Decompensation Prediction for Hospitalized COVID-19 Patients

Since the start of the COVID-19 pandemic, hospitals have been overwhelmed with the high number of ill and critically ill patients. The surge in ICU demand led to ICU wards running at full capacity, with no signs of demand falling. As a result, resource management of ICU beds and ventilators has been a bottleneck in providing adequate healthcare to those in need. Short-term ICU demand forecasts have become a critical tool for hospital administrators. Therefore, using the existing COVID-19 patient data, we build models to predict if a patient's health will deteriorate below safe thresholds to deem admission into ICU in the next 24 to 96 hours. We identify the most important clinical features responsible for the prediction and narrow down the health indicators to focus on, thereby assisting the hospital staff in increasing responsiveness. These models can help the hospital staff better forecast ICU demand in near real-time and triage patients for ICU admissions as per the risk of deterioration. Using a retrospective study with a dataset of 1411 COVID-19 patients from an actual hospital in the USA, we run experiments and find XGBoost performs the best among the models tested when tuning parameters for sensitivity (recall). The most important feature for the four prediction tasks is the maximum respiratory rate, but subsequent features in order of importance vary between models predicting ICU transfer in the next 24 to 48 hours and those predicting ICU transfer in the next 72 to 96 hours.