructure care processes to support the rollout before clinicians get involved.
Ordinarily, a complex project like deploying AI or analytics could take months or even years. Of late, though, hospitals are making things happen in as little as a week. Given their lack of experience with the virus, doctors desperately need guidance on which patients are at high risk, in part to be sure that ventilators are available to patients who might need them.
For example, at Parkview Health, a nine-hospital system using Epic’s tool, hospital leaders are working to see how the risk scores created by the deterioration predict where patients’ condition will be heading. The index, which draws on data such as patient vital signs, lab results and nursing assessments, assign patients a score from 0 to 100. Higher scores indicate higher risks that the patient will deteriorate.
After analyzing about 100 cases, doctors found that 75% of hospitalized patients who scored between 38 to 55 eventually ended up in the ICU. While this isn’t terribly precise, doctors are still bearing these scores in mind as they decide how to treat patients.
Meanwhile, at the University of Michigan, an analysis of 200 patients found that the deterioration index makes its best predictions for patients at the high and low ends of the scale. For example, the 9% of patients whose scores remained low during their first 48 hours of hospitalization proved unlikely to experience a life-threatening event. Meanwhile, the 10% to 12% of patients scoring on the high end were far more likely to need ICU care, according to SN.
Of course, these measures are imprecise at best, and eventually doctors will demand better tools for tackling the pandemic. That being said, at least for the near future, hospitals are likely to experiment with technologies they hadn’t considered before in an effort to meet the huge challenges COVID-19 poses.
Normally as a caregiver or patient I am reluctant to rely on AI & use providers that are well educated and board certified in a large university/medical school setting.
In the case of interns and residents AI is a good tool that points them in the right direction when lacking years of experience.
However, the Covid pandemic needs every tool it can find to help predict the direction of acuity in their hospitals and lower the death rate. Since the standards of care for patients with Covid are being developed in real time, many tools are required to provide accurate care. And lacking the proliferation of reliable large scale testing, Covid patients are landing in the hospital later in the disease progression, causing higher acuity rates and stress on frontline healthcare providers.
Carmela,
It’s amazing how many things COVID-19 has impacted. it brings up an interesting question. Which is faster to update in a crisis? Frontline healthcare clinicians or an algorithm?
Of course, it’s also worth noting that the algorithm doesn’t replace the frontline healthcare providers. It’s more about informing and triaging patients. At least at this point.
great article!!