Food Safety Analysis, LLC


Laboratory Audit

Perhaps one of the best ways to head off problems with conflicting laboratory results is to routinely test your laboratory's performance against a number of other laboratories all performing the same analysis on distributed aliquots of an identical sample. The results obtained by all laboratories is then analyzed statistically. A laboratory whose results show a statistically significant deviation from the mean of all laboratories is naturally suspected of having made an error in its analysis and so is asked to review its work. Such so-called "Correlation Studies" are quite common, are paid for by the laboratories themselves and are often used as a measure of the quality that a laboratory delivers. Such "Proof" of the quality of a laboratory's deliverables is the point of the whole thing. It is well worth reviewing the results of any correlation studies a given laboratory has participated in given that the value of such correlation studies is widely recognized.

Not widely recognized, however, is the downfall of all correlation studies. Such studies have no way to evaluate the excellence or mediocrity of the group as a whole and make no attempt to do so. All they can do is assume that the statistically identified majority MUST be right. So, if the majority of laboratories in a correlation group are making the same mistake in a given analysis they nevertheless define what is "right" by the preponderance of their numbers. One or more laboratories but a minority, doing the analysis correctly, will produce a result that will then be see to "deviate" from the majority value and so will be treated as statistical "outliers" and therefore assumed to have delivered a result that is "wrong."

A clear example of this issue may be found in the measurement of the density of gasoline. Gasoline, in broad terms, is distilled from crude oil at temperature that range roughly between 40 and 200 degrees Centigrade. It contains may chemicals, including some one would normally think have no business being there, like butane, which has a boiling point of -0.5 degrees Centigrade. However, the concentration of this compound in gasoline is often used to adjust volatility which helps with cold-weather starting, etc. Gasoline is know to stratify upon standing. That is why the method used for its density measurement specifies that the sample be well-mixed before a density measurement is made. On the other hand, mixing is also thought to promote the evaporation of butane and other so-called "light ends" resulting in their release to the atmosphere. Since the manner and extent of mixing is not specified by the densitometric method, analysts feel free to adopt their own opinions. The result is that it is impossible to tell who was right, the light set or the heavy set.

The solution, of course, would be for the management running the correlation study and sending out the samples to have determined ahead of time exactly what the densities of the samples were. That way, the accuracy of the measurements made by all participants would be known and would not have to be left to individual (or group) surmise. In this example, the density of the contents of a sample bottle could have been (but was not) determined radiometrically with great precision and without opening the bottle so that the question of whether or not light ends were lost became moot.

A seasoned laboratorian well understands the considerabe difference between the accuracy and precision of an analytic procedure. Such a person is also cognizant that while precision may be easily calculated from some number of repeat analyses of the same sample, accuracy is a much more difficult quantity to get obtain and, in some cases, may be impossible to obtain. In today's regulatory environment, the customer needs to understand these crucial distinctions as well.

The laboratory results your operations and product safety are based upon, whether obtained by an in-house lab or an outside analytic facility, may have to stand up in court one day. Any manufacturer needs to make certain that such results (unlike the example given above) do indeed have legs to stand on. The best time to do this is well before you have to rest your reputation on the lab results used to define your product.

Copyright © 2012 by M. Mychajlonka, Ph. D.