Health, Normality, and the Ghost of Gauss

Abstract

It has been suggested that "clinical normal values" can be estimated from routine laboratory records by purely statistical methods, so that accumulation of values for healthy persons is unnecessary. The basic premise for this view is that the routine laboratory values give a mixture of two gaussian (or "normal") distributions (one for the sick and one for the healthy) and that these can be identified. The results of the proposed method are compared with observed values for healthy persons tested in the same laboratories during the same period, and are found to be unsatisfactory. The distributions in healthy persons cannot be adequately described by a normal distribution. We propose that the misleading phrase "normal limits" be dropped, and that the phrase "clinical limits" be used to refer to the 2 1/2 and 97 1/2 percentage points of the distribution in healthy persons.