Predictive Value Of Chest CT Combined With Clinical Data In Severity And Mid-and-long Term Prognosis Of COVID-19 Pneumonia Patiens

Part Ⅰ:The role of chest CT findings and clinical data in predicting the severity of COVID-19 pneumonia inpatientsObjective:To compare the accuracies of quantitative computed tomography(CT)parameters and CT semiquantitative visual score in predicting severity of coronavirus disease 2019(COVID-19)inpatients,and to establish a prediction model for the severity of COVID-19 pneumonia inpatients by combined chest CT findings and clinical data.Materials and Methods:187 COVID-19 pneumonia inpatients were retrospectively enrolled.Patients were divided into group A(mild/moderate)and group B(severe/critical)according to the severity.Demographic data,laboratory findings,and subjective CT imaging features were collected.The CT semiquantitative visual score was used to estimate the lesion extent.The volume and CT values of whole lung,ground-glass opacity(GGO)and consolidation were accurately quantified by 3D Slicer software(version 4.10.2,https://www.slicer.org/),and the percentage of lesion volume in whole lung volume was calculated.The differences of clinical data and imaging data between group A and group B were compared.Receiver operating characteristic(ROC)curve was used to compare the accuracies of quantitative CT findings,CT semiquantitative visual score,and clinical data in predicting the severity of COVID-19 pneumonia inpatients.Logistic regression was used to establish a predictive model for the severity of COVID-19 pneumonia inpatients based on combined chest CT findings and clinical data.P<0.05 was considered statistically significant.Results:There were 59 patients(30 males,age 55.40±2.30)in group A and 128 patients(65 male,age 57.90±1.30)in group B.All quantitative CT findings and CT semiquantitative visual scores related to lesions volume in group B were significantly higher than those in group A(P<0.001).Among the parameters used to evaluate the severity COVID-19 pneumonia inpatients,the area under the curve(AUC)of the percentage of lesions was the largest(0.807;95%confidence interval,0.744～0.861:P<0.001),and the AUC of the quantitative CT findings was significantly larger than that of the CT semiquantitative visual score(P<0.05).Multivariate Logistic regression showed that the percentage of GGO,percentage of lesion and interleukin-6 were independent risk factors for the severity of COVID-19 pneumonia inpatients,with OR values of 6.125,6.026 and 3.039,respectively.The classification accuracy of this predictive model was 82.89%.Conclusion:The predictive accuracy of quantitative CT findings was significantly superior to that of CT semi quantitative visual score and clinical laboratory findings in evaluating the severity of COVID-19 pneumonia inpatients,especially the quantitative CT findings related to lesion volume were better.For the severity of COVID-19 pneumonia inpatients,the predictive model of quantitative CT findings combined with interleukin-6 has a good performance.Part Ⅱ:Mid-and-long term longitudinal changes of chest CT findings and predictive analysis of post-discharge dyspnea in COVID-19 pneumonia patientsObjective:To analyze mid-and-long term longitudinal changes of chest CT findings in COVID-19 pneumonia patients and to evaluate the role of chest CT findings and clinical data in predicting persistent dyspnea in COVID-19 pneumonia patients after discharge.Methods:The retrospective study included 337 COVID-19 pneumonia patients who underwent CT scans during hospitalization,at discharge,and between 3 months to 1 year after onset.Subjective CT imaging features,lesion volume(measured by 3D Slicer software),therapeutic measures during hospitalization,and laboratory findings at peak disease were collected.The severity of the dyspnea in COVID-19 pneumonia patients was determined by a follow-up questionnaire.The mid-and-long term longitudinal changes of the CT features from the peak period to discharge to follow-up were analyzed,and the predictive ability of CT findings and clinical data for patients with or without dyspnea were evaluated by ROC curve.Results:From the peak period to discharge,and then to follow-up,the lesion volume gradually decreased(451.55 cm3,298.27 cm3,and 54.17 cm3,respectively),the residual lesions at follow-up were mainly GGO and interstitial abnormalities.Compared with the patients less than six months after discharge,the lesion extent of the patients more than six months after discharge was further reduced(P<0.05).Of the 337 COVID-19 pneumonia patients,91 remained dyspnea at follow-up.Among the quantitative CT findings,subjective CT features and clinical data used to predicting dyspnea in COVID-19 pneumonia patients,the largest AUC was the lesion volume at discharge(0.820;95%confidence interval,0.772～0.867),interstitial abnormalities at follow-up(0.752;95%confidence interval,0.693～0.811)and C-reactive protein(0.746;95%confidence interval,0.681～0.811),respectively.The AUC of quantitative CT findings was significantly larger than that of subjective CT features and clinical data.Conclusions:From the peak period to discharge to follow-up,the chest lesions of COVID19 pneumonia patients were gradually absorbed,and most of the lesions were absorbed after discharge.The lesions were still absorbing six months later after discharge.Quantitative CT findings related to lesion volume were superior to conventional chest CT features and clinical data in predicting persistent dyspnea in COVID-19 pneumonia patients after discharge,of which lesion volume at discharge was the most accurate.Part Ⅲ:The role of chest CT findings and laboratory findings in predicting pulmonary fibrosis in COVID-19 pneumonia patientsObjective:Laboratory findings and chest CT findings were compared between COVID-19 pneumonia patients with and without pulmonary fibrosis and between lobes with and without pulmonary fibrosis.To identify laboratory findings or CT imaging findings that can better predict pulmonary fibrosis in patients or lobes.Methods:Forty-six COVID-19 pneumonia patients were enrolled,all of whom underwent CT scans during hospitalization,six months after onset,and more than one year after onset.Laboratory findings during hospitalization,as well as subjective CT imaging features and CT semi quantitative visual scores for the patients and each lobe were collected.The lesions of patients and lobes were segmented with 3D Slicer software,and then histogram features of the lesions were extracted with the Pyradiomics module.Pulmonary fibrosis is defined as parenchymal band,subpleural curvilinear line,reticular pattern,traction bronchiectasis or honeycombing on chest CT one year after onset.The collected parameters were compared between patients with and without fibrosis and between the lobes with and without fibrosis.The accuracy of laboratory findings and chest CT findings in predicting pulmonary fibrosis was analyzed by ROC curve.Results:Pulmonary fibrosis occurred in 15 of 46 patients and in 51 of 230 lobes,with the highest incidence of pulmonary fibrosis in bilateral lower lobes(both 28.26%).When predicting pulmonary fibrosis in patients,the CT semi quantitative visual score at 6-month follow-up was the most accurate(0.928;95%confidence interval,0.852～1.000;cutoff value of 3),the sensitivity of D-dimer was the highest(100%),and the specificity of bronchiectasis at 6-month follow-up was the highest(100%).Among the CT findings of the patients during hospitalization,the AUC of the CT semiquantitative visual score was the largest(0.746;95%confidence interval,0.596～0.897);among the laboratory findings,the AUC of Interleukin-6 was the largest(0.802;95%confidence interval,0.649～0.956).When predicting pulmonary fibrosis in lobes,the CT semiquantitative visual score at 6-month follow-up was the most accurate(0.923;95%confidence interval,0.888～0.958;cutoff value of 1),and its sensitivity was the highest(100%).The interstitial abnormalities at 6-month follow-up had the highest specificity(93.92%).Among the CT findings of the lobes during hospitalization,the AUC of the CT semiquantitative visual score was the largest(0.746;95%confidence interval,0.667～0.815).Conclusions:CT semiquantitative visual score during hospitalization and 6-month followup,and Interleukin-6 are good indicators for predicting pulmonary fibrosis in COVID-19 pneumonia patients,among which the CT semiquantitative visual score at 6-month followup has the highest accuracy.The probability of fibrosis is higher in the lobes with lesions at 6-month follow-up,especially in thoes with interstitial abnormalities.