Grossi Enzo, Colombo Roberto, Cavuto Silvio, Franzini Carlo
Research & Development, Centro Diagnostico Italiano, Milan, Italy.
Clin Chem. 2005 Jul;51(7):1232-40. doi: 10.1373/clinchem.2005.047787. Epub 2005 May 26.
In a primary healthcare center concerned more with maintaining wellness than with diagnosing and monitoring illness, it is particularly important to compare patients' results with reference intervals derived from a matched population by use of defined statistical methods.
Laboratory results over a 3-year period (approximately 15,000,000 records; 197,350 individuals) were retrieved from our laboratory information system. An inclusion/exclusion procedure for individual patients was applied based on (a) presence of at least 1 of 23 previously defined "basic tests"; (b) only 1 measurement per test by the laboratory over the 3-year period; (c) for each test, absence of any abnormality in the correlated tests. Before the third step, correlations among quantities were assessed by a Spearman correlation matrix, comparing each of the 23 basic tests with all remaining tests by use of a novel multivariate algorithm.
The initial sample group (n = 197,350) was reduced stepwise by the selection criteria outlined above to 166,027, then to 93,649, and finally to 61 246 individuals constituting our reference sample group. Results from the last 2 groups were used to calculate sex-specific, and in some cases age-related, reference limits for the 23 basic tests and for 13 additional quantities. Reference limits were calculated throughout this study by nonparametric estimation of percentiles.
Reference values derived by retrospective analysis of large samples of data obtained at a given institution are particularly suitable for the evaluation of results for the presenting patient population at that institution.
