Comparison of observed mortality by 4C mortality risk group for recover dataset of SARS-CoV-2+ patients hospitalised from the emergency department
| 4C mortality risk group (score range) | RECOVER dataset with CRP | RECOVER dataset without CRP | ||||
| Mortality predicted by 4C* | Observed mortality | Prediction error | Mortality predicted by 4C* | Observed mortality | Prediction error | |
| Overall | 22.9% | 16.9% | 6.0% | 17.6% | 16.9% | 0.7% |
| Low (0–3 points) | 1.4% | 0.8% | 0.6% | 1.2% | 1.6% | −0.5% |
| Intermediate (4–8 points) | 9.7% | 5.3% | 4.4% | 9.5% | 8.7% | 0.8% |
| High (9–14 points) | 29.9% | 22.3% | 7.6% | 28.6% | 27.4% | 1.2% |
| Very high (≥15 points) | 60.2% | 50.6% | 9.6% | 56.8% | 58.2% | −1.4% |
| AUROC | ||||||
| 4C validation | 0.763 (0.757–0.769) | √Calibration error | √Brier Score | |||
| RECOVER dataset with CRP | 0.786 (0.773–0.799) | 0.066 | 0.350 | |||
| RECOVER dataset without CRP | 0.776 (0.762–0.79) | 0.017 | 0.346 | |||
*The mortality predicted by 4C is constant for each individual score, but when the scores are grouped into ranges (as they are here), the predicted mortality varies based on the proportion of patients from the test dataset with each individual score within the range.
CRP, C reactive protein; RECOVER, REgistry of suspected COVID-19 in EmeRgency care.