Table 2 Measures of model performance in predicting 28-day risk of COVID-19 mortality, using the “selected” models
Approach; measures of infection prevalence included; model form | Validation period | COVID-19 deaths/sample size | C-statistic | Observed mean risk (%) | Predicted mean risk (%) | Calibration | |
|---|---|---|---|---|---|---|---|
Intercept (95% CI) | Slope (95% CI) | ||||||
(A), none, Cox | 1 | 455/11,972,492 | 0.924 | 0.0038 | 0.0038 | 0.00 (−0.10, 0.09) | 0.95 (0.86, 1.05) |
2 | 4471/11,955,296 | 0.934 | 0.0374 | 0.0038 | 2.30 (2.27, 2.32) | 1.02 (0.99, 1.05) | |
3 | 1246/11,942,608 | 0.941 | 0.0104 | 0.0037 | 1.03 (0.97, 1.08) | 1.05 (1.00, 1.11) | |
(B), modelled estimates, Poisson | 1 | 455/11,972,492 | 0.925 | 0.0038 | 0.0044 | −0.15 (−0.24, −0.06) | 0.93 (0.84, 1.02) |
2 | 4471/11,955,296 | 0.937 | 0.0374 | 0.0354 | 0.06 (0.03, 0.09) | 1.00 (0.97, 1.03) | |
3 | 1246/11,942,608 | 0.944 | 0.0104 | 0.0128 | −0.20 (−0.26, −0.15) | 1.03 (0.98, 1.09) | |
(B), A&E COVID-19 attendances, Poisson | 1 | 455/11,972,492 | 0.921 | 0.0038 | 0.0145 | −1.34 (−1.43, −1.25) | 0.92 (0.83, 1.02) |
2 | 4471/11,955,296 | 0.933 | 0.0374 | 0.0420 | −0.12 (−0.15, −0.09) | 0.99 (0.96, 1.02) | |
3 | 1246/11,942,608 | 0.943 | 0.0104 | 0.0197 | −0.64 (−0.69, −0.58) | 1.05 (0.99, 1.10) | |
(B), Suspected COVID-19 in primary care, Poisson | 1 | 455/11,972,492 | 0.921 | 0.0038 | 0.0085 | −0.80 (−0.89, −0.71) | 0.90 (0.81, 1.00) |
2 | 4471/11,955,296 | 0.935 | 0.0374 | 0.0378 | −0.01 (−0.04, 0.02) | 1.00 (0.97, 1.03) | |
3 | 1246/11,942,608 | 0.942 | 0.0104 | 0.0156 | −0.41 (−0.46, −0.35) | 1.04 (0.98, 1.09) | |