Validated COVID-GRAM clinical risk calculator for optimizing management of critical illness

  • Liang W, & et al.
  • JAMA Intern Med
  • 13 may. 2020

  • de Liz Scherer
  • Clinical Essentials
El acceso al contenido completo es sólo para profesionales sanitarios registrados. El acceso al contenido completo es sólo para profesionales sanitarios registrados.

Takeaway

  • The COVID-GRAM web-based calculator offers an estimate of individual patient risk for developing severe illness.
  • Risk score predictors: chest radiography abnormality (CXR), age, hemoptysis, dyspnea, unconsciousness, number of comorbidities, cancer history, neutrophil/lymphocyte ratio (NLR, 0-80), lactate dehydrogenase (0-1500 µL), direct bilirubin (0-24 µmol/L).
  • Risk calculator accuracy is 0.88.

Why this matters

  • Accurate identification of who is at highest risk for severe disease could greatly aid clinical management, patient counseling, and resource use.
  • Risk score is based on 10 independent, validated predictors.

Key results

  • Development cohort: 1590 patients, mean (standard deviation [SD]) age 48.9 (15.7) years, 57.3% (n=904) male. 
  • Logistic analysis identified 10/72 variables independently/significantly predictive (ORs) of critical illness:
    • Unconsciousness: 4.71 (P=.01).
    • Hemoptysis: 4.53 (P=.01).
    • Cancer history: 4.07 (P=.02).
    • CXR abnormality: 3.39 (P<.001>
    • Dyspnea: 1.88 (P=.01).
    • Comorbidities (number): 1.60 (P<.001>
    • Direct bilirubin: 1.15 (P=.001).
    • NLR: 1.06 (P=.003).
    • Age: 1.03 (P=.002).
    • Lactate dehydrogenase: 1.002 (P<.001>
  • Validation cohort: 710 patients, mean (SD) age 48.2 (15.2) years, 53.8% (382) men.
  • Mean area under the curve: 0.88 (95% CI, 0.85-0.91; P<.001>

Study design

  • Retrospective nationwide cohort analysis of epidemiological, clinical characteristics associated with developing critical COVID-19 illness to construct a risk prediction score.
  • Funding: China National Science Foundation.

Limitations

  • Modest sample size.
  • Limited generalizability.