A large multicenter model for predicting deterioration of Covid-19 inpatients
2Departments of Internal Medicine “C”, “E”, Tel-Aviv Sourasky Medical Center, Israel
3Infectious Diseases Unit, Sheba Medical Center, Israel
4Department of Critical Care Medicine, Ashdod University Hospital, Israel
5Sackler Faculty of Medicine, Tel-Aviv University, Israel
The COVID-19 pandemic remains threat to global health. Due to the limited
understanding of disease progression and despite the recent advances in treatments, to
date, it is still a clinical challenge to detect which hospitalized patients will deteriorate
and hence early warning tools are required. Moreover, several studies suggested that
taking early measures for treating patients under risk of deterioration could prevent or
lessen condition worsening and the need for mechanical ventilation.
We developed a predictive model for early identification of patients at risk for clinical
deterioration by analyzing electronic medical records of 5,523 COVID-19 inpatients from
three hospitals: Tel-Aviv Medical Center, Sheba Medical Center and Assuta Ashdod
Medical Center. Our model employs machine learning methods and uses routine clinical
features such as vital signs, lab measurements, demographics, medications, and
background diseases. Deterioration was defined as mortality or as the start of mechanical
ventilation. In prediction of mortality within the next 24 hours, the model achieved an
area under the ROC curve of 0.89 and area under the precision-recall curve of 0.45.
Validation of the model was performed across cohorts and sequentially utilizing "wave-
fold" cross validation, i.e, the model was trained using data from earlier waves and
evaluated using subsequent wave data. Our results confirmed that the model was
agnostic to the effect of different variants, new treatments, and vaccinations.