Skip to main content
. 2021 Mar 2;10:586054. doi: 10.3389/fcimb.2020.586054

Table 4.

The ablation experiments of the second predictive model.

Accuracy (%) AUC (%) 95% CI for AUC (%) Sensitivity (%) 95% CI for sensitivity (%) Specificity (%) 95% CI for specificity (%) F1 (%)
hs-CRP 68.75 81.28 71.2–91.27 66.67 50.95–79.56 72.00 50.40–87.13 73.17
+IL-6 74.29 79.56 68.92–90.19 80.00 64.95–89.91 64.00 42.62–81.29 80.00
+IL-2R 75.71 83.38 74.00–92.76 77.78 62.52–88.29 72.00 50.41–87.12 80.46
+PCT 75.71 79.56 68.96–90.15 75.56 60.14–86.61 76.00 54.48–89.84 80.00
+TNF-α 74.29 83.91 74.60–93.22 71.11 55.48–83.16 80.00 58.87–92.39 78.05
+LYMPHA 78.57 87.38 78.88–95.88 73.33 57.79–84.90 88.00 67.66–96.85 81.48
+NEUTP 80.00 90.58 83.59–97.56 73.33 57.79–84.90 92.00 72.50–98.60 82.50
+RBC 82.86 92.98 86.91–99.04 82.22 67.41–91.49 84.00 63.08–94.75 86.05

The first experiment only used the hs-CRP value as a single input feature of the model. Then IL-6 was added to the second column, that is, the model used hs-CRP and IL-6 as input features. Finally, when taking eight features, including Red Blood Cell Count (RBC), the model’s performance is very close to the prediction model with 45 features.