AUTHOR=Xu Jiqian , Yang Xiaobo , Huang Chaolin , Zou Xiaojing , Zhou Ting , Pan Shangwen , Yang Luyu , Wu Yongran , Ouyang Yaqi , Wang Yaxin , Xu Dan , Zhao Xin , Shu Huaqing , Jiang Yongxiang , Xiong Wei , Ren Lehao , Liu Hong , Yuan Yin , Qi Hong , Fu Shouzhi , Chen Dechang , Zhang Dingyu , Yuan Shiying , Shang You TITLE=A Novel Risk-Stratification Models of the High-Flow Nasal Cannula Therapy in COVID-19 Patients With Hypoxemic Respiratory Failure JOURNAL=Frontiers in Medicine VOLUME=Volume 7 - 2020 YEAR=2020 URL=https://www.frontiersin.org/journals/medicine/articles/10.3389/fmed.2020.607821 DOI=10.3389/fmed.2020.607821 ISSN=2296-858X ABSTRACT=Background: High-flow nasal cannula (HFNC) has been recommended as a suitable choice for the management of COVID-19 patients with acute hypoxemic respiratory failure (AHRF) before mechanical ventilation (MV), however, delaying MV with HFNC therapy is still a dilemma between the technique and clinical management during the ongoing pandemic. Methods: Retrospective analysis of COVID-19 patients treated with HFNC therapy from 4 hospitals of Wuhan, China. Demographic information, clinical variables before, at, and shortly after HFNC initiation were collected and analyzed. A risk-stratification models of HFNC failure (the need for MV) was developed with the 324 patients of Jin Yin-tan Hospital, and validated its accuracy with 69 patients of other hospitals. Results: Among the training cohort, the median duration of HFNC therapy was 6 (range 3-11) and 147 experienced HFNC failure within 7 days of HFNC imitation. Early predictors of HFNC failure on the basis of a multivariate regression analysis included age older than 60 years (OR 1.93; 95%CI 1.08-3.44, p = 0.027; 2 points), respiratory rate-oxygenation index (ROX) < 5.31 (OR 5.22; 95%CI 2.96-9.20, p<0.001; 5 points) within the first 4 hours of HFNC initiation, platelets < 125 × 10⁹ /L (OR 3.04; 95%CI 1.46-6.35; p=0.003; 3 points) and interleukin-6 (IL-6) > 7.0 pg/mL (OR 3.34; 95%CI 1.79-6.23; p<0.001; 3 points) at HFNC initiation. A weighted risk-stratification model of these predictors showed sensitivity of 80.3%, specificity of 71.2% and a better predictive ability than ROX index alone (AUC = 0.807 vs. 0.779, p < 0.001). Six points was used as a cut-off value for the risk of HFNC failure stratification. The HFNC success probability of patients in low risk group (84.2%) was 9.84 times than that in high risk group (34.8%). In the subsequent validation cohort, the AUC of the model were 0.815 (0.71-0.92). Conclusions: Aged patients with lower ROX index, thrombocytopenia and elevated IL-6 values are at increased risk of HFNC failure. The risk-stratification models accurately predicted the HFNC failure and early stratified COVID-19 patients with HFNC therapy into relevant risk categories.