Bioequivalence and Bioavailability Forum • Tlag

d_labes
★★★

Berlin, Germany,
2010-03-03 10:23
(5517 d 08:18 ago)

Posting: # 4850
Views: 10,764

Tlag - subtleties [NCA / SHAM]

Dear all,

beside the discussion we had already in this thread about some variants of defining Tlag I came across some data which made me wondering.

Here an example:


planTime time   CStr   Conc.

     0.0  0.0  <LLOQ   0.00

     0.5  0.5  12.82  12.82

     1.0  1.0  <LLOQ   0.00

     1.5  1.5  <LLOQ   0.00

     2.0  2.0 120.51 120.51

     2.5  2.5 247.52 247.52

     3.0  3.0  99.21  99.21

     3.5  3.5  76.85  76.85

     4.0  4.0 160.08 160.08

     6.0  6.0  35.26  35.26

... (truncated)

LLOQ was 9.6 ng/mL in that study.

My questions:

What do you think where to locate Tlag? Between 0 and 0.5 h or between 1.5 and 2 h?
Is it reasonable to state a definition like:
Tlag was defined as the time prior to (or at the first of) two consecutive Concentrations >LLOQ?
Do you know of any software which has implemented such?

—
Regards,

Detlew

ElMaestro
★★★

Denmark,
2010-03-03 11:49
(5517 d 06:51 ago)

@ d_labes
Posting: # 4852
Views: 8,674

Tlag - subtleties

Post reply

Ahoy d_labes,

❝ • What do you think where to locate Tlag? Between 0 and 0.5 h or between 1.5 and 2 h?

❝ • Is it reasonable to state a definition like: Tlag was defined as the time prior to (or at the first of) two consecutive Concentrations >LLOQ?

I do not have a really useful input to your question, but as often before I have an unqualified opinion.
I think your situation would merit a case-by-case assessment. I do not consider it likely that the true concentration rises to 12 after a <LLOQ and then truly drops back to <LLOQ twice and then skyrockets.
Assay variability may cause a concentration truly <LLOQ to be measured as >LLOQ etc, and that could be a likely explanation in your case. Along the same lines it could also be that the last <LLOQ is truly >LLOQ but recorded <LLOQ due to assay variation.
I am thus sure in your dataset random variation plays a role. If I were a regulator I'd probably have accepted if an applicant argued that lag time should be reflecting the time of the last <LLOQ before the median Tmax :blahblah:

. Or sumfin like that.

❝ • Do you know of any software which has implemented such?

You are equipped with the power to write clever scripts :-P

?

Best regards from the Behring Sea
EM.

d_labes
★★★

Berlin, Germany,
2010-03-03 14:44
(5517 d 03:56 ago)

@ ElMaestro
Posting: # 4853
Views: 8,497

Tlag - R qualified

Post reply

Ahoy Old Saylor,

❝ ... but as often before I have an unqualified opinion.

Thanx for that! :-P

❝ ... I do not consider it likely that the true concentration rises to 12 after a <LLOQ and then truly drops back to <LLOQ twice and then skyrockets.

But the only truth I have are the data.

❝ I am thus sure in your dataset random variation plays a role.

Sure with probability almost one!
Random variation as always. Otherwise we would not need any statistics.

❝ If I were a regulator I'd probably have accepted if an applicant argued that lag time should be reflecting the time of the last <LLOQ before the median Tmax :blahblah: .

Interesting suggestion.
BTW: You were my favorite regulator if.

❝ ❝ • Do you know of any software which has implemented such?

❝

❝ You are equipped with the power to write clever scripts :-P ?

As you know I have the POWER TO KNOW.
Here is on. Of course in R (cleverness of code may be improved).
But your suggestion yet not considered because of time.
Should be ELM1.

# DEF1: time prior to (of two consecutive) Conc.>LLOQ

# DEF2: time of the first (of two consecutive) Conc.>LLOQ

# DEF3: mean of Def1 and Def2

fTlag <- function(Ctdata1, method="DEF3", consec=2)

{

  #find first of 'consec' consecutive Ct points with Conc>LLOQ  (if available)

  Ctdata1 <- Ctdata1[!is.na(Ctdata1$Conc),]


  n <- length(Ctdata1$time)

  

  if (n<2) return(NA)

  

  Czero <- Ctdata1$Conc[1] # to have it also if multiple dose

                           # but this should be rare

  if (consec<1) consec <- 1

  if (consec>3) consec <- 3

  

  Ctdata1$on <- 0

  on <- 0

  for (i in c(1:n)) {

    if (Ctdata1[i,"Conc"]-Czero > 0){

      on <- on+1

    } else { #Conc=0

      if (on>0) on <- 0

    }

    Ctdata1$on[i] <- on

    if (on==consec) break

  }

  

  ion1 <- which(Ctdata1$on==1)    # indexes with on=1

  if (length(ion1)==0) return(NA) # we have no on -> tlag not est.

  

  ion2 <- which(Ctdata1$on==2)

  if (length(ion2)==0){

    # we have only on=1, may be multiple?

    # take the first? or the last? 

    # take time of maximum Conc?

    ion1 <- ion1[1]


    t0 <- Ctdata1$time[ion1-1] # time prior to

    t1 <- Ctdata1$time[ion1]   # time at Conc>LLOQ

  } else {

    ion3 <- which(Ctdata1$on==3)

    if (length(ion3)==0) {

      # we have only 2 consec.

      ion2 <- ion2[1] # take first if multiple?

        t0 <- Ctdata1$time[ion2-2] 

        t1 <- Ctdata1$time[ion2-1] 

    } else { # 3 consec. values Conc>0

        t0 <- Ctdata1$time[ion2-3] 

        t1 <- Ctdata1$time[ion2-2] 

    }

  }

  tlag <- NA

  method <- toupper(method)

  if (method=="DEF1") tlag <- t0

  if (method=="DEF2") tlag <- t1

  if (method=="DEF3") {

    if (!is.na(t1)) tlag <- (t1+t0)/2 else tlag <- t0

  }

  return(tlag)

}

BTW2: How thick is the ice nowadays?

—
Regards,

Detlew

Helmut
★★★

Vienna, Austria,
2010-03-03 14:55
(5517 d 03:45 ago)

@ d_labes
Posting: # 4854
Views: 8,511

Tlag - subtleties

Post reply

Dear D Labes,

I agree with or seafaring friend’s post. Make yourself clear that at the LLOQ a bias of ±20 % and a precision of ±20 % is generally acceptable (even higher for ligand binding assays or other difficult methods, if justified). Such a value (12.82>9.6) might be just chance. Now for the stupid part. For the last 30 years I applied a blinded plausibility review – such a value would trigger reanalysis. But according to the new BE-guideline and the drafted bioanalytical GL this procedure is considered blasphemy.
I’m fine with your definition.
No off-the-shelf-software I know of.

—
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

d_labes ★★★ Berlin, Germany, 2010-03-03 15:31 (5517 d 03:09 ago) @ Helmut Posting: # 4855 Views: 8,498	Tlag - subtleties Post reply
	Dear Helmut, ❝ ... Such a value (12.82>9.6) might be just chance ... Fully agreed. Therefore I asked what to do. ❝ ... - such a value would trigger reanalysis ... In this nasty dataset there are many of such points present. So a big part had the need to reanalysis because of this cause. And this would be heretic indeed. — Regards, Detlew