Using a dataset of 270 critically ill patients with COVID-19 we have identified a set of variables that predict the mortality for patients on the day of admission to the ICU. The use of this practical prediction model should affect clinical decision-making and medical management of COVID-19 patients.
- We can predict the chance of survival based on values measured on the day of arrival in the ICU. This timely prediction should improve the treatment strategies and outcome of patients.
- There is a paucity of ICU severity predictor scales overall. No risk scales are yet available for the general prediction of mortality in severe COVID-19 patients.
- Using the local interpretable model-agnostic explanations (LIME) model we are able to indicate which specific clinical condition is responsible for the prediction. Thus, physicians can apply this machine learning model to individual patients and can tailor treatments toward medical conditions knowing the key variables underlying the prognosis.
In order to validate our model we will require a similar number of datapoints (about 300 patients) to externally confirm the prediction model. The preprint of our work can be found under: