Analytical Goals for the Estimation of Non-Gaussian Reference Intervals

Abstract

Goals for analytical quality are postulated for the situation in which a number of laboratories will share common reference intervals for some quantities for which the population is homogeneous. These goals are evaluated on the assumption that the acceptable combined analytical bias and imprecision should not allow the percentage of the population outside either upper or lower reference limits to exceed the interval 1.3 to 4.4%.

The goals are evaluated for log-Gaussian biological distributions, and for distributions which can be transformed into log-Gaussian by addition of a constant value to all elements. Further analytical goals for non-parametric distributions are discussed. For log-Gaussian distributions the maximum allowable bias and imprecision are dependent on the ratio of the upper and lower reference limits. When this ratio is c. 1.25, the bias alone must be less than 1.5% or the coefficient of variation alone less than 3.5%. For a ratio of 10, these values may be as high as 15% and 34%, respectively.

Key Words:

• analytical bias
• analytical imprecision
• confidence intervals
• log-Gaussian distributions
• non-parametric distributions
• percentage of population outside reference limits
• probit transformation