Abstract P233: Estimating Rates of Cardiovascular Death Using Royston-Parmar Survival Models: An Example in a High Risk Patient Population in Sweden
INTRODUCTION: The aim of this retrospective cohort study was to estimate the hazard rates of cardiovascular (CV) death in a high risk population in Sweden as inputs to a cost effectiveness model. We evaluated the use of Royston-Parmar (RP) flexible parametric models to estimate time-varying event risks.
HYPOTHESIS: We assessed the hypothesis that due to time-varying risks of patients with acute coronary syndrome, the RP model is a suitable method to estimate the hazard and event rate of CV death.
METHODS: De-identified primary care electronic medical records were obtained for a 10% sample of the adult (>18yrs) Swedish population from 2003 – 2011 (N = 49,857). Patients with incident acute coronary syndrome were selected and followed through time. The hazard of CV death was estimated for each cohort overall as well as for the sub-populations of high dose statin users (>40 mg simvastatin or equivalent) and non-statin users. RP flexible (spline-based) parametric survival models were used to estimate the time-varying hazard of CV death.
RESULTS: Figure 1 shows an example of output from the RP model showing the hazard of CV death in each month post-ACS event in high-dose statin users and non-statin users, separately. In the top two panels, the flexibility of 3 knots clearly shows short term high acute risk in the first year followed by a period of lower risk roughly 1-3 years after the initial event, and a long term stable risk over time for high dose statin users. Among non-statin users, the risk of CV death is high but dropping in the first year, and rises more dramatically in the medium and long-term time periods. The lower two panels show consistent increased risk of CV death for diabetic patients over time.
CONCLUSION: RP splines are a more flexible alternative to the Cox model and are useful for generating cause-specific event rates over time. They are particularly useful when the assumption of proportional hazards is not met and thus traditional Cox methods may not be appropriate.
Author Disclosures: J. Sandberg: A. Employment; Significant; Greater than 10,000 in research funds from Amgen Inc. L. Jorgensen: A. Employment; Significant; Greater than 10,000 in research funds from Amgen Inc. M. Danese: A. Employment; Significant; Greater than 10,000 in research funds from Amgen Inc. B. Taylor: A. Employment; Significant; Amgen Inc pays 100% of Salary. R. Kilpatrick: A. Employment; Significant; GSK Pays 100% of salary. P. Sobocki: A. Employment; Significant; Greater than 10,000 in research funds from Amgen Inc..
- © 2015 by American Heart Association, Inc.