Analyzing Event History Data with nlme in S
José Pinheiro(1), Chyi-Hung Hsu(1), and Fei Chen(2)
(1) Novartis Pharmaceuticals, USA; (2) Fair Isaac Co., USA
Pharmacokinetic (PK) and pharmacodynamic (PD) data are often collected in event history format, in which records representing different types of events (concentration measurements, dose administration, etc) are included in chronological order. This is the standard data format used with NONMEM, for example, because it lends itself to efficient event-driven, recursive representations of PK/PD models as used in that software. The event history format, however, is not particularly well suited for use with the S language, which is most efficient when operations can be vectorized. This incompatibility between the standard format of PK/PD data and the desire for efficient vectorized model representations in S has been one of the main obstacles preventing broader use of the nlme function within the PK/PD community. This talk describes a set of functions and methods developed in S to extend the capabilities available in the NLME library to efficiently handle event history data. These include functions for reading NONMEM-like event history text files and creating a special class of data.frames extending the groupedData class in NLME; plot methods suitable for displaying event history data; and recursive representations of common compartment models (written in C) which can be used with the nlme function for efficient and flexible population PK modeling of event history data. The resulting fitted objects can be used as any other nlme object which can be used with all available nlme methods (e.g., diagnostic plots, simulations, etc.) Real and simulated event history PK data will be used to illustrate the different methods.