b.
Pooled Quality Control Results
Another approach utilizes a large number (e.g., 50) of the most recent quality control sample results. By pooling uncertainty values, a running method evaluation can be effected. The result is (1) a direct measure of the analytical uncertainty of the method as implemented in the lab, and (2) a means of detecting any problems that may creep into a method during routine use. Note that a running uncertainty average is similar to a partial method evaluation and not to a method re-evaluation at each measurement because consecutive running averages are strongly correlated. This approach is adopted within a current MSHA procedure for the analysis of silica. A sampling filter is dissolved and re-deposited onto an analysis filter where the silica is quantified by infrared absorption. From each batch of samples to be analyzed, an analysis filter is retained for re-dissolution, re-deposition, and re-analysis within a subsequent batch. The result is a large number of pairs of nearly identical samples, which can give a running estimate of the method’s analytical uncertainty. c.
Continual Method Re-Evaluation
A third possibility, the closest to the original measurement approach of GUM uses a large number (e.g., 30) of independent control samples for each application measurement. This is the most expensive approach, but also may give the best estimate of the analytical uncertainty, especially in cases where uncertainties may be measurement-dependent. Because many more evaluative measurements per application measurement are needed, this approach is not easily implemented for most industrial hygiene applications. As an example of this approach [2], suppose that a lab estimates only a 30-day average concentration of a given gas or vapor. Further, every day a measurement is taken of a known calibration gas concentration. Then, if the method is expected to behave similarly for measurements of gas and field samples calibrations, the 30 control samples give analytical uncertainty estimates that differ month-to-month and from field measurement to field measurement. d.
Limit of Detection and Detection Limit
When the concentration is low, approaching the method uncertainty uc , concepts of the limit of detection (LOD) and a related detection limit LD may be useful. LOD is used for controlling false positives when asserting the presence of a substance. On the other hand, the detection limit specifies what measurand value (e.g., concentration) is required so that the false negative rate is negligible when the substance is actually present. The limits can refer to the analytical measurement only or, as in this section, to the entire sampling and analytical measurement method. In the following examples, several often realistic assumptions are made. The standard deviation in concentration estimates is assumed constant (i.e., independent of the sampled concentration), unlike the commonly occurring constant relative standard deviation at larger concentrations. Also, bias (or uncertainty in its correction) is assumed to be negligible. More
3/15/03
218
NIOSH Manual of Analytical Methods
