A comparison of some different methods for purifying core temperature data from humans.

Abstract

Nine healthy females were studied about the time of the spring equinox while living in student accommodations and aware of the passage of solar time. After 7 control days, during which a conventional lifestyle was lived under a 24h "constant routine," the subjects lived 17 x 27h "days" (9h sleep in the dark and 18h wake using domestic lighting, if required). Throughout the experiment, recordings of wrist activity and rectal (core) temperature were taken. The raw temperature data were assessed for phase and amplitude by cosinor analysis and another method, "crossover times," which does not assume that the data set is sinusoidal. Two different purification methods were used in attempts to remove the masking effects of sleep and activity from the core temperature record and so to measure more closely the endogenous component of this rhythm; these two methods were "purification by categories" and "purification by intercepts." The former method assumes that the endogenous component is a sinusoid, and that the masking effects can be estimated by putting activity into a number of bands or categories. The latter method assumes that a temperature that would correspond to complete inactivity can be estimated from measured temperatures by linear regression of these on activity and extrapolation to a temperature at zero activity. Three indices were calculated to assess the extent to which exogenous effects had been removed from the temperature data by these purification methods. These indices were the daily variation of phase about its median value; the ratio of this variation to the daily deviation of phase about midactivity; and the relationship between amplitude and the square of the deviation of phase from midactivity. In all cases, the index would decrease in size as the contribution of the exogenous component to a data set fell. The purification by categories approach was successful in proportion to the number of activity categories that was used, and as few as four categories produced a data set with significantly less masking than raw data. The method purification by intercepts was less successful unless the raw data had been "corrected" to reflect the direct effects of sleep that were independent of activity (a method to achieve this being produced). Use of this purification method with the corrected data then gave results that showed least exogenous influences. Both this method and the purification by categories method with 16 categories of activity gave evidence that the exogenous component no longer made a significant contribution to the purified data set. The results were not significantly influenced by assessing amplitude and phase of the circadian rhythm from crossover times rather than cosinor analysis. The relative merits of the different methods, as well as of other published methods, are compared briefly; it is concluded that several purification methods, of differing degrees of sophistication and ease of application to raw data, are of value in field studies and other circumstances in which constant routines are not possible or are ethically undesirable. It is also concluded that such methods are often somewhat limited insofar as they are based on pragmatic or biological, rather than mathematical, considerations, and so it is desirable to attempt to develop models based equally on mathematics and biology.