New York University researchers have developed a model to predict favorable four-day outcomes among COVID-19 patients.
Based on real-time lab values, vital signs and oxygen support variables, the model – which researchers say has 90% precision – may help clinicians to determine which patients can be safely discharged.
“Discharging patients safely to free up beds for incoming patients is optimal as it does not require expanding human … or structural … resources,” wrote researchers from the NYU Grossman School of Medicine and NYU’s Courant Institute of Mathematical Sciences in the study, published this week in the journal npj Digital Medicine.
“Given clinical uncertainty about patient trajectories in this novel disease, accurate predictions could help augment clinical decision making at the time the prediction is made,” they continued.
WHY IT MATTERS
As one of the early epicenters of the U.S. novel coronavirus pandemic, New York City facilities faced a flood of COVID-19 patients in April and May. Although the initial surge has abated somewhat, case numbers continue to rise around the country, and officials are making dark predictions about infection rates in the colder months to come.
With that in mind, NYU researchers sought to leverage artificial intelligence in conjunction with patient data to address population management in hospitals.
“During the COVID-19 pandemic, the operational needs of frontline clinicians have rapidly shifted,” wrote the researchers. For instance, as the disease response progressed, triage and cohorting became less of a priority.
“Similarly, while predicting deterioration is clinically important, our health system had already implemented a general clinical deterioration predictive model and did not have an immediate use case for a COVID-19-specific deterioration model,” they continued.
“Furthermore, since ICU beds were already limited to patients in immediate need of requiring higher levels of care, predicting future needs would not dramatically change clinical management,” they added.
Using 3,345 retrospective and 474 prospective hospitalizations, researchers leveraged machine learning and laboratory values to identify patients with favorable outcomes within four days.
“We defined a favorable outcome as absence of adverse events: significant oxygen support (including nasal cannula at flow rates >6 L/min, face mask or high-flow device, or ventilator), admission to ICU, death (or discharge to hospice) or return to the hospital after discharge within 96 [hours] of prediction,” wrote the research team.
A plurality (45 percent) of the patients studied were white, and the majority (61 percent) were men, with an average age of 63.5 years.
The tool was incorporated into clinicians’ EHRs in May, and preliminary results suggest that providers are adopting the scores into their workflows.
THE LARGER TREND
Researchers have leaned on artificial intelligence throughout the COVID-19 pandemic, often using it in conjunction with analytics to try to anticipate needs such as personal protective equipment or ICU bed availability.
Diagnostic AI has also played a role in the response, particularly where radiology is concerned. This past week, researchers at the University of Minnesota, in conjunction with Epic, built a tool to detect COVID-19 in lung X-rays.
ON THE RECORD
“By identifying patients at low risk of an adverse event with high precision, this system could support clinicians in prioritizing patients who could safely transition to lower levels of care or be discharged,” wrote the NYU researchers in the study.
“By contrast, using published models that predict occurrence of adverse events to guide discharge decisions may not be as effective,” they continued.
Kat Jercich is senior editor of Healthcare IT News.
Twitter: @kjercich
Email: [email protected]
Healthcare IT News is a HIMSS Media publication.
Source: Read Full Article