Abstract
Given a prognostic model based on one population, one may ask: Can this model be used to accurately predict disease in a different population? When the underlying rate of disease differs in the new population, the model must be calibrated. van Houwelingen (Citation2000) considered this calibration problem focusing on proportional hazards models. We extend the validation by calibration to the log-logistic accelerated failure time model. We use calibration of proportional hazards models and log-logistic accelerated failure time models to examine whether a survival model based on the Framingham Heart Study can be applied to diverse studies around the world.
Mathematics Subject Classification:
Acknowledgments
Jeannete Simino's research was partially funded by a pre-doctoral fellowship from the Florida/Puerto Rico affiliate of the American Heart Association. Myles Hollander's and Dan McGee's research were partially funded by grants DK 52329 and HL 67460 from the National Institutes of Health.
Notes
*Reject the cohort-specific hazard ratio test at the 0.05 level.
*Reject the cohort-specific acceleration factor test at the 0.05 level.