Skip to main content
Fig. 3 | BMC Medical Research Methodology

Fig. 3

From: Comparison of correctly and incorrectly classified patients for in-hospital mortality prediction in the intensive care unit

Fig. 3

Decision Fusion. Results from all rounds and models are combined, as shown in the upper table. The table shows the results from four fictive patients. The A and D represent ‘Alive’ and ‘Dead’, respectively. The hyphens ‘-’ represent the rounds where the patient is not in the test set. Patient 1 is predicted alive at discharge by three models for round 1, while the LR model predicts dead at discharge. All models predict alive at discharge for round 2 and round 100. Round 3-99 are not shown. The lower table shows the patient’s discharge status, the percentage of predicted ‘Alive’ and ‘Dead’ across all models and rounds, and the final group. The percentage of ‘Alive’ predictions is 92%, and ‘Dead’ is 8%. With an agreement threshold of 90%, the models agree on an ‘Alive’ prediction. However, as seen in the lower table, the patient is dead at discharge, thus, is placed in the FN group. Patient 2 is placed in the MIXED group since neither ‘Alive’ nor ‘Dead’ are predicted more than 90% of the patient’s occurrences in the test set. Patients 3 and 4 are placed in the TN and TP group respectively since all models agree on the prediction across all rounds, and the prediction is the same as the discharge status

Back to article page