B-145 General Linear Model Statistical Techniques for Evaluating the reliability of Performance Between Two Instruments after Major Maintenance. A Practical Example with hs-cTnT Gen5 Performed with two Cobas e601

Background. CLIA’s regulations require verification of calibration and linearity when an instrument receives major maintenance or has critical parts replaced. If the instrument is the sole operating analyzer, univariate statistics are sufficient. However, when the instrument is used interchangeably with other instrument(s), comparisons with multivariable statistics are more robust and parsimonious. Additionally, multivariable statistics allow immediate, simultaneous comparisons between instruments and time of operation (e.g. years). This poster illustrates a practical example of multivariable statistical techniques employed in evaluating the reliability of high sensitivity Troponin-T performed on two Cobas e-601 instruments, one of which had major maintenance. Methods. Instruments: Two Cobas e-601 ™ (Roche Diagnostics). Due to mechanical failure e601.2 had major maintenance with replacement of most of the microfluids and the measuring cell . Assay: hs-cTnT Gen5® electrochemiluminescence (Roche Diagnostics). Short-term precision was assessed with QC material (Cardiac Markers Plus® level 1, Liquichek® level 3, (Bio-Rad)) assayed with five independent assays for five consecutive days. Calibration/Linearity was assessed with five levels of calibration/linearity material (Validate®, Maine Standards) assayed with three independent assays within one day, three times per year for two years. Patient specimens, collected by phlebotomy in lithium heparin plasma separator vacutainers (Becton-Dickinson), were assayed in parallel and within 15 minutes with both instruments. Data analysis: Data were transferred either electronically or manually to Minitab ® (Version 22, Minitab Inc) statistical software and analyzed with General Linear Model (GLM) techniques, their appropriate exploratory/diagnostic techniques, and graphic representations. Results. The short-term precision evaluation with GLM while did not show probabilistically significant inequality of mean performance by day (Level 1: P=0.16, Level 2: P=0.38) and instrument (Level 2: P=0.09) it showed probabilistically significant mean difference by instrument (Level 1: P<0.001). However, the mean difference of 0.4 ng/L (95% CI: 0.2-0.6 ng/L) was not significant for either QC nor clinical applications. Parallel box plots, normal probability plots, standard deviation with Bonferroni’s 95% CI plots and autocorrelation plots visually corroborated these results. For the calibration/linearity evaluation, the weighted polynomial regression analysis model did not show probabilistically significant inequality of regression lines between e-601 instruments (P=0.9) after repair of e601.1 and for two consecutive years prior failure of e-601.2 (P=0.7) in the interval 5-10,00 ng/L. The standardized deleted residual plots clearly showed a quasi-normal distribution, a quai-linear relationship with the fitted values and no abnormal consecutive distribution patterns. The relative differences between the observed and the reference values and patient specimens were within the acceptable total error (target value ± 10%). Conclusions. The GLM analysis demonstrated for both instruments post maintenance similar and stable QC performance, and pre and post maintenance quasi-linear dose-response curve within 5 and 10,000 ng/L . The relative difference between assayed and reference values and between patient specimens were within ±10%. the performance of e-601.2 was acceptable and harmonized with that of e-601.1. The two instruments could be used interchangeably for patient care. Finally, this practical example illustrated the power of GLM multivariate statistical techniques and their graphics, offered by statistical software such as Minitab.