A predictive analytics algorithm can determine which COVID-19 patients are most likely to deteriorate while in the hospital.
Using predictive analytics, researchers were able to accurately identify COVID-19 patients at risk of quickly deteriorating up to 24 hours in advance, a study published in British Journal of Anaesthesia revealed.
Throughout the pandemic, the only constant has been the virus’s wildly different impact on individual patients. While some present with only mild respiratory symptoms, others have severe illness and need supplementary oxygen or ventilators.
Researchers noted that using invasive mechanical ventilation to treat COVID-19 respiratory failure have shown mortality greater than 85 percent. However, there is little information available about which patients admitted to the hospital who don’t require mechanical ventilation will progress to mechanical ventilation. Researchers also have limited data on which clinical factors are associated with that progression.
“You can see large variability in how different patients with COVID-19 do, even among close relatives with similar environments and genetic risk,” said Nicholas J. Douville, of the Department of Anesthesiology, one of the study’s lead authors. “At the peak of the surge, it was very difficult for clinicians to know how to plan and allocate resources.”
To build the predictive analytics algorithm, researchers from Michigan Medicine looked at a set of patients with COVID-19 hospitalized during the first pandemic surge from March to May 2020 and modeled their clinical course.
The team used inputs such as a patient’s age, whether they had underlying medical conditions, what medications they were on when entering the hospital, and variables that changed while hospitalized – including vital signs like blood pressure, heart rate, and oxygenation ratio. The group aimed to discover which of these data points would help best predict which patients would require a mechanical ventilator or die within 24 hours.
Of the 398 patients in the study, 93 required a ventilator or died within two weeks. The model was able to predict mechanical ventilation most accurately based on key vital signs, including oxygen saturation ratio, respiratory rate, heart rate, blood pressure, and blood glucose level.
The team examined data points of interest at four-, eight-, 24-, and 48-hour increments to identify the optimal amount of time necessary to predict and intervene before a patient deteriorates.
While the algorithm worked best at shorter increments, researchers noted that the model maintained accuracy even two days before an adverse event.
“The closer we were to the event, the higher our ability to predict, which we expected. But we were still able to predict the outcomes with good discrimination at 48 hours, giving providers time to make alterations to the patient’s care or to mobilize resources,” said Douville.
For example, the algorithm could quickly identify a patient on a general medical floor who would be a good candidate for transfer to the ICU, before their condition deteriorated to the point where ventilation would be more difficult.
Going forward, the team expect that the algorithm can be integrated into existing clinical decision support tool already used in the ICU. In the short term, the study highlights patient characteristics that clinicians should consider when treating patients with COVID-19.
The study also raises new questions about which COVID-19 therapies, such as anti-coagulants or anti-viral drugs, may or may not alter a patient’s clinical trajectory.
“Our algorithm can be integrated into a clinical support software with the ultimate goal of identifying patients prior to clinical decompensation. Our primary target (24-hour prediction window) was selected to allow appropriate time for interventions, while still providing evidence of deterioration in dynamic features,” researchers said.
The algorithm could help providers better manage patient care and allocate necessary resources.
“While many of our model features are well known to experienced clinicians, the utility of our model is that it performs a more complex calculation than the clinician could perform ‘on the back of the envelope’ – it also distills the overall risk to an easily interpretable value, which can be used to ‘flag’ patients in a way so they are not missed,” said Douville.