Methodological problems in characterizing an individual's plasma glucose level.

Abstract Two methodological problems in characterizing an individual's plasma glucose level are examined in this study. First, how large is the intra-individual variation of an individual's 1-hr post-load glucose level and for this estimated intra-individual variation what are the probabilities of misclassifying individuals based on a one-time measurement only of glucose level? Second, do different tests—i.e. fasting, 1-hr, 2-hr post-load, GTT—yield consistent ranking for the same individual? The first of these was explored with data on subsamples from the Chicago Peoples Gas Company (PG) study and the Chicago Heart Association Detection Project in Industry (CHA) study; the second, with data from the Chicago Coronary Prevention Evaluation Program (CPEP). For both the PG and CHA studies, the estimated ratios of the intra- to inter-individual variances were generally higher for post-load plasma glucose than blood pressure, heart rate, weight and serum uric acid. The conditional probabilities of misclassifying individuals into quintiles or deciles based on one measurement of 1-hr post-load glucose were also estimated from these data. These estimated probabilities indicate that the possible attenuation due to intraindividual variation cannot abolish a strong association; however, it may create some problem if the relationship is not very strong. Furthermore, both rank correlation and quintile classification analyses show that fasting, 1-hr and 2-hr plasma glucose level characterize individuals differently. Thus it is possible that the inconsistent results of previous studies, all using a one-time measure of plasma glucose, are partially due to the large intra-individual variation of this variable, and the use of methods that are not highly consistent in their classification of individuals.