Choice of Statistical Tools for Outlier Removal Causes Substantial Changes in Analyte Reference Intervals in Healthy Populations

Abstract

Background: Reference intervals are an important aid in medical practice as they provide clinicians a guide as to whether a patient is healthy or diseased. Outlier results in population studies are removed by any of a variety of statistical measures. We have compared several methods of outlier removal and applied them to a large body of analytes from a large population of healthy persons. Methods: We used the outlier exclusion criteria of Reed-Dixon and Tukey and calculated reference intervals using nonparametric and Harrell-Davis statistical methods and applied them to a total of 36 different analytes. Results: Nine of 36 analytes had a greater than 20% difference in the upper reference limit, and for some the difference was 100% or more. Conclusions: For some analytes, great importance is attached to the reference interval. We have shown that different statistical methods for outlier removal can cause large changes to reported reference intervals. So that population studies can be readily compared, common statistical methods should be used for outlier removal.