Skip to main content
Advertisement
Browse Subject Areas
?

Click through the PLOS taxonomy to find articles in your field.

For more information about PLOS Subject Areas, click here.

< Back to Article

Fig 1.

Example of prediction of daily severity during admission.

The target is actual daily severity, and day 0 to day 2 represent the model predictions. The patient’s status aggravated to severe on day 2, recovered to moderate on day 9, and became mild on day 13. The model (day 2) predicts disease aggravation and recovery 2 days before the infection.

More »

Fig 1 Expand

Table 1.

Baseline characteristics of included patients.

More »

Table 1 Expand

Table 2.

The best model performance for daily COVID-19 severity.

More »

Table 2 Expand

Fig 2.

The AUROC and AUPRC for daily prediction of COVID-19 severity.

The upper figures denote the receiver operating curve for predicting day 0, day 1, and day 2. The lower figures denote the precision-recall curve for each severity outcome.

More »

Fig 2 Expand

Table 3.

Comparison of the model performances between DNN and transformer.

More »

Table 3 Expand

Fig 3.

The comparison of the model by decision curve analysis.

The figure represents the decision curve analysis for severity predictions for the period two days ahead. The baseline comparison (All positive) is premised on the assumption that every two-day event will be mild, moderate, or severe. All of our models demonstrated benefits exceeding the basic assumption made without any model (All positive).

More »

Fig 3 Expand

Fig 4.

The feature importance among the models.

The three model’s feature importance were compared by the SHAP method. The tree-based model places a greater focus on steroid usage, SpO2 count, and vital signs (Pulse rates, and diastolic BP). The DNN model placed a greater focus on demographic data and symptoms.

More »

Fig 4 Expand