Chromatography Forum

Look through the past posts on this site. There are a number of discussions in GC and LC threads on making these computations.

Hi Don,
thank you so much for the response! I am so stressed out trying to understand this. There were different ways to calculate them, however I do not know which applies to me. I am doind a project on developing a method to detect illicit drugs using GC MS. I started with simple compounds such as anisaldehyde and biphenyl, and then moved on to Benzoylecgonine, which i was able to quantify at ppm concentrations without an internal standard. I have plot of conc against area. Do i use the equation of a my line to calculate LOD? or do i have to go back to the lab and run 3 blanks for each compound as i saw in another method ?

Differing regulatory bodies require differing ways for computing LODs. The first question is are you working under a particular regulatory body or is your work applicable to people working under a regulatory body. Find out what they require. The way I learned to handle LODs was from the EPA methods - but EPA doesn't do drug analysis, unless someone is dumping drugs into the environment.

I am a master student doing my final year project on GCMS. I dont think my requlatory body has any specific requirements

I just qant to understand how to calsulate both of them.....please

Okay, since you don't have any bureaucratic issues to meet, you can go back to fundamentals.

LOQ is the lowest level at which you can quantitate with the required level of precision. Soooo:

1. What kind of precision do you need or want? 25% RSD? 10% RSD? 5% RSD? 1% RSD? (for the record, RSD = "relative standard deviation" = standard deviation divided by the mean and expressed as a percentage). I can't tell you what to shoot for, that depends on the purpose of the analysis; you will have to confer with your mentor/advisor. Once you have established that value, you can go on to

2. Run an "RSD vs A" plot, where A = "amount of analyte". A can be in concentration, mass, moles, whatever (up to you). What you do is a series of replicate analyses at different levels covering the range of interest and going down to below what you think you will need. Probably 5-6 replicates at each level would be good. Calculate the RSD at each level and then plot log(RSD) on the y-axis as a function of the log(A) on the x-axis. That plot should be pretty much flat over most of the linear range, and curve upward as A becomes very small. Now you find your required RSD on the y-axis, draw a horizontal line across to your plot, and drop a vertical line down to the x-axis. That tells you the LOQ for your required level of precision.

The procedure for LOD is similar; you can extrapolate your plot up to 100% RSD.

This is time-consuming, but thorough and gives you a meaningful value of LOQ and a justifiable value for LOD. Most of the other ways (e.g. signal/noise ratios) are only approximations in any case.

There are some useful references you should chase up:

Obviously you need to have a look at things like the ICH's Q2(R1) document, although its "Based on Visual Evaluation" paragraphs make it look as though someone on the committee took the attitude "I've been doing this for 40 years, and I know when it looks right".

There's an excelent paper by Epshtein in Pharmaceutical Chemistry Journal 38:212-225

The US DoD environmental data quality work-group produced a fact-sheet "Detection and Quantitation - what project managers and data users need to know" in 2009 which is worth a look.

Also LC-GC did a 2-part article on analytical method validation in November 2009, the second part of which dealt with LODs and LOQs.

I personally like the method based on the standard error of a calibration curve at low concentration. I'm not keen on the signal:noise method because it's not safe on instruments where there is no noise (e.g. MRMs on an ion trap with a cut-off for recording data), and because signal:noise values are very subjective (Thanks Thermo: your Genesis and Isis algorithms give values orders of magnitude different on the same data! And where do we take the noise region when working in a peakful chromatogram?). Although Tom's RSD vs A approach is good (see Epshtein), you are plotting errors, which are more variable than the values of which they are errors, and you're doing so at low concentrations, just where the errors get big. Looking at errors-in-errors is quite a messy business and requires copious, copious replication to get a meaningful result. Schemes based on measuring the error in a blank are, of course, nearly always hopeless in LC. How do you integrate a peak that isn't there?

but EPA doesn't do drug analysis, unless someone is dumping drugs into the environment.

Don, look up some of the publications by Chris Daughton

Mary,

It is amazing what is turning up in the water. And I do expect to see EPA methods showing up for personal care products and pharmaceuticals in soil and water before long.

And in the event that the EPA comes around monitoring the sewer line from my house, I've already been working up the arguments on how the unmetabolized medications that pass from me to the environment are actually a beneficial contribution, not dumping of waste. Do you think the judge will buy the argument: We do want to keep fish from having high blood pressure, or those other well-advertised side effects of ageing -- don't we?

There's an excelent paper by Epshtein in Pharmaceutical Chemistry Journal 38:212-225

I was unable to locate it. Do you have a link? Or better - can you share a copy?

Try Epstein

Kristof: really sorry, it's on SpringerLink. I didn't notice it's not free access, so if your institution hasn't subscribed, you're probably doomed not to read it. Don't worry, practically you won't miss out on much, given the abundance of freely-available information out there, including LC-GC.