I have been reflecting on what I have learned over the years and what drives my perspective, how I approach problems. Spending so much time in the lab has enabled me to incorporate an analytical perspective into my personality, in my daily life I approach everything with an effort to be efficient. On a side note, I have become very efficient at conserving calories and can prove it with the inactivity tracker currently on my wrist, which warns me if I have become too active!
Kidding aside, my goal is to be efficient. If I am on my way to the kitchen and see an empty cup, I pick it up and take it to the kitchen. If I am about to fill the car with gas and have garbage, I take the garbage out and throw it away. You get the idea.
So back to the lab, in a previous article, “Why the Current Quality System Doesn’t Matter”, I suggested that D6299, doesn’t really meet analytical needs, it is more about administrative needs. The sole purpose is documentation, and the basis of the quality control charts is essentially to monitor precision. If the instrument presents as out of control, we must take the instrument out of service once verified. The statistics are all based on normally distributed data. Now, I know that there are people out there in the Quality Industry that really understand how this all works, but I have a sense that most people really, just know how to use the software, the skill of interpreting the data and what it means in terms of analytical instrumentation is usually lacking.
For example, I have argued that the instrumentation being produced today is very precise, one example to showcase is the VGA detector by VUV Analytics, Inc. The way that the VGA data is processed enables extremely precise data. This is because it is based on Beer’s Law, relative response factors and normalization, this quantitation eliminates sample introduction errors, the main source of variability. The result of adopting this technology is increased precision, D6299 is not as useful if an instrument has good precision. For example, if my instrument reads 0.01 over the mean nine times in a row, the instrument is out of control, despite the fact that it is essentially equivalent to being on the mean, this is part of D6299 and built into our quality control software.
I had the opportunity to speak to the author of D6299 about this at an ASTM meeting last year, and his answer to me was that D6299 gives guidance on what to do in this situation, you can create a custom chart, it is the user's responsibility to determine how to apply D6299. Essentially, this type of data is not normally distributed, so, we have the option to use an approach that we deem appropriate. The initial guidance is to increase the number of significant digits, which will increase variability. Never mind the fact that there was quite a bit of time during our education emphasizing that significant digits must be limited to the least accurate measurement tool, this fact alone should have eliminated it as an option, but it was an easy solution and to a statistician, perfectly acceptable. The result of which is a standard deviation that is so tight, instruments are put out of service, despite having more than acceptable precision. It may be time to incorporate chemists into the discussions of quality.
I shared this with our quality department and their response was, “well, we need to figure out why it is not normally distributed and correct it”. Unfortunately, the people that I have encountered in the places I have worked would likely have had the same response. So, while those of you might take exception to my comments about those who work in the Quality Industry, keep in mind that my opinion is based on a small population. But this signals a potential problem, and increases the challenge of how to correct it. If D6299 has guidance for these situations, then why is there resistance to apply the guidance, a topic for another article…
Why does Variability Matter?
The presumption is that analytical instruments are a black box, and they just do what they do, it’s all random. From my analytical perspective and drive for efficiency, I reject this presumption. Physics applies to everything, matter can neither be created nor destroyed. Therefore, there must be a scientific reason for why the QC data appears random. One other thing about instruments that you will hear is that no two instruments will ever give the same result. In terms of GC, many years ago someone once told me that you have different syringes, different inlets, different columns, different detectors, there is no way to make them read the same number. My immediate question was, “then what is the purpose of calibration?”.
It is not as difficult to get two instruments to read the same number as people assume. And if you have read my previous articles about sample introduction, sample points, etc., then you should have a clue. You must mitigate variability; variability is what makes the data seem random. For example, if you calibrate your instruments on the same day with the same standards, you likely will get the same result on your QC when analyzed on both instruments. The problem is that we rarely do this, calibration involves bringing the instrument down, and “if it ain’t broke, don’t fix it”. Of course, from my perspective, if the numbers don’t match, it’s “broke”. I figured out the solution shortly after my calibration question, I hadn’t fleshed out the concept as to why it worked yet.
I have worked in brand new laboratories and laboratories that could double as a garage. This enabled me to understand variability in terms of analytical instrumentation and I sincerely hope that by sharing it here, it helps to improve your data.
Variability matters because it is the source of all errors, if you can successfully mitigate the variability, you could potentially flat line your QC charts! I have a plan to do this, that I will share eventually. Essentially, I figured out how to merge precision and accuracy that will render D6299 obsolete.
But if you want to improve your data now, look at your sample introduction for flaws, are you using the right solvent wash? Are you rinsing enough? Look for bubbles in your syringe. Most detectors are mass sensitive, so sample introduction errors are likely your main source of variability. Look at the location of your instrument, is there a draft that could be affecting temperature, is it under a vent? Remember, temperature matters! Once you mitigate variability, you will increase precision, and you will be ready to join me in the fight against the dreaded 9 point rule!