Is there any peak height limit to have good linearity?

[moonchips]

As I know, some people strictly limit the absorbance to be within 1 AU. Do we have to limit peak height to be within 1 AU? I thought if we check linearity, and if the correlation factor is bigger than, say 0.995, then we are good.

Thanks

[JGK]

Depends on the type of analysis. In previous companies I've worked at the criterion for correlation coefficient was generally >0.9995 (for non-bioanalytical methods).

However, general unhappiness with the method would set in at 0.9999 or less.

[danko]

I’m afraid you won’t believe this, but there are people checking linearity between two levels and guess what. They never experience problems related to non-linearity

And now to the question: I won’t have any problems with peak heights over 1 AU, provided the calibration is carried out properly and the samples’ peaks do not exceed the actual calibrated range. By the way the same (proper calibration) goes for peaks under 1 AU as well

Best Regards

[Uwe Neue]

I am sure that there is a suitable and valid outlier test for the highest concentration...

In general, I would think that 1 AUFS is not a problem, unless your lamp is dying.

[tom jupille]

First of all, if you can demonstrate adequate linearity for your compound, go for it (it is always dangerous to reason from the general to the particular!).

Second, assuming that you are not overloading the chromatography (which would typically show up as peak tailing problems), your UV detector linearity will be limited at the high end by stray light. Remember that spectrophotometers (and UV detectors) don't measure absorbance, they calculate it. What they measure is transmittance; absorbance is the negative log. Which means that if you have 1% stray light, the signal will be absolutely flat at 2.0 AU.

The "cutoff" depends on how much nonlinearity you will tolerate and on how well your detector keeps out stray light. I'm paranoid, so I don't like to go much above 1 AU. As indicated, other opinions exist.

[anirbanrc]

I don't forsee any problems in quantitataion.I have seen linearity being achived upto a 2 AU units also.Actually criteria for linearity with only Regression coeffiecient /correlation is very lenient .
However you may face problems in acheiving peak purity at such high concentration.Non linearity does creap in at high concentration and the deviation may not be same directional or same extent for all wavelengths.

[grzesiek]

"Do we have to limit peak height to be within 1 AU?" - short answer is no, many people do this probably because qualification of detectors linearity is usually done in this region (up to about 1 - 1.5 AU)

"I thought if we check linearity, and if the correlation factor is bigger than, say 0.995, then we are good. " - depends for whom it is done and what he expects, remember that methods are generally developed and validated not for oneself but for the custmer whoever he is, R2 of more than 0.995 can be good for one customer, and the other could say "what do I need this R2 for, anyway?"

[lmh]

sorry to be the grumpy old man again, but over-reliance on R2 values can be a really bad thing.

The important thing is that your method yields the right result. In this instance you are looking at a systematic deviation from a straight line at high absorbance. If you put a straight line through a set of slightly curvy points, you can get quite reasonable R2 values, but the points at an extreme end of the line will be wrong.

The worst offenders are obviously likely to be very low concentration points in the case where the calibration curve has gone to excessive concentrations and started to curve downwards, meaning that the straight-line fit y-axis intercept starts to move upwards from zero. This leads to huge percentage errors on low points. Hey, and fixing the y-axis intercept to zero doesn't help much either.

Incidentally, the size of the error (being systematic) is completely unrelated to any LOQ that's been measured!

The simple test is that your method must reproducibly return measured values that lie within a certain % of "correct" over the whole range of concentrations for which you plan to use it. Slight deviation from linearity is only a problem if you are determined to plaster a straight line through curved points come what may; an R2 value can be a very insensitive quality check.

[tom jupille]

R2 value can be a very insensitive quality check.

lmh is correct. You get more information from looking at a plot of the residuals from the least-squares fit. A good "quick-and-dirty" test is to plot the response factor as a function of the amount of analyte.

All of which circles around the answer to the original question, which is that so long as you can demonstrate adequate linearity (however you define it!), then there is nothing magic about 1.0AU

[Rob]

You just need to make sure that you are in the linear range. I always used to tell analysts not to go over 1500mAU as most detectors were potentially non-linear above this due to stray light effects (see the above posts).
More recently I have changed my advice as modern detectors are much better and I recently did a linearity check on a DAD which was linear up to 2600mAU. I would still feel uncomfortable at this level so would probably keep it below 2000mAU even with this. Bear in mind that it will change with compound, mobile phase, age of lamp etc so as always validate your method including linearity check and leave yourself some headroom in setting your method conditions.

[tom jupille]

One additional comment that's worth making: if you are checking linearity chromatographically, remember that most of your peak (and most of the area) is well below the maximum amplitude. That means that signficant non-linearity at the top of the peak can be masked by the bulk of the peak lower down. If you're checking under static conditions, or infusing a constant concentration of your probe/chromophore, this comment does not apply.

[HW Mueller]

One shouldn´t forget that the absorbance of the mobile phase is a hidden part of the linear range, since one zeroes the baseline at the start. In the extreme case, if the mobile phase "blacks out" the detector you are not going to even see peaks.