Chromatography Forum

Hi Everyone,
I've recently developed a GC-uECD method for Trichloroacetone in one of the raw materials.
I’ve run standards prepared from stock at 25,50,100,200 and 400ppm in respect to 20mg/ml sample preparation. I've also run spiked samples at the mentioned levels.

1,1,3-Trichloroacetone / Standard Response /Spiked sample Resp
ppm
19.845 /358 /556
39.69 /901 /1327
79.38 /2320 /2647
158.76 /5568 /5916
317.52 /12780 /13584
slope 42.16705001 43.84924021
intercept -802.7916667 -589.1666667


As you can see significant negative intercept was obtained for both lines.
Can someone please give me any advice about this negative intercept?
In addition injection of my spike samples gives me higher response (although no peak detected in unspiked sample).
I read that ECD detector is not really linear what approach should I take to analyse my samples.
Thank you in advance

Maggie

It looks to me like you have a good bit of nonlinearity in your data. I took the liberty of plotting it and fitting a 2-parameter parabola (R = a1*C + a2*C^2) to it:



It wasn't clear to me what the 3rd column really represented so I just plotted the same concentrations vs. that response as well (green curve). For the red data: a1 = 27.0188, a2 = 0.0421. For the green data: a1 = 31.1424, a2 = 0.0367.

You need to know that the intercept is a fudge factor that is dictated by the average value of the data points (both x and y). The best-fit regression line must pass through xbar and ybar. I've always had trouble allowing for it in calibration data because it really doesn't make physical sense to me to have the calibration standards that I choose (somewhat arbitrarily) dictate the value of the instrument response at zero concentration. At zero concentration, you should have zero (or if you prefer, "not detected") response.

Since you have curvature in your data (concave up), you will get a negative intercept because of what I stated above.

If this type of concentration-response curve can be generated reliably for your analyte, I'd use the R = a1*C + a2*C^2 transfer function to calculate the concentrations in your unknowns. After determining the regression constants, you can use Newton's Method (see any undergraduate calculus text) to calculate the concentrations. Essentially, you'll be finding the pertinent root of a quadratic equation - the one that lies in your calibrated range. Once calibrated, you can only use the regression data to calculate concentrations that are in the calibrated range.

Linear algebra can be used to determine a1 and a2 by correlating C and C^2 to R. If you don't see how that goes readily, I can show you an easy way to do it.

Thank you for your response

I've tried this as I though quadratic fits better to my data. I did not detect any peak in my sample, so I am not sure how to calculate from quadratic equation.What should i report as my final result?

In addition the problem lies in a fact that there is a difference between my standard responses and spiked sample response (green graph) suggesting that TCA might actually be present in a sample.

M

If you don't measure a response in your sample, doesn't it stand to reason that there is not detectable TCA in your sample? If that is the case, then I would call that "undetectable". If you know your detection limit, you can say "less than X ppm".

Do you think you're having more of a precision issue? Here's what happens when I plot all of your data in the same fit:



a1 = 29.0806
a2 = 0.03942

I used these regression constants and the peak areas you supplied to back calculate what all of those concentrations should be. Here's the data in tabular format:

http://s1285.photobucket.com/user/rb6ba ... 3.jpg.html

The recoveries on your spiked sample are quite good. To me, that indicates that there may have been an issue with the lower concentration standards.

Could it be that within the precision of your injection technique, these are the same? What is your sampling technique (direct injection, headspace, SPME, etc.)?

This is direct injection method.I do not think this is precision issue.I not have a clue what that maight be.

I will run the calibration line. If any peak of TCA will be detected then calculate it from quadratic equation, if not report results as below lowest Std (25ppm). Hope that will work.

Thanks
M

I think it is precision. If you inject your lowest concentration standard and your lowest spiked sample 5 times each, what are the peak areas for TCA?

Hi

I have found from experience that ECDs tend to give non linear responses and funnily enough the issues were at the lower concentrations, Whether this was an inlet issue or the ECD I was not sure. In general as ECDs are so sensitive you tend to lower your concentrations ( or raise your split ratio ) which I think adds to the problem.

Here's what I'm saying. You make your low standard at 20 ppm. If you measure that 5 times and get a relative standard deviation of 5% (maybe it's better than that, let's just take that as an estimate). That means a given measurement could give you values like this:

20 x 1.05 = 21
20 x 0.95 = 19

The standard deviation of three measurements (take 20 as the third) is 1 ppm. Now, if you choose to only do triplicate analyses of your sample to determine the unknown concentration then you are 95% sure the true concentration is (t value for 3-1 degrees of freedom is 4.303):

20 +/- 4.303*1

or

20 +/- 4 ppm

Given the precision of the measurement, you're 95% sure that the true concentration lies somewhere between 16 and 24 ppm. If the precision is worse, the range is wider (you're not as sure) so 12 can become not so different from 20.

I think I figured out what you're after and I believe it proves my point. What you want to know, is the residual concentration to TCA in your sample and you want to accomplish that task by running external standards and samples spiked with the same concentrations of TCA as the standards. In the first step (red data above) you generate a calibration model for TCA in your matrix.

a1 = 27.0188
a2 = 0.0421

for the model: R = a1*C + a2*C^2. In your line of thinking, this is the absolute.

In the second step (green data), the measured concentration is actually Co + C, where Co is the residual concentration of TCA in the sample. So, if you use the regression model to calculate the "apparent" concentration of TCA it is actually:

Cmeasured = Cadded + Co

Cmeasured was calculated from the peak area (556 gives 19.95 for the measured concentration) and the regression constants for the red curve.

For the 5 standards, you get:

19.95 = 19.85 + Co, Co = 0.1
45.84 = 39.69 + Co, Co = 6.15
86.35 = 79.38 + Co, Co = 6.97
172.56 = 158.76 + Co, Co = 13.8
331.54 = 317.52 + Co, Co = 14.02

The average value of Co is 8 ppm and the standard deviation is 6 ppm. For 5 measurements, the confidence interval is 2.776 x 6 = 17 ppm. If you're working at the detection limit here, then it's really saying that your measured 8 ppm residual in the sample is not different from zero.

It's a precision issue. If you don't see a peak for TCA in the sample, must say it's not detectable. You can't legitimately rely on your overspike analysis to get at the residual concentration in the sample.

It's seems to be you love statistics; your comments are really helpful.
Anyway my customer should be happy as they requested method down to 200ppm.
M

Yep. I'm glad you have a method that suits your needs. If you can reliably see 20 ppm, you're way below a 200 ppm limit.

Do you know how that molecule is made in the raw material you are analyzing? I've never heard anyone worry about trichloroacetone before.