Pharmacokinetics in clinical oncology

Transcription

1 Pharmacokinetics in clinical oncology Carlos Ochoa Biostatistics Journal Club Zurich, 27 March 2013

2 Pharmacokinetics in clinical oncology Anticancer drug effects: Shrink tumour size Slow tumour growth Protect normal cells during chemotherapy Pharmacokinetics (PK): What happens to the drug once is in the body Pharmacodynamics (PD): What happens to the body once the drug is introduced New fields: Pharmacogenetics: Genetic factors that influence PK and PD variation in a population

3 PK of anticancer agents How? Repeated measurements of drug concentration in the blood During and after infusion Drug concentration Vs. Time: Differential Equation [? ] Incorporate compartments Change of concentration Compartmental model

4 PK outcome parameters CL: Clearance Volume/Time (e.g. ml/min) AUC: Area-under-the-concentration-time curve Concentration units Time (e.g. µg/ml h) C ss : Steady-state concentration Plateau concentration after continuous infusion at constant rate Q:Distribution clearance between compartments [? ]

5 Other PK parameters Elimination rate: e.g: Elimination half-life Time to eliminate half of the administered dose Volume of distribution: Hypothetical volume of the main compartment Bioavailability: Of the active compound in the body Estimation: Proportion between active compound in the blood after oral and IV administration; Values: Normally between 0 and 1 Important relationships CL = d i.v. /AUC (d i.v. : Dose intravenous) CL = InfRate/C ss (InfRate: Infusion rate) Q = k 12 V 1 = k 21 V 2 (k ij : Exchange rate between compartments)

6 Population modelling Parametric model Factors affecting PK: Genetic, environment, diet, age, other drugs,... Variation on subject-specific model parameters use of mixed-effects models and hierarchical modelling within and between variation [? ] Population modelling y ij = f (t ij, Θ i ) + e ij, e ij F with E(e ij ) = 0 Θ i Θ 0 G(Θ i ) with E(Θ i ) = Θ 0 Θ 0 H ( )

7 Population modelling y ij = f (t ij, Θ i ) + e ij, e ij F with E(e ij ) = 0 Θ i Θ 0 G(Θ i ) with E(Θ i ) = Θ 0 Θ 0 H ( ) y ij : j-th drug concentration at time t ij for the i-th patient Θ i : Patient s own model parameters (e.g. log(cl), log(q), log(v 1 )) G distribution: Allows inclusion of unknown parameters and regression-type parameters e ij : Residual difference between measured and modelled concentrations for the i-th patient at time t ij H: Hyperprior distribution

8 Parametric example Anticancer agent for children with leukaemia Different PK in children under 1 year of age Drug elimination variability in PK may be due to kidney function (e.g. Glomerular filtration rate, GFR)[? ] Dose 4g/m 2 in 24h in week 4,5,11,12 of treatment Sample at end of infusion and every 24h until a concentration of 0.18µM Full hierarchical model and a two-compartment model Result: No relationship between age and clearance

9 Non-parametric model Previously a parametric model was assumed sometimes is restrictive [? ] [? ] A Bayesian non-parametric model removes assumptions of normality from the subject-specific parameters Dirichlet process [? ] for the distribution of subject-specific parameters full Bayesian inference with non-linear regression functions Dirichlet process is a distribution on the space of distributions inference on underlying distribution parameters

10 Non-parametric example Dirichlet process mixture model Dataset of white blood cell count (chemotherapy) [? ] Cytotoxic chemotherapy low blood counts PD endpoint Non-linear model for repeated measurement data Drug dose served as pharmacological covariate Same model can be fit to other repeated measurements

11 Data combination Meta-analysis Combining data from multiple sources in studies of population PK Merging different population PK and PD studies leads to learn more about the distribution of PK parameters For a more precise inference of Θ i effects on PK

12 Dose individualization The possibility of modelling a drug s PK while considering between-individual variation Prediction of a person s own PK Possibility of PK guided dosing This widen the therapeutic window of drugs systemic exposure Other applications are e.g. the use of antibiotics [? ]

13 Example Patients with leukaemia ultra-high doses of chemotherapy kills all cancer cells...but also all white cells Bone marrow transplant Drug given in fix doses (body surface)[? ] any change in PK and PD the dose can be too high or too low Define the best range of exposure e.g. via AUC Dependent Dirichlet Process prior[? ] Based on: test dose PK analysis and previous studies the AUC posterior predictive distribution was found

14 Example (continuation) d = arg max u(d, y, Θ)p d (y Θ)p(Θ Data)dydΘ Values for each individual patient d : Optimal dose d: Dose y: Concentration over time Θ: Patient s own PK parameters u(d, y, Θ): Utility function p d (y Θ): Sampling distribution of future concentrations The result Ideal dose for each patient

15 Software One individual Non-linear regression nls() in R Several individuals non-linear repeated measurements Specialized Software ADAPT: Fits PK and PD models MCSIM: Fits one s model and performs Bayesian inference via MCMC quantitative/mcsim/mcsim.php MONOLIX: For non-linear repeated measurements PKBUGS For WinBUGS: For full Bayesian PK analysis [? ] PKSolver: Add-in program for PK and PD analysis [? ] Also: PK tutor for MS Excel

16 Thank you for your attention