Table 3. Sensitivity, specificity, PPV, NPV, and accuracy of clinical, radiology, and AI findings in predicting the critical cases.

Findings	Sensitivity	Specificity	PPV	NPV	Accuracy
AI − %CL	55.6 [40.00 to 70.36]	68.9 [53.35 to 81.83]	64.1 [51.81 to 74.78]	60.78 [51.43 to 69.41]	62.2 [51.38 to 72.23]
Radiologist-%CL	53.3 [37.87 to 68.34]	64.4 [48.78 to 78.13]	60.0 [48.16 to 70.77]	58.0 [48.56 to 66.89]	58.9 [48.02 to 69.16]
SpO2 < 88%	69.8 [53.87 to 82.82]	90.9 [78.33 to 97.47]	88.4 [74.27 to 95.12]	75.5 [65.93 to 83.03]	80.5 [70.57 to 88.19]
Immunocompromised	24.4 [12.88 to 39.54]	95.6 [84.85 to 99.46]	84.6 [56.36 to 95.91]	55.8 [51.43 to 60.17]	60.0 [49.13 to 70.19]
Comorbidities	60.0 [44.33 to 74.30]	64.4 [48.78 to 78.13]	62.8 [51.58 to 72.78]	61.7 [51.46 to 71.00]	62.2 [51.38 to 72.23]
Clinical model(SpO2 < 88%, Immunocompromised, Comorbidities)	79.1 [63.96 to 89.96]	86.4 [72.65 to 94.83]	85.0 [72.62 to 92.37]	80.8 [70.01 to 88.42]	82.8 [73.16 to 90.02]
Radiology and clinical model	76.7 [63.96 to 89.96]	88.9 [75.44 to 96.21]	86.8 [74.60 to 94.03]	80.0 [70.60 to 88.66]	82.9 [74.48 to 90.91]
AI and clinical model	79.1 [61.37 to 88.24]	88.6 [78.33 to 97.47]	87.2 [76.16 to 95.52]	81.2 [69.75 to 87.40]	83.9 [74.48 to 90.91]

PPV: positive predictive value; NPV: negative predictive value; AI = artificial intelligence; %CL = fraction of compromised lung parenchyma; SpO2 = oxygen saturation.