A-351 Reliability of POCT AccuChek Inform II® Performance Assessed with The Laboratory Method. A Practical Example for a System with Fourteen Facilities Monitored for Nine Years

Abstract

Abstract Background Reliability is acceptable performance over time and by location. The reliability of AccuChek Inform II® glucose meters was assessed in fourteen Sentara Healthcare (SH) facilities for nine years by comparison with the laboratory method. Methods Nine hundred and twenty-one patient specimens collected in parallel by fingerstick and by phlebotomy in lithium heparin tubes (Becton Dickinson) were assayed with the AccuChek Inform II glucose meter (Roche Diagnostics) and cobas c-501® (Roche Diagnostics). The data were stored in RALS® (Abbott Diagnostics) and electronically transferred to Minitab® (Version 19, Minitab Inc.) statistical software. The data, in the interval 30–600 mg/dL, were analyzed with orthogonal, ordinary least squares (OLS), weighted least squares polynomial (WLSP) and locally weighted scatter plot smoother (lowess) regression models. The aptness of the OLS model was evaluated by comparison with the orthogonal model and that of the WLSP model was evaluated with the standardized deleted residuals (Std.Del.Res) diagnostics. The equality of the regression lines, as obtained for each year and for each location, was assessed with the F statistical test. The linearity of the relationship of the two methods was assessed with the pure error test. The absolute and relative differences between the paired values as obtained with the two methods were evaluated using CLIA's acceptance limits (target value ±6 mg/dl or 10%, greater). Results The regression lines obtained with the OLS model (Inform II = 3 + 0.97 c-501) and the orthogonal model (Inform II = 2 + 0.97 c-501) were very similar and the differences were not clinically significant. The standardized residuals for both models showed an increase in variance for increasing levels of the values as obtained with the c-501 method. Consequently, a WLSP model was employed. The analysis of the Std.Del.Res. for the WLSP model showed a quasi-normal distribution with twelve possible outliers (Std.Del.Res.>|3|) and no influential observations (Leverage < 0.5). The test for equality of regression lines did not show probabilistically significant differences for either year (P = 0.93) or facility (P = 0.46). The pure error test showed no probabilistically significant lack-of-fit (P = 0.07). The lowess of the Std.Del.Res. plotted by the fitted value showed a quasi-linear relationship between 30 and 500 mg/dL. For values between 30 and 100 mg/dL the majority of the differences between methods was within ± 6 mg/dL. Similarly, for values between 101 and 600 mg/dL the majority of the relative differences were within ±10%. Comments The reliability of the POCT method has been demonstrated. The POCT and the laboratory methods demonstrated a linear, similar and within the CLIA's acceptance limits of total error performance in fourteen SH facilities for nine years. The application of regression models to a large data base was possible due to the availability of Minitab statistical software. Additionally, these results demonstrated the organizational reliability of the SH POCT protocols for QC, QA, acceptance of new reagents and instruments, designed by the POCT leadership and performed by the POCT personnel and nurses. This reliable performance could not have been possible without the POCT team dedication to quality.