Clinical cases data
The baseline demographic and initial symptoms of the study population are summarized in Table 1.
A total of 71 cases confirmed diagnosed with COVID-19 were retrospectively studied. 14 cases were excluded due to poor image quality or without clear CT lung abnormalities. 57 cases (38 men and 19 women, average age, 67.00 years±18.89, range from 31-97 years) were enter into the final analysis. According to the Protocols-Ver8, 36 cases (63.2%) were classified as severe type, while 21 cases (36.8%) as mild and moderate type. 15 cases (26.3%) died during hospitalization and 42 cases (73.7%) were discharged from hospital. All the discharged cases were survival at 30 days follow-up survey. The mean LOS for survival population was 35.29±18.04 days, while the mean overall survival (OS) of died population was 31.67±11.87 days.
The patients in this study population were present at least one respiratory symptoms symptom, and/or more symptoms onset of illness. The most common initial symptoms were fever (48 cases, 84.2%), cough (33 cases, 57.9%) the other initial symptoms include hypodynamia, chest tightness, muscle pain, diarrhea, headache, etc. Noting that, 5 in 7 cases present with diarrhea were died in this study population.
Chest CT findings and Semi-quantitative CT score
The frequencies of major chest CT findings, GGO and/or CLO, were analyzed in whole study population, each group (survival vs. died, severe vs. mild/moderate) of population independently, summarized in Table 2. Overall, 13 pure GGO, 8 pure CLO and 36 GGO mixed with CLO were present in the whole study population. The most highly frequency of lung abnormality was GGO mixed with CLO in both survival population (26 in 42, 61.9%) and died population (10 in 15, 66.7%). In terms of the predominance of the lung opacity, survival population were more likely to show more cases of GGO predominance over CLO (GGO vs. CLO, 61.9% vs 38.0%), while died population showed almost equal cases between GGO and CLO predominance. In survival population, the highly frequency infected lobes were at the right and left lower lobe (92.9% and 83.3%). In died population, both right and left lower lobes were infected in all 15 cases (100%). The CT severity score was significantly higher in died cases than survival cases [died vs survival cases: 7.41±4.18 vs 4.26±2.68]. Only died population shown the score 4 [right upper lobe: 3(20%); right lower lobe: 3(20%); left upper lobe: 2(13.3%); left lower lobe: 4 (26.7%)]. The frequencies and CT semi-quantitative score were also analyzed in severe, mild/moderate populations (Table 2).
Semi-quantitative CT score for clinical classification
ROC curves were presented in Fig. 2. The CT severity score has the ability to discriminate the clinical classification and outcome. The sensitivity and specificity for diagnosis severe type from mild/moderate counterparts were 88.24% and 56.52%, respectively (AUC, 0.753). For outcome, the sensitivity and specificity were 86.67% and 52.38%, respectively (AUC, 0.729).
Histogram quantitatively analysis for clinical classification
ROC curves are presented in Fig. 3 and details of each histogram parameter are presented in Table S1（Supplementary materials）. The histogram parameters have great ability to discriminate severe type from mild/moderate type COVID-19. The significant diagnostic value of histogram parameters derived from ROC analysis was as following: GGO_skewness (specificity=66.67%, sensitivity=78.12%, AUC=0.706, p=0.0094), GGO_mode(specificity=77.78%, sensitivity=59.38%, AUC=0.707, p=0.0059), GGO_mean (specificity=83.33%, sensitivity=62.5%, AUC=0.742, p=0.0008), CLO_mode (specificity=80.00%, sensitivity=57.69%, AUC=0.680, p=0.0024), and CLO_mean (specificity=70.00%, sensitivity=76.92%, AUC=0.746, p=0.001).
COX proportional hazards mode evaluates for individual risk of death prediction in COVID-19 and nomogram
The Cox proportional hazard analysis with ten variables (8 histogram parameters with age and gender) is presented in Table S2（Supplementary materials）. Significant results were obtained for GGO_kurtosis (OR=7.607 p=0.002), GGO_skewness (OR=0.122, p=0.026), GGO_mode(OR=3.566, p=0.007), GGO_mean (OR=0.254, p=0.005), CL_mode (OR=0.609, p=0.036) , and CL_mean (OR=2.016, p=0.007). The result of nomogram on the basis of Cox proportional hazard analysis was shown in Figure.4. Internal validation using the bootstrap method showed that the C-index for the model was 0.763. The calibration curve is shown in Fig. S1. It reveals no deviations from the reference line and no need for recalibration. We have shown two examples of severe type pneumonia from died and survival cohort respectively (Fig.5). A 60 years old female patient of died cohort with a total of histogram parameters ≈137 points, which corresponds to a 30-day risk of death probability of 79%, while a 57 year old male patient of survival cohort with a total of histogram parameters ≈126 points, which corresponds to a 30-day risk of death probability of 35%.
COX proportional hazards mode evaluates for individual prolonged median LOS prediction in COVID-19 and nomogram
The Cox proportional hazard analysis with ten variables (8 histogram parameters with age and gender) is presented in Table S3（Supplementary materials）. Significant result was obtained for CLO_kurtosis (OR=0.110 p=0.017). Trend result was obtained for GGO_mode (OR=0.290 p=0.071). The result of nomogram on the basis of Cox proportional hazard analysis was shown in Fig. 4. Internal validation using the bootstrap method showed that the C-index for the model was 0.888. The calibration curve is shown in Fig. S2. It reveals no deviations from the reference line and no need for recalibration.