Article Text

Download PDFPDF
P256 Prognostication in COVID-19: a prospectively derived and externally validated risk prediction score for in-hospital death
  1. F Chua1,
  2. A Draper1,
  3. M Knight2,
  4. R Mogal2,
  5. J Singh2,
  6. LG Spencer3,
  7. E Thwaite3,
  8. T Vaghela2,
  9. H Mitchell2,
  10. S Calmonson2,
  11. N Mahdi2,
  12. S Assadullah2,
  13. M Leung2,
  14. A O’Neill2,
  15. C Popat2,
  16. R Kumar2,
  17. S Raghunath3,
  18. TJ Humphries3,
  19. R Talbutt3,
  20. M Schechter2,
  21. J Lowe2,
  22. A Barlow2,
  23. R Vancheeswaran2
  1. 1Royal Brompton Hospital, London, UK
  2. 2West Hertfordshire NHS Trust, Watford, UK
  3. 3Liverpool University Hospitals NHS Foundation Trust, Liverpool, UK


Introduction and Objectives The disease spectrum of COVID-19 ranges from mild viral illness to devastating lung injury that heralds the acute respiratory distress syndrome. Different risk factors of adverse outcomes have been identified but prospectively stratified and externally validated studies of prognosis are lacking. We set out to identify independent predictors of mortality and to develop and validate a clinically applicable risk prediction model of COVID-19.

Methods 983 consecutive patients with COVID-19 were prospectively recruited over an 11-week period for an outcome of in-hospital death. Multiple imputation was used to address randomly missing data. 12 independent mortality predictors were identified by multivariate regression and internally validated by bootstrapping. A prognostic score was constructed and validated in an external cohort (N=277) and assessed for predictive accuracy including goodness-of-fit by the Hosmer-Lemeshow test.

Results The median age of the derivation cohort was 70 (IQR: 53-83). Among non-survivors (29.9%; 294/983), the highest odds ratios for death (with 95% confidence intervals) were age >70 (7.65; 4.89–11.98; P<0.001), BMI >30 (2.39; 1.88–3.03; P<0.001), baseline hypoxia (2.24; 1.78–2.79; P<0.001), chronic kidney disease stage 5 (2.00; 1.18–3.41; P<0.05) and tachypnoea (1.79; 1.43–2.24; P<0.001). White ethnicity accounted for 85% of all non-survivors (P<0.01 vs. non-White ethnicities). Care home residency was associated with an increased risk of COVID-19 death on univariate analysis (OR 3.14; 95% CI: 2.28–4.32). A linear relationship between increasing COVID-19 severity and in-hospital mortality was derived from the development dataset. Evaluation of a risk score (ranging 1–19 points) disclosed good discriminatory ability (area under the receiver operating characteristic 0.855), sensitivity (59.7%), specificity (87.6%), positive predictive (70.2%) and negative predictive value (81.6%). Subsequent validation of the score in an age and mortality-matched independent cohort showed robust performance parameters: accuracy/AUC 0.797, calibration slope (R2) of 0.882 (see calibration belt figure 1).

Conclusions Integration of key variables including age, indices of acute respiratory illness and comorbidities into a clinical risk score allows in-hospital death due to COVID-19 to be reliably predicted. The ability to risk stratify may help frontline clinicians in decision processes in respect of escalation and de-escalation strategies during resurgent COVID-19.

Statistics from

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.