Prediction and significance of the temporal pattern of hormone secretion in disease states

G Brabant,K Prank
DOI: https://doi.org/10.1002/0470870796.ch7
Abstract:Comparison of the temporal pattern of hormone secretion in health and disease reveals distinct differences in many systems. Analysis of these visually apparent differences conventionally rests on computer-assisted programs based on either model assumptions, or estimations of hormonal decay rates or threshold values, all of which may not accurately reflect physiological and/or pathophysiological states. Only recently have new methods evolved which are independent of preexisting knowledge of the system under study. Apart from the widely used approximate entropy statistic (ApEn), a measure for the regularity of a time-series, artificial neural networks are able to capture temporal structures in endocrine rhythms without any previous assumptions. In particular, non-linear dynamical systems may be delineated and separated from random behaviour. This is achieved by mapping complex input data to a given complex output by propagating data from the input layer to the output layer through a larger number of interconnections, so-called hidden layers. The networks are capable to extract relevant features from training samples and store this information in the distributed structure of interconnections. Using this approach on growth hormone (GH) rhythms of healthy controls, fasted healthy subjects, untreated acromegalic patients and acromegalics under octreotide suppressive therapy we were recently able to demonstrate the power of this approach to differentiate the temporal pattern of GH secretion following normalization of the data for absolute amplitudes. In a second approach we were able to significantly reduce the number of data points required to characterize the temporal structure of these rhythms. This latter quality of the networks may help to transfer analysis of changes in the temporal pattern of hormone secretion on a more routine basis.
What problem does this paper attempt to address?