Individualized principal component analysis of endocrine circannual variability

Abstract

The technique of principal component (PC) analysis (PCA) of multivariate observations is a method that allows dimension reduction of multivariate data for further analysis. It is here introduced as a means of selecting chronobiologically important variables that can be further studied by an analysis of variance. The use of PCA is illustrated for a study of major temporal sources of human endocrine variability. Contributions to temporal variability by seven steroidal and six nonsteroidal hormones are compared in samples available at 100-min intervals for 24 hr in three seasons for each of three clinically healthy individuals: an adolescent woman, a menstrually cycling woman, and a postmenopausal woman. On an individualized basis, it is ascertained that the first principal component, a new variable, is primarily determined by steroids and that PCA can single out variables displaying interseasonal (circannual) differences validated as statistically significant by a subsequent analysis of variance. The variables here scrutinized and identified as contributing to the PC, however, need not all differ with statistical significance along the scale of the seasons. The steroids contributing the first principal component are DHEA-S and an estrogen in all three individuals studied, cortisol and aldosterone in two of them, and 17-OH progesterone in one case.