Chromatography Forum

Good day

Could somebody help to solve the following problem

I work with LC/MS/MS with API mass-sprcctrometer, Analyst software

In my calibration I can use both linear and non-linear cases with weighting or without.

Last time I used quadratic calibration with weighting 1/y and got r=0.9999
with Accuracy from 96 % to 100.04%

I wonder what formulas does the Software use to estimate Accuracy?
can I make a report which can show it?

P.S.: earlier I used linear calibration with no weighting and estimated error of calibration according to Accuracy and ISO 8466-1, but got too narrow linear interval (0.5 -5 ng/ml)
Using of quadratic calibration with weighting gave me an interval 0,5 -100 ng/ml with practically the same accuracy.
I wonder if it is reliable to use such weighting and can I use data Accuracy to estimate error of calibration.

It has been a while since I used analyst software.
However...
I do not believe we ever used 1/y weighting/quadratic.
I think our Y axis was Internal Standard area /Internal Standard area and the X axis was Concentration of the standards.
We used linear regressions only and used 1/x or sometimes 1/x^2 weighting.

While personally I saw nothing wrong with a calibration curve that was quadratic, the company was moving towards linear regressions, and if quadratic regressions were required we looked into lessening the analytical range and using sample dilutions in order to achieve a linear regression.
At the time it was felt that having linear regressions was more "pleasing" to the FDA.

The fact that you are having a quadratic response to such a small range (0.5 ng to 100 ng /mL) may suggest that your recover varies with concentration or perhaps you are saturating your detector.

Are you using an Internal Standard in your samples?
You might also try switching over to 1/x weighting and see if that helps but I would not be surprised if it did not.

Alp

Thank for your answer

Thank for your answer

I have a quadratic response to the range 0.5 ng to 5000 ng /mL, but use more narrow range to save detector. There is no oversaturating of detector.

I do not use internal standarts in this experiment, but work with calibration with external standarts.

Linear regression is posible for narrow range 0.5 ng to 100 ng /mL, but "r" and "Accuracy" is not so good as for quadratic.
I need to make estimation of calibration uncertainty, and I am very interested why FDA required linear only?

It is known that linear regession has lowest uncertainties, but in my method I propose to use any type of calibration with "r">0.999 and "Accuracy">95 %

That is why I want to see formulas which Analyst uses in calculations of "r" and "Accuracy"

A quadratic response doesn't only suggest that you are saturating the detector but you could also be suffering from the effects of ion suppression.

If your non-linear response is not due to either of these I would look more closely at how you prepare the samples eg non specific binding at the lower concentrations

I talk about standart solutions, no prepatation of samples

I need to solve the problem

if it is enough to say in my method - let's use any calibration with r>0.999 and Accuracy> 95 %

or let's use linear calibration only