Table 2 Mean metrics for each fold of 10-fold cross-validation over the 8 6-hour epochs under study (48 hours total). After 3 epochs, performance drops as marked by the stark decrease in the Matthew’s Correlation Coefficient Metric, illustrating the difficulty of predicting outcomes for the subset of individuals that have indeterminate predictions in the preceding epochs.

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