Modeling concentration data with imprecise sampling time

All: I have a dataset following a constant rate 1-hour infusion where the time of 1-hour sampleis imprecise. In some cases the "1-hr" sample was taken prior to infusion termination as it was intended; however, in others it was quite apparently take after the infusion ended. I'm thinking of modeling with the time of the "1-hr" sample as a "theta". Does anyone have a NONMEM coding example showing how to do this? Any other suggestions as to how to handle this would be appreciated as well. Thanks, Michael

Mike, You could use a MIXTURE model to determine of the 1-hour nominal time point is before or after the infusion. $PK CALLFL=1 EST=MIXEST IF (MIXNUM.EQ.2) THEN D1 = THETA(1) ELSE D1=THETA(2) ENDIF ... $MIX P(1)=THETA(3) P(2)=1.-THETA(3) NSPOP=2 ... $THETA (0.9 FIXED) ; Duration of Infusion for Population #1 (1.1 FIXED) ; Duration of Infusion for Population #2 (0 0.5 1) ; Proportion of Population having sampled in infusion phase.

Michael, I do not think you can change observation time, but you can change infusion duration using RATE=-2 (in the data file) D1=THETA(*)*EXP(ETA()) (in the control stream) Alternative is to exclude data points that are inconsistent with the profiles. Leonid -------------------------------------- Leonid Gibiansky, Ph.D. President, QuantPharm LLC web: www.quantpharm.com e-mail: LGibiansky at quantpharm.com tel: (301) 767 5566

Hi Michael, Another method to consider would be to fit the individuals who obviously were correctly sampled and then estimate the true infusion rate for the other individuals iteratively. * Subset data to IDs with correct sampling * Fit model to those individuals * Subset points for incorrect sampling individuals to later times (time >> 1 hr) to estimate other parameters * Assume 1 hr total infusion and estimate Cmax/Tmax for these individuals (= actual end of infusion time) * Adjust data file so that time for incorrect sampling individuals is time + (Tmax - 1) * Fit model with new data file. Thanks, Bill

Maybe I'm missing something, but would it work to say that the infusion start time is imprecise instead? For example, if the infusion starts at 0, and the sample is supposed to be at 1, move everything back 0.5 hours, give the dose at 0, with a 0.5 hr lag time, then the sample at 1.5 TIME AMT DV RATE 0 100 . 100 ; infusion is lagged - estimated infusion lag time, initial 0.5 hours 1.5 . . . ; observation recorded as 1 hr, will be 1 hour after infusion is ALAG1 is 0.5 $PK ALAG1 = THETA(1) $THETA (0,0.5) But, probably everything else would need to be off by the ALAG1 time as well. If that isn't the case (e.g., other samples are timed correctly), you could do the admittedly strange double model, with one infusion into one pk model, but only predicts that first (1 hr) observation), and all the rest based on a completely different observation. TIME AMT DV RATE CMT 0 100 . 100 1 0 100 . 100 2 1.5 . . . 1 ; this is the observation, recorded at 1 hour, but is at unknown time 2 . . . 2 ; this is the observation, recorded at 2 hours, and really is 3 . . . 2 ; this is the observation, recorded at 3 hours, and really is $PK ALAG1 = THETA(1) ALAG2 = 0 K10 = THETA(2) K20 = THETA(2) Mark Sale MD President, Next Level Solutions, LLC www.NextLevelSolns.com 919-846-9185 A carbon-neutral company See our real time solar energy production at: http://enlighten.enphaseenergy.com/public/systems/aSDz2458

Dear Michael I think you might want to use TSCALE (can also be written as XSCALE). This PREDPP parameter lets you scale time, and you can also add an eta (should you wish). The parameter is described in Guide VIII (Feb 2010). With appropriate IF statements you should be able to scale the time around the 1 hr time point only. Kind regards Janet

Dear Janet, I am not sure that tscale parameter would allow you to change the observation time relative to the end-of-infusion time. As stated in the manual, this parameter multiplies all rate constants (and only in linear models). "The time scale parameter does not scale the duration parameter D of a duration-modeled zero-order bolus dose." I assume it also does not change RATE value. Thus, it may change rates (internal individual times) but not the record times (or their relative positions). Thanks Leonid -------------------------------------- Leonid Gibiansky, Ph.D. President, QuantPharm LLC web: www.quantpharm.com e-mail: LGibiansky at quantpharm.com tel: (301) 767 5566