AUTHOR=Meng Zirui , Wang Minjin , Zhao Zhenzhen , Zhou Yongzhao , Wu Ying , Guo Shuo , Li Mengjiao , Zhou Yanbing , Yang Shuyu , Li Weimin , Ying Binwu 

TITLE=Development and Validation of a Predictive Model for Severe COVID-19: A Case-Control Study in China

JOURNAL=Frontiers in Medicine

VOLUME=Volume 8 - 2021

YEAR=2021

URL=https://www.frontiersin.org/journals/medicine/articles/10.3389/fmed.2021.663145

DOI=10.3389/fmed.2021.663145

ISSN=2296-858X

ABSTRACT=Background: Comprehensively understanding and early predicting the risk of progression to severe COVID-19 can facilitate personalized diagnosis and treatment options, optimizing utilization of medical resource.
Methods: In this prospective study, 206 COVID-19 patients were enrolled from regional medical institutions between 20th December 2019 and 10th April 2020. Demographics, clinical characteristics, laboratory findings and cytokine levels were adequately described and analyzed for deriving and validating a predictive model. Variation analysis, lasso and boruta algorithm were used for modeling. Performances of models were evaluated by specificity, sensitivity, AUC, AIC, calibration plot, Decision Curve Analysis e and Hosmer-Lemeshow test.
Results: Predictive model including ALT, IL6, Expectoration, Fatigue, LYMR, AST and CREA established by using LASSO algorithm and logistic regression can predict the progression risk of severe COVID-19 accurately. It yielded satisfactory predictive performance with an AUC of 0.9104 and 0.8792 in derivation and validation cohort, respectively. This model finally was visualized in the form of a nomogram plot and packaged into an open-source and free predictive calculator for clinical use easily and is available online at https://severeconid-19predction.shinyapps.io/SHINY/. 
Conclusion: The model combining demographics, clinical characteristics, laboratory findings and cytokine levels can effectively predict the progression to severe COVID 19, promoting early personalized management and the allocation of appropriate medical resources.