To manage scant resources, University of Michigan researchers developed a machine learning model to predict which COVID-19 patients are likely to become severely ill. The model outperformed Epic's deterioration index tool, according to a study published Feb. 17 in The BMJ.
Three details:
- The tool was tested on 33,119 adult COVID-19 patients. It was tested at first at Ann Arbor-based Michigan Medicine and then 12 other hospitals.
- The model was designed to predict clinical deterioration within the first five days of hospital admission, defined as death or requiring mechanical ventilation, intravenous vasopressors or heated high-flow nasal cannula.
- To calculate its prediction, the tool analyzes nine personal and clinical characteristics from patients' EHR: age, respiratory rate, oxygen saturation, oxygen flow rate, pulse oximetry type, head-of-bed position, position during blood pressure measurement, venous blood gas pH and partial pressure of carbon dioxide in arterial blood.