One of the key challenges for hospitals during the COVID-19 pandemic has been being able to predict the availability of open beds, particularly during viral spikes. To that end, researchers in a study at Michigan Medicine have developed a machine learning algorithm that can accurately identify patient deterioration for both general ward and COVID-19 patients.
In a study published in JMIR: Medical Informatics, researchers describe the development and performance of the Predicting ICU Transfer and other Unforeseen Events (PICTURE) algorithm.
The team trained and validated the PICTURE model on a cohort of hospitalized COVID-19 patients using EHR data from 2014 to 2018. Researchers then applied the model to two test sets, including non-COVID-19 patients from 2019 and COVID-19-positive patients in 2020.
Their conclusion?
“The PICTURE model is more accurate in predicting adverse patient outcomes for both general ward patients and COVID-19 positive patients in our cohorts compared to the EDI,” or Epic Deterioration Index (EDI), an existing model that has recently been assessed for use in patients with COVID-19.
The new tool is able to analyze a range of data, from vital signs and lab results to demographic information. Using this data, PICTURE is able to flag patients at highest risk of decline, as well as explain what risk factors influence the prediction. This can help clinicians respond faster.
“The PICTURE model is able to integrate data from the electronic health record and transform it into meaningful predictions based on the patient’s risk of experiencing an adverse outcome,” said Brandon Cummings, a data scientist at Michigan Center for Integrative Research in Clinical Care (MCIRCC).
“This is especially important in the case of COVID-19 patients, who can deteriorate rapidly and unexpectedly. By predicting these events before they occur, PICTURE can give clinicians time to react and stabilize the patient before more drastic measures are required.”
With COVID-19 cases continuing to surge and wane, researchers believe the PICTURE tool could serve as a valuable resource for ICU staff making triage decisions, but they noted that “this analysis is limited to a single academic medical center, and its generalizability to other healthcare systems will require future study.”
“The PICTURE model is more accurate in predicting adverse patient outcomes for both general ward patients and COVID-19 positive patients in our cohorts compared to the EDI.”
Michigan Medicine researchers are working to test PICTURE in other health systems and develop specialized versions of it for other populations, including rehabilitation and sepsis patients.
“The ability to anticipate these events will be valuable when considering potential future waves of COVID-19 infections,” says Kevin Ward, MD, executive director of MCIRCC. “However, the real value will be the continued use of PICTURE in all hospitalized patients no matter what the situation is.”