Log in
Register|
auditor ★ India, 2012-01-20 12:15 (4879 d 22:00 ago) Posting: # 7973 Views: 5,599 |
|
|
Hi, We would like to know how to plot calibration curve if there are duplicate standards used at LLOQ & ULOQ levels? For example: if we take 8 standards (each of diferent level) for plotting a curve, wherein 2 levels are in duplicate (LLOQ & ULOQ) i.e. total 10 no of samples (standards). In such case if all the standards are paasing the criteria including both the duplicates then all 10 standard needs to be considered or mean of duplicates (LLOQ & ULOQ) needs to be considered while plotting the curve? Regards, Auditor |
|
Ohlbe ★★★ France, 2012-01-20 12:59 (4879 d 21:15 ago) @ auditor Posting: # 7975 Views: 5,088 |
|
|
Dear Auditor, What most people do is that they only use one of the two replicates of the LLOQ and ULOQ samples. Usually the first replicate is used, unless it fails (> 20 % deviation from nominal at the LLOQ, 15 % at the ULOQ) or there is a clear analytical reason (instrument failure, chromatographic interference). This should be defined in a SOP. Some people take whichever value gives the highest R2, but I don't like that (I don't consider R2 to be a relevant parameter). The idea of running the LLOQ and ULOQ samples in duplicate is to avoid having to truncate the curve if one of these samples is failing. Regards Ohlbe — Regards Ohlbe |
|
Helmut ★★★   Vienna, Austria, 2012-01-20 14:09 (4879 d 20:06 ago) @ Ohlbe Posting: # 7980 Views: 5,083 |
|
|
Dear Ohlbe! ❝ What most people do is that they only use one of the two replicates of the LLOQ and ULOQ samples. Usually the first replicate is used,… Really most people? That’s strange to me. Why the first and not the second? Why not both? ❝ Some people take whichever value gives the highest R2, but I don't like that (I don't consider R2 to be a relevant parameter). Right. R² is meaningless in calibration (if you prefer the statistical term: inverse regression). A nice reference (based on backcalculated concentrations): Almeida AM, Castel-Branco MM, and AC Falcão — Dif-tor heh smusma 🖖🏼 Довге життя Україна! ![[image]](https://static.bebac.at/pics/Blue_and_yellow_ribbon_UA.png) Helmut Schütz ![[image]](https://static.bebac.at/img/CC by.png) The quality of responses received is directly proportional to the quality of the question asked. 🚮 Science Quotes |
|
Ohlbe ★★★ France, 2012-01-20 15:26 (4879 d 18:48 ago) @ Helmut Posting: # 7982 Views: 5,003 |
|
|
Dear Helmut, ❝ ❝ What most people do is that they only use one of the two replicates of the LLOQ and ULOQ samples. Usually the first replicate is used,… ❝ ❝ Really most people? In my experience, yes. I mean, most people analysing only the first and last samples in duplicate. ❝ That’s strange to me. Why the first and not the second? Why not both? Their rationale is only to avoid having to truncate the curve if the first or last samples fail. If there is no objective reason to use the second value, they use the first to avoid nasty discussions with regulators (such as: the run passes with the second injection but fails with the first, why did you choose the second ? Did the first one fail by accident, or did the second one pass by chance ?). Using both repetitions would make sense to improve the fit (the two ends of the curve are where you have the more uncertainty). But then you get into other discussions such as: if you consider you need duplicate analysis to get a correct fit, then you should also analyse the subject samples in duplicate... Some other labs analyse the whole curve in duplicate, once at the start of the run and once at the end, to compensate for possible drifts. Personally if there is a drift I would rather shorten the run... But I think that's a discussion we had previously on the forum. Regards Ohlbe — Regards Ohlbe |
|
Helmut ★★★ Vienna, Austria, 2012-01-20 19:02 (4879 d 15:13 ago) @ Ohlbe Posting: # 7983 Views: 5,038 |
|
|
Dear Ohlbe! ❝ ❝ ❝ What most people do is that they only use one of the two replicates of the LLOQ and ULOQ samples. Usually the first replicate is used,… ❝ ❝ ❝ ❝ Really most people? ❝ ❝ In my experience, yes. I mean, most people analysing only the first and last samples in duplicate. Ah, now I get your point! ❝ […] if you consider you need duplicate analysis to get a correct fit, then you should also analyse the subject samples in duplicate... From a statistical point of view the best weighting scheme would be 1/s² (1/x or 1/x² are crude attempts to deal with multiplicative error). I don’t want to start a discussion whether it makes sense to calculate s² from n=2.  But there is a drawback: 1/s² can only be used for pretty ‘stable’ methods where one would never drop any calibrator.❝ Some other labs analyse the whole curve in duplicate, once at the start of the run and once at the end, to compensate for possible drifts. Personally if there is a drift I would rather shorten the run... But I think that's a discussion we had previously on the forum. Yes… — Dif-tor heh smusma 🖖🏼 Довге життя Україна! Helmut Schütz The quality of responses received is directly proportional to the quality of the question asked. 🚮 Science Quotes |
|
auditor ★ India, 2012-01-21 06:45 (4879 d 03:30 ago) @ Helmut Posting: # 7984 Views: 4,920 |
|
|
Thanks for the comments. However, we still not get the precise response for above said concern i.e. if we are taking duplicates (LLOQ & ULOQ) then we should consider both the duplicates (10 no. of samples) or mean of duplicates (LLOQ & ULOQ) for plotting? |
|
Helmut ★★★ Vienna, Austria, 2012-01-21 13:35 (4878 d 20:40 ago) @ auditor Posting: # 7985 Views: 4,947 |
|
|
Dear Auditor! ❝ […] if we are taking duplicates (LLOQ & ULOQ) then we should consider both the duplicates (10 no. of samples) or mean of duplicates (LLOQ & ULOQ) for plotting? Don’t plot means of duplicates (loss of visual information about their variability). Plot all data points you have used in the calibration, regardless whether you have used all n=10, dropped one (n=9), or two (n=8). Overlay the calibration curve based on the estimated slope and intercept. — Dif-tor heh smusma 🖖🏼 Довге життя Україна! Helmut Schütz The quality of responses received is directly proportional to the quality of the question asked. 🚮 Science Quotes |
|
Ohlbe ★★★ France, 2012-01-21 14:44 (4878 d 19:31 ago) @ auditor Posting: # 7986 Views: 4,915 |
|
|
Dear Auditor, Sorry I had misunderstood your question, I thought you were asking which point to use for calculations. Regarding plotting usually HPLC or LC-MS/MS software can plot all points, whether used or not. Points excluded from the calculations will look differently (e.g. clear circles instead of black dots). I would plot all points, including unused duplicates. Don't plot means, unless you are using means for calculations. Regards Ohlbe — Regards Ohlbe |
|
Helmut ★★★ Vienna, Austria, 2012-01-22 22:48 (4877 d 11:27 ago) @ Ohlbe Posting: # 7988 Views: 4,936 |
|
|
Dear Ohlbe! ❝ […] usually HPLC or LC-MS/MS software can plot all points, whether used or not. Points excluded from the calculations will look differently (e.g. clear circles instead of black dots). I would plot all points, including unused duplicates. Hey, that’s even better! — Dif-tor heh smusma 🖖🏼 Довге життя Україна! Helmut Schütz The quality of responses received is directly proportional to the quality of the question asked. 🚮 Science Quotes |
Ing. Helmut Schütz