Fig. 3
ROC curves analysis for single variables and adjusted for the Charlson Comorbidity Index. The figure shows ROC curve analysis for PaO2/FiO2 (A), for CPK (B), INR (C), and AT III (D). A Score model [PaO2/FiO2] = − 0.0028 * (PaO2/FiO2), cut-off values > − 0.3271 (J-index = 0.3015), Se 54.9% (CI95% 36.6–71.2%) and Sp 75.9% (CI95% 56.5–89.7%); score model [Charlson + PaO2/FiO2] = 0.4705 * (Charlson) − 0.0072 * (PaO2/FiO2), cut-off values > 0.2659 (J-index = 0.5586), Se 80.0% (CI95% 63.1–91.6%) and Sp 75.9% (CI95% 56.5–89.7%). The AUCs for single and adjusted model were, respectively, 0.586 (CI95% 0.456-0.708, p-value 0.2400) and 0.819 (CI95% 0.702–0.904, p-value < 0.0001). The adjustment for the Charlson Comorbidity Index showed a statistically significant difference for the AUC (p-value < 0.0001). When the Charlson Comorbidity Index and adjusted model score ROC curves were compared, the p-value was not statistically significant (0.9254). B Score model [CPK] = 0.0007 * (CPK), cut-off values ≤ 0.0558 (J-index = 0.2112), Se 65.8% (CI95% 49.4–79.9%), Sp 55.3% (CI95% 38.3–71.4%); score model [Charlson + CPK] = 0.2984 * (Charlson) + 0.0009 * (CPK), cut-off value > 0.9267 (J-index = 0.6149), Se 87.8% (CI95% 73.8–95.9%), Sp 73.7% (CI95% 56.9–86.6%). The AUCs for single and adjusted models were, respectively, 0.522 (CI95% 0.407–0.636, p-value 0.7390) and 0.807 (CI95% 0.703–0.887, p-value < 0.0001). The adjustment for Charlson Comorbidity Index showed a statistically significant difference for the AUC (p-value 0.0008). When the Charlson Comorbidity Index and adjusted model scores ROC curves were compared, the p-value was not statistically significant (0.3234). C Score model [INR] = 0.9573 * (INR), cut-off values > 1.1296 (J-index = 0.2997) Se 53.7% (CI95% 37.4–69.3%), Sp 76.3% (CI95% 59.8–88.6%); score model [Charlson +INR] = 0.2623 * (Charlson) + 0.8398 * (INR), cut-off value > 1.5220 (J-index = 0.5847), Se 92.7% (CI95% 80.1–98.5%), Sp 65.8% (CI95% 48.6–80.4%). The AUCs for single and adjusted model were, respectively, 0.621 (CI95% 0.505–0.728, p-value 0.0590) and 0.815 (CI95% 0.712–0.894, p-value < 0.0001). The adjustment for Charlson Comorbidity Index showed a statistically significant difference for the AUC (p-value 0.0051). When the Charlson Comorbidity Index and adjusted model scores ROC curves were compared, the p-value was not statistically significant (0.0805). D Score model [AT III] = − 0.0299 * (AT III), cut-off values > − 2.3655 (J-index = 0.3055), Se 46.3% (CI95% 30.7–62.6%), Sp 84.2% (CI95% 68.7–94.0%); score model [Carlson + AT III] = 0.1980 * (Charlson) − 0.0227 * (AT III), cut-off value > − 1.5922 (J-index = 0.4891), Se 80.5% (CI95% 65.1–91.2%), Sp 68.4% (CI95% 51.3–82.5%). The AUCs for single and adjusted models were, respectively, 0.662 (CI95% 0.547–0.765, p-value 0.0092) and 0.776 (CI95% 0.668–0.862, p-value < 0.0001). The adjustment for the Charlson Comorbidity Index showed a statistically significant difference for the AUC (p-value 0.0053). When the Charlson Comorbidity Index and adjusted model scores ROC curves were compared, the p-value was not statistically significant (0.8428)