Accuracy of Residual Solvent Determination

[PeterK]

Hi All,

I am performing a residual solvent method validation, which is testing Toluene using GC/HS FID. 50%--150% of 890 ppm toluene solutions were spiked into the my sample to make 5 different concentration solutions, also the sample does have about 100ppm toluene and the concentration of spiked sample (100%) was about 1000ppm.

My question is what recovery of results or the accuracy should be? I mean that the result I got was about 1150ppm for 100% spiked sample, in another meaning, the recoveries of all results were about 1.15 to the theothetical value. Is that acceptable for residual solvent determination? The RSD for Standards was pretty good--1%, by the way.

Thanks for advice in advance.

Peter K

[chromatographer1]

First, if you are doing a validation, then this one rabbit test is inadequate. But assuming you will be repeating the method you want a proper way to report your results.

First your 1% variation of your stds is encouraging and is in the ballpark for a good method (0-2% is adequate reproducibility - in other words - GREAT )

I assume you made multiple preparations of the unspiked sample.

Take your regression line (0.995+ hopefully, if not then troubleshoot and determine why it is less than that and correct the problem) and determine the response per ppm (the slope) and calculate the amount in your sample.

Take the x-axis intercept of the regression line and see what value that gives you for the toluene content of the spiked samples. The difference is the recovery of toluene.

Now if you made your spiked samples from different weighings of the sample, then you may have variation in the content of the toluene within your sample. It is not surprising if one weighing of sample contains 100 ppm of toluene and another has 120 or 80 ppm, so I hope your spikes were made with the same sample solution, made with a single dissolution and weighing of your sample.

If you wish a more precise answer post the raw data here or to me privately and I would be glad to review the data and the calculations.

best wishes,

Rod

[PeterK]

Thank you, Rod.

I am going to do the STD linearity and will see how a regression line goes.

I used 5 different concentration toluene solutions to spike sample even though sample weights may not be same, the 5 final spiked sample solutions should not be affected too much, because the sample result was about 100ppm and then the spiked sample was about 1000ppm, 10 times than the original. All results I got are about 115% for recovery and the R2 of regression line for the spiked sample is 0.993.

I am just wondering that the accuracy of this kind of test may not be good enough, which is not like +-2% difference for regularly validation? Anybody who has this experience would like to share with me, it will be greatly appreciated.

Best regards,

Peter K

[chromatographer1]

Peter K.

I believe a regression of 0.993 isn't outstanding. Some may consider it adequate.

In my paper in Analytical Chemistry my regression line for extremely small concentrations for 18 solvents including toluene, from a mixture of water and DMAc, were all > 0.9995 except for very basic pyridine which was 0.995. That was at concentrations of 0.5, 10, 25, 50, 137, 484, and 1000 ng per sample vial. (nanogram is correct, it is not microgram)

These were not exceptional results, but results that were gathered during routine work period between regular daily samples. I know another researcher has done much better than this; right, Peter A. in Africa?

When preparing stds one should STANDARDIZE as much as possible. Putting the spiked additions into five potions of the same sample solution is one way to minimize errors and to demonstrate the true accuracy of the method.

You have good reproducibility from all appearances and I believe you will demonstrate even better results.

I have always found HS to be extremely precise and it would actually show my errors in preparation when properly executed.

best wishes,

Rod

[Peter Apps]

In my paper in Analytical Chemistry my regression line for extremely small concentrations for 18 solvents including toluene, from a mixture of water and DMAc, were all > 0.9995 except for very basic pyridine which was 0.995. That was at concentrations of 0.5, 10, 25, 50, 137, 484, and 1000 ng per sample vial. (nanogram is correct, it is not microgram)

These were not exceptional results, but results that were gathered during routine work period between regular daily samples. I know another researcher has done much better than this; right, Peter A. in Africa?

When preparing stds one should STANDARDIZE as much as possible. Putting the spiked additions into five potions of the same sample solution is one way to minimize errors and to demonstrate the true accuracy of the method.

You have good reproducibility from all appearances and I believe you will demonstrate even better results.

I have always found HS to be extremely precise and it would actually show my errors in preparation when properly executed.

best wishes,

Rod

I was focussing almost exclusively on repeatability rather than linearity (this was in the SA National Metrology lab with a very specific focus of being able to compare between samples). If I recall correctly headspace in other applications and other labs could give r-squared of 0.999 if everyting was well tuned (but this was usually at higher levels than Rod was using). As a last resort I would sometimes skirt around repeatability by running duplicates or triplicates.

1% rsd is actually better than I would expect from an off-the shelf headspacer (I never tried a Perkin Elmer) so the high recovery is not a repeatability issue.

How certain are you that the unspiked material has 100 ppm (not a proper unit by the way, I've been nagging another poster about it !) ?

By implication you are adding different volumes of toluene solution to give you different spiked concentrations (and presumably that volume was zero for the unspiked samples). If the diluent for the toluene has a matrix effect in the samples this might account for both the high recovery and the somewhat poor linearity. Best practise is to add a constant volume of a spike solution whose concentration is varied to give the different spike concentrations.

Peter

[PeterK]

Best practise is to add a constant volume of a spike solution whose concentration is varied to give the different spike concentrations.

Hi Peter and Rod, your help is greatly appreciated.

I exactly spiked the material in the same way as that Peter mentioned, and I assayed the material to get about 100 ppm ( or 0.01% w/w toluene). Also the Thermo head space system was used.

I absolutely agreed "If the diluent for the toluene has a matrix effect in the samples this might account for both the high recovery and the somewhat poor linearity".

I found that toluene content of the spiked sample solution, which I test after 2 weeks, are getting higher than that I tested before. That is partly why I got 115% recovery, I believe.

Peter K

[Peter Apps]

I found that toluene content of the spiked sample solution, which I test after 2 weeks, are getting higher than that I tested before. That is partly why I got 115% recovery, I believe.

Peter K

This is alarming - it points to either a very slow equilibration between the sample and the solution, the slow formation of new toluene, or contamination from the environment.

Peter

[chromatographer1]

Or it points to the original determination being in error, which is what I believe Peter K. is saying, Peter.

As Peter A has noted, it is required to achieve a consistent partitioning the amount and content of the matrix and the spike must also be CONSISTENT, where many approximations are adequate, for true accuracy good science must be exercised.

I wish more analysts would read a book before jumping into a technique. But then they would not come here and ask questions, would they?

maybe they would. If it were easy, anyone could do it.

I am glad Peter K. has a good feel for the work.

best wishes,

Rod

[Peter Apps]

In the plot of area vs quantity with the r-squared of 0.993, is the scatter around the line random, or is there actually a non-linear relationship that could be better fitted by a curve ? It would have to be a pretty sharp curve to give a recovery of 115%, but it could be a contributing factor.

Peter