Type of predictions* | Statistical model used | |||

Cox proportional hazards model† | Accelerated failure model(using generalised gamma distribution) | |||

No covariates | Covariates included | No covariates | Covariates included | |

Composite outcome‡ | ||||

True favourable predictions, n (%) (out of a total of 5970 predictions made)§ | 2658 (44.5) | 2910 (48.7) | 2658 (44.5) | 2918 (48.9) |

True unfavourable predictions, n (%) (out of a total of 5970 predictions made) | 1115 (18.7) | 1174 (19.7) | 1115 (18.7) | 1197 (20.0) |

Total number of true predictions, n (%) (out of a total of 5970 predictions made) | 3773 (63.2)§ | 4084 (68.4) | 3773 (63.2) | 4115 (68.9) |

All-cause death
| ||||

True favourable predictios, n(%) (out of a total of 5970 predictions made) | 4768 (79.9) | 4585 (76.8) | 4768 (79.9) | 4563 (76.4) |

True unfavourable predictions, n (%) (out of a total of 5970 predictions made) | 0 (0)¶ | 392 (6.57) | 0 (0) | 401 (6.72) |

Total number of true predictions, n (%) (out of a total of 5970 predictions made) | 4768 (79.9) | 4977 (83.4) | 4768 (79.9) | 4964 (83.1) |

*For each patient at time (T) equal to 3, 6 and 9 years and using the patient’s individual survival curve the predicted outcome (patient ‘alive at T’ (favourable outcome) compared with patient ‘not alive at T’ (unfavourable outcome)) was read off the survival curve and the results compared with the observed outcome. When no covariates were included in the model, the Kaplan-Meier survival curve was used to calculate the predictions.

†Time-dependent covariates were not allowed in the SAS V.9.4 program used. To allow for the fact that age violated the proportional hazard assumption we stratified by age categories (see text) in addition to centre and excluded age from the covariates.21

‡Comprising acute myocardial infarction, unstable angina pectoris, cerebrovascular disease and all-cause death.

§n is the number of true predictions made. The per cent is n over the total number of predictions made. 5970 predictions are made, namely 1998 patients with complete covariates times three time points (at 3, 6 and 9 years) minus 24 predictions not checkable due to patients lost to registries. Of these predictions, a total of 3773 were true which amounts to 3773/5970=63.2%.

¶As total mortality is <50%, all predictions are automatically favourable (and 79.9% of them pertain to a so far survivor).