1
5. BRIDGING THE GAP between
Pharmacometricians and
Biostatisticians
‘The battle lines were clear’
One the one side were the
forces of light:
One the other side were the
forces of darkness:
those who liked models used
biological insights, generally
welcomed data from disparate
sources and were not afraid to try
various bold and ingenious
strategies for putting models and
data together
a bunch of dice throwers and
hypothesis testers with an
inane obsession with intention
to treat
2
Bridging the gap
Pop PKPD: the start
• Continuous variables
• Short time scale
• Exploratory studies
• Early phases in drug development
3
 Mainly learning
Pharmacometrics now
• Clinical end points
•
•
•
•
Longer time scale
Pivotal/confirming phases
Discrete variables and time to event
Disease progression
• Results use for prediction / simulation
Extrapolation
Planning / Design evaluation
Clinical trial simulation
Decision making…
 More attention on model building /
estimation / uncertainties
4
•
•
•
•
NLMEM and statisticians (1)
Last decades
• Development of good estimation methods and fast
algorithms
Present/ Future for estimation
• More complex statistical models
•
•
More complex mechanistic models
•
•
discrete data, RTTE, Markov, joint models, dropouts,
confounding, nonparametric, distributions, mixtures ….
ODE, PDE, SDE….
Bayesian approaches
• Better use of computers (cloud, GPU,…)
 Enhanced (and new) software tools
5
 Engineers, Computer scientists, Mathematicians,
Statisticians….
NLMEM and statisticians (2)
Research topics….
• Model evaluation
• Covariate model building
Bertrand & Balding. Pharmacogenetics Genomics, 2013
• Optimal design
• Tests for small samples
• Model averaging
• Joint models: prediction of event from
biomarker evolution
6
EU Project: IDeAL
WP5: Optimal design in mixed models to analyse studies
in small population groups (France Mentré)
Pitfalls in pharmacometrics
• Handling of data (per protocol, ITT, missing,
dropout)
• problem especially in confirmatory analysis
• Multiple testing in model building, covariates
analysis …
• Lack of control of type I error
• Model evaluation, checking assumptions
• Often lacking model based analysis plan
• Design / sample size
7
• No standard procedure
Pitfalls in statistics
• Stick to standard linear or empirical models
• Few assumption models whereas based on
centuries of physiology in pharmacology
• Reluctance to use of new software/ tools,
and not totally pre-specified analysis
8
• ‘fear’ for NLMEM
Benefits: evolution of both groups
9
• More standardization in pharmacometrics
• More modelling in biostatistics (analysis of
longitudinal data in clinical trials)
ASA and ISoP Joint Interest Group on
Pharmacometrics and Statistics
10
• Chairs: Alan Hartford (Abbvie) and Matthew
Rotelli (Lilly)
• First meeting: Feb 9, 2016 at ASA
6. CONCLUSION
• Pharmacometrics increasingly used in drug
developments and drug use
• Rare diseases, children, …
• Good estimation method available for NLMEM
11
• Analysis of complex PKPD interactions (system
pharmaocology)
• Dose finding
• Clinical Trial Simulation
• Therapeutic drug monitoring
• Analysis of longitudinal data in clinical trials:
more power for treatment comparison/
evaluation
Personal perspectives & hopes ….
1. Model-based analysis of pivotal trials
2. Model-based treatment individualization
3. Model-based evaluation of treatments in the
developing world
Pharmacometricians AND Statisticians
• better drugs/ treatments
• better targeted to each patient
12
 Help decrease disease burden in the world