Comparison of performance of different models for prediction of total stroke and stroke pathological types in men and women in China Kadoorie Biobank
| Men | Women | |||||
| Discrimination | ΔAUC | Calibration | Discrimination | ΔAUC | Calibration | |
| AUC (95% CI) | P value | χ2 | AUC (95% CI) | P value | χ2 | |
| Total stroke | ||||||
| 2017 FSRP | 0.78 (0.77 to 0.79) | – | 1825 | 0.77 (0.76 to 0.78) | – | 3053 |
| +Recalibration | 0.78 (0.77 to 0.79) | – | 156 | 0.77 (0.76 to 0.78) | – | 506 |
| +Refitting | 0.79 (0.79 to 0.80) | +0.01 (<0.001) | 51 | 0.78 (0.77 to 0.78) | +0.01 (<0.001) | 148 |
| +Area stratification | 0.82 (0.82 to 0.83) | +0.04 (<0.001) | 124 | 0.82 (0.82 to 0.83) | +0.05 (<0.001) | 178 |
| +Additional risk factors | 0.83 (0.82 to 0.84) | +0.05 (<0.001) | 101 | 0.83 (0.82 to 0.84) | +0.06 (<0.001) | 177 |
| Ischaemic stroke | ||||||
| 2017 FSRP | 0.77 (0.76 to 0.78) | – | 1200 | 0.76 (0.76 to 0.77) | – | 2406 |
| +Recalibration | 0.77 (0.76 to 0.78) | – | 118 | 0.76 (0.76 to 0.77) | – | 479 |
| +Refitting | 0.78 (0.78 to 0.79) | +0.01 (<0.001) | 21 | 0.77 (0.76 to 0.78) | +0.01 (<0.001) | 74 |
| +Area stratification | 0.82 (0.81 to 0.83) | +0.05 (<0.001) | 70 | 0.82 (0.82 to 0.83) | +0.06 (<0.001) | 124 |
| +Additional risk factors | 0.83 (0.82 to 0.84) | +0.06 (<0.001) | 55 | 0.83 (0.82 to 0.84) | +0.07 (<0.001) | 90 |
| Haemorrhagic stroke | ||||||
| 2017 FSRP | 0.79 (0.78 to 0.81) | – | 136 | 0.78 (0.76 to 0.80) | – | 70 |
| +Recalibration | 0.79 (0.78 to 0.81) | – | 58 | 0.78 (0.76 to 0.80) | – | 65 |
| +Refitting | 0.80 (0.78 to 0.81) | +0.01 (0.007) | 23 | 0.80 (0.78 to 0.82) | +0.02 (<0.001) | 33 |
| +Area stratification | 0.81 (0.80 to 0.83) | +0.02 (<0.001) | 22 | 0.81 (0.80 to 0.83) | +0.03 (<0.001) | 11 |
| +Additional risk factors | 0.82 (0.81 to 0.84) | +0.03 (<0.001) | 14 | 0.82 (0.80 to 0.84) | +0.04 (<0.001) | 9 |
ΔAUC values were calculated as changes from the 2017 FSRP. Modifications to models were applied cumulatively. 2017 FSRP: 2017 FSRP. +Recalibration: Baseline hazard functions re-estimated in China Kadoorie Biobank.+Refitting: Model coefficients from FSRP re-estimated in China Kadoorie Biobank.+Area stratification: Stratification by study area and area-specific estimation of baseline hazard functions.+Additional risk factors: Further sociodemographic, health and lifestyle risk factor indicators selected using LASSO regularisation.
AUC, area under the curve; FSRP, Framingham Stroke Risk Profile; LASSO, least absolute shrinkage and selection operator.