Pyry Välitalo 1.10.2009 Nonlinear mixed effects models in pharmacokinetic modeling Lecture notes.

Slides:



Advertisements
Similar presentations
Introduction Simple Random Sampling Stratified Random Sampling
Advertisements

Orion Corporation Pyry Välitalo SSL Presentation A case example: Building a population pharmacokinetic model for theophylline using NONMEM program.
Polynomial Regression and Transformations STA 671 Summer 2008.
FTP Biostatistics II Model parameter estimations: Confronting models with measurements.
Regression Analysis Once a linear relationship is defined, the independent variable can be used to forecast the dependent variable. Y ^ = bo + bX bo is.
HSRP 734: Advanced Statistical Methods July 24, 2008.
Correlation and regression
Cox Model With Intermitten and Error-Prone Covariate Observation Yury Gubman PhD thesis in Statistics Supervisors: Prof. David Zucker, Prof. Orly Manor.
Mechanistic PBPK as an aid in identifying the size of covariate effects and design of POPPK studies: An example focusing on haematocrit as a determinant.
Some Terms Y =  o +  1 X Regression of Y on X Regress Y on X X called independent variable or predictor variable or covariate or factor Which factors.
Verify or refute the use of Non Linear Mixed Effect Model for Interferon effect on HCV Hila David Shimrit Vashdi Project Advisors: Prof. Avidan Neumann.
Chapter 13 Introduction to Linear Regression and Correlation Analysis
Inferences About Process Quality
PY 427 Statistics 1Fall 2006 Kin Ching Kong, Ph.D Lecture 6 Chicago School of Professional Psychology.
Introduction to Regression Analysis, Chapter 13,
Chapter 12 Section 1 Inference for Linear Regression.
Marshall University School of Medicine Department of Biochemistry and Microbiology BMS 617 Lecture 12: Multiple and Logistic Regression Marshall University.
Chapter 8 Introduction to Hypothesis Testing. Hypothesis Testing Hypothesis testing is a statistical procedure Allows researchers to use sample data to.
Psy B07 Chapter 1Slide 1 ANALYSIS OF VARIANCE. Psy B07 Chapter 1Slide 2 t-test refresher  In chapter 7 we talked about analyses that could be conducted.
Inference for regression - Simple linear regression
Statistics for Managers Using Microsoft Excel, 4e © 2004 Prentice-Hall, Inc. Chap 12-1 Chapter 12 Simple Linear Regression Statistics for Managers Using.
STA291 Statistical Methods Lecture 27. Inference for Regression.
Population Pharmacokinetics
1 Least squares procedure Inference for least squares lines Simple Linear Regression.
Gokaraju Rangaraju College of Pharmacy
Confidence Intervals for the Regression Slope 12.1b Target Goal: I can perform a significance test about the slope β of a population (true) regression.
Modelling and Simulation Group, School of Pharmacy Pharmacokinetic design optimization in children and estimation of maturation parameters: example of.
T tests comparing two means t tests comparing two means.
Analysis of Zidovudine pharmacokinetics to determine whether there is a genetic component to the variability and to determine the bioequivalence of seven.
Review of Chapters 1- 5 We review some important themes from the first 5 chapters 1.Introduction Statistics- Set of methods for collecting/analyzing data.
CLEARANCE CONCEPTS Text: Applied Biopharm. & PK
CHAPTER 14 MULTIPLE REGRESSION
+ Chapter 12: Inference for Regression Inference for Linear Regression.
PK/PD Modeling in Support of Drug Development Alan Hartford, Ph.D. Associate Director Scientific Staff Clinical Pharmacology Statistics Merck Research.
1 Chapter 3 Multiple Linear Regression Multiple Regression Models Suppose that the yield in pounds of conversion in a chemical process depends.
The Examination of Residuals. Examination of Residuals The fitting of models to data is done using an iterative approach. The first step is to fit a simple.
POPULATION PHARMACOKINETICS RAYMOND MILLER, D.Sc. Pfizer Global Research and Development RAYMOND MILLER, D.Sc. Pfizer Global Research and Development.
+ Chapter 12: More About Regression Section 12.1 Inference for Linear Regression.
PHARMACOKINETIC MODELS
Chapter 5 Parameter estimation. What is sample inference? Distinguish between managerial & financial accounting. Understand how managers can use accounting.
Population Pharmacokinetic Characteristics of Levosulpiride and Terbinafine in Healthy Male Korean Volunteers Yong-Bok Lee College of Pharmacy and Institute.
1-Compartment Oral Dosing 400 mg of moxifloxacin is administered orally to Mr BB, a 68 yr old male who weighs 75 kg. Blood samples were drawn following.
11/23/2015Slide 1 Using a combination of tables and plots from SPSS plus spreadsheets from Excel, we will show the linkage between correlation and linear.
Model structure  Law of mass action applied to describe the reversible solifenacin-AGP, solifenacin-albumin and solifenacin-VBC binding  VBC positioned.
Rivaroxaban Has Predictable Pharmacokinetics (PK) and Pharmacodynamics (PD) When Given Once or Twice Daily for the Treatment of Acute, Proximal Deep Vein.
Sampling Design and Analysis MTH 494 Lecture-22 Ossam Chohan Assistant Professor CIIT Abbottabad.
28. Multiple regression The Practice of Statistics in the Life Sciences Second Edition.
Chapter 14: Inference for Regression. A brief review of chapter 4... (Regression Analysis: Exploring Association BetweenVariables )  Bi-variate data.
Statistics for Managers Using Microsoft Excel, 4e © 2004 Prentice-Hall, Inc. Chap 14-1 Chapter 14 Multiple Regression Model Building Statistics for Managers.
Model-based dose selection for next dose- finding trial 1. Introduction Exploratory clinical development trials often include biomarkers or clinical readout.
The Practice of Statistics, 5th Edition Starnes, Tabor, Yates, Moore Bedford Freeman Worth Publishers CHAPTER 12 More About Regression 12.1 Inference for.
BIOPHARMACEUTICS.
Measurements and Their Analysis. Introduction Note that in this chapter, we are talking about multiple measurements of the same quantity Numerical analysis.
1.Andersson, T, et al. Clin Pharmacokinet 2001;40: Hassan-Alin, M, et al. Eur J Clin Pharmacol 2000;56: Population Pharmacokinetic Modelling.
The Practice of Statistics, 5th Edition Starnes, Tabor, Yates, Moore Bedford Freeman Worth Publishers CHAPTER 12 More About Regression 12.1 Inference for.
Chapter 10: The t Test For Two Independent Samples.
Marshall University School of Medicine Department of Biochemistry and Microbiology BMS 617 Lecture 13: Multiple, Logistic and Proportional Hazards Regression.
Pharmacokinetics 3rd Lecture
Compartmental Models and Volume of Distribution
Stats Methods at IC Lecture 3: Regression.
CHAPTER 12 More About Regression
Chapter 13 Simple Linear Regression
Anticonvulsants: Valproic acid
CHAPTER 12 More About Regression
Background and Objectives
ANALYSIS OF POPULATION KINETIC DATA
CHAPTER 12 More About Regression
CHAPTER 12 More About Regression
Presentation transcript:

Pyry Välitalo Nonlinear mixed effects models in pharmacokinetic modeling Lecture notes

Pyry Välitalo Pharmacokinetics: Traditional, standard two-stage* approach Recruit subjects from a homogenous (healthy?) population. Collect lots of blood samples from each patient. Estimate pharmacokinetic parameters for each patient separately. Calculate means and variances for pharmacokinetic parameters. Use regression to investigate the effects of covariates. ** *Not to be confused with the term ”two-stage estimation procedures” in statistics ** Riviere, J. Comparative Pharmacokinetics: Principles, Techniques and Applications, page 260. Iowa State Press, 1999.

Pyry Välitalo Problems associated with the traditional approach: Parameter variabilities are inflated.* Since recruitment is usually done from a homogenous (healthy?) population, it is harder to extrapolate into target population. Difficult to study special populations who would not handle the blood loss well (neonates, AIDS/cancer patients, critical care patients, etc). *Sheiner LB, Beal SL. J Pharmacokinet Biopharm Dec;8(6):

Pyry Välitalo A solution: population pharmacokinetics ( = Pharmacokinetics using nonlinear mixed effects modeling) Build a pharmacokinetic model with fixed effects, between-subject variabilities, and residual variabilities. Differences to traditional pharmacokinetic modeling: – One model explains all data – Between-subject variability is included in the model as a new kind of parameter: A random effect that varies between patients but stays constant within the patient.

Pyry Välitalo Advantages of a population approach to pharmacokinetics 1/2 Less blood samples per patient are needed – Special patient groups can be studied (children, cancer patients, etc) – Samples taken during routine treatment can be used in studies. Cost-effectiveness increases The results naturally reflect the patient group that is usually receiving the drug.

Pyry Välitalo Advantages of a population approach to pharmacokinetics 2/2 Because data are combined into a single model, more detailed models can be used. E.g. nonlinearity can be detected better than with standard two-stage approach* Easier to design future clinical trials with one single model. *EN Jonsson, JR Wade, MO Karlsson. AAPS PharmSci. 2000;2(3):E32

Pyry Välitalo Advantages of a population approach to pharmacokinetics: an example Docetaxel, a chemotherapeutic agent Main problem: Patients with poor liver function – FDA: do not dose due to unpredictable PK! Aim: Build a clinically relevant model to predict docetaxel clearance in patients with poor liver function. Evaluate CYP3A activity spesifically as a predictor of docetaxel clearance. A population PK model was built with following covariates: – Liver functioning classification based on a few markers – Plasma protein binding – CYP3A liver enzyme activity After these covariates had been accounted for, the unexplained variability in clearance actually became lower for patients with poor liver function than with normal-liver-function patients. Hooker et al. Clin Pharmacol Ther Jul;84(1):111-8.

Pyry Välitalo Using the parameters of NLME models in pop-PK Fixed-effects parameters (Thetas, θ) – Can be for example a typical value for volume of distribution, typical value for clearance or the effect of a covariate (e.g. sex) on a parameter. – A capital theta (Θ) denotes a vector of all fixed-effects parameters in model. – A lowercase theta (θ) denotes an element of Θ, one specific parameter.

Pyry Välitalo Using the parameters of NLME models in pop-PK Between-subject variability (Etas, η) (Empirical Bayes Estimates) – These describe unexplained differences in parameter values between individuals. – E.g. the individual value for clearance could be described as θ CL *e η(CL). – Can also be used to describe inter-occasion variability, between-study variability. – It is expected that etas are distributed N(0, ω 2 )

Pyry Välitalo Using the parameters of NLME models in pop-PK Residual random effects (Epsilons, ε) – The unexplained residual error (different for each observation). – Sources: Misrecording the time of sampling, mistreatment of samples, error induced by analytical methods, model misspesification, etc. – It is expected that epsilons are distributed N(0, σ 2 ) – E.g. observation=prediction+ε

Pyry Välitalo In summary: The population model y ij =f(Θ i )+ε ij Where y ij is the jth observation of ith individual f is a model that describes all observations Θ i is a vector of individual i’s parameter values ε ij is residual error of individual i’s jth observation The elements of Θ i are usually θ i = θ*e η, where θ is the typical value for a parameter ω 2 is the variance of η values

Pyry Välitalo Components of a population pharmacokinetic model Fixed effects Mostly random effects

Pyry Välitalo A hypothetical example: Building a 2-compartment model Let’s say we have pharmacokinetic data from 40 individuals, of which 20 received oral dosing and 20 received intravenous injection. A total of 200 plasma concentrations, 2-8 per individual Known covariates: Weight, sex (SEX=0 to indicate male, SEX=1 to indicate female) This example takes many shortcuts and should not be viewed as a reference of how to build a pop-PK model.

Pyry Välitalo Let’s start by building a single-compartment IV model (and ignore the oral treatment group for now) We use three parameters: Volume of distribution for central compartment (V)= θ V Clearance (CL) = θ CL Residual error (with standard deviation σ) The starting amount (A) in central compartment is Dose. After dosing, amount (A) of drug in central compartment starts to get eliminated. A t0 =Dose dA/dt=-A*CL/V IPRED=A/V The prediction (PRED) is compared to observation (Y) Y=IPRED+ ε ;σ is standard deviation ;of all epsilons

Pyry Välitalo The model failed to converge. Let’s add between-subject variabilities to CL and V. We estimate five parameters: V= θ V *e η(V) ;Volume of distribution and BSV for it CL= θ CL *e η(CL) ;Clearance and BSV for it Residual error (σ) A t0 =Dose dA/dt=-A*CL/V IPRED=A/V Y=IPRED+ ε ; σ is standard deviation ;of all epsilons …Success! The model now converges.

Pyry Välitalo The ”weighted residuals vs time” graph hints that a 2-compartment model might perform better. On the left: WRES vs Time. Notice the red line (locally weighted scatterplot smoothing) starting above zero, falling below and rising back above zero. Below: An illustration of what it usually means if WRES vs time has a shape like that on the left. The black line represents predictions and red dotted line represents observations.

Pyry Välitalo Let us try the two- compartment model. V central = θ V *e η(V) V peripheral = θ V2 ;Peripheral compartment volume Q= θ Q ;Intercompartmental clearance CL= θ CL *e η(CL) A t0, central =Dose dA central /dt=-A central *CL/V central – A central *Q/V central + A periph *Q/V periph dA periph = A central *Q/V central – A periph *Q/V periph IPRED=A central /V central Y=IPRED+ ε

Pyry Välitalo Let us try adding weight as a covariate into the model now. V central = θ V *(WT/70)*e η(V) ; Linear scaling V peripheral = θ V2 *(WT/70) Q= θ Q CL= θ CL *(WT/70) θscale *e η(CL) ; Allometric scaling A t0,central =Dose dA central /dt=-A central *CL/V central – A central *Q/V central + A periph *Q/V periph dA periph = A central *Q/V central – A periph *Q/V periph IPRED=A central /V central Y=IPRED+ε When adding covariates, we should also check if all the BSV’s are still necessary. It might be that covariates can explain most of the between-subject variability.

Pyry Välitalo Is sex a covariate? Let’s find out (remember, SEX=0 for male, SEX=1 for female). V central = θ V *(WT/70)* (1+SEX*θ sex1 )*e η(V) ;affects only ;females V peripheral = θ V2 *(WT/70)* (1+SEX* θ sex2 ) Q= θ Q CL= θ CL *(WT/70) θscale * (1+SEX* θ sex3 )*e η(CL) A t0,central =Dose dA central /dt=-A central *CL/V central – A central *Q/V central + A periph *Q/V periph dA periph = A central *Q/V central – A periph *Q/V periph IPRED=A central /V central Y=IPRED+ε The testing of covariate relationships should be done one at a time. In this example we didn’t find any significant improvement when adding sex as a covariate in any of the parameters.

Pyry Välitalo Once satisfied with the IV model, we add the oral treatment group Ka= θ Ka ;Rate of absorption F= θ F ;Oral bioavailability V central = θ V *(WT/70)*e η(V) V peripheral = θ V2 *(WT/70) Q= θ Q CL= θ CL *(WT/70) θscale *e η(CL) A t0,depot =F*Dose oral A t0,central =Dose IV dA depot /dt= - A depot *Ka dA central /dt=-A central *CL/V central – A central *Q/V central + A periph *Q/V periph +A depot *Ka dA periph = A central *Q/V central – A periph *Q/V periph IPRED=A central /V central Y=IPRED1+ε

Pyry Välitalo Another example: Modeling flurbiprofen pharmacokinetics in children (real case) Data from 64 patients, 1-7 samples per patient Oral dose given to 37, intravenous dose to 27 patients.

Pyry Välitalo Flurbiprofen pharmacokinetics: At the beginning… Observations of flurbiprofen CSF concentrations (60), and observations of both total (304) and unbound (62) flurbiprofen concentrations in plasma Prior knowledge: The doses given for each patient and the volume of CSF compartment.

Pyry Välitalo Flurbiprofen pharmacokinetics: What we ended up with All the parameters were estimated to best describe concentrations in central compartment and CSF. The number of parameters in the final model was: 13 fixed-effect parameters 4 between-subject variability parameters 2 residual error variability parameters

Pyry Välitalo Flurbiprofen pharmacokinetics: Most significant findings Bioavailability of oral flurbiprofen syrup for children was estimated. The model includes children from 3 months to 13 years (previous study: children aged 6-12 years). There was no impairment of clearance seen in infants. Flurbiprofen distributes into cerebrospinal fluid very effectively.

Pyry Välitalo Conclusion With flurbiprofen data, population pharmacokinetic approach yielded several benefits: – Estimating bioavailability of oral flurbiprofen syrup was made possible. – More credibility when describing CSF kinetics with model parameters than with a summary of raw data (e.g. mean ratio of unbound flurbiprofen in plasma versus flurbiprofen in CSF). – More thoroughly investigated covariate model. – Only one model: Possible to use in simulations in future.

Pyry Välitalo Resources NONMEM. Currently the ”golden standard” in population pharmacokinetic modeling. Requires license. – Xpose: An R package that helps in deciphering the output of NONMEM. Free. – R: A program needed by Xpose to operate. Free. – Census. A helpful program for keeping record of NONMEM runs. Free. – PsN (Perl-speaks-Nonmem). A collection of helpful Perl scripts for NONMEM that make life easier in a lot of ways. Free. MONOLIX. Another population pharmacokinetic program. Free. – Advantages: Shorter runtimes than in NONMEM, provides also graphical output by itself – Disadvantages: Currently not as flexible as NONMEM. –

Pyry Välitalo EXTRA: An example of simulatory model diagnostics: Visual Predictive Check Using the model parameters (including random effects), simulate a number of observations, e.g. 200 simulated observations for every true observation. Calculate the prediction intervals for these simulated observations -> see if they agree with real observations. Blue dots: Real observations Black lines: 95th percentile prediction intervals In this text, prediction interval means an interval, inside which a certain percentile of simulated observations fall.

Pyry Välitalo EXTRA: An example of simulatory model diagnostics: Visual Predictive Check Usually a better alternative is to plot confidence intervals for prediction intervals and see if the intervals for real observations fall inside the confidence intervals for prediction intervals. Red lines: Intervals of real observations Blue area: Confidence intervals for prediction intervals

Pyry Välitalo EXTRA: Features of the flurbiprofen model: Intravenous infusion The intravenous dosage had to have an absorption rate constant. The reason for this is that the intravenously given drug is a prodrug and takes some time to hydrolyze into active flurbiprofen (see figure below).

Pyry Välitalo EXTRA: Features of the flurbiprofen model: Implementing unbound observations Central compartment included two kinds of observations: Total flurbiprofen concentrations Unbound flurbiprofen concentrations If the observation was marked as unbound observation, the prediction was multiplied by fraction unbound (FU) before it was compared to the observation. FU= θ FU *e η(FU) … IPRED=A(2)/V2 ;TOTAL IF (FLAG.EQ.3) IPRED=A(2)/V2*FU ;UNBOUND

Pyry Välitalo EXTRA: Features of the flurbiprofen model: CSF kinetics Modeling the distribution of flurbiprofen into CSF was challenging. Only unbound flurbiprofen can enter CSF. However, in CSF the concentrations of flurbiprofen were circa sevenfold compared to unbound flurbiprofen in plasma. This happened probably because of protein binding in CSF (the CSF observations reflect the total amount of flurbiprofen in CSF).

Pyry Välitalo EXTRA: Features of the flurbiprofen model: CSF kinetics An intercompartmental clearance QCSF was estimated to describe the movement between central compartment and cerebrospinal fluid (CSF). The rate from central to CSF was adjusted by fraction unbound and an uptake factor (UPTK). QCSF= θ QCSF UPTK= θ UPTK K25=QCSF*FU/V2*UPTK K52=QCSF/V5 ;K25 and K52 represent rate constants from ;central compartment to CSF and from CSF to ;central compartment