We have developed improved methods for detection and characterization of episodic hormone secretion, and to estimate the instantaneous rate of hormone secretion using deconvolution analysis. These methods have been applied to several clinical investigations of the dynamics of LH, FSH, prolactin, ACTH, cortisol and beta-endorphin. The DETECT program and algorithm have been tested extensively, using computer simulation (""""""""Monte Carlo"""""""") studies. The sensitivity and specificity of these methods have been estimated for a wide variety of peak shapes, height, duration, signal-to-noise ratio, pulse frequency, sampling frequency, variability of peak shape, and other parameters. Receiver-Operating-Characteristics (R-O-C) curves have been constructed for DETECT and for another popular program, CLUSTER. Results indicate that for any desired level of specificity (false-positive rate), DETECT shows better sensitivity (lower false-negative rate), and sensitivity is better than 90% in most cases. A new method for computer simulation assumes that inter-pulse interval behaves as a Poisson process. This analysis was useful to examine sensitivity as a function of sampling frequency and led to the concept of 'visible' or observable peaks. New methods have been developed to evaluate coincidence of pulses in two hormonal time series. These include the concept of """"""""specific concordance"""""""" as a function of threshold levels and adjustable lag times, and also uses the Kappa and McNemar's chi- square statistics. Program DETECT and is being extensively revised to improve speed and user-friendliness. A new algorithm for deconvolution has been developed based on use of digital filters. A new program, EXPFIT, has been developed to analyze exponential decay curves.
