Accuracy problems

Basic questions from students; resources for projects and reports.

16 posts Page 1 of 2
Hi

I’m developing a LC-MSMS method. Finally, it seems that it is ok, but now I’ve problems with ACCURACY!!!! I don’t understand at all.
I prepared some QC samples for the first test, and I froze the QC remaining. The first test was ok either for selectivity…. and for accuracy.
Two days after I repeated the test, using the QC frozen. The results for accuracy are not good!!! There is around -40% of underestimation for all the levels!!!!
I don’t understand it at all!!! The accuracy is mainly regarded to the preparation of the samples, isn’t it? Is it possible that the samples can’t be frozen????
I’m really lost….
It is possible that freezing changes the samples. For example, with biological samples, you may have proteins precipitate - and take adsorbed analyte with them.
Hi,

I've "discovered" that my accuracy problems are regarding the lack of linear relation (and trying to estimate the concentrations of the QC samples using a linear relation).
But my doubt now, is WHY?????
In some cases I've a linear relation, and in some cases I don't have it!!!! Is it possible? Does it mean that my method isn't enough robust?
In the case it was always non-linear, could I adjust my data using a non-linear equation? How is it done?
Please, could somebody help me?

Thank you very very much!!!
Hi smartin,

Actually, a thread I posted here in Student Projects asking for help with weighted linear regression has a partial answer to your troubles, perhaps. There's a couple really nice threads in the Archives (see above Menu Bar, at Right Side) where you can learn about that type of regression...as well as an LC-GC article, the year was 2004, can't recall the month at the moment.

Wide concentration ranges often lend themselves to weighted linear regression...sometimes the detector type can do this (I think of ELSD, and sometimes conductivity and integrated pulsed amperometry, though in all cases generally with wide calibration ranges).

Best Wishes!
MattM
Hi Matt,

Thank you very much for your reply.
I've checked your helping in Students Projects; it's usefull for me in some ways, it's true. However, I have read that you said:

"It is possible, though, to have wonderful calibration curves....and when the samples are measured, the data returned Don't Make Any Sense...at least, in my experience"

And that is exactly my problem!!! So, taking into account your experience.... what did you do in this cases? Have I to search for a "good" model to explain my data? Or have I to improve my method?....

Oh!! I'm searching for the 2 articles you cited me. Could you give me some more reference to find them he through the site?
Hi Smartin,

It was the LC-GC North America Pittcon 2004 issue, Vol. 27, No. 2, February. That was the issue that held the Dolan discussion of weighted linear least squares regression and how to analyze calibration data to determine if that fitting is appropriate.

A reference within the July 2009 LC-GC North America, also by Dolan, may be helpful

http://www.chromatographyonline.com/lcg ... ?id=613591

as well as,

A.M. Almeida, M.M. Castel-Branco, and A.C. Falcão, J. Chromatogr., B 774, 215–222 (2002).

This also may be useful:

http://www.stat.cmu.edu/~cshalizi/350/l ... ure-18.pdf

The answer to your other main question is...maybe both the method as well as the fitting of the calibration data need to be carefully looked at. The main trouble I've had in my past with calibrations was in making an unfounded assumption that there was an absence of a "matrix effect" in the calibration standards. Generally a placebo or matrix matched to the samples...and occasionally treatment of the calibration standards as samples (like treatment) helped to repair poor correlation between calibration curve and sample data, whatever the data model that ended up as the chosen one.

Some detectors do "like", or seem to "like", polynomial fitting of data as opposed to straight line(s), weighted or not, but these are exceptions...often narrowing the range of the calibration can "cure" this.

I hope this helps a bit, and best wishes.
MattM
Thqnk you very much for the lectures!!! I'll read them!!

Regarding the matrix effect, I've think about it, but I think that it won't be the case.... as I prepare my calibration curve using blank matrix, aso (as quality controls).
I'm thinking about something regarding stability of the analite during the assay....
Hi, I had similar problem before.
I guess you are doing the project in MRM mode, right?
I would suggest you to lower the dwell time or simply turn it to auto-tune, it may help on accuracy of weak analyte.
Hope it can help you.
Hi!!

Thank you very much!!! Yes, I'm using MRM mode.
I did it before!! I changed the dwell time; of course it change a lot the definition of the peak..... but it didn't solved the problem :cry:
Thank you...

But.... what temperature are you using in autosampler? Some suggestion? Some experience with this parameter?

Thank you!!
Hi Smartin,

Maybe I misunderstood...you're actually OK with fitting equations already based at least a bit on the other thread in Student Projects.

I have used autosamplers as cold as five degrees Celsius in my past...this depends on the solvent composition of the solutions you are analyzing and stability of the analytes you are looking for/at. Watch out for condensation around and about the autosampler in this type of work!! :eye:

And you are preparing your standards in the same matrix as your samples as well. Perhaps it is entirely a case of preparation of standards/samples, their storage and their behavior upon storage.

As to MS detection with LC, I've virtually no experience other than what I have read or talked about with others...

Best Wishes!
MattM
smartin wrote:
Hi!!

Thank you very much!!! Yes, I'm using MRM mode.
I did it before!! I changed the dwell time; of course it change a lot the definition of the peak..... but it didn't solved the problem :cry:
Thank you...

But.... what temperature are you using in autosampler? Some suggestion? Some experience with this parameter?

Thank you!!


The sample temperature is room temperature, column temp is 40 degree C.
I am Newbie to LCMSMS too. Last time I am so lucky that the linearity resume after turn the dwell time down. :roll:

What is your target analyte? I am doing OP pesticide and don't really sure if the condition suitable for you.
I also tried to frozen the standard solution for weekends, and it can still give R^2 of 0.99. But the signal intensity drop comparing with calibration curve of another day.
I prepare fresh calibration standard from frozen stock everytime in need. And it seems provide a little better accuracy with "new" calibration curve.

p.s. I vented the MS last week because of electric checking of the buliding and I am not really sure if it will affect the MS performance very much.
I'm a bit worried.

(1) If your qc samples and your standards are both analyte prepared in the same blank matrix, then if you run them together, they should give similar results. There is no reason why analyte should degrade faster in a qc sample than a standard if the two are in the same background solution, stored the same way, for the same time.

(2) You said you felt the problem was caused by lack of linearity, but also that all your qc samples are down by 40%. If your qc samples span the entire concentration range of the calibration curve, and the problem is lack of linearity, I would expect some to be high, some low, as the straight-line fit will go through the middle of the curve, with some above and some below. But if your qc samples are from a narrower, realistic range of concentrations, but your cali curve is extended a lot further, then it's quite possible the qc samples will be wrongly estimated - particularly if your curve curves downwards at high concentration, so the straight-line fit is below it in the middle where your qc points are.

(3) Don't get confused about weighting and linear fits. If your points are on a curve, no matter how cleverly you weight them in regression, a straight line cannot fit through points on a curve, and no matter how carefully you fit the straight line, it will be wrong. Many people get in a complete panic about nonlinear response. If you're in an environment with people like that, you'll just have to narrow the working range of your assay until a linear fit is "good enough". Unfortunately LC-MS very often shows curvy relationships between area and concentration, and a quadratic fit often works better.

Good luck!
I apologize for tailgating...I concur with Imh on all points.

In general in my past as well, if I had quadratic or even cubic fittings for, say, saccharides by Integrated Pulsed Amperometric Detection, narrowing the calibration range could afford a nice straight line fit through the data, most times without weighting. For amino acid runs, I was kind of unable to get away from using large calibration ranges (sample runs were already taking weeks as it was without dilutions of samples to increase the experimental time), so the only way I could get about was with weighted linear regression. Same kind of thing for trace level analysis at, say, ppb levels...weighting was necessary for the calibration to be proper.

As to the fitting...it all depends on the residuals of the data points, and the repeatability (more or less) of coefficients for quadratic/cubic fittings. That indeed, the coefficient values, is where much trouble may be wrought, and why most people will shy away from polynomial fittings for calibration.

Hey Imh--is it really true that LC-MS lends itself at times to quadratic data fitting?! Wow.

Reminds me of working with ELSD...but those days are well in my past, and design of those detectors have likely improved greatly, if I believe what I read.
MattM
well, I can't swear that all LC-MS instruments give curvy responses, or that all analytes behave that way! It's just my experience so far that I'm far more likely to get a curving-off calibration curve from an electrospray MS than from a UV signal (mostly using Thermo ion trap or Agilent single quad as they're the instruments where I also have PDA).
Yes, absolutely, it's about residuals: if they are non-randomly distributed (or too big) then you have a problem.
Hi everybody !!

Thanks a lot for all the discussion, it’s really helpfully for me !!
Returning to my “accuracy” problem….. I think that there isn’t a problem of linearity or adjusted curve.
Now, I think that it could be a problem of “contamination” of my column, as I think that the response of my IS is decreasing along the sequence (I’m going to check it).
Here is my doubt: I’m using a column of 2.6µm, with a precolumn, of course. I’ve noticed, also that I have problems with high pressure on it (although I centrifuge my samples and I use a precolumn). Do you think it could be related with the “accuracy” and “contamination” I explained before?
As I don’t need a lot of resolution, I’ve thought to change to a 5µm…. Do you think it could be the solution of my problem?

Thank you very much!!
16 posts Page 1 of 2

Who is online

In total there is 1 user online :: 0 registered, 0 hidden and 1 guest (based on users active over the past 5 minutes)
Most users ever online was 1117 on Mon Jan 31, 2022 2:50 pm

Users browsing this forum: No registered users and 1 guest

Latest Blog Posts from Separation Science

Separation Science offers free learning from the experts covering methods, applications, webinars, eSeminars, videos, tutorials for users of liquid chromatography, gas chromatography, mass spectrometry, sample preparation and related analytical techniques.

Subscribe to our eNewsletter with daily, weekly or monthly updates: Food & Beverage, Environmental, (Bio)Pharmaceutical, Bioclinical, Liquid Chromatography, Gas Chromatography and Mass Spectrometry.

Liquid Chromatography

Gas Chromatography

Mass Spectrometry