Regression models in the study of circadian rhythms in nursing research
GERALANDE FELTON, J. E. WARD · Nursing Research · 1978
In studying the effects of change in activity cycles on biological rhythms, there need to be strictly controlled experimental designs and objective methods for analyzing results in order that the data are not subject to different interpretations by different viewers (Reinberg, 1972). The design should include a A (the interval, whether equal or unequal, between measurements for a given subject on any given day) small enough, and a T (the total number of days of study), large enough to provide sufficient data for analysis. For purposes of analyses, special electronic computer programs have been developed and tested by Chronobiology Laboratories at the University of Minnesota under Franz Halberg (1969). Using these programs, estimates of rhythm parameters such as rhythm-adjusted level, the amplitude (a measure of the extent of rhythmic change) and the acrophase (the peak of the function used to approximate the rhythm), along with their respective (0.95) confidence intervals are derived. This procedure, called the Cosinor is presented as a multiple regression, permitting the detection of a rhythm and the objective and quantitative evaluation of its amplitude and phase. Recently two critical theoretical studies of the procedure suggest that Cosinor be used cautiously, as its application may lead to a wrong interpretation of the data. The use of other methods might be both more profitable and safer (Yuruza, 1971; Van Cauter and Huyberechts, 1973). Nevertheless, the fitting of multiple linear regression models gives evidence of being a tool to help nurse researchers obtain meaningful answers to research questions. It may be well that more researchers consider the advantage of seeking assistance to exploit the power of general linear models and to use the computer for