Page 1 of 1

Estimate an AUC?

Posted: Wed Apr 11, 2012 11:12 am
by mneely
The easiest way is to look at the Pmetrics help file for makeAUC (type "?makeAUC" into R). This uses the trapezoidal approximation to calculate an AUC between any two time points, given a wide variety of inputs - observations (PMop objects), predictions (PMpost objects), simulations (PMsim objects), or custom.

AUCs are also calculated directly in NPAG using a trapezoidal approximation on the individual, Bayesian posterior predicted concentrations. The mean, median, or mode of the Bayesian posterior distribution for each patient is used for the predictions, as specified in the run instructions. The interval for these predictions is also specified in the run instructions, and the predicted concentrations used in these calculations can be found in the post object loaded with the NPload() command, which is a processing of the PRTBxxxx file located in the /outputs subfolder. In the OUTTxxxx or OUTxxxx files, which normally you don't need to look at, you can find tables of AUCs calculated by NPAG for each patient, divided into intervals as specified in the run instructions (default 24 hours). There will be minor numerical differences between the values from these tables and the values from makeAUC because there is an additional low-pass filter applied in NPAG which I have omitted in favor of a straight trapezoidal approximation in Pmetrics. The difference in calculated AUC using the two methods is <0.5%.