Jiakai Liu1, Chin Hon Tan1, Tze Ping Loh2, Tony Badrick3. 1. Department of Industrial and Systems Engineering, National University of Singapore, Singapore; 2. Department of Laboratory Medicine, National University Hospital, Singapore; tploh@hotmail.com. 3. Faculty of Health Sciences and Medicine, Bond University, Queensland, Australia.
Abstract
BACKGROUND: Between-reagent lot verification is a routine laboratory exercise in which a set of samples is tested in parallel with an existing reagent lot and a candidate reagent lot (before the candidate lot is committed to test patient samples). The exercise aims to verify and maintain consistency in the analytical performance of a test. We examined the limitations of a routine between-reagent lot verification procedure in detecting long-term analytical drift and looked for a more sensitive alternative. METHOD: Via numerical simulations, we examined the statistical power of the current regression-based (weighted Deming regression) procedure for between-reagent lot verification in detecting proportional bias and constant bias. An alternative procedure applying the Student t-test to separately examine cumulative regression slopes and intercepts across multiple reagent lots was proposed and evaluated by numerical simulations. RESULTS: The regression-based procedure had poor statistical power in detecting proportional bias and constant bias when small numbers of samples were used in each between-reagent lot verification exercise. Furthermore, the method failed to detect long-term drifts in analytical performance. The proposed approach based on the Student t-test can detect long-term (cumulative) drifts in regression slopes and intercepts. This method detected a mild downward drift in the serum sodium assay in our hospital that was missed by routine between-reagent lot verification. CONCLUSIONS: The proposed method objectively and systematically detects long-term proportional and constant bias separately. However, the statistical power of this procedure remains unsatisfactory when used with small sample sizes. Sharing of information between laboratories may provide sufficient statistical power to detect clinically important analytical shifts.
BACKGROUND: Between-reagent lot verification is a routine laboratory exercise in which a set of samples is tested in parallel with an existing reagent lot and a candidate reagent lot (before the candidate lot is committed to test patient samples). The exercise aims to verify and maintain consistency in the analytical performance of a test. We examined the limitations of a routine between-reagent lot verification procedure in detecting long-term analytical drift and looked for a more sensitive alternative. METHOD: Via numerical simulations, we examined the statistical power of the current regression-based (weighted Deming regression) procedure for between-reagent lot verification in detecting proportional bias and constant bias. An alternative procedure applying the Student t-test to separately examine cumulative regression slopes and intercepts across multiple reagent lots was proposed and evaluated by numerical simulations. RESULTS: The regression-based procedure had poor statistical power in detecting proportional bias and constant bias when small numbers of samples were used in each between-reagent lot verification exercise. Furthermore, the method failed to detect long-term drifts in analytical performance. The proposed approach based on the Student t-test can detect long-term (cumulative) drifts in regression slopes and intercepts. This method detected a mild downward drift in the serum sodium assay in our hospital that was missed by routine between-reagent lot verification. CONCLUSIONS: The proposed method objectively and systematically detects long-term proportional and constant bias separately. However, the statistical power of this procedure remains unsatisfactory when used with small sample sizes. Sharing of information between laboratories may provide sufficient statistical power to detect clinically important analytical shifts.
Authors: Angela C Rutledge; Anna Johnston; Ronald A Booth; Kika Veljkovic; Dana Bailey; Hilde Vandenberghe; Gayle Waite; Lynn C Allen; Andrew Don-Wauchope; Pak Cheung Chan; Julia Stemp; Pamela Edmond; Victor Leung; Berna Aslan Journal: Pract Lab Med Date: 2022-08-10