I have a question about the calculation of partial AUC using linear up log down method. I am wondering if anyone could help with that.
In one of my studies, the Tlast is about 23.96 hr. To calculate AUC0-24, we used lambda z and intercept to get the predicted C24h, which is a little higher than the observed Clast. I thought WNL should use linear method for the calculation of AUC23.96-24h since the concentration is increasing but it looks like WNL used log method for the calculation. I am a little confused. I searched the guide but didn’t anything about that. I am using Winnonlin 7.0.
If I recall correctly the interpolation rule is set by the OBSERVED points. In this case you’ve had to extrapolate from your previous time e.g.16h to 23.96 and then on to 24 h. Your using the Lz to get there, which is assuming a decline so it seems reasonable to me.
overall differences should be very small as the time intervals are very short, so I think this is a practical enough rule to use.
Yes, we use the observed points to calculate the Lz and then predicted the C24h. When we use other software, like SAS, to calculate AUC23.96-24h, we compare the C23.96h and the C24h first. If C23.96h is larger than C24h, then we use log method. If C23.96h is equal or less than C24h, then we will use linear method. According to what you mentioned above, does it mean that WNL decides the calculation method according to the trend of some phase, not the comparison between individual concentrations? In my current case, the difference is small because the Tlast is close to 24 hr. But if the Tlast is not so close to 24 hr and Clast is smaller than predicted C24h, I am not sure how big the difference will be. What is your thought?
From Page 48 of the Phoenix WinNonlin User’s Guide.pdf (8.1)
If the start time for a partial area is before the last numeric observation and the end time is after
the last numeric observation, then the log trapezoidal rule will be used for the area from the last
observation time to the end time of the partial area. However, if the last observation is non-positive
or is equal to the extrapolated value for the end time of the partial area, then the linear trapezoidal
rule will override the log trapezoidal rule.
Perhaps a little cumbersome but works as designed.
I would expect mpst designs should have the actual sampling time quite close to the nominal 24H as a predose sample.
Actually it reminds me of an idea I’ve had in the past to ‘automatically’ duplicate a predose day 2 as a day 1, 24 h sample etc. I’ve made workflows to do it in the past.