Two Survival Tree Models for Myocardial Infarction Patients

for providing the data set and clearing the way through the jungle of medical terminology as well as for their input on the medical feasibility of the models. We would also like to thank our colleague Ursula Berger for her helpful hints and suggestions on parts of the paper. 2 Summary In the search of a better prognostic survival model for post-acute myocardial infarction patients, the scientists at the Technical University of Munich's " Klinikum rechts der Isar " and the German Heart Center in Munich have developed some new parameters using 24-hour ECG (Schmidt et al 1999). A series of investigations were done using these parameters on different data sets and the Cox – PH model (Schmidt et al 1999, Ulm et al 2000). This paper is a response to the discussion paper by Ulm et al (2000), which suggests a Cox model for calculating the risk stratification of the MPIP data set patients including the predictors ejection fraction and heart rate turbulence. The current paper suggests the use of the classification and regression trees technique for survival data in order to deduct a survival stratification model for the NIRVPIP data set. Two models are compared: one contains the variables suggested by Ulm et al (2000) the other model has two additional variables, namely presence of couplets and number of extra systolic beats in the longest salvo of the patient's 24-hour ECG. The second model is shown to be an improvement on the first one.