USP Resolution Problem

[champangepowder]

I have a question about resolution. We have a method that yields several peaks. One of the current criteria for the method is resolution > 1.2. The two peaks that this calculation is performed on are not baseline resolved. The first peak is much smaller than the second and is a shoulder of the first so the first peak is tailing into the second one. This leads to the peak width at the base being more of a guess than an accurate calculation. If you draw the tangents off the first peak down to the baseline you get width values that are not true representations because you don't really know where the peak ends. Has anyone encountered this type of situation? What other calculations have you used to address this? I have heard of one alternative that measures the height of the valley between the two peaks and compares this to the height of the smaller peak. If anyone has any opinions or other recommendations I'd love to hear them.

Thanks!!

[mbicking]

You have a difficult problem, caused by poor resolution. First, you need to recognize that with overlap, and the fact that one peak is much smaller than the other, you are going to get integration errors for the small peak. Your values could be off by 20 - 50%, or more, depending on the specifics.

Your data system should be able to measure resolution for you. Manual methods are not very accurate. And if the valley is too high compared to the small peak, then even the data system may not give a reliable estimate.

The equation for resolution is fixed. There are no generally accepted alternatives - the only options seem to be how the width is measured (e.g, using half width, tangents, moments, etc.).

The obvious answer is to improve the resolution. If that isn't an option, then you will have to find a consistent way to do the measurements, perhaps using some standards as a guide.

[Mattias]

If we consider only the quantification of small peaks eluting in the tail of another peak - I remember there was an article in LC-GC many years ago about this.

The article showed that (obvious) the small peak would be overestimated with a "drop baseline", and underestimated with a tangent integration. The surprising part was that the drop baseline alternative was much closer to the true value than the tangent version. I'll see if I can find the reference if you are interested.

I don't know how the different integrations affected the calculated resolution though.

[Uwe Neue]

Procedure for the quantification of rider peaks
Journal of Chromatography A, Volume 929, Issues 1-2, 21 September 2001, Pages 165-168 Simon Jurt, Martin Schär, Veronika R. Meyer

Abstract
Rider peaks are small peaks which are not well resolved from a large and asymmetrical neighbour but sit on its trailing side. The usual case is a large, tailed peak which is eluted just in front of the small peak, although the opposite situation can also occur (a small peak in front of a large peak with fronting). The common integration techniques, i.e. separating the peaks by vertical drop or by a tangent and determining area or height, give erroneous results. We propose a method for their quantification with low error. It is necessary to set up a ‘‘two-dimensional’’ calibration by varying both concentrations, i.e. of the large peak and of the rider. This leads to a series of linear equations which describe the rider size, as found by the
integrator, as a function of the size of the large peak. The y-axis intercepts i of these equations show a linear relationship with the concentration x of the rider analyte, whereas the slopes s follow a quadratic relationship. These equations can be used to solve the equation y5s(x) ? z1i(x) for x ( y and z are the integrated peak size of the rider and the large peak,
respectively). The procedure was tested with computer-generated peak pairs as well as with HPLC separations of 2,3-dimethylaniline (large tailing peak) and 2,3-dimethylphenol (symmetrical rider peak).

[champangepowder]

I wanted to post a visual to help people see what I am talking about. The insert image isn't working so here's the link. The way the tangents are drawn in this image show that the peak width is not very accurate.



http://www.flickr.com/photos/31201328@N ... 3/sizes/o/

[mbicking]

I published two of those articles on integration errors. The links are:

"Integration Errors in Chromatographic Analysis, Part I: Peaks of Approximately Equal Size," LCGC Magazine, 24,402 - 414 (April, 2006). To access the on-line version
http://www.lcgcmag.com/lcgc/article/art ... ?id=318543.

"Integration Errors in Chromatographic Analysis, Part II: Large Peak Size Ratios," LCGC Magazine, 24,604 - 616 (June, 2006). To access the on-line version .
http://www.lcgcmag.com/lcgc/article/art ... 9&pageID=1

These continued the earlier work of Meyer and extended the results to explain why the various integration methods produce inaccurate results in certain situations. As Mattias says, some errors are obvious; some are not. For example, many analysts assume that the tangent/exponential skim methods are always the best approach, when in fact there are only certain situation where this is true.

[Bruce Hamilton]

I know that you will not like my opinion, but seeing it's Friday 13th...

1. If possible, modify the method to improve resolution - 42 min? is a long time. How variable is the impurity concentration, is that image typical?. I'd also worry about the effect of any changes in both peak shapes as the column ages.

2. Have you got pure samples of the main peak and impurity?. If so, I would prepare a calibration curve using appropriate mixtures. That way you can ascertain the best integration regime.

3. If you have to rely on your software, you may find that the presence of nearby small peaks ( as in your image ) may affect integration consistency, so you will need careful selection of parameters.

I believe the best solution would be to improve the method and resolution.

Please keep having fun,

Bruce Hamilton

[tom jupille]

I believe the best solution would be to improve the method and resolution.

Bruce hit the nail on the head.

I'll add one more quote from Norman Dyson's book on integration techniques:

. . . errors arising from peak overlap are introduced by the algorithms of perpendicular and tangent separation and cannot be eliminated by anything but better chromatography. Integrators are able to generate a highly precise and totally inaccurate set of results for all the foregoing examples.

Dyson, Chromatographic Integration Methods, 2nd ed. , pg 67; RSC Monographs (1998)

[HW Mueller]

Long ago I bought some software to resolve overlapping peaks (not only chromatographic). When I saw the assumptions on peak shape made by this software I shelfed it (never used it).
The link given by Champagne...... illustrates this clearly. Running standards might give you a good idea of what the peak shape should be, but doesnt tell you whether it doesn´t change due to all the crud that seems to be present here. If this example is a standard the chromatography is really totally unacceptable in my view. Rather it appears that this is a sample with maybe (can´t even tell this for sure) 4 peaks riding on dirt. So the baseline can´t even be determined accuratly. Having "grown up" with the rule that a single chromatography (even a very good one) can´t be used as proof of the identity and amount of a substance I go with Dyson on this.
Thanks, Tom, for the citation.