Validation of deconvolution analysis of LH secretion and half-life

Deconvolution methods constitute a class of analytic tools to quantitate hormone secretion and/or clearance in vivo. Although mathematically rigorous, deconvolution techniques have assumed, rather than proven, validity. Accordingly, we tested the validity of deconvolution analysis on true-positive human, animal (sheep and monkey), and computer-simulated data using the luteinizing hormone (LH) pulse signal as a relevant paradigm. We found that multiparameter deconvolution analysis has high discriminative sensitivity (human data 91%, animal 81%, computer-stimulated 95%) and specificity (human 90%, animal 81%, computer-simulated 100%). Sensitivity was impaired by low secretory burst amplitude (< 0.1 IU.l-1.min-1), short interpulse interval (< 60 min), infrequent venous sampling (every 20-30 min), and high random experimental variation (e.g., noise > 15%). Specificity was hindered by noise. Deconvolution accurately characterized the unknown hormone half-life (r = +0.994) and production rate (r = +0.990). Interoperator reliability was high when statistically based criteria for secretory pulse detection were applied. We conclude that multiparameter deconvolution, within recognizable constraints, is a valid and reliable tool for in vivo investigation of hormone secretion and half-life.