estimated AUC72 [NCA / SHAM]
Hi Ratnakar,
I think that your approach is not optimal, and the sponsor’s one is simply wrong. Let’s start with the sponsor’s. If a concentration was measured at 74 hours one cannot simply shift it to 72 hours (the “as-if” approach) ⇒ AUC72 will be biased. Guidelines require the actual time, not the scheduled one.
Your approach is better, but biased as well. Imagine that F=100% and tlast after reference was 72 hours and after test 74 hours. Naturally AUC74 (apples) > AUC72 (oranges). The apples-to-oranges ratio will be >1 and you get a positive bias for F. It is a weak argument that due to randomization this should happen equally often to T and R (~similar number of apples and oranges in the fruits’ basket) and therefore means out. Why not calculate AUC72 for both? In Phoenix/WinNonlin request a partial AUC (Start Time 0, End Time 72). C72 will be lin/log-interpolated between the timepoint preceeding tlast and tlast.
Example: t½,abs 1 hour, t½,el 36 hours, F 90%, lin-up/log-down trapezoidal rule.
F at 52.52 hours is 90.10% (bias +0.11%). If one uses AUClast for both formulations (your method) one gets 91.24% (bias +1.38%). The sponsor’s “method” of shifting 21.65 measured at 74 hours forward to 72 hours would give 3362.33 (bias –0.22%). The interpolated concentration of test at 72 hours is 22.50 (note: 22.50/25.00=0.90!) and AUC72 3371.96 – which would give F 90.06% (bias +0.07%). If you don’t have suitable software, the interpolation formula is:
I think that your approach is not optimal, and the sponsor’s one is simply wrong. Let’s start with the sponsor’s. If a concentration was measured at 74 hours one cannot simply shift it to 72 hours (the “as-if” approach) ⇒ AUC72 will be biased. Guidelines require the actual time, not the scheduled one.
Your approach is better, but biased as well. Imagine that F=100% and tlast after reference was 72 hours and after test 74 hours. Naturally AUC74 (apples) > AUC72 (oranges). The apples-to-oranges ratio will be >1 and you get a positive bias for F. It is a weak argument that due to randomization this should happen equally often to T and R (~similar number of apples and oranges in the fruits’ basket) and therefore means out. Why not calculate AUC72 for both? In Phoenix/WinNonlin request a partial AUC (Start Time 0, End Time 72). C72 will be lin/log-interpolated between the timepoint preceeding tlast and tlast.
Example: t½,abs 1 hour, t½,el 36 hours, F 90%, lin-up/log-down trapezoidal rule.
t R T
0.00 0.00 0.00
0.50 28.33 25.50
1.00 48.09 43.28
2.00 71.22 64.10
3.00 81.89 73.70
4.00 86.34 77.70
5.50 87.74 78.97
7.50 86.00 77.40
10.50 81.63 73.46
14.50 75.64 68.07
20.00 68.04 61.24
27.50 58.89 53.00
38.00 48.11 43.30
52.25 36.57 32.91
72.00 25.00 NA
74.00 NA 21.65
AUC52.25 3142.42 2831.29
AUC72 3744.03 NA
AUC74 NA 3416.11
F at 52.52 hours is 90.10% (bias +0.11%). If one uses AUClast for both formulations (your method) one gets 91.24% (bias +1.38%). The sponsor’s “method” of shifting 21.65 measured at 74 hours forward to 72 hours would give 3362.33 (bias –0.22%). The interpolated concentration of test at 72 hours is 22.50 (note: 22.50/25.00=0.90!) and AUC72 3371.96 – which would give F 90.06% (bias +0.07%). If you don’t have suitable software, the interpolation formula is:
Ĉ = ℯlog(Ci–1)+|ti–ti–1|/(ti+1–ti–1)×log(Ci+1/Ci–1)
Now for the nasty parts. The study failed if evaluated according to protocol. I don’t think that the FDA will accept the sponsor’s “method” – cherry-picking and simply wrong. BTW, I’m surprised to see such a large difference. You can present BE based on AUC72 as a sensitivity analysis and discuss the impact on BE (if the conclusions differ). For the next studies describe what you intend to do in the protocol and stick to it. If you don’t feel comfortable with an estimated AUC72 (i.e., prefer mixed fruits), decrease the time allowance window. As an example EMA requires for studies in steady-state (τ 24 hours) a maximum deviation of 10 minutes.—
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
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
Complete thread:
- Actual time for PK analysis ratnakar1811 2013-10-07 08:07
- estimated AUC72Helmut 2013-10-07 13:36
- estimated AUC72 ratnakar1811 2013-10-08 08:20
- estimated AUC72 sam 2013-10-08 13:15
- estimated AUC72 Helmut 2013-10-08 13:51
- estimated AUC72 ratnakar1811 2013-10-10 11:30
- estimated AUC72 jag009 2013-10-13 21:17
- estimated AUC72 ratnakar1811 2013-10-15 06:23
- estimated AUC72 Dr_Dan 2013-10-15 12:24
- estimated AUC72 jag009 2013-10-15 21:53
- estimated AUC72 ratnakar1811 2013-10-17 09:34
- estimated AUC72 ratnakar1811 2013-10-15 06:23
- estimated AUC72 jag009 2013-10-13 21:17
- estimated AUC72 ratnakar1811 2013-10-10 11:30
- estimated AUC72 ratnakar1811 2013-10-08 08:20
- estimated AUC72Helmut 2013-10-07 13:36