Assessment of canonical diurnal variations in plasma glucose using quantile regression modelling and Chronomaps.

Objectives: Diurnal variation of plasma glucose levels may contribute to diagnostic uncertainty. The permissible time interval, pT(t), was proposed as a time-dependent characteristic to specify the time within which glucose levels from two consecutive samples are not biased by the time of blood collection. A major obstacle is the lack of population-specific data that reflect the diurnal course of a measurand. To overcome this issue, an approach was developed to detect and assess diurnal courses from big data.

Methods: A quantile regression model, QRM, was developed comprising two-component cosinor analyses and time, age, and sex as predictors. Population-specific canonical diurnal courses were generated employing more than two million plasma glucose values from four different hospital laboratory sites. Permissible measurement uncertainties, pU, were also estimated by a population-specific approach to render Chronomaps that depict pT(t) for any timestamp of interest.

Results: The QRM revealed significant diurnal rhythmometrics with good agreement between the four sites. A minimum pT(t) of 3 h exists for median glucose levels that is independent from sampling times. However, amplitudes increase in a concentration-dependent manner and shorten pT(t) down to 72 min. Assessment of pT(t) in 793,048 paired follow-up samples from 99,453 patients revealed a portion of 24.2 % sample pairs that violated the indicated pT(t).

Conclusions: QRM is suitable to render Chronomaps from population specific time courses and suggest that more stringent sampling schedules are required, especially in patients with elevated glucose levels.