Table 4 Key feature differences between TP and FN groups predicted during epoch 2. SpO₂/FiO₂ measured in epochs 1 and 2, the type of oxygen delivery patients received in epochs 1 and 2, as well as the albumin and estimated glomerular filtration rate (GFR) test results from epoch 2 were found to be the most important for distinguishing patients for whom a false negative and true positive prediction was made at this stage of analysis. Feature importance was calculated via KL divergence and statistical testing for differences of means for laboratory and vital signs were conducted via ANOVA (* p < 0.05, ** p < 0.01, *** p < 0.001).

From: Using machine learning on clinical data to identify unexpected patterns in groups of COVID-19 patients