Statistical Correction for Non‐parallelism in a Urinary Enzyme Immunoassay

Abstract

Our aim was to develop a statistical method to correct for non‐parallelism in an estrone‐3‐glucuronide (E1G) enzyme immunoassay (EIA). Non‐parallelism of serially diluted urine specimens with a calibration curve was demonstrated in an EIA for E1G. A linear mixed‐effects analysis of 40 urine specimens was used to model the relationship of E1G concentration with urine volume and derive a statistical correction. The model was validated on an independent sample and applied to 30 menstrual cycles from American women. Specificity, detection limit, paral‐lelism, recovery, correlation with serum estradiol, and imprecision of the assay were determined. Intra‐and inter‐assay CVs were less than 14% for high‐ and low‐urine controls. Urinary E1G across the menstrual cycle was highly correlated with serum estradiol (r = 0.94). Non‐parallelism produced decreasing E1G concentration with increase in urine volume (slope = −0.210, p < 0.0001). At 50% inhibition, the assay had 100% cross‐reactivity with E1G and 83% with 17β‐estradiol 3‐glucuronide. The dose–response curve of the latter did not parallel that of E1G and is a possible cause of the non‐parallelism. The statistical correction adjusting E1G concentration to a standardized urine volume produced parallelism in 24 independent specimens (slope = −0.043 ± 0.010), and improved the average CV of E1G concentration across dilutions from 19.5% ± 5.6% before correction to 10.3% ± 5.3% after correction. A statistical method based on linear mixed effects modeling is an expedient approach for correction of non‐parallelism, particularly for hormone data that will be analyzed in aggregate.

Keywords:

• E1G
• EIA
• Urinary reproductive steroids
• 3F11 clone
• Assay validation
• Linear mixed effects model

Acknowledgments

This research was supported by NIH (NIA 5 T32 AG00208, NIA RO1 AG15141, NIA RO1 AG14579, NICHD 2 P30 HD28263, NICHD R24 HD42828), the Population Research Institute of Pennsylvania State University and the Center for Studies in Demography and Ecology of the University of Washington. Thanks to J. Aranda, D. Clifton, D. Holman, F. Kohen, D. Schechter, C. Stroud, and G. Tomblingson for their contributions.