Linearity and non-linearity question

[AdrianF]

As I read the ICH, it says nothing about requiring 0 intercept for major component assay. In essence, your standards must bracket the sample concentration. -Tom J

This is an important point, your regression line doesn't need to go through zero- I think you might overcome the problem if you took more measurements at lower concentrations and then used weighted linear regression.

Another concern with your RS being as much as 6%, you need to be calibrating against the actual related substance unless you have very good evidence that the response is the same as the main compound

[AdrianF]

I note that one of your impurities is as much as 6%. With this very significant level I think you should be calibrating with the actual impurity in question rather than comparing it with the main peak.

[bartjoosen]

We have very good evidence about the similarity of response of the impurity as high as 6% and the response of our main component.

If we use weighted linear regression, our residuals are unacceptable high for the lower concentrations.

Thanks for your valuable input

Bart

[DR]

Since linearity appears to be good and go close to zero for 0.2366mg/mL and lower standards, I would be inclined to prepare assay samples so that they were ~0.18mg/mL (final dilution), then I would inject more concentrated preps for related substances.

Have the main peaks of interest for injections of <1mg/mL preparations been overlayed on injections of >3mg/100mL? I'm betting that they don't look the same. It could be that either your column or your detector just can't handle as wide a range of concentrations as we're all used to...

[mbicking]

This is a very interesting discussion, and raises many philosophical issues, as well as chromatographic ones.

First, we seem to be putting the "cart before the horse" when we worry too much about what ICH (or GMP, or GLP, or XYZ) seem to think about it, rather than what the science tells us to do. While ICH guidelines seem to have been written by people who know what chromatography is, I think most of the information is a "suggestion" not a "requirement." To paraphrase GMP (in my own non-regulatory way), "set up reasonable rules and then follow them." If you do something a little different (e.g., using a different calibration method), simply justify and explain it. I doubt most auditors would understand the details. They are more often looking for documentation/procedural errors.

Remember, we are fitting a mathematical model to our data, not the other way around. The best model comes from the best fit (however you choose to define it). My only concern is that most labs get hung up on r2 only, when that is a very poor indicator of fit quality. I can show you curves with "three-nines" that generate analytical errors of greater than 50%.

There are many great suggestions in this thread about how to evaluate data, so don't be afraid to use them. Just keep good notes!

[wanda50]

Our customers are asking for the relative response factors in increasing numbers. this applies even if RRF's aren't used in the testing method. they just want to know the response differences between the main peak and the rel subs. Since you are doing linearity anyway, it is a simple slope calculation.

[mbicking]

If I understand you correctly, you want to use the relative response factor along with calibration information to calculate an amount. This approach is certainly more accurate than the traditional area % calculations (which can be off by 100X), but the results are only as good as the RRF you use. Make sure it is carefully and correctly defined and determined.

[vep]

In our trichlorethanol assay on GC, we use a quadratic standardization curve, because we are near the saturation of the detector. If we dilute, we loose sensitivity. We had no problem explaining that to the auditors of the external audit team. The dilution tests work out perfectly.