Abstract Shewhart X and R charts were used to maintain and validate data quality of percent recovery estimates for 8 analytes determined by 4 procedures used routinely in 4 commercial laboratories over a 2-year period. However, because range (R) estimates of uncertainty did not Include lot-to-lot calibration variability, approximately 24% of the lots were "out-of-control." We extracted pooled standard deviations for So (repeatability within lot), SL (calibration variability), and SR (reproducibility), which represents the total variability. Values of So and SL were generally similar In size although there were some substantial differences between analytes and between laboratories for a given analyte. When control limits were based on reproducibility rather than repeatability, only about 6% of the lots were "out-of-control." However, these limits are less convenient to compute at the bench, wlthln-lot precision estimates are still required, and there Is still no Information on data acceptability. Capability estimates from the grand mean ±3 SR were surprisingly consistent for the 8 analytes. These values coupled with data quality objectives suggested the 82-115% range as the specifications for acceptable individual recoveries. A combination of repeatability limits plus modified limits anchored to specifications retains the simplicity of range computations while offering substantial administrative advantages. Examples are given to illustrate these points.
Read full abstract