5-point Calibration curve

[tom jupille]

Bruce, you make a good point: what is the purpose of the analysis? and what are the consequences of making an error?

Even for pharmaceuticals (ICH), you only have to demonstrate linearity over 70-100% of the stated value for something like content uniformity. Yes, you have to run replicates, but those are to establish that the repeatability is suitable.

If the method will not come under regulatory scrutiny, then you have to balance the benefit and the additional effort.

Once you have established that the response is linear and that the plot goes approximately through zero, then a single-point calibration to establish the response factor on a day-to-day basis could be acceptable. That said, my personal philosophy is "any experimental result should be bracketed by standards".

[HW Mueller]

Peter, because of the problem of overlap I don´t know whether I would trust that, I probably would concoct a second method (luckily I never needed to do headspace, it scares me).
Maybe I should "sign" once in a while:
Hans

[pcummins]

Ok, the curve is linear, not s-shaped and does go through 0,0.
I am ungrading to a 5-point curve simple because the samples that I am analyzing have a huge range of nucleotide content and the baseline tends to be more 'noisy' with higher dilutions. If I have a curve from 10-50uL, this covers the whole range.

[Peter Apps]

Hello Hans

If I had to ensure that the peak assigned to the analyte really is the analyte, and is not co-eluting with something else I would run the separation (only once - no need to do the whole calibration) on a different column. For anything critical or really tricky I would use an MS rather than an FID.

Headspace does give a few extra things to worry about - matrix effects being the worse of them - but you do not have to worry about extraction solvent purity ! It's main attraction is that sample prep could not be simpler; put sample in vial, crimp cap, drop in autosampler, press start.

Peter

[Bruce Hamilton]

Ok, the curve is linear, not s-shaped and does go through 0,0.
I am ungrading to a 5-point curve simple because the samples that I am analyzing have a huge range of nucleotide content and the baseline tends to be more 'noisy' with higher dilutions. If I have a curve from 10-50uL, this covers the whole range.

I'm disappointed, I wanted somebody to have an S-shaped curve

I'm curious whether the 10-50ul is referring to creating a calibration curve by injecting different volumes of the same standard solution. If your baseline is messy, varying the std injection volume has the potential to mislead.

Please keep having fun,

Bruce Hamilton

[bartjoosen]

When you have a validated method, which has linearity in the validation, you can show that a single point calibration is enough.
You first show your linearity, then you look at the confidence intervals of the slope (0 not in interval), and intercept (0 in interval).

If 0 is in the interval of the intercept, you can use a single point calibration going from zero till your point.

As others already said: If you use 1 point, you have only one chance to do a right calibration, and you wouldn't know if it's wrong.

Bart

[LewisC]

As a clinical chemist analysing human plasma samples by HPLC using fluoresence detection, we have looked into the possibilities of using single point calibration curves in particular for one of our anayltes.

We have carried out the assay of the cardioactive drug flecainide for many years, and currently continue to do so using a 7-point calibration curve, with each standard injected before and after any clinical samples (typically batches of 5-6 samples per analysis). We know the assay to be linear over our calibration range, and only accept assays with r values greater than 0.99. Over a series of assays, with different (yet highly competant!) analysts, we have retrospectively looked at the differences in the results that would have been achieved using a variety of different calibrators.

The analysis proved that despite the competence of the analyst, the use of single standard calibration is not acceptible even for this assay. As an example, our IQC, spiked at 800µg/L ranged from 486-894µg/L when back calculated using different single point calibration curves.

To drop to a five point calibration for this assay may be easily achievable, although to have confidence in each batch assay, were they carried out using single calibrators, would be very difficult when dealing with clinically important samples.

LewisC

[bartjoosen]

LewisC,

a range of 486-894µg/L is a quit broad range (at least to me, working in a pharmaceutical company). A range from 60 to 110% recovery seems a problem with accuracy and or precision.
What's the use of the 7 points of the calibration curve? To really have a range of points, or just to average the results to get a more precise estimate of your actual result?

In case where accuracy and precision is good, linearity can be proved and if the confidence interval of the intercept includes 0, it should be fine to work with a single point calibration. Although a multi point calibration can increase your accuracy and precision! But this is a choice between quality and quantity (and price).

Bart

PS: This is my way of view, not neccesairy reflecting the view of my company.

[Beaker]

Hi.

Can I throw something in to the mix here - what if your one point calibrant isn't the concentration it's meant to be, say due to a mis-calculation? We all have bad days after all. How would you know?

Also, complex sample matrix + different analysts = multiple point calibration (in my view). I think at the end of the day it depends very much on sample type, purpose, how closely you want to nail the 'true' value, and the implications of getting it wrong.

Phill

[bartjoosen]

What about 1 analyst, and multiple preparations of a standard with about the same concentration? In my opinion this is also a multiple point calibration. Some errors can be avoided in this way other (like systematic errors) can't.

[LewisC]

I was just wondering whether people are using separately prepared IQC samples to check these single-point calibration curves?

We apply strict assay acceptance criteria for all our clinical samples based on the calculated IQC concentrations, and single point calibration curves seem to produce far fewer acceptable assays.

I can only agree with Phill that the sample matrix needs to be taken into account, and for my purpose, single point calibration is not a viable shortcut.

Lewis

[ym3142]

Hi, LewisC,
it is really interesting. So would you let us know what the key element was which caused the large bias in your method/sample?
noisy base line? low sample concentration? your detector property? Please.

[Alfred88]

Dear all:
I would love to have a 10-point calibration curve! However, I have a problem.

The problem is that we don't have ample machine-time, so normally we use only a single standard calibration. In this set up, the method is demonstrated (by validation) to have linear response from 50-150% of the intended concentration.

Actually, we have two exceptions to the single-level standard 'rule' above. During method development, when we observed that results for the range 50-100-150% (of target conc.) gave a line that did not go thru the origin, we chose to use three-level calibration. This way we still get good results, even though there is a little bias in the calibration curve.

To sum up, you should use one standard if you know the approx. conc. (like in a QC lab). If your samples may differ significantly (e.g. environmental or blood samples), you can use three points.
Two points would be insufficient (you can establish a line, but not confirm linearity), while five points would be overkill.

Alfred.

[LewisC]

Hi Excel,

The chromatography for the assay is fine in terms of peak shape, and we are generous with our detection limit of 10 µg/L.

For the mentioned example of our 800 µg/L IQC sample, the range of results calculated using different single point calibrants did show bias, and this varied with the choice of single calibrant used. The very low results arose from using our lowest 20 µg/L standard as the single calibrant. Any slight variation at this low concentration will of course be amplified when calculating results 40 times the concentration of the standard. The results calculated for the same IQC when using, say, our standard 600 are far more acceptible. In general, the curves we produce are very shallow, and so small percentage differences can produce unacceptible error in clinical samples.

With biological matrices (which include our prepared standards) these slight variations do occur, and for that reason we feel single point calibration is not a possibility, especially when a batch of clinical samples may include samples where non-compliance to the drug is suspected in one case, and borderline drug toxicity is expected in another. In this instance, which single calibrant should be used?

Maybe seven-point calibration is excessive, but with good chromatography in a short run time, the improved accuracy and precision offered, especially when patient treatment depends largely on the results we produce, outweighs the small amount of extra time needed to prepare and run the assay.

LewisC

[ym3142]

Hi, LewisC,
Defenitely, you had a good job here. Do you mind letting us know what is the concentration range of your samples? What is your concentration range for those standards, according?
what was the linarity from 1200ug/L to 10ug/L if you want cover the range of 800 to 20 asssuming 20 ug/L was the concentration of one of your samples?
Also, comparing 800 to 20ug/mL, the latter is in the range of impuity therefore I'd think it is more important to get a accurate meaurement for the former than the latter such as accepitable criteria for former is +/_ 5% and 50% for latter.
If we go for the one point calibration, will it be better we prepare one standard at 800 level and the other at 20 level. Thanks for you sharing.