Register update: there will be a delay adding new records during our holiday period (Dec 17 - Jan 4)
Cochrane COVID-19 Study Register
Study record
Levi 2022aFirst Published: 2022 Jul 4Updated Date: 2022 Jul 4

Early Oxygen Treatment Measurements Can Predict COVID-19 Mortality: a Preliminary Study

  1. Study Type
  2. Observational
  1. Study Aim
  2. Diagnostic/Prognostic
  3. Treatment and Management
  1. Study Design
  2. Case series/Case control/Cohort
  1. Intervention Assignment
  2. Not Applicable
Reference record

Early Oxygen Treatment Measurements Can Predict COVID-19 Mortality: a Preliminary Study

Levi Y, Yamin D, Brandes T, Shmueli E, Patalon T, Peretz A, Gazit S, Nahir B
Journal article
Report Results
Halting the rapid clinical deterioration, marked by arterial hypoxemia, is among the greatest challenges clinicians face when treating COVID-19 patients in hospitals. While it is clear that oxygen measures and treatment procedures describe a patient's clinical condition at a given time point, the potential predictive strength of the duration and extent of oxygen supplementation methods over the entire course of hospitalization for a patient death from COVID-19 has yet to be assessed. In this study, we aim to develop a prediction model for COVID-19 mortality in hospitals by utilizing data on oxygen supplementation modalities of patients. We analyzed the data of 545 patients hospitalized with COVID-19 complications admitted to Assuta Ashdod Medical Center, Israel, between 7 March 2020, and 16 March 2021. By solely analyzing the daily data on oxygen supplementation modalities in 182 random patients, we could identify that 75% (9 out of 12) of individuals supported by reservoir oxygen masks during the first two days died 3-30 days following hospital admission. By contrast, the mortality rate was 4% (4 out of 98) among those who did not require any oxygenation supplementation. Then, we combined this data with daily blood test results and clinical information of 545 patients to predict COVID-19 mortality. Our Random Forest model yielded an area under the receiver operating characteristic curve (AUC) score on the test set of 82.5%, 81.3%, and 83.0% at admission, two days post-admission, and seven days post-admission, respectively. Overall, our results could essentially assist clinical decision-making and optimized treatment and management for COVID-19 hospitalized patients with an elevated risk of mortality