Models And Methods For Hormone Pulse Analysis.

Abstract

A hormone pulse is a release of hormone from a gland into the blood of an organism. This event causes a relatively rapid rise, followed by a somewhat slower decline, in the serum concentration of the hormone. These fluctuations in concentration are involved in regulating basic physiological processes, such as growth and reproduction. Variations in the sizes and timing of the pulsatile release events are of interest in understanding the functioning of these regulatory mechanisms. The data required to study hormone pulse patterns are obtained in pulse bleed experiments, in which blood samples are taken from a common site of a single organism at regular time intervals and assayed for the concentrations of one or more hormones. A statistical pulse detection method is designed to estimate the timing and sizes of the pulses in such a data series. Several methods are currently in use, including those of Goodman and Karsch (1980), Van Cauter (1981), Merriam and Wachter (1982), and Clifton and Steiner (1983). In this thesis, assumptions about the nature of the biological processes involved in pulsatile hormone release are discussed, and a first-order differential equation describing the variation of hormone concentration over time is derived. A discrete time version of this equation leads to a parametric model for hormone pulse data. Based on this model the existing pulse detection methods are critiqued, and a new pulse detection method is developed. The new method involves fitting the proposed parametric model to a pulse bleed data set by alternating between (a) nonlinear parameter estimation while holding fixed the times at which pulses are estimated to occur, and (b) stepwise addition and deletion of pulse times while holding the nonlinear parameter(s) fixed. Techniques for generating simulated hormone pulse data are developed, and criteria for assessing pulse detection methods are discussed. The results of simulation experiments comparing the new method with the methods referenced above are then presented. Finally, examples of the application of the various pulse detection methods to data from actual pulse bleed experiments are presented and discussed.