[Derivation of latent reference intervals from routine laboratory database: its principle, validity and methodological implications for laboratory informatics]

Abstract

A new method to derive reference from routine laboratory database intervals was developed. It depends on (1) availability of multiple screening tests results, (2) initial estimates for the reference intervals (RI) and (3) exclusion of data with multiple abnormalities in "related" test items. We designated those data which remained after exclusion as "latent reference values". RI was calculated from them parametrically employing a modified Box-Cox power transformation and maximal likelihood method. After RI were derived for all the items, they were used as a new set of RI estimates and the series of exclusion/calculation procedures were repeated for 7 times. The validity of the latent reference intervals (LRI) derived from routine biochemical screening test results were evaluated in three ways, by comparing them (1) with RIs determined conventionally from healthy individuals, and (2) with LRIs derived from qusi-normal individuals (blood donors), and (3) with LRIs obtained from a simulated healthy population which was generated as multivariate normal random values.