LOD based on calibration curve

[Edu]

Hi everyone,
I'm new to the forum and wanted to check the following:
I'm trying to determine the limit of quantification and detection of ions by ion chromatography,
I made a calibration curve with concentrations ranging from 0.004 mg - 2 mg / L.
my blank (distilled water) gives a signal for the ion of interest, so I included it in the calibration curve by assigning a concentration of 0 mg / l.
From this curve I estimate the values of LOD and LOQ (by interpolation in the curve - "Statistics and Chemometrics for Analytical Chemistry" Miller and Miller).
From these calculations gives me the LOQ >> LOD > lowest point on the curve calibration, this is correct?
what could be the explanation? because if I am able to see the signal corresponding to a concentration greater than the blank (for example, of 0.004 mg / l ), the LOD should not be a concentration close to that ?
Thanks in advance

[tom jupille]

If your data are non-linear, that can easily happen. A calibration plot is a simplified description of our assumptions about reality.

[Edu]

Thanks for your response,
but the adjustment seems to be good. the linear regressions gives me a r of > 0,99 and no tendecy of the residual plot is obtained

[JGK]

The ICH method validation guidelines offer a number of methods for determining LOQ and LOD

http://www.ich.org/fileadmin/Public_Web ... deline.pdf

[MSCHemist]

LCGC validation viewpoint determining LOD LOQ
http://www.suregmp.com/forum/file_down_ ... LODLOQ.pdf

Here are three guidelines I use.
1. the calculation as recommended by LCGC. Set up 2 calibration curves running below and above where you suspect the LOQ is and inject each one 3 times. Set it up in excel concentration vs response. Use the linest function to get the standard error of the Y intercept and the slope.
LOD = 3.33*Se Yint/slope
LOQ = 10 * Se Yint/slope

This typically gives you a very low value

2. The concentration where the %RSD of repeated injections is ~10% at the LOQ.

3. Signal to noise ratio at least 10:1 at LOQ (this is typically a pretty high value)

[lmh]

Naive reply:

I think your concern is that your calculated LOD is greater than the lowest point on your calibration curve, and yet you know you can see a peak at the lowest cali point, so surely you can detect something? So surely the LOD is wrong?

One answer: the limit of detection isn't the lowest point at which you can see a peak. It is the lowest amount of analyte that you can distinguish from a blank sample, at some level of statistical confidence. In your case, you have a detectable peak in the blank. So the important thing is: how much bigger does the peak have to be, before you can be absolutely certain that there is analyte present? And how much bigger could it get randomly, and it still be quite possible there wasn't any analyte present?

The answers to these two questions determine your LOD, and they depend on how variable your measurements are at low concentration. If you estimate the LOD from the calibration curve, what you are actually doing is measuring a y-axis error at low concentration by getting a standard deviation of the cali curve (using suitably low concentration points), and using this to estimate ranges in which a blank measurement could fall.

Practically, it's a good idea if you're estimating LOD from a cali-curve to use a curve that has lots of points in the expected region of the LOD, and doesn't have loads of points at vastly higher concentrations. Otherwise you may get a s.d. of the curve that reflects its linearity at high concentration and errors at high concentration, which aren't relevant to the reliability of measurements close to the LOD.

[Markus Laeubli, Metrohm]

Thanks for your response,
but the adjustment seems to be good. the linear regressions gives me a r of > 0,99 and no tendecy of the residual plot is obtained

r > 0.99 does not imply a linear calibration curve.
As an example: take points from a segment of a circle curve and apply regression calculation --> your result will be r = 1.0000. I assume that you agree that that curve is not linear at all.
If you have a closer look to the distribution of your calibration points, you will realize that lowest and highest points are on one side of the calibration curve and the poits of medium concentration are located on the other side. Therefore not randomly distributes as is should be in a linear curve.

[Edu]

Hello,
Thank you all for the answers.
imh your supsociones are correct and it is also true that an r> 0.99 does not mean a good fit.
My initial question was that the value I get from LOD and LOQ by the regression analysis considering the residual standard deviation is high relative to what I had expected. In this case my LOD (0.03 mg / L) and my LOQ (0.1 mg / L) and considering that my curve starts at 0.004 mg / L, my question is whether this can happen?
May be I need to use concentrations close to 0.03 mg / L and recalculate my curves? as says imh. When analyzing the response function (RF), I note that my first points are too high and then remain within a certain range, this always happens?
With regard to the relationship between the blank and the standard, there is some "area ratio" rule? is there any value in area that is considered good? for example an area of 1000 or 10, is the same (while the signal cunpla analyte / white signal is> 3 for the LOD, for example)?
I apologize if my questions are too basic
Thanks again