Is manual integration acceptable?

[lmh]

(1) I suppose the answer is this:

In any sequence of samples, you will have QC samples to check that the method is performing within specifications. For example, if I'm measuring samples that are really, really low, and the answers that I will report are "present"/"absent", it makes sense for me to include low calibration points and assess my LOD in the same sequence as the samples.

What I need to do is make sure that I process my samples in exactly the same way as the LOD standards/QC samples or whatever else I need to validate my method.

The integration parameters can be different from day to day, or lab to lab, but each day I am choosing the parameters in a defined way, and (most important!) testing that what I have chosen will do the job to a certain standard.

This is why I'm not happy about manually tweaked peaks. If I start tweaking, how can I be sure I don't try just a little bit harder on my LOD standards than I do on the real samples? So does my LOD really still apply when I quote "absent" from a particular sample?

By the way, there's a big difference between manually chosen integration parameters (used on all samples in a sequence), and manual integration (drawing baselines by hand to where they "look right").

(2) Alex Buske, exactly my point. It seems wrong to be overconcerned about smoothing and then gloss over integration, when the latter is worse-defined and can do much more harm than the former. I'm certainly not advocating throwing away, or hiding, the original unsmoothed data. I'm just advocating well-defined/validated smoothing as part of the process, and as a tool that can help the integrator towards doing a better job, and therefore create a more reliable end result.

But the smoothing debate will no doubt go on for years.

[mbicking]

Having visited many different labs over the years, and seeing many reports cross my desk, my impression is that the reason many people manually integrate is that they don't know how to use the integration program properly. I could write a couple pages of stories, but I don't want you falling asleep while you read this!

This statement is independent of the software program. While I have the most knowledge and experience with ChemStation, while working recently in Empower, it only took me a short time to figure out how to get the program to properly set the integration limits properly. Yet the "experienced" users had to do it manually.

Having the software make the initial decisions saves time and removes any bias, whether intentional or not, on the part of the manual integrator. As noted above, consistency is the key. But as also noted above, computers are stupid, and someone always needs to check the data using some chemical common sense.

Given the publicity and consequences associated with scientific "cheating" I am reluctantly forced to recommend that manual integration be forbidden in all but the most intractable cases.

[HW Mueller]

Well, some people have the "most intractable cases" all the time.
Also, if one wants to cheat one can do it splendidly with automation.

[mbicking]

HWM:
Yes, I agree that the most clever people can always find a way around the rules. There will never be a way to catch them. It's the clumsy and foolish that we try to prevent from doing bad things.

I would argue that "intractable" is a relative term. If you say an integration problem is intractable, I would believe you, because you have the experience to make that decision. But a problem that is simple for you and I to solve may seem intractable to someone who does not understand integration or their software. In such cases, they simply do not have the tools to solve it, which is a different situation.

[HW Mueller]

Merlin, if you are saying that there is not enough professionalism in analytical chemistry I agree with you completely. Thus, clever people, who always think it´s advantageous to cheat, are actually stupid as sooner or later another lab, etc., will notice.

Of course, professionalism is hard to define, still, I think most people will know what I mean, so I will refrain from a tortuous attempt to explain.

[mbicking]

Hans:
I agree completely.

The bigger issue I suppose is that we try to prevent bad behavior by adding rules, but we often do not choose the rules carefully. The result is that we honest scientists have more restrictions on our work, and the cheating still continues. However, I do not think it is common in analytical chemistry (at least I hope not).

Professionalism is one problem, but experience is another. There are too few highly experienced scientists still in the lab. Who is teaching the next generation? Companies will not pay for training, so the analysts learn it on their own, sometimes with unfortunate results.

[Ron]

One of the situations I have run across numerous time is a scientist or operator that says "This can't be done" when what they should be saying is "I don't know how to do this". It is amazing the number of people that can't properly identify the correct phrase, and they are very different.

I tend toward the position that in a well designed method with well characterized samples there should be little need for manual integration, or there should not be a need for manual integration on a routine basis for routine samples.

[Jade.Barker]

...the reason many people manually integrate is that they don't know how to use the integration program properly.

... it only took me a short time to figure out how to get the program to properly set the integration limits properly. Yet the "experienced" users had to do it manually.

... I am reluctantly forced to recommend that manual integration be forbidden in all but the most intractable cases.

Merlin, I totally agree with your points. When I took an Empower class last fall this is pretty much what the trainer said. Usually, the software should meet your needs without having to go manual.

My 2 cents,
Jade

[wrighty]

I also hail from a cGMP site and from time to time have to maunal integrate data such as to correct baseline integration, change report format to generate performance data.

Our guidelines for re-integrating is thus:

-Only specific indivudals with an acceptable security level on Chemstation can perform this task.
-The reason for re-integration must be obvious
-the original chromatogram with a re-print of the original and the re-integrated chromatogram should be stapled together so all changes can be easily seen.
-Any changes must written on the re-integrated chromatogram, signed and dated by the analyst who has performed this task.

This is a dangerous area where data can be manipulated to give the "correct" result which is why we follow the above guidelines. If we find that the same cause for re-integration is consistently occuring then we review the method under use.

This method for manual integration was in place during the most recent MHRA audit and no comments were made.

[mbicking]

I like this approach. It offers an option for those rare problem chromatograms, but it requires additional effort and documentation, so as to discourage its frequent use.

[Jade.Barker]

Maybe I'm confused but here's my logic: If the integration works for the check standard, it should work for the sample, and it would just be part of the processing method for the sample, right? Therefore you *shouldn't" need to do manual integration at all (in theory)? Because the same rules should be applied to all the samples... I'm so new at this, help me understand - Can anyone share some examples of the situation that would cause the integration to work for the standard but NOT the sample?

I'm imagining maybe a situation where their is interferance present in the sample that isn't in the standard? I never got to dig into it, but I had a suspicion some samples in my previous lab had interferance due to trace amounts of cleaning agents interfering with my ion exchange... But still, I don't think I would try to fix the wonky chromatogram with manual integration... So guess I'm still lost...

[mbicking]

Jade:
You're not lost. This is all very common.

Ideally, our standards are well resolved with good peak shape and no baseline disturbances (remember, I said ideally). But for some sample types, particularly those of environmental or biological origin, it is common to have changes in the baseline, either from individual peaks, or so many unresolved components that the baseline is a broad "hump" (variously referred to as a "humptane" peak in GC, or generally, a "humpogram!" ) Integration in these cases can be difficult, but as many of us have noted, difficult does not mean impossible.

[lmh]

Jade, I don't think your lost. If, for instance, you've assessed a LOD or LOQ using a particular integration, and then use a manual approach on the samples, your LOD or LOQ is no longer valid. The reason is that you are changing the areas on small, difficult peaks (if you weren't changing the areas, there would be no reason to integrate manually...), and if you are changing the areas, you are changing the errors on the areas, and it is these errors that define the LOD/LOQ.

Unfortunately, resorting to manual integration throws a lot of standard quality control procedures out the window, so there are wide ranges of work where it shouldn't be accepted.

On the other hand, if you've carried out a phenomenally expensive study and produced lots of difficult biological samples, there may be a strong financial case for lowering standards a bit to assess whether the result of the study is ever going to be interesting enough to justify developing a more robust assay.

[Jade.Barker]

... and if you are changing the areas, you are changing the errors on the areas, and it is these errors that define the LOD/LOQ...

That makes sense - thanks for the plain language answer

[shahis77]

no guideline says it is not acceptable but try to do it with system as most of the auditors accept only automated integration.....with manual integration they think you manipulate the data.