11 Replies Latest reply on Mar 2, 2017 1:53 AM by gchaplain

    Sudden shift from linear to quadratic calibration?

    mkpbphd

      I work in an environmental lab doing wastewater analyses based on EPA methods which do not currently allow for any calibration model beside linear or avg response factor. Recently, we've run into some issues where two different analytes, using two different instruments and two different methods will no longer calibrate linearly but exhibit excellent quadratic calibration (R2=0.999). Are there any common factors that can contribute to this observation? We've ruled out curve preparation, autosampler performance, and column installation. In both cases inlet maintenance has been performed with no improvement (liners, gold seals, septa, etc have been changed and the columns have been trimmed).

       

      Here's what we know:

       

      Kepone by EPA 8081B (GC-ECD):

      Problem seems to have begun after installation of new column (01/16), but did not become unacceptable until recently.

       

      Benzidine by EPA 625/8270 (GC-MS):

      No obvious change associated with the beginning of the problem.

      New column has been installed since, but no improvement observed.

       

      Has anyone observed anything like this before? I'm not sufficiently familiar with other factors that affect calibration to guess at why we are observing this.

        • Re: Sudden shift from linear to quadratic calibration?

          Hi Megan,

           

          Your situation does sound troubling. Could you please tell us what model instruments and what version software you're using? Did you recently upgrade or patch any software?

           

          Thanks. - Josh

            • Re: Sudden shift from linear to quadratic calibration?
              mkpbphd

              Hi jkurutz,

               

              The Kepone is being run on a 6890N with a uECD detector. Benzidine is being analyzed on a 7890A/5975C. Both systems are running ChemStation (EnviroQuant) version E.02.01.1177 and have not been updated recently. One of the more interesting aspects is that other analytes run on these two instruments still exhibit normal behavior. Pesticides by 8081B and semi-volatiles by 625 still show RSDs<10-20%.

               

              In the case of Kepone, I just noticed that the total response is much lower than it was a few months ago (after installing the new column) but my response for other analytes have not changed.

               

              Thanks!

               

              Megan

            • Re: Sudden shift from linear to quadratic calibration?
              ryan_ohearn

              Have you tried cleaning the source on the 5975C?  I've seen twice with two different customers running 524/624 a couple compounds start to exhibit non-linear calibrations when the source is dirty, but all other analytes are unaffected.

               

              As as for the uECD, have you tried baking it out at a high temperature?  You can take the column out and plug the end with a no-hole ferrule and increase the makeup flow to 50-100 and the temperature to 375-400 and watch the signal until it stabilizes.

               

              Hope this helps!

              • Re: Sudden shift from linear to quadratic calibration?
                ryan_ohearn

                There is also a mixing liner inside the makeup gas adapter for the uECD (Part G2397-20540).  If that gets contaminated it may be why the calibration has changed.  It can be replaced by taking out the makeup gas adapter (maybe replace the nut and ferrule while it's out) and unscrewing the top portion (you may need two pairs of pliers to get a good grip if it hasn't been taken apart in awhile).

                  • Re: Sudden shift from linear to quadratic calibration?
                    mkpbphd

                    Hi Ryan-

                     

                    Thanks for the suggestions! I did check the mixing liner, but it was neither broken nor visibly dirty and as only one compound was being affected, I'm fairly confident that was not the cause. We cleaned the source and baked the ECD and neither helped. However, time seems to have somewhat resolved our problems. I purchased a new Kepone standard from a different manufacturer and it is calibrating acceptably. It's still not as good as previous, but I won't be picky. Similarly, Benzidine is back to calibrating just barely above acceptable. Both still tend towards quadratic, though, so I can't say for sure if the good will last.

                     

                    In the case of Kepone it appears we were also having trouble with the column. For Kepone only, the response dropped dramatically over the first few months of use and that seems to have contributed to our troubles.

                     

                    At any rate, for now that is not an issue. I really appreciate the helps/suggestions!

                     

                    Megan

                     

                    EDIT: In case anyone happens upon this in the future, it now appears that time and the new standard did not fix the problem, as the second time I ran the new calibration curve I observed once again the perfectly quadratic curve with an unacceptable linear calibration. Luckily, we noticed that EPA 8000D (we were referencing 8000C previously) has a much more generous allowance for the use of non-linear calibration models, so we'll just be using a quadratic regression moving forward.

                  • Re: Sudden shift from linear to quadratic calibration?
                    frank's

                    Hi, were are investigating similar behaviour on a brand new machine.

                    Our main suspect is the splitter tot the detectors.

                    • Re: Sudden shift from linear to quadratic calibration?
                      gchaplain

                      Hi Megan,

                       

                      I see you say the responses are OK. What concentrations are you using for the calibration and what area responses are you getting for the high end concentration?

                      Also how many data points does it take for the linear model to turn quadratic?

                       

                      You probably have already tried this but have you tuned the instruments?

                        • Re: Sudden shift from linear to quadratic calibration?
                          mkpbphd

                          Hi Grant-

                           

                          Regarding Kepone:

                          I am calibrating from 40-300 ppb (40, 100, 150, 200, 250, and 300 ppb standards). The area response for the 300 ppb standard is around 370 million now.

                           

                          I can obtain an acceptable linear calibration (r2 > 0.99) on average from 150-300, but the x intercept is very negative at that point (so any non-zero response is >LOD) which is not a reasonable representation of reality. From the low end, it's obvious after 3 points that the curve is not linear.

                           

                          I am still not sure as to why this shift occurred, but after 9 months of utilizing a quadratic calibration model, I can say that it has been incredibly consistent, with r2 > 0.997 on average.

                           

                          This is by GC-ECD, so tuning is not an option.

                           

                          Regarding Benzidine:

                          The calibration for benzidine has returned to linear with the same behavior as prior to the shift, so that issue was cleared up. That instrument had been tuned/source cleaned a few times with no resolution, but as is the way of these things sometimes, it went back to normal as suddenly and inexplicably as it had deviated.

                            • Re: Sudden shift from linear to quadratic calibration?
                              gchaplain

                              Hi Megan,

                               

                              I think it is to do with the LOD over time. The detector/column wears with age and with the low end becoming more of a curved model rather than a linear model. It maybe worth having the lowest standard at 0 ppb and see where it falls on your curve or see if it fits a linear model.

                               

                              Of course one thing is for sure, if your curve model still works, it's still going to be correct, but the precision of your results would be more variable at lower concentrations.

                                • Re: Sudden shift from linear to quadratic calibration?
                                  mkpbphd

                                  Hi Grant-

                                   

                                  While I can't rule it out completely, I doubt that column or detector wear is responsible. The calibration model remained quadratic after replacing the column and the ECD on this instrument is under two years old. Additionally, LOD verification has confirmed the validity of our LOD. I'm not entirely sure what you mean by placing the lowest standard at zero (running a blank and including it, creating a (0,0) point or falsely setting my lowest point equal to zero). In any case, I don't see that it would fit a linear model. The quadratic model generates an x-intercept that is very near zero, while the linear model generates a large, negative x-intercept (which dramatically inflates near-zero values). If I used the linear calibration from the higher point standards, near-zero responses would calculate as hits higher than my LOD, which is not an accurate result.

                                   

                                  Additionally, if the concern is precision at lower concentrations, I can verify by repeated injections that the precision remains excellent.

                                    • Re: Sudden shift from linear to quadratic calibration?
                                      gchaplain

                                      Hi Megan,

                                       

                                      The LOD may have changed over time with this method on this instrument, even with it being under 2 years old. It maybe worht confirming that the LOD is still true, or at least the LOQ. I've worked on GCMS's that were under 6 months old and they had detector/HED problems.

                                       

                                      If you set teh lowest standard at 0 ppb, it will give you what the background is rather than just setting a 0,0 point proving that the relationship is quadratic or that the background signal at the lower end is subject to background and noise influence. Once you have the signal response for a blank at 0 ppb you can subtract that from all your values on your calibration effectively removing the background signal giving the true response. This may change your quadratic to a linear relationship.

                                       

                                      With the precision i meant that for example if you ran 15 injections at 20 ppb and 15 injections at 40 ppb, are they statistically different on a t-test. I expect at lower concentrations the uncertainty of ± ppb is greater than at highre concentrations.