Reut Kassif Lerner, MD
Senior Physician
Sheba Medical Center, United States
Disclosure information not submitted.
Nadav Baharav, MD
Senior Physician
Sheba Medical Center, United States
Disclosure information not submitted.
Ofer Chen, MD
VP Clinical Development
Clew Medical, United States
Disclosure information not submitted.
Ari Lipsky, MD
Chief of Emergency Medicine
HaEmek Medical Center, United States
Disclosure information not submitted.
Gershon Celniker, M.Sc
Head of Data and AI Research
CLEW Medical, United States
Disclosure information not submitted.
Itai Pessach, MD, PhD, MHA
Director, The Edmond and Lily Safra Children's Hospital
Sheba Medical Center, Israel
Disclosure information not submitted.
Title: Using Artificial Intelligence Models to Predict Deterioration in Critically Ill COVID-19 Patients.
Introduction:
The COVID-19 pandemic resulted in a surge of critically ill adults that presented significant challenges to critical care providers. These challenges included the need to treat a new disease and in the same time to endure the high demand for critical care capacity. In this study we aimed to evaluate the performance of previously established, FDA approved, artificial intelligence based predictive models that were developed to detect respiratory failure and hemodynamic instability prior to the COVID-19 era, in predicting these events in critically ill COVID-19 adults.
Methods:
29 adult ICU stays from the Sheba COVID-19 Critical care unit in the period between 3-5 2019 were included. Respiratory and hemodynamic deterioration events were prospectively tagged by on-site clinicians. The prediction models were trained and validated to predict either respiratory failure requiring mechanical ventilation or hemodynamic instability requiring inotropic/pressor support. As all the patients in this cohort were mechanically ventilated and required pressors we evaluated the models performance in accurately predicting secondary deterioration i.e the need for increased respiratory support (Desaturation, higher FiO2 or PEEP etc.) or increased pressor/inotropic support (higher dosage, additional drug, fluid boluses etc.).
Results:
During the study period 289 respiratory deterioration events were tagged by the clinical team. We excluded 2 events since they occurred during the first 6 hours of the stay as well as 63 other events that were determined to be mild in severity. Our model had an AUC of 0.76 for predicting these events 3 hours in advance. Furthermore, 177 hemodynamic instability events were detected by the clinical team, of which 14 were determined to be mild events and were excluded. Our model had an AUC of 0.78 for predicting these events 3 hours in advance.
Conclusions:
We have demonstrated that predictive models that were developed, trained and validated to predict respiratory failure and hemodynamic instability in non-COVID patients had reasonable performance in predicting deterioration in COVID-19 critically ill patients. Furthermore, the models that were trained to predict initial deterioration in non-ventilated patients without pressor support were able to predict secondary deterioration in this cohort.