AI Clinical Decision Support Tool Can Help Predict COVID-19 Prognosis

Researchers from The Feinstein Institutes for Medical Research, the research arm of Northwell Health, have developed an artificial intelligence (AI)-based clinical decision support tool that can predict COVID-19 patient prognosis and severity of the disease using blood work and EHR data.

According to the study describing the tool, clinical prognostic models can assist in patient care decisions, but their performance can dip or drift because of shifts in time and location. The researchers state that these models necessitate regular monitoring and updating to address this.

Further, the authors note that prognostic models for COVID-19 do not account for these changes in performance, despite rapid, significant changes across variants and disease waves. The research team set out to develop a model that accounts for quick changes in patient conditions and outcomes.

“COVID-19 was one of the most dynamic diseases we’ve witnessed in modern history and information about how to care for patients was constantly evolving,” said Theo Zanos, PhD, senior author of the paper and associate professor at the Feinstein Institutes’ Institute of Health System Science and Institute of Bioelectronic Medicine, in the press release. “By harnessing data and developing a real-time auto-updating clinical tool, we set out to create a tool that accounts for these developments and helps clinicians make the decisions they need to deliver better care.”

To develop their model, the researchers leveraged EHR data from nearly 35,000 patients hospitalized with COVID-19 across 13 Northwell Health hospitals between April 2020 and May 2022.

From these data, the model was trained to predict 28-day survival using five data points routinely collected early in a patient’s hospitalization: age, serum urea nitrogen, lactate, serum albumin, and red cell distribution width. The model is also designed to continuously monitor its predictive performance and automatically update when it detects performance drifts.

Overall, the model achieved high performance and was accurate throughout the two-year study, which spanned four COVID-19 waves and three dominant variants: Alpha, Delta, and Omicron. The researchers also found that the model performed equally well regardless of gender, race, and ethnicity.

These findings highlight the importance of updating prognostic models in settings with rapidly changing clinical dynamics and indicate the potential for their methodology to be extended to clinical prognostic models for diseases other than COVID-19, the authors concluded.

This is among many recent efforts to leverage AI and predictive analytics to shed light on COVID-19 severity and outcomes.

In February, researchers found that machine learning techniques achieved 88.5 percent accuracy in predicting the severity of disease in 300 patients who tested positive for COVID-19 at JinYanTan Hospital in Wuhan, China. The models leveraged 23 features from patient records to make predictions, including chest computed tomography, fever, malignant tumor presence, heart rate, and systolic blood pressure.