"Good" Peak Integration

[gbalock]

This may churn up a lot of comment, but I am looking for a description of what qualifies as proper peak integration. I know when I look at a chromatogram whether it is properly integrated or not, but how would you decribe that. I am looking to set some criteria that analysts in our company can use that are not so restrictive that they become paralyzing. I'm looking forward to a livley discussion.

[ccyting]

If you have a sharp peak, various preset peak width and threshold values will not affect the precision that much. Please pay attention to both standard and sample peaks. For a relative braod or small peak, it is more tricky to define the peak width and threshold value. Again, it is difficult to use the same integration values for those peaks. The most important is the reproducibility of the separation (peak shape). We normally start with a pre-set values (from method and validation experience) and evaluate the results and make required adjustment as needed. Many times we need manual integration for the peaks not completely separated. Good luck.

[HW Mueller]

Proper peak integration is that which gives you the required accuracy with various standards, spiked samples. Or, even more simply: It´s that which gives you correct results.

[Rafael Chust]

I may add... consistent, reproducible, linearizable results!

[Sunjay]

The best way is to use same parameters for the standard and test peaks, wherever posible.

Regards

[KC]

George, I don't want to be critical here but it sounds as though you are either not the one doing these analyses or associated with a QA department. Setting universal criteria for integration of peaks in various analyses sounds like a nightmare to me. As analyst and study director, I will decide what is proper integration and if it does not meet with your criteria then your criteria are wrong. Been doing this about 12 years with HPLC, GC, LC/MS/MS and GC/MS and never had an agency question integration of a peak.

[Paulhurleyuk]

I agree, this one will churn up a lot of comment....First, I don't know of any definitive guidelines for what is 'good' integration. I would check out some of these references for general background:

• Felinger, A. & Guiochon, G. (2001), 'Validation of a chromatography data analysis software', Journal of Chromatography A, vol. 913, no. 1-2, pp. 221-231.
  Papas, A. N. & Tougas, T. P. (1990), 'Accuracy of Peak Deconvolution Algorithms within Chromatographic Integrators', Anal.Chem., vol. 62, no. 3, pp. 234-239.
  Dyson, N. (2000), 'Towards a quantitative definition of perfection in chromatographic analyses', Journal of Chromatography A , vol. 868, no. 2, pp. 305-312.
  Dyson, N. (1990), Chromatographic Integration Methods, Royal Society of Chemistry.

I would especially recomend the last book by Dyson, which contains a good overview of methods of integration, and why you should have more than twenty data points across a peak.

Once you've read those, I would have a few guide's
Let your analysts decide what is good intergration, sometimes it is just an art !
Try to always get analysts to use automatic integration, but let them tweak it. Manual integration is to variable between samples.
Paul.

--
Paul Hurley http://www.paulhurley.co.uk

[Mark Tracy]

I think he is not asking to set fixed integration parameters, but what are the criteria for judging whether or not the integration is correct. This can be tough when you are testing for impurities. Quantitative criteria probably will fail to include all the possibilities. I would suggest visual training of the analysts using a set of examples, and testing their proficiency on a set of standard chromatograms. The human visual cortex has an amazing capacity for data analysis when that data is presented in the right way to a trained mind.

[gbalock]

Mark,
This is exactly what I'm looking for. It is a training problem. The criteria that we are exploring is to say that the peak must be integrated from baseline to baseline when the peak is viewed at 1/10 of full scale.

[KC]

And what do you do when you are looking at a day 180 soil sample from somewhere and low and behold there is another component present preventing you from integrating from baseline to baseline. Your peak of interest is still 95% resolved and 100% resolved in most of your chromatograms but you have a handful out of a thousand where you must drop a tangent and integrate from baseline to this point. Then must the analyst or study director prepare an SOP deviation or write some sort of note-to-file explaining what he or she did? If so, I feel that would be a waste of time and would fight tooth and nail to prevent such a rule from being incorporated. Like I said in my original comment, it sounds like something that a QA dept would initiate.

[Paulhurleyuk]

KC, I don't think anyone is suggesting hard rules written into an SOP. What we're talking about is how to traina nalysts so everyone does approximatley the same.

In your example, with a new, very small rider peak on your main peak there are several ways to handle it, some people would drop a vertical line and integrate the two peaks as main peaks, some would integrate the rider as a rider with a straight skim and some would integrate it as a rider with an exponential (curved) skim. All could be argued to be correct, the point is that everyone in a particular lab or company should be doing the same thing

I agree with Mark that the best way is to demonstrate what 'good' integration looks like on several examples, introducing some 'guidelines' if possible (like always use automatic integration first, try and resolve baselines visually at 1/10 scale etc), and then get trainees to do some integration and compare the results.

Paul.

--
Paul Hurley