Tipping points in cardiac surgical performance: the application of dynamic risk prediction modelling


We compare dynamical, periodically refitted and static cross-sectional multiple logistic regression models for fitting prediction models for in-hospital mortality following cardiac surgery that adjust for case-mix in a heterogeneous patient population. Data from 300,000 consecutive procedures performed at all NHS and some private hospitals in England and Wales between April-2001 to March-2011 were extracted from the Society for Cardiothoracic Surgery in Great Britain and Ireland database. The study outcome was in-hospital mortality. Covariate adjustment was made in each model using risk factors included in the logistic EuroSCORE model. The association between in hospital mortality and the risk factors varied with time. Notably, the intercept coefficient has been steadily decreasing over the study period consistent with decreasing observed mortality. Some risk factors such as operative urgency and post infarct ventricular septal defect have been relatively stable overtime, whilst others such as left ventricular dysfunction and surgery on the thoracic aorta have been associated with lower risk relative to the static model. It is known that prediction models can lose calibration. Periodic model updating is necessary but may be better implemented using a less arbitrary modelling approach such as dynamical modelling.

In: Bissell J, Caiado C, Goldstein M, Straughan B, Curtis SE (Eds.), Tipping Points: Modelling Social Problems and Health, Part II: Mathematical Modelling in Healthcare. Wiley: London. Chapter 4.