Error repack

Samples that are in a dry form require tumbling or thorough mixing to minimize the effects of repack error. Repack error is an effect that is demonstrated for a given sample as the differences between sample spectral measurements of several aliquots of a sample. If this effect is not minimized, the correlation between the optical data and the reference chemical information is drastically reduced and the sensitivity of NIR for any particular application is reduced. Repack averaging, or compression, and other methods of reducing these effects have been demonstrated in the literature [66]. 

A key step in the application of rapid analysis methods is the collection of high quality NIR spectra. To minimize the effect of water in the biomass spectra, each sample was air-dried to less than 10 percent moisture prior to NIR spectroscopic analysis. Spectroscopic techniques were used that enable a high quality, reproducible, and representative NIR reflectance spectrum to be obtained. For each sample, a total of 35 spectra were collected and averaged to compensate for sample heterogeneity. Each calibration sample was subsampled three times. Sample cells were emptied and repacked between subsamples. Instrument reproducibility tests demonstrated that the reproducibility limits of the NIR spectrometer contributed less than 0.1 percent to the absolute prediction errors in the rapid analysis method. 

The Food and Drug Administration (FDA) implemented a final rule in April 2004 requiring bar coding on certain products administered to patient in hospitals in an effort to reduce medication errors. This rule applies to manufacturers, repackers, relablers, and private label distributors of the following products ... 

In the case of powder samples, a mixing process is required to minimize the effects of repack error due to variation in packing density. When there are differences... 

When defining instrument performance for a specific NIR spectroscopic application, the terms detection limit and sensitivity (or signal-to-noise) can be used. Detection limit can be loosely approximated for any NIR spectroscopic method as equal to three times the Standard error of calibration (SEC) for the specified application. The detection limit for a specific application may also be described as three times the repack error for the sample spectrum representing the lowest constituent concentration. This value approximates a 99% confidence that a difference in concentration between two samples can be correctly determined. 

Questions often arise as to which mathematical treatments and instrument types perform optimally for a specific set of data. This is best addressed by saying that reasonable instrument and equation selection composes only a small quantity of the variance or error attributable to the NIR analytical technique for any application. Actually, the greatest error sources in any calibration are generally reference laboratory error (stochastic error source), repack error (nonhomogeneity of sample — stochastic error source), and nonrepresentative sampling in the learning set or calibration set population (undefined error). 

Comments The repack error, sometimes referred to as the SDD or standard error of differences for replicate measurements (SED replicates), is calculated to allow accurate estimation of the variation in an analytical method due to both sampling and presentation errors. It is a measure of precision for an analytical method. 

However, the noise of current NIR instrumentation is so extraordinarily small that it is invariably smaller than almost anything else in the calibration experiment. That fact might be considered justification for concluding a priori that the reference laboratory results are always the dominating factor. However, that simplistic conclusion ignores the fact that the total error of the optical data depends on sample-induced errors as well as instrumental errors, and these can be much larger than the noise level of the instrument. Such errors include particle size effects, repack variations, effects of impurities, and effects due to changing physical characteristics of the samples (e.g., crystallinity). 

---