Local Polynomial Regression Modeling of Human Plasma Melatonin Levels

Abstract

Accurate characterization of features of the melatonin production curve is important for research in chronobiology. Local polynomial regression is used to model the plasma melatonin concentration curve and obtain consistent and asymptotically normally distributed estimates of synthesis onset and offset times, duration, peak concentration, and area under the curve. A simple production-clearance model for melatonin kinetics provides the connection between plasma concentrations and the melatonin production curve. This model identifies onset and offset times with critical points of derivatives of the concentration curve. The method proposed has the advantage of flexibility and ease of implementation. Local polynomial regression can be shown to produce consistent estimates of the regression curve and parameters of interest. Furthermore, the method can be implemented in any software package with weighted least squares routines.