Predictive Performance of Geoadditive Survival Models to Study Geographical Patterns in Coronary Heart Disease

Acute coronary syndrome (ACS) represents the most common cause of death in the western world. Numerous prediction models exist for the different types of ACS. Most of these models have been developed from large populations by means of the classical (parametric) Cox proportional hazard model, in which the geographic area has not been taken into account as a health determinant. However, this statistical Cox model may not be enough to capture some flexible effects of covariates on survival, and does not allow to include spatial effects. In this study, we used flexible extensions of the Cox model, such as Structured Geoadditive Survival Models, to evaluate geographical inequalities in survival of patients admitted to a tertiary hospital, with a diagnosis of ACS. The predictive performance of the survival models were assessed through time-dependent Receiver Operating Characteristic (ROC) curves computed by the incident sensitivity and dynamic specificity for each time point.