5 replies to this topic

[csheme]

Hi All,

 

Recently I have been asked to look into the generation of a "NCA best practices" document for clinical PK analysis. This seems to be a tricky subject, as most of the time the analysis is very subjective, with the need to be looked at on a case-by-case basis. Nevertheless we would like to have a uniform best-practices starting ground to work from. The best practices NCA would be applied to biologics (mAbs, ADCs, etc) and small molecules.

 

I was wondering if you might be able to provide any input, or if you may happen to have any reference documents you could share with me to assist in my endeavor.

 

Some of the items I was considering to possibly include are described below:

 

1) Rsq < 0.80, HL lambda Z is excluded.

2) %extrapolation of AUCinf >20 is excluded.

3) At least 3 timepoints in the terminal phase (not including cmax) to calculated HL lambda Z.

4) Actual timepoint differences from nominal , >30% are excluded (not sure about this rule, in my past experience we followed this rule, but currently we have a much less stringent criteria of 200% or so, because we do not want to exclude observations, what do you typically apply).

5) Actual dose differences from nominal , >30% are excluded.

6) Outlier exclusion test for NCA parameters, exclude extreme statistical outliers, defined as those records outside the range of (first quartile -3.0X interquartile range, third quartile +3.0X interquartile range), where interquartile range is calculated using individual patient parameter values stratified by treatment arm.

7) The test described above is great at catching outliers whose records are >3.0X IQR, but is not good at capturing <3.0X IQR, because the lower limit is usually negative, as such we usually may exclude individuals with extremely poor exposure due to X rationale, for example (AUCinf <40X the median) etc. 

8) pre-dose positive samples (prior to the first dose) are usually excluded from the analysis.

 

What is your take on the criteria, are there any other items that are worthy of mentioning, or things you agree with or disagree with, or what general guidance do you follow for clinical PK analysis?

 

Not sure if there are any colleagues working in any regulatory setting, but if so, what are the health authorities typically looking for? Would appreciate any comments.

 

Thanks greatly for your assistance. 

Edited by csheme, 16 January 2018 - 05:37 PM.

[Simon Davis]

Hi Colby, my input would be to be cautious of writing an SOP that sets hard limits; (is 0.799 really that much worse than 0.801?).  try to phrase that your limits are guidance, not a yes/no cut off.  Eg. the analyst will aim to select points that result in a Lambda_z with an Rsq of 0.8 or greater. (I have see 0.85 and 0.75 in SOPs too).

 

Same for  %extrapolation of AUCinf.

 

for your point (4), I am not sure I follow why you would exclude a late timepoint? if you knwo the time accurately it's immaterial for anything but summary statistics, in which case I think i'd look at ssomething like 5 mins inthe first hour and 5% thereafter

 

 

regarding (5) Actual dose differences, this is a little more concerning but again if you knw the actual amount it should matter with excluding from summary statistics

 

Lastly I'm not a statistician but I'm always wary of (6) Outlier exclusion, what if this individual is how you identify some sub population phenotype that needs to be identified and studued furhter in future experiments

 

(7)  Excluding individuals with extremely poor exposure due to X rationale, seems reasonable, but I woudl definitely want to know why they were non-compliant in dosing.

 

Remmeber in Phoenix 8 many of these business rules can be set in the NCA tool!

https://youtu.be/Ubzx7DG1dEw?t=845

 

I also try to suggest that the analyst looks at the other profiles and tries to consider how the group was responding as well as the individual to try to achieve a consistent set of results.

 

 Simon

 

 Simon.

[csheme]

Dear Simon,

 

Thanks greatly for your input, I am much obliged!

 

Best wishes.

[PJLARGER]

Hello,

Further to Simon's comments.

for 6) and 7) you could consider to do stats on log (or Ln) transformed data, that will avoid a negative lower CI. I also do not see the point of excluding the subjects: In a regulatory (BE) study exclusion will not be accepted (in most cases). For other type of studies you may just be hiding out real variability that is relevant. That would be true for high exposure (e.g. do you have low metabolizers for your drug); I would also consider it true for low exposure, unless non-compliance can be proven, do you have part of the population with high clearance or low absorption...

8) These I guess one could exclude but it may cast some doubt on the bioanalytical assay and the whole PK evaluation. If there is a spurious response it may be true of all samples, what would be the impact on your estimates? If your drug is an endogenous substance you may need to consider how much of the exposure is due to exogenous vs endogenous origin. That is do you need to consider a "baseline" for all samples

Best wishes

[csheme]

Hi all, thank you greatly for your input, how about the following guidance below?

 

·       Before beginning, plot the cycle 1 profiles, or any additional cycle profiles, and overall (peak & trough) exposures across multiple cycles, pool patients together and stratify by cohort/treatment arm to get a sense of how the group is responding as well as the individual to arrive at a consistent set of results. Note any observations which may seem odd.

 

·       Actual timepoint differences (%) may need to be evaluated, use (actual time-nominal time)*100/nominal time, pay attention to any samples with large deviation in timepoints (> or <30% or larger deviation) which may need to be excluded. Larger deviation in earlier timepoints might have a greater impact on results compared to later observations.

 

·       Actual dose differences (> or <30% difference from nominal dose) may also need to be excluded from the descriptive summaries statistics.

 

·       Pre-dose positive samples (prior to the first dose) which are several fold-larger than the BLQ may need to be excluded from the analysis. Pre-dose samples throughout treatment should be screened, and any C_max like observations may need to be excluded.

 

·       Post-infusion concentrations throughout the treatment should be screened and any BLQ or substantially low values may need to be excluded.

 

·       The following PK parameters may be calculated, additional parameters may be needed

o   T_max

o   C_max

o   C_max/Dose (for dose escalation studies)

o   AUC_0-tau (may have to define a partial area)

o   AUC_0-inf (if adequate washout), most important for single-dose

o   AUC_all                                                                                                                         

o   AUC_0-t (AUC_last)

o   AUC_0-inf/Dose if needed

o   t_1/2λz

o   CL_ss (or CL_ss/F)

o   V_ss (or V_ss/F)

 

·       Aim for at least 3 timepoints in the terminal phase (not including C_max) to calculate t_1/2λz. Aim for an Rsq > 0.80, t_1/2λz may need to be excluded if a poor Rsq is observed. Observation range should span >2Xt_1/2λz for a reliable estimate of t_1/2λz. For t_1/2λz calculation, use the built-in regression formula (unweighted) and let WinNonlin (WNL) decide. Let WNL decide what is appropriate to permit for reporting of half-life and relevant parameters. Analyst should review all plots to ensure appropriate calculation of terminal half-life (check for missing data, adequate sampling time, outliers, etc). Outliers may be excluded from the regression if they can be clearly identified and justified.

 

·       Aim for a % extrapolation of AUC_inf <20%, and at the very most <50% based on the amount of data available, otherwise AUC_inf may need to be excluded. Check the percent extrapolated area. If above the noted window AUC_inf or V_ss may not be reported. CL_ss may be reported if steady-state and the exposure over the dosing interval is characterized.

 

·       If a very large, unexplained, or unexpected variability is observed within PK parameters, consider looking into an outlier exclusion test. As such extreme statistical outliers may need to be excluded, these are defined as those records outside the range of (first quartile -3.0X interquartile range, third quartile +3.0X interquartile range), where interquartile range is calculated using individual patient parameter values stratified by treatment arm. Log-transformed data may be used for the exclusion test. Check if poor exposure is related to non-compliant dosing. Limitations with this approach are evident, as you may be potentially excluding some sub-population phenotype that needs to be identified and studied further.

 

·       BQL rules- Set LLOQ to zero, add LLOQ line to plots if a single value

o   For NCA, always set BLOQ to zero, and WNL will exclude as needed

o   For plots, set BLOQ to ½ the LLOQ, and use a dotted line to distinguish imputed data

o   Tables and listings. Set to ½ LLOQ and follow the 1/3 rule

 

·       All plots may be linear scale and as appropriate on log-linear scale (i.e. concentrations spanning a wide range >10-fold or including the terminal phase) for visualization. The LLOQ may be shown with a horizontal dotted line. For linear and log-linear scale, plot the mean + Stdev.

 

·       Summary stats always includes the following: Mean, Stdev, CV%, Geometric Mean, Stdev of Logs, Median, Min, Max.

 

·       Formal comparisons of C_max or AUC may use the geometric mean (90% CI). For any exposure comparison the geometric mean of test to geometric mean of reference with a back-transformed 90% CI on the log scale could be applied.

 

·       Comparisons of concentrations (C_xhr, C_max, C_min, C_trough) may use the mean.

 

·       Steady-state: Day 1 Ratio (based on individual patients), Note: accumulation might imply carry-over of drug which does not always explain higher steady-state exposure (i.e. TMDD). 

Edited by csheme, 23 January 2018 - 05:19 PM.

[spic]

Hi all,

 

Thank you greatly for your input about NCA rules.

I'm discovering the 8.0 version of Phoenix and the addition of criteria for compliance with rules (very useful).

Is it possible to download the Phoenix project used for the demo on this video:

https://youtu.be/Ubzx7DG1dEw?t=845

 

I would like to exclude parameters according to the defined rules, for example:

- exclude Kel,t1/2, AUCext, AUCinf,Cl and V if R²<0.7 and if span <2

- exclude AUCext, AUCinf,Cl and V if R²<0.7 and if AUCextr>20%

I can apply the rules in the NCA object but I would like then to define a  template worflow to automate these rules in order to be able to export an.xpt file in which values that do not respect the rules are excluded.

 

Thanks for your help,