Reference interval computation: which method (not) to choose?

Abstract

Background: When different methods are applied to reference interval (RI) calculation the results can sometimes be substantially different, especially for small reference groups. If there are no reliable RI data available, there is no way to confirm which method generates results closest to the true RI.

Methods: We randomly drawn samples obtained from a public database for 33 markers. For each sample, RIs were calculated by bootstrapping, parametric, and Box-Cox transformed parametric methods. Results were compared to the values of the population RI.

Results: For approximately half of the 33 markers, results of all 3 methods were within 3% of the true reference value. For other markers, parametric results were either unavailable or deviated considerably from the true values. The transformed parametric method was more accurate than bootstrapping for sample size of 60, very close to bootstrapping for sample size 120, but in some cases unavailable.

Conclusions: We recommend against using parametric calculations to determine RIs. The transformed parametric method utilizing Box-Cox transformation would be preferable way of RI calculation, if it satisfies normality test. If not, the bootstrapping is always available, and is almost as accurate and precise as the transformed parametric method.