On dividing reference data into subgroups to produce separate reference ranges

Abstract

We consider statistical criteria for partitioning a reference database to obtain separate reference ranges for different subpopulations. Using general formulas relating population variances, sample sizes, and the normal deviate test for the significance of the difference between two subgroup means, we show that partitioning into separate ranges produces little reduction in between-person variability, even when the differences between means are highly significant statistically. However, when there is a clear physiological basis for distinguishing between certain subgroups, simulation studies show that partitioning may be necessary to obtain reference limits that cut off the desired proportions of low and high values in each subgroup. Guidelines based on these results are provided to help decide whether separate ranges should be obtained for a given analyte.