Although there has been considerable attention paid to the prognostic significance of the heart rate rise during exercise, only recently has it been noted that the heart rate fall after exercise, or "heart-rate recovery," may be an even more powerful predictor of outcome. Heart-rate recovery after exercise is a consequence of central reactivation of vagal tone. As impaired parasympathetic function has been associated with increased risk of death, our hypothesis is that an impaired heart-rate recovery is a powerful and independent predictor of mortality. Our group recently published the first report linking heart-rate recovery to long-term mortality in the New England Journal of Medicine. The proposed project will extend upon that report by determining the optimal definition of heart-rate recovery, by showing that an abnormal heart-rate recovery is an independent predictor of mortality in diverse patient groups, and by developing accurate survival models that incorporate heart-rate recovery. The overall aim of this project is to use heart-rate recovery to substantially improve the prognostic value of the exercise test. The specific aims of this project are: 1) Derive biologically meaningful mathematical models of heart-rate recovery. Data from over 20,000 patients who have undergone exercise testing at Cleveland Clinic Foundation between 1990 and 1998 will be used; all of these patients had their tests performed on exercise workstations which recorded heart rates every 10 seconds during and after exercise. Heart-rate recovery measures will be the difference between heart rate at peak exercise and heart rate at different points during recovery. Modeling will be based on exponential families, using stepwise selection, bootstrapping, and information theory approaches. Correlates of different patterns of heart rate recovery will be determined. 2) Using the results of modeling of heart-recovery derived from the work in Specific Aim 1, determine a prognostically defined optimal definition of abnormal heart rate recovery and demonstrate that an abnormal heart rate recovery is a powerful and independent predictor of mortality in diverse patient groups. Data from exercise tolerance tests of over 40,000 patients studied at the Cleveland Clinic Foundation between 1990 and 1999 will be analyzed. Statistical methods to be used will include the nonparametric Kaplan-Meier product limit method and the Cox proportional hazards model with bootstrap validation, which will include use of the random forest technique. 3) Using completely parametric techniques. develop predictive survival models in which heart-rate recovery is included along with clinical data and other exercise findings, including exercise capacity and heart rate changes during exercise. The advantages of the parametric technique include: a) it allows for modeling of nonproportional hazards that may permit differential strength of effect at different follow-up times for different sets of risk factors; b) it generates absolute risk, not just relative risk; and c) it permits patient-specific prediction. All the data needed for this project are already electronically available; that fact, along with the work already done, lends confidence to the feasibility of this project.