The Statistical Basis of Laboratory Data Normalization

Abstract

The problem of laboratory data normalization in clinical trials is considered in the statistical framework. Leading principles for laboratory data normalization are proposed and practical normalization formulas are derived using as general assumptions as possible. The location-scale family and the scale family are used to model the assay value distributions. These assumptions lead to straightforward normalization formulas that have been proposed as ad hoc solutions. The preferred normalization is found to depend on the distribution of an assay value. We also consider the case where some normalized values become negative and propose that the negative values indicate that an unsuitable normalization is applied.