Adaptation of the robust method to large distributions of reference values: program modifications and comparison of alternative computational methods. Issue 3 (4th May 2019)

• Record Type: Journal Article
• Authors: Beasley, Charles M.; Crowe, Brenda; Nilsson, Mary; Wu, Lieling; Tabbey, Rebeka; Hietpas, Ryan T.; Dean, Robert A.; Horn, Paul S.
• Is Part Of: Journal of biopharmaceutical statistics. Volume 29:Issue 3(2019)
• Page Start: 516
• Page End: 528
• Publication Date: 2019-05-04
• Subjects: Reference limits -- nonparametric method -- roust method -- Dixon-Reed method -- Horn algorithm -- outliers; Pharmacy -- Statistical methods -- Periodicals; Drugs -- Testing -- Statistical methods -- Periodicals; Biometry -- Periodicals; Biopharmaceutics -- Periodicals; Pharmacokinetics -- Periodicals; 615.19
• DOI: 10.1080/10543406.2019.1579223
• Languages: English
• ISSNs: 1054-3406

Abstract

ABSTRACT: The objective of this research was to compute reference limits using reference values from patients entering pharmaceutical development clinical trials by the nonparametric method and the robust method of Horn and Pesce, with and without outlier exclusion, and compare the methods with respect to influence on the limits. Reference limits were computed for 38 analytes with over 130, 000 subjects contributing reference values. Subjects were partitioned into 10 demographic strata for limit computation. Limits were computed for both 95- and 98-percentile reference intervals by both methods. For each reference interval and method, the limits were calculated with and without outliers. Outliers were excluded by the Horn algorithm. Irrespective of method, reference limits were expanded with the 98-percentile interval, but some expansions were small. Outlier exclusion contracted limits with more influence on the upper limit. The robust method contracted the upper limit to a meaningful degree and slightly expanded the lower limit for many analytes. Outlier exclusion and computation by the robust method have an increasing influence on analytes with right-skewed distributions of reference values from large populations not screened to exclude common, stable diseases and environmental factors that might affect analyte variability. The method has advantages for computation of reference limits used in clinical trial analyses.