COVID-19: scoring tool identifies 4 patient risk groups to guide clinical management

  • BMJ
  • 10 Sep 2020

  • curated by Liz Scherer
  • Clinical Essentials
Access to the full content of this site is available only to registered healthcare professionals. Access to the full content of this site is available only to registered healthcare professionals.

Takeaway

  • A newly validated 4C (Coronavirus Clinical Characterisation Consortium) Mortality Score prediction tool appears to accurately identify patients with high mortality risk at hospital admission for COVID-19.

Why this matters

  • 4C Mortality Score (range, 0-21 points) is based on 8 readily available variables: age, sex, number of comorbidities, respiratory rate, peripheral oxygen saturation, level of consciousness, urea level, C-reactive protein.
  • Score can guide clinical decision-making:
    • Low risk score (0-3): manage in community.
    • Intermediate (4-8): ward level monitoring.
    • High (9-14): consider aggressive treatment.
    • Very high (≥15): critical care, isolation, invasive ventilation, other supportive measures.

Key results

  • Derivation cohort: 35,463 patients; mortality rate, 32.2%. 
  • Validation cohort: 22 ,361 patients; mortality rate, 30.1%. 
  • 4C demonstrated high model discrimination vs machine learning (areas under the curve; 95% CIs):
    • Derivation cohort: 0.786 (0.781-0.790) vs 0.796 (0.786-0.807).
    • Validation cohort: 0.767 (0.760-0.773) vs 0.779 (0.772-0.785).
  • Low risk (0-3):
    • Mortality rate: 1.2%. 
    • Sensitivity: 99.7%. 
    • Negative predictive value (NPV): 98.8%.
  • Intermediate risk (4-8):
    • Mortality rate: 9.9%.
    • NPV: 90.1%.
  • High risk (9-14):
    • Mortality: 31.4%.
    • NPV: 68.6%.
  • Very high risk (≥15):
    • Mortality: 61.5%.
    • Positive predictive value: 61.5%.
  • 4C discriminated better than tools developed with other COVID-19 cohorts (areas under the curve, 0.61-0.76).

Study design

  • Prospective, observational cohort study.
  • Funding: National Institute for Health Research.

Limitations

  • Inability to assess predictive performance of other tests.
  • Selection bias.
  • Limited generalizability.