Log in
Register|
Beholder ★ Russia, 2014-03-25 07:23 (4045 d 21:28 ago) Posting: # 12700 Views: 12,683 |
|
|
Dear all, Please could you help me in adjusting nominal time of blood sampling to actual time of blood sampling in WNL for BE study. I will try to explain. If we have nominal time, for example, 1.25 h, 1.50 h, 1.75 h and 1 minute time deviation on 1.50 h time point for 1 volunteer from 30 then we have following actual time for that volunteer as 1.25 h, ~1.517 h, 1.75 h. After this I get following average graph: ![[image]](img/uploaded/image86.jpg) Then you see that I got 2 points (1.50 h for 29 volunteers and 1.517 h for 1 volunteer) on one time interval (see the red circle). How could I handle it? Thank you in advance. — Best regards Beholder |
|
fno ☆  Belgium, 2014-03-25 10:47 (4045 d 18:04 ago) @ Beholder Posting: # 12703 Views: 11,259 |
|
|
Dear Beholder, Descriptive statistics at the concentration level, and therefore average graphs, should be based on nominal sampling times, not on actual ones. However, that implies that you should have a predefined rule in your SAP (or at least in your SOPs) on how to handle points deviating from the scheduled sampling time. IMHO, such a rule can exclude a point with a too large time deviation from descriptive statistics (but not from the calculation of PK parameters). BTW, the absorption profile of your test formulation looks quite weird  — Kind regards, Fabrice |
|
Beholder ★ Russia, 2014-03-27 08:50 (4043 d 20:00 ago) @ fno Posting: # 12722 Views: 11,198 |
|
|
Dear Fno, ❝ Descriptive statistics at the concentration level, and therefore average graphs, should be based on nominal sampling times, not on actual ones. Applying this approach the graphs will not reflect the real PK profiles. In terms of time-concentration matching. ❝ IMHO, such a rule can exclude a point with a too large time deviation from descriptive statistics (but not from the calculation of PK parameters). What about little deviations? 1-3 min, for example. Either little or big deviations, anyway according to this approach we need to use nominal time in order to eliminate the points from this graph... I thought just that there would be any special tool in WNL for accounting time deviations that would also allow us to make ordinary graphs. I mean without additional points which were caused by time deviations. — Best regards Beholder |
|
Helmut ★★★  Vienna, Austria, 2014-03-27 14:47 (4043 d 14:03 ago) @ Beholder Posting: # 12728 Views: 11,250 |
|
|
Hi Beholder, I agree with what Fabrice said: ❝ ❝ Descriptive statistics at the concentration level, and therefore average graphs, should be based on nominal sampling times, not on actual ones. ❝ Applying this approach the graphs will not reflect the real PK profiles. In terms of time-concentration matching. No kind of data-reduction (e.g., mean plots) will reflect the entire information contained in original data. That’s trivial. ❝ ❝ IMHO, such a rule can exclude a point with a too large time deviation from descriptive statistics (but not from the calculation of PK parameters). ❝ ❝ What about little deviations? 1-3 min, for example. Either little or big deviations, anyway according to this approach we need to use nominal time in order to eliminate the points from this graph... Yep. BTW, mean plots (I hope you are using geometric means) are not relevant in BE – they give just a first impression. Initiates will preferably look at individual plots (treatments per subject overlayed) and spaghetti plots (subjects per treatment overlayed); both with actual times. Since means of ratios ≠ ratios of means everybody had to deal with boring questions from newbies of the type: “The ratio of Cmax is given in the report with 90%. But from the mean plots I calculated the ratio as 95%. Please explain.” ❝ I thought just that there would be any special tool in WNL for accounting time deviations that would also allow us to make ordinary graphs. You can generate any plot in WNL. ❝ I mean without additional points which were caused by time deviations. As Fabrice said: Calculate descriptive statistics of concentration ordered by nominal times. Hint: Data binning. — 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 |
|
Beholder ★ Russia, 2014-03-28 09:08 (4042 d 19:42 ago) @ Helmut Posting: # 12733 Views: 11,114 |
|
|
Dear Fabrice and Helmut, Thank you very much for your lucid explanations. — Best regards Beholder |
|
d_labes ★★★ Berlin, Germany, 2014-03-28 12:30 (4042 d 16:21 ago) @ Helmut Posting: # 12734 Views: 11,111 |
|
|
Dear Helmut, ❝ Yep. BTW, mean plots (I hope you are using geometric means) are not relevant in BE – they give just a first impression. If mean curves are not relevant (full ACK with this statement  ) but only an illustration why should the type of mean matter? IMHO the arithmetic means (mostly much easier to obtain via built in functions, SAS f.i. doesn't have a function for the geomean) will do also.Of course you are right from a 'theoretical' point of view. The geomean is more appropriate if one presumes a log-normal distribution for the concentrations. But an Austrian friend of mine is not totally convinced of that  .What to do with values <LLOQ in calculating the geometric means? Setting to zero as usual would result in geomean=0 if any value at the considered time point is <LLOQ, according to the definition geomean=n-th root of the product of the valuesBut usually the geomean is calculated via logs. And if I remember a quite recent discussion correct WNL/Phoenix will produce silently missings, like my SAS beast also. That corresponds to "leave values <LLOQ out". Only R which treats log(0) = -Inf comes natively out with geomean=0 if calculated via mean of logs and then back-transformed.Using the geomean may further require different strategies for handling <LLOQ values depending on the time point. We had a discussion about this here back in the year 2008. — Regards, Detlew |
|
AngusMcLean ★★ USA, 2014-03-29 14:06 (4041 d 14:44 ago) @ fno Posting: # 12738 Views: 11,092 |
|
|
Dear Beholder, The way to deal with an outlier in your profile in US as follows:
Angus |
|
mittyri ★★ Russia, 2014-03-29 15:45 (4041 d 13:05 ago) @ AngusMcLean Posting: # 12739 Views: 11,020 |
|
|
Dear Angus, ❝ The way to deal with an outlier in your profile in US as follows: The subject with one time deviation (+1min) doesn't seem to be an outlier. What kind of SOPs for identifying pharmacokinetic outliers do you mean in this case? — Kind regards, Mittyri |
|
mittyri ★★ Russia, 2014-03-31 22:31 (4039 d 07:19 ago) @ fno Posting: # 12746 Views: 10,979 |
|
|
Dear Fabrice and All, ❝ IMHO, such a rule can exclude a point with a too large time deviation from descriptive statistics (but not from the calculation of PK parameters). What kind of excluding rule do you prefer in SAP for such cases? "The obtained samples with the time devitaions more than 5% of the current length of sampling time should be excluded from descriptive statistics" Could this rule be used? — Kind regards, Mittyri |
|
fno ☆ Belgium, 2014-04-01 19:48 (4038 d 10:03 ago) @ mittyri Posting: # 12750 Views: 10,965 |
|
|
Dear Mittyri, ❝ What kind of excluding rule do you prefer in SAP for such cases? ❝ "The obtained samples with the time devitaions more than 5% of the current length of sampling time should be excluded from descriptive statistics" ❝ Could this rule be used? It could. However, the acceptance rule should be adapted to the expected PK profile (and from your sampling schedule  ). For instance, if you are sampling in a phase with an apparent half-life of 8 h: taking a sample 10% before the planned time will result in a bias of less than 1%… no real concern here !However, if you are in a phase with an apparent half-life of 15 min (quick distribution phase for instance), taking a sample 10% too early would result in a bias > 30%… more difficult to justify keeping the concentration in descriptive statistics  .In this case, the time deviation should be kept <3% to limit the bias <10%. So all of this is about finding a good (not saying "the best") compromise that could work for most of the expected individual PK profiles (considering the inter-individual variability). — Kind regards, Fabrice |
|
Beholder ★ Russia, 2014-04-02 10:23 (4037 d 19:28 ago) @ fno Posting: # 12754 Views: 10,892 |
|
|
Dear Fabrice, ❝ For instance, if you are sampling in a phase with an apparent half-life of 8 h: taking a sample 10% before the planned time will result in a bias of less than 1%… no real concern here ❝ However, if you are in a phase with an apparent half-life of 15 min (quick distribution phase for instance), taking a sample 10% too early would result in a bias > 30%… more difficult to justify keeping the concentration in descriptive statistics Could you please clarify how did you calculate the bias for first (<1%) and second (>30%) example? — Best regards Beholder |
|
fno ☆ Belgium, 2014-04-02 19:16 (4037 d 10:34 ago) @ Beholder Posting: # 12760 Views: 10,902 |
|
|
Dear Mittiry, ❝ Could you please clarify how did you calculate the bias for first (<1%) and second (>30%) example? Assuming a local monoexponential phase around a sample planned at 1 h post-dose (sorry if that was not clearly specified): C(t+t.dt)) = C(t)*exp(-k.dt) [t = planned sampling time, dt = relative time deviation, and k = local rate constant = ln(2)/(local half-life)]. In the examples, assuming a true concentration C(t) = 100 at the planned sampling time: Case 1, half-life = 8 h: t1/2 = 8 -> k = 0.086643398 dt = -0.1 (sample taken 10% too early) -> C(t+t.dt) = 100.8701984 -> bias = 0.87% Case 2, half-life = 15 min: t1/2 = 0.25 -> k = 2.772588722 dt = -0.1 (sample taken 10% too early) -> C(t+t.dt) = 131.9507911 -> bias = 32% dt = -0.03 (sample taken 3% too early) -> C(t+t.dt) = 108.6734863 -> bias = 8.7% — Kind regards, Fabrice |
|
Helmut ★★★ Vienna, Austria, 2014-04-02 13:57 (4037 d 15:54 ago) @ fno Posting: # 12757 Views: 11,030 |
|
|
Dear Fabrice, I wonder if you are not overdoing standardization. I am definitely a friend of laying down applied methods in the protocol, which – contrary to SOPs – are reviewed by the IEC and approved by an agency, but I think that you are going a little bit too far here. Descriptive statistics of concentrations are as relevant in BE as the listing of anthropometric data of subjects. I would be reluctant to use the term “bias”, since we don’t have good (any?) evidence of the underlying distribution of concentrations. Which location parameter should we use? The arithmetic or geometric mean, the median (see this post)? BTW, most (all?) people connect data points by straight lines. Does that make sense (see this post)? ❝ […] taking a sample 10% before the planned time […] Sampling too early would ring my alarm bell concerning the proper performance of the clinical part. — 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 |
|
fno ☆ Belgium, 2014-04-04 11:49 (4035 d 18:02 ago) @ Helmut Posting: # 12765 Views: 10,953 |
|
|
Hi Helmut, ❝ I wonder if you are not overdoing standardization. Damned, I am discovered ! I agree this is close to nitpicking.The intention was just to give some math arguments to the fact that a fixed rule like "±5%" is not always the best choice. The best choice would be not allowing any deviation, but not sure that the clinical staff would appreciate it  , therefore the need for a compromise...Practically, I often use a ±10% acceptance (for a classical IR formulation). However, this can be adapted depending on the context. ❝ I would be reluctant to use the term “bias”, since we don’t have good (any?) evidence of the underlying distribution of concentrations. Which location parameter should we use? The arithmetic or geometric mean, the median (see this post)? I was not thinking of the deviation from any point estimate (for which I am generally more in favor of the geometric mean) but of the shift expected for the true concentration (so "bias" looks appropriate to me) due to the time deviation. ❝ Sampling too early would ring my alarm bell concerning the proper performance of the clinical part. Indeed, this unusual case was just a theoretical consideration to maximize the math impact of the example. — Kind regards, Fabrice |
|
SDavis ★★ UK, 2014-04-24 17:10 (4015 d 12:41 ago) @ Beholder Posting: # 12890 Views: 10,974 |
|
|
Dear Beholder, There is quite some flexibility about how to present your data in a plot; a nice trick is using the option overlay information from different worksheets, e.g. raw data is plotted against actual time (red circles) whilst the mean conc summarised by Nominal time is plotted as the black line with error bars ![[image]](img/uploaded/image87.jpg) ❝ (the set up looks like this) ![[image]](img/uploaded/image88.jpg) You could of course set up a workflow using e.g. a Data Wizard tool to codify your SOP on excluding from summaries samples that were >5min late in 1st hour or 10% late thereafter. Simon. — Simon Senior Scientific Trainer, Certara™ [link=https://www.youtube.com/watch?v=xX-yCO5Rzag[/link] https://www.certarauniversity.com/dashboard https://support.certara.com/forums/ |
Ing. Helmut Schütz