A simple question from somebody who is quite new to chromatography....

Should I force my calibration curve through the origin or not. If I do my slope changes, and eventually leads to variations in predicted concentrations, as I'm sure you all know!

Any help is appreciated.

I would geuss that at low levels, the curve starts to deviate from the regression line. This is the point what you can consider as the Limit of Quantitation. I would not include points on the calbration curve that have a too big variance. So I would not include the origin in my calibartion curves.

hope this helps

Gilbert Staepels

Basically my problem is this....
For a seven point calibration curve my equation is y= 6390.5x+6519.6 r2 = .999

Some of my samples are giving me peak areas of between 500 and 900. As the line intercepts the y axis at 6519.6, any peak area below this will give me a negative concentration??
However if i force the line through the origin the slope changes slightly and my intercept becomes 0, therefore every peak area will yeild a positive concentration.

Is this allowed??
Any more to the point is this correct ?

Thanks in Advance.

I don't know about your SOPs, but ours say DO NOT force through the origin, and anything less than the lowest calibration point is written as "less than (the value)." If you are trying to extend your curve below the lowest (or above the highest) point, you're not going to get valid results. A better idea would be to make a standard slightly lower than the concentration your sample is, then shoot it and see how it fares on your curve. If it lands on the curve, then you can calculate your unknown value correctly. If it doesn't fall on the curve and you really want to know the exact concentration of that unknown, you'll have to make a curve of lower concentration standards and run it.

Sound comments!

In the absence of SOP's, you might want to check out the article by Paul King in the January, 1999 issue of LC/GC (LC/GC, 17(1), 46-50). It provides a straightforward way of identifying the range over which you can safely assume a zero intercept.

-- Tom Jupille / LC Resources

I believe the correct answer here is that you should use whatever works best with the method you are running. Ideally your curve should pass through the origin, however, the world is not an ideal place. My experience has been that the larger the calibration range, the more the curve deviates from the ideal. In other words, if you need to calibrate over a very broad range (e.g., drug concentrations in plasma), you may need to force the intercept through zero to obtain acceptable accuracy at low levels. On the other hand, if your method has a fairly narrow range at high concentrations (e.g., a drug in a finished pharmaceutical formulation) the better choice may be to not force the intercept through the origin. You may even find that the most appropriate technique varies from run to run using the same method. In any case, the calibration procedure to be followed should be established during methods development and written into the final SOP. It may even be appropriate for the SOP to allow the analyst to use either technique as long as the decision rules are clearly delineated and based upon good science.

Best Regards,

Michael Hinsberg

Hinsbar Laboratories, Inc.
http://www.hinsbarlabs.com

When you say "My experience has been that the larger the calibration range, the more the curve deviates from the ideal," what you are really saying is that you're trying to force a quadratic curve into a linear fit. I *DO* quantitate drug levels in biological fluids and it's really NOT acceptable practive in the forensic community to extend curves past your lowest standard. We never force curves through the origin; rather, we make lower standards and repeat the quantitation.
The main problem in assuming linearity through the origin is that the blood usually has some interferences that cause higher y-intercepts.

After reviewing the thread, it appears as if my response may not have been appropriate to the initial question. I fully agree with Lisa that it is not acceptable practice to extend calibration curves below the lowest standard and I did not mean to imply that it was. On the other hand, when calibrating over a large range (> two orders of magnitude) there are instances where the y-intercept, calculated using a least squares quadratic fit, is higher than the areas of the lowest one or two standards and no detectable peak is observed in the blank. Certainly, recalibrating over a smaller range is the best option, however, one doesn't always have that choice, either due to procedural restrictions or simply insufficient sample. In that case, I don't see any choice other than forcing the intercept through zero to bring the lowest standards onto the curve, keeping in mind that the lowest standard, not zero, is the limit of quantitiation. By the way, this is not a theoretical excercise, we have actually had to confront this scenario in our laboratory. If you can think of a better way around this problem, I'm open to suggestions.

Best Regards,

Michael Hinsberg

Hinsbar Laboratories, Inc.
http://www.hinsbarlabs.com

It's been longer than I care to admit but if anyone can remember back to their P-Chem classes, the relationship between concentration and "activity" of solutes in dilute solutions is largely affected by weak molecular interactions so the relationship between concentration and response for many solutions used in liquid chromatography is predicted to be second order. This means that for a given solution the dynamic range of a linear plot will not extend through the origin. An exception might be allowed in some cases by calibrating with very dilute standards so as to effectively redraw the slope of the line. Quantifying at such levels is usually an approximation anyway which will be challenged by the need for an acceptable signal to noise ratio.

Good point, but I was under the impression that "analytical scale" chromatography was essentially carried out under "infinite dilution" conditions anyway. If the sample is concentrated enough to have higher-order interactions, you would expect to see significant tailing and other "overload" related problems.

Or am I over-simplifying here?

Actually, peak shape goes to a fronting condition as a result of overload. Gross overload reels in RTs. For assays of pharm. actives, we generally shoot for 0.5 to 1.0mg/ml for our principal analyte(s), assuming it doesn't have a huge extinction coefficient at our wave length of choice. From there, we do linearity up to ~150%, down below 0.05% w/w of our target assay values. LOD & LOQ are calculated at 3x, 10x S/N and things (sample conc., inj. vol., AU/V) are tweeked if not all desired limits are met.