Validation_RSD variation

[montana]

Hi,

During the devlopping of HPLC-DAD method for determination of creatinine, for intra-day validation method , i found that the RSD (3 injections/standard) varied as below:
Concentrationmg/L///RSD %
0,125 ///8,35551708%
0,25///8,51927337%
0,5///1,71051364%
0,75///3,3519658%
1///0,36496194%
2///1,53302757%
3///0,97283056%
4///0,65727449%
5///0,62996642%

LOD (3.3 x SE / slope)=0,000148074 mg/L
LOQ (10 x SE / slope) =0,00044871 mg/L

My question is , are these results logic, and how can i improve the RSD.

[tom jupille]

are these results logic,

No. If you are getting almost 10% RSD at 0.125 mg/L, there is no way you are getting an LOQ that is 277 times lower.

The RSDs are bouncing around, but trending higher at lower concentrations, which suggests that the limiting factor is signal/noise ratio. Only two ways to improve that:
- more signal (better wavelength? larger injection volume? narrower peaks?)
- less noise (spectral bandpass? time constant? new lamp? clean detector cell? better degassing?)

[DR]

Only two ways to improve that:
- more signal (better wavelength? larger injection volume? narrower peaks?)
- less noise (spectral bandpass? time constant? new lamp? clean detector cell? better degassing?)

or maybe a bit of smoothing...

[montana]

Thank you, i will try to improve the RSD.

[montana]

Hi,

One more question, i want to know if the method that i use to calculate the LOD -LOQ is right or not,The same sample i inject it 3 times , and i calculate the mean of area peak value , RSD. Or I have to replicate the same sample 3 times.

[tom jupille]

If you are doing replicate injections of the same standard (or sample, for that matter), the RSD you are seeing is the "repeatability". It is a measure of the variability in the chromatography part of you analytical process (i.e. from injection through integration). It does not include any measure of variability from sample preparation.

When you are estimating LOD or LOQ, the RSD values do not enter directly into the computation. Using Excel, for example, you would enter the results from all 27 of your injections (9 levels x 3 replicates) and do a linear regression (you can use the Excel LINEST function). What you are interested in is the slope, m (otherwise known as the response factor) and the standard error of the y-intercept b ("seb" in Excel speak). You can go ahead and calculate the RSD at each level in the usual way (calculate the standard deviation and divide by the mean). Note, however, that 3 replicates is the minimum you need to calculate standard deviation. 5 or 7 would give you a better measure.

The reason I said your results were inconsistent is that LOQ amounts to a peak that is 10 times larger than the standard error of the y intercept. LOD amounts to a peak that is 3.3 times larger than the standard error of the y intercept. If you think about it, those imply (very, very roughly!) that RSD at LOQ is somewhere in the vicinity of 10% and RSD at LOD is somewhere in the vicinity of 30% (I'm ignoring a lot of statistics theory here, but you get the idea). If your 0.125 mg/L standard gave and RSD of 8%, you were close to LOQ right there.

[montana]

Thank you very much .

I get the idea, i will try to improve my results.

Thank you again.