Log in
Register|
sciguy ☆ Canada, 2011-12-16 21:45 (5305 d 10:04 ago) Posting: # 7787 Views: 5,207 |
|
|
Hi everyone, I've been trying to develop an SOP for pk repeats and wanted to ask the forum about a few issues. In general, repeats are not recommended but may be warranted if suspected problems in the clinic (sample switching) or physiologically implausible data are shown. To identify an anomalous value, I want to build in 2 checks:
My first question is does anyone have any suggestions on how to do the second check? I'm thinking to derive a range for that value across subjects from the 25-75th percentile and check if the anomalous value falls outside this range. Moving forward, if re-analysis is issued and the re-analysed value is within xx% of the original anomalous value (ie: confirmed), it can be dropped from the stats analysis. If it is outside xx%, the re-analysed value (or mean of these values assuming they were re-analysed in duplicate) will be used in the analysis - and hope the results don't change too much. In either case, the stats analysis would need to be presented with both the anomalous value and without (or with the re-analysed values). My second question is does this sound like a valid approach? Or am I out in left field  ?Thanks, sciguy |
|
Helmut ★★★   Vienna, Austria, 2011-12-17 14:44 (5304 d 17:05 ago) @ sciguy Posting: # 7788 Views: 4,572 |
|
|
Hi sciguy, luckily you are not from the EU! Over here this is just history.  See this thread and others linked from within. ❝ My first question is does anyone have any suggestions on how to do the second check? I'm thinking to derive a range for that value across subjects from the 25-75th percentile and check if the anomalous value falls outside this range. Will only work if the subject don’t show genetic polymorphism. Imagine the suspected value comes from a subject belonging to the ‘majority population’. Your IQR will be so wide that the value will be within. Another example are delayed release formulations with short half-lives. If you look at any given time point the variability between subjects will be extreme. In my SOP I’m using something similar to your (1), analyzing neighbouring concentrations (2), and eyeball-PK (3). In order to come up with a reasonable assessment of (3) at least ⅓ of the subjects have to be analyzed before the first plausibility review of analytical data. ❝ Moving forward, if re-analysis is issued and the re-analysed value is within xx% of the original anomalous value (ie: confirmed), it can be dropped from the stats analysis. If it is outside xx%, the re-analysed value (or mean of these values assuming they were re-analysed in duplicate) will be used in the analysis - and hope the results don't change too much. Yes, why not. I have another (!) SOP – not for the EU, of course – allowing to drop an anomalous value even if it was confirmed. The most common example is sample mix-up in the clinics which you failed to track down. We can re-analysed such a sample 100times and will never be able to reject it based on analytics alone. The most nasty combination are studies of drugs with low within-subject variability (=small samples size) and high between-subject variability (e.g., polymorphism). Mix-up of a single sample in the area of Cmax will blow the entire study. Guaranteed. ❝ In either case, the stats analysis would need to be presented with both the anomalous value and without (or with the re-analysed values). Of course you have to present the data; not sure whether it makes sense to evaluate both datasets for BE. IMHO it is better to have an unambiguous decision-tree which data are considered valid. Don’t mention just your SOP(s) in the protocol; give a concise description and probably a flow-chart as well. ❝ My second question is does this sound like a valid approach? Or am I out in left field 1. Makes sense; think about my remarks. 2. Oh no, you aren’t! — 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 |
|
sciguy ☆ Canada, 2011-12-19 18:16 (5302 d 13:32 ago) @ Helmut Posting: # 7792 Views: 4,470 |
|
|
Hi Helmut, ❝ In my SOP I’m using something similar to your (1), analyzing neighbouring concentrations (2), and eyeball-PK (3). In order to come up with a reasonable assessment of (3) at least ⅓ of the subjects have to be analyzed before the first plausibility review of analytical data. Great thread - thanks for the link. So my 2) looks to be your 3). Not all the studies that my SOP will apply to will be for BE. For exploratory PK for NDA where things like polymorphism haven't been figured out, maybe an IQR would be too restrictive and eye-balling would be the way to go. Still, it would be nice to have a less subjective approach... Just to clarify on analyzing the neighbouring concentrations though, I guess you are referring to analytical variability. So neighbouring conc. are assayed as well and if they (and the suspected value) are all within some specified precision, we can rule out bioanalytical mess-ups? ❝ The most nasty combination are studies of drugs with low within-subject variability (=small samples size) and high between-subject variability (e.g., polymorphism). Mix-up of a single sample in the area of Cmax will blow the entire study. Guaranteed. Oh yes, hope to not run into one of those! Many thanks, sciguy |
|
Helmut ★★★ Vienna, Austria, 2011-12-19 19:32 (5302 d 12:16 ago) @ sciguy Posting: # 7793 Views: 4,414 |
|
|
Hi sciguy! ❝ […] For exploratory PK for NDA where things like polymorphism haven't been figured out, maybe an IQR would be too restrictive and eye-balling would be the way to go. Even for an NDA, you can see (sic!) polymorphism.  ❝ Still, it would be nice to have a less subjective approach... Think about the way we estimate λz in NCA. Except for the most simple cases (one compartment) automated methods fail more or less often – visual inspection is mandatory. ❝ Just to clarify on analyzing the neighbouring concentrations though, I guess you are referring to analytical variability. So neighbouring conc. are assayed as well and if they (and the suspected value) are all within some specified precision, we can rule out bioanalytical mess-ups? Exactly. I can’t give you my entire SOP.  Basic steps:
❝ ❝ The most nasty combination are studies of drugs with low within-subject variability (=small samples size) and high between-subject variability (e.g., polymorphism). Mix-up of a single sample in the area of Cmax will blow the entire study. Guaranteed. ❝ ❝ Oh yes, hope to not run into one of those! Have a look at this one… before the new European guidelines were applicable. We were able to justify the mix-up in the clinical phase by looking at the subjects’ lab-values and comparing them to their pre- and post-study values. The anticoagulant was citrate, so only γ-GT and albumine could be measured in the biosamples. Luckily these two volunteers had different values. When I presented this example, the answer by a regulator was: “We know that things like this happen. You should have expected it and powered the study accordingly.” CV of Cmax of this drug is ~15% (in my twelve studies and about the same number of studies published). Would have needed 98 instead of 12 subjects (CV 15⇒50%). Ethical?— 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 |
|
sciguy ☆ Canada, 2011-12-21 21:24 (5300 d 10:25 ago) @ Helmut Posting: # 7803 Views: 4,434 |
|
|
❝ Exactly. I can’t give you my entire SOP. By "certain range", you mean precision of the duplicate values? ❝ Have a look at this one… before the new European guidelines were applicable. We were able to justify the mix-up in the clinical phase by looking at the subjects’ lab-values and comparing them to their pre- and post-study values. The anticoagulant was citrate, so only γ-GT and albumine could be measured in the biosamples. Luckily these two volunteers had different values. Nice.. ❝ When I presented this example, the answer by a regulator was: “We know that things like this happen. You should have expected it and powered the study accordingly.” CV of Cmax of this drug is ~15% (in my twelve studies and about the same number of studies published). Would have needed 98 instead of 12 subjects (CV 15⇒50%). Ethical?You mean they made you keep the original values and 98 would have been needed for it to pass? That's crazy..how the do you plan for completely erroneous values into your sample size estimation? |
|
Helmut ★★★ Vienna, Austria, 2011-12-21 22:04 (5300 d 09:45 ago) @ sciguy Posting: # 7804 Views: 4,422 |
|
|
Hi sciguy! ❝ ❝ […] If the neighbouring values (i.e. before/after the suspect) are within a certain range (generally ±20%), the repeated values are considered meaningful. ❝ ❝ By "certain range", you mean precision of the duplicate values? No. I mean deviation of the two means (before/after the suspect) to their original counterparts. You see that I avoid the term ‘accuracy’ because we don’t know the ‘true’ concentration. ±20% is our general rule, unless we have justified in method development and validation that we can’t meet that (e.g. in some LBAs). Of course the actual range is stated in the protocol. ❝ ❝ Have a look at this one… ❝ Nice.. Well, we were not amused. ❝ You mean they made you keep the original values and 98 would have been needed for it to pass? That's crazy..how the do you plan for completely erroneous values into your sample size estimation? Exactly. Leaving ethics aside (that’s my interpretation of EMA’s POV) one can’t seriously plan for that. If we face true polymorphism (let’s say AUC-ratio of slow/fast metabolizers of 20:1) we would need hundreds (!!) of subjects to have enough statistical power to cover mix-up of a single sample. Gerald Beuerle of Ratiopharm presented a similar example at the joint EGA/EMA workshop (London, June 2010). Mix-up of a single sample (obviously a pipetting error like in my study). But his example was actually the other way ’round: The study would pass if the obviously erroneous sample was left ‘untouched’ and fail if the value was rejected. Even then members of CHMP’s PK group agreed upon that “it is not acceptable to exclude a value based on pharmacokinetics/statistics alone.” Weird. Defending their standpoint even though it would result in approval of a product which is not bioequivalent (needing to rub the dazzle from their eyes). To quote Dirk Barends: In BE we forget the only important person, the patient! See the compiled Q&A-document: Q Could individual implausible values in a plasma profile trigger an investigation? Can they be considered “rogue” and under which circumstances would a particular value be selected? Another quote may answer EMA’s burning question from above: Even though it’s applied science we’re dealin’ with, P.S.: I love counterquestions as part of an answer. Reminds me on my therapist. @ ElMaestro: Yes, I’ve taken my pills today. — 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 |
|
sciguy ☆ Canada, 2011-12-23 17:24 (5298 d 14:25 ago) (edited on 2011-12-23 19:08) @ Helmut Posting: # 7817 Views: 4,313 |
|
|
❝ See the compiled Q&A-document Thanks for sharing Helmut - I'll try to avoid the EMA if I can.. Happy holidays all! Edit: Full quote removed. Please delete anything from the text of the original poster which is not necessary in understanding your answer; see also this post. Happy Xmas! [Jaime] |
Ing. Helmut Schütz