Reference Intervals: Comparison of Calculation Methods and Evaluation of Procedures for Merging Reference Measurements From Two US Medical Centers.

Abstract

To analyze consistency of reference limits and widths of reference intervals (RIs) calculated by six procedures and evaluate a protocol for merging intrainstitutional reference data. The differences between reference limits were compared with "optimal" bias goals. Also, widths of the RIs were compared. RIs were calculated using Mayo-SAS quantile, EP Evaluator, and four International Federation of Clinical Chemistry and Laboratory Medicine methods: parametric and nonparametric (NP) with and without latent abnormal values exclusion (LAVE). Regression parameters from cotested samples were evaluated for harmonizing intrainstitutional reference data. Mayo-SAS quintile, LAVE(-)NP, and EP Evaluator generated similar RIs, but these RIs often were wider than RIs from parametric procedures. LAVE procedures generated narrower RIs for nutritional and inflammatory markers. Transformation with regression parameters did not ensure homogeneity of merged data. Parametric methods are recommended when inappropriate values cannot be excluded. The nonparametric procedures may generate wider RIs. Data sets larger than 200 are recommended for robust estimates. Caution should be exercised when merging intrainstitutional data.