Understanding LOD and LOQ

This is a topic that's really been baffling for me.  I've been trying to run some calibration curves and quantify compounds in some samples on a GC/MS (5977B MSD/ 8890 GC) under SIM settings.  I know the concentrations of some of the compounds in the samples based on previous HPLC data.

To start off, I was unsure how much I should dilute my samples in order to get them in the ideal range on the HPLC.  Some compounds are rather abundant and some are not.  I've done a series of dilutions of the known samples with a full injection as well as 1:20, 1:100, 1:200 and 1:400 dilutions.  I then attempted to quantify the compounds to see what worked best.  

My hope was to get them all onto the calibration curve, and in general I can get them all there using the 1:100 dilution, but the compounds that fall in the lowest two calibrators don't seem to give me good concentrations.  Looking at the batch table in Quant, it generally shows accuracy values of ~200% for the next to last calibrator and ~300-400% for the final calibrator.  Knowing this I find that I need to analyze many of the compounds at 1:20 dilution and the rest at 1:100.  And even then, some of the compounds in 1:20 fall into these lower points and don't give good data.  But sometimes I do have points that fall in this range that do seem to give good results, I just don't entirely understand why (dumb luck?), and when I'm working with compounds where I don't know the concentration, how would I know when I could trust the data?

This got me thinking about LOD and LOQ, and many of the discussions I've read leave me quite confused.  Some use a formula of LOD = SD*3.3/slope, and the LOQ is 10x that.  When I try this it gives me values that essentially makes compounds that fall in the lower half of the calibration curve unusable.  Even when I know many of these points do seem to be giving me accurate data.  So I'm either doing the math wrong, or I don't understand what it means.

I found other sources that say to look at the signal to noise ratio.  If I'm interpreting this correctly, I searched for my compounds in a blank.  If the blank gave a reading of 9x10^1 then I would multiple this by 3 and I would get 2.7 x 10^2.  And the LOQ would be 10 times that, or 2.7 x 10^3.  Based on this I would look at the size of the peaks and see if they are over this measurement.  When they are, I could trust that the compound is really there and/or I could quantify it.  But when I use this technique to screen against my data sometimes it seems to work and sometimes it doesn't.

The LOD and LOQ do seem like things that I should be able to quantify directly in Quant, but I can't seem to figure it out.  I can get columns in my batch table to include these, but I can't get it to populate with numbers.  If anyone knows how to do that I would appreciate that.

Anyway, if anyone can help me figure this all out I would very much appreciate your help.

Parents
  • Hello, 

    Using signal to noise ratio of 3:1 (analyte to noise) for LOD and 10:1 for LOQ is sometimes a regulatory guideline and sometimes just a recommendation. It is not necessarily the best way to determine your LOD and your LOQ. This depends on your compounds and your method. You can always determine these values empirically.

    Whenever you are dealing with a lot of analytes with varying concentration ranges it gets difficult to have one sample dilution that will work for all compounds. But there are ways of manipulating the response of certain compound when operating in SIM mode. For example you could pick ions of lower abundance to get compounds at higher concentrations to have a lower response and fall into a similar calibration range as compounds at lower concentration. 

    But before getting into all that I would backup and ask some questions. 

    What is the linear range of the analytes you are trying to analyze? Have you made calibration curves for all your analytes (the ones you have standards for at least) to get an idea what this range might be? If not, make some a calibration curve and see how low you can go. This would be the first step in figuring out your LOD and LOQ. 

  • So I'm pretty new to all of this.  I have run many cal curves.  I've typically been running them with a highest concentration of 45ppm and a lowest concentration of 0.35ppm (basically I do a serial dilution and get 8 points, halving the concentration at each point.

    That said, I find that the last 2-3 points typically aren't entirely linear with the rest of the curve (starting at around 1.4ppm, 0.7ppm and 0.35ppm).  The last two points usually have a calculated concentration greater than where it should be.  And when I have samples that I test with compounds that fall into this range, they give back values that are too high and I find the calculated concentrations unusable.

    The thing that I find is weird when looking at these points is that the peaks are noticeably smaller as the samples get more dilute, and the response is also changing, but the calculated concentrations don't reflect this.  I've tried different variations with my curve, trying a linear fit, a quadratic fit, forcing the origin, etc.  No luck there.

    At the moment I'm attempting a new curve with the concentrations ending above these later points, it would be nice to be able to calculate at these lower concentrations, but I need to know I can trust the calculations I'm getting back.

  • Have you tried linear curve fits with weighting? 

  • So I've mostly just played around with linear vs. quadratic and forcing the origin.  I just tried changing the weighting, and found an improvement using 1/x as a quadratic.  It seems like it improves the accuracy of the lower end of my curve.  It's not perfect, but it's far better than it was.  Thank you.  This gives me something to work with.

  • How are you assessing accuracy? 

    Often when you are covering a few orders of magnitude calibration range weighting helps with linear regressions at the low end of the curves accuracy. But in general its good to follow principle of use the simplest standard curve you can. 

  • So I have an independent analysis of the compounds I'm trying to identify done by HPLC.  I'm comparing the results I'm getting from the protocol I'm developing to the previously generated data (that I'm confident in).  When I get into the lowest points on the calibration curve the concentrations that are being given are far too high for what should be there.  If I stay above that range then things are good.

Reply
  • So I have an independent analysis of the compounds I'm trying to identify done by HPLC.  I'm comparing the results I'm getting from the protocol I'm developing to the previously generated data (that I'm confident in).  When I get into the lowest points on the calibration curve the concentrations that are being given are far too high for what should be there.  If I stay above that range then things are good.

Children
Was this helpful?