Flowgraph Models for Recurrent Events

• Activity Number: 413
• Type: Contributed
• Date/Time: Thursday, August 12, 2004 : 8:30 AM to 10:20 AM
• Sponsor: Biometrics Section
• Author(s): C. Lillian Yau*+ and Aparna V. Huzurbazar
• Companies: Tulane University and University of New Mexico
• Keywords: survival analysis ; saddlepoint approximations ; counting process ; aging ; semi-Markov model ; repeated falls

Abstract

Multistate models have gained popularity for analyzing event history data. Recurrent events are one example of such data. Recurrent events can be studied by observing the number of occurrences over a prescribed period of time, or by observing the waiting times between consecutive events. Semi-Markov models have played important roles in modeling the transition waiting times between events. The emphasis of a semi-Markov model is on specifying appropriate transition intensity functions. They are then converted to survival functions. However, intensity functions are never observed, and in practice, assumptions on these functions are simplified. Alternatively, flowgraph models relax these assumptions. They model the waiting time distributions directly. They are naturally apt for multivariate survival data. Flowgraphs operate on moment-generating functions. Saddlepoint approximations are employed to convert the MGFs to waiting time distributions, survival functions, and hazard funcitons. We illustrate the methods by applying a series flowgraph model to data collected from New Mexico Aging Process Study. The data consist of number of falls and survival time between falls.