Best practices in human PK prediction: which method should I use?
(An introduction to ADME WorkBench)
May 7, 2013
Conrad Housand
www.admewb.com
Context
• What exactly do we need to predict?– NCA descriptors? – PK parameters?– Plasma concentration profiles?– Tissue cell or interstitia concentrations?
Context
• What data do we have with which to make predictions?– Preclinical species in vivo?– Physicochemical parameter values?– In vitro values?
Context
• What predicive accuracy do we require?– Plasma AUC and AUMC within 3-fold error for
75% of drug-like compounds?– Accurate prediction of curve shape for a small set
of compounds?– Within 50% of observed values for a single
chemical?
Best Practices
• PhRMA CPCDC Initiative on Predictive Models of Human PK– Working group comprised of representative of 12 PhRMA
member companies– Goal: “to assess the predictability of human
pharmacokinetics (PK) from preclinical data and to provide comparisons of available prediction methods from the literature, as appropriate, using a representative blinded dataset of drug candidates”
– Findings published in series of five articles in J Pharm Sci (2011)
Best Practices
• PhRMA Initiative study components– Assembly of a diverse data set
• 108 compounds• IV, PO PK data in humans and preclinical species• In vitro and physchem data
– Assessment of predictive methods based on this data set• Methods for predicting human CL, VDSS• Wajima (allometric) approach• Physiologically-based (PBPK) approach
Prediction Methods
• Prediction of human CL– Evaluated 29 different methods including
allometric and IVIVE techniques– In vivo performed slightly better than in vitro– FCIM and two-species allometry performed best
among in in vivo methods– IVIVE using hepatocyte data w/o binding and
microsomal data with plasma and mic binding performed best among IVIVE methods
Prediction Methods
• Prediction of human Vdss– Evaluated 24 methods including empirical, semi-
mechanistic and mechanistic– No single method was better for all compounds,
but limitations in data precluded thorough evaluation of some methods
– But methods based on in vivo preclinical data generally performed better
– Best in vivo: Øie–Tozer, two-species scaling (rat/dog) and Arundel (lumped PBPK)
Prediction Methods
• Allometry (Wajima)– Uses CL and VDSS prediction techniques
described above– Conc scaled by Css, time scaled by MRT
• Equivalently, can scale microconstants
– Human Ka, Fabs predicted by averaging values from preclinical species (determined by comparmental PK analysis)
– Predictions were within 3-fold error for IV compounds, but ability to predict PO parameters and overall curve shape was poor
Prediction Methods
• PBPK– Combinations of absorption, distribution and
clearance models were evaluated• Absorption: avg. preclinical, ACAT• Distribution: Jansson, Arundel, tissue composition• Clearance: IVIVE, in vivo allometric methods
– Inputs based on in vitro and in vivo methods showed similar accuracy
– In general, IV kinetics were predicted much more accurately than PO
Implementation in ADME WorkBench
• Models– CSL files, M language scripts
• Computational engines (acslX)– ODE solution, parameter estimation
• User Interface– Spreadsheet-based inputs– Tabular and graphical results– Interactive tools
Roadmap
• 2013 Product Roadmap– Coming soon:
• Gut metabolism, transporters• Permeability-limited tissues
– Later this year:• DDI, mixtures, metabolites