Posted by Sten Westgard, MS
About a month ago, we posted an article titled Would you drive your car the way you run your QC?
The article was trying to highlight some of the QC practices in the laboratory, that, if we view them in a different light, are really counter-productive and sometimes outright absurd.
Another thought like that recently occurred to me, as someone was discussing how they "adjust" their laboratory's standard deviation.
The question is, would you ever consider "adjusting" your car's speedometer?
In other words, what if you could "adjust" your speedometer so that when you're really traveling 75 miles per hour, the speedometer reads just 55 miles per hour. [Side note to our global readers: 55 miles per hour is a common speed limit here in the US]
How does this help you? Well, you can always pretend you're going the speed limit. And if the highway patrol pulls you over, you can always claim, "According to my speedometer, I was still going the speed limit."
Of course, the highway patrol isn't going to accept that excuse. No matter what you've chosen to believe your speed is, there is an actual speed that can be measured, and if you exceed the limit, it doesn't matter if you think you were under that limit, you've violated the law. As they say, "ignorance of the law is no excuse." Particularly willful, deliberate ignorance.
Now, to the parallel situation in the laboratory. Laboratories often find themselves unhappy with their "2 SD" control limits and seek to find ways to adjust those limits wider. Sometimes this occurs when the laboratories look up other standard deviations that can be found outside their own facility: a peer group SD, a proficiency testing SD or EQA SD, or perhaps a manufacturer's "range." All of these standard deviations involve larger numbers of laboratories, which means more varation, thus, a larger SD. So if a laboratory uses the larger SD from the external source, their "2 SD" control limits are wider.
But, as you clearly can see, their "2 SD" isn't really 2 SD anymore. It's someone else's SD. If you used the actual laboratory SD, the "2 SD" might be more like 3 or 4 of the actual laboratory SD.That's not necessarily bad, except in cases where you need better error detection. Then the laboratory may not be able to detect medically important errors.
When labs "adjust" their SD to widen them, no matter the rationalization, they are essentially lying to themselves, purposely pretending that someone else's SD is their own. It gives the laboratory the illusion that they are performing "good" QC while in fact they may not really know what level of quality assurance they are actually achieving.
A better question for the laboratory might be: Why are we stuck on "2 SD" as the standard for QC practices? True, it is a practice that has been around for about half a century. But that practice is more of a tradition than a scientific practice. Laboratories have often handed down the tradition from one generation of medical technologists to the next. But that doesn't mean the tradition isn't obselete at this point.
It would be far more efficient to make an evidence-based, data-driven choice for control limits and numbers of controls (i.e. use OPSpecs charts to choose the right QC), rather than fudge the limits by using someone else's SD.
There is a saying, "Two wrongs don't make a right. But three lefts will." Rather than setting the wrong control limits with the wrong SD, it's time for labs do the right QC right.
Thanks for the good article.
I agree that it's a losing battle, not to mention concerning in regard to patient care, adjusting SDs frequently. The only time we use "someone else's SD" (manufacturer's) is when we have a new lot# and need a guage on an acceptable starting mean until we can establish our own through multuple runs. Manufacturer's ranges are really just that, a range of means and not a range to be used for any particular lab. What we've discovered with modern instrumentation is that from lot to lot our lab's SDs stay esentially the same. Only the mean must be adjusted and typically only by a slight amount.By the way, 2SD still works for us :).
Posted by: Kathy Walker | September 20, 2012 at 02:46 PM
Great article. We generally use 60 to 100 data points to calculate a new range when working with a new lot #. Aftera while, there may be a shift. Could additional data points be collected, ex: another 60, and a new 2sd established? This will be done only after all other efforts have been looked at to solve the problem.
Posted by: Randy Sparrow | September 22, 2012 at 09:42 PM