Federal Institute for Drugs and Medical Devices The BfArM is a Federal Institute within the portfolio of the Federal Ministry of Health (BMG) The use of modelling and simulation in drug approval: A regulatory view Norbert Benda Federal Institute for Drugs and Medical Devices Bonn Disclaimer: Views expressed in this presentation are the author's personal views and not necessarily the views of BfArM
2/20 N Benda: M&S in Drug Approval Federal Institute for Drugs and Medical Devices Overview Principles in drug approval Challenges Modelling ? Simulation ? Problems Longitudinal analysis Small population dilemma Conclusions
3/20 N Benda: M&S in Drug Approval Federal Institute for Drugs and Medical Devices General principles in drug approval Demonstrate efficacy Show favourable benefit risk Additional requirements Additional claims to be demonstrated after general efficacy (1) has been shown Homogeneity Subgroups to be excluded / justified Relevant dose / regimen
4/20 N Benda: M&S in Drug Approval Federal Institute for Drugs and Medical Devices Statistical principles in drug approval Independent confirmatory conclusion no use of other information type-1 error control limiting false positive approvals Internal validity blinded randomized comparison assumption based External validity relevant population to study random sampling, etc
5/20 N Benda: M&S in Drug Approval Federal Institute for Drugs and Medical Devices Areas that may challenge approval principles Paediatrics Orphan drugs Integrated benefit risk assessments Dose adjustments (body weight, renal impairement, etc.) Individualized dosages / therapies
6/20 N Benda: M&S in Drug Approval Federal Institute for Drugs and Medical Devices Example: Limitations in paediatric drug approvals Sample size small Treatment control placebo unethical / impossible Endpoints different from adults / between age groups Dosages age / weight dependent
7/20 N Benda: M&S in Drug Approval Federal Institute for Drugs and Medical Devices General use of M&S Prediction dose response dose adjustment impact of important covariates identification of subgroups of concern Optimization of development program identification of optimal / valid methods informed decision making accelerating drug development
8/20 N Benda: M&S in Drug Approval Federal Institute for Drugs and Medical Devices Impact of M&S on the regulatory review Low impact internal decision making (hypothesis generation, learning) more efficient determination of dose regimen for phase III optimise clinical trial design Medium impact identify safe and efficacious exposure range dose levels not tested in Phase II to be included in Phase III inferences to inform SPC (e.g. posology with altered exposure) High impact extrapolation of efficacy / safety from limited data (e.g. paediatrics) Model-based inference as evidence in lieu of pivotal clinical data
9/20 N Benda: M&S in Drug Approval Federal Institute for Drugs and Medical Devices Model based inference Models = assumptions Models with increasing complexity random sampling from relevant population variance homogeneity proportional hazard generalisability of treatment differences (scale) longitudinal model for the treatment effect PK models / population PK models PK / PD models models on PK – PD – clinical endpoints
10/20 N Benda: M&S in Drug Approval Federal Institute for Drugs and Medical Devices Modelling Modelling = Model building + model based inference Model building aspects biological plausibility extrapolation from animal models healthy volunteers adults interpretational ease robustness evidence based derived from / supported by data
11/20 N Benda: M&S in Drug Approval Federal Institute for Drugs and Medical Devices Problems with modelling Model selection bias if model selection and inference based on same data Ignored pathway Dose PK PD clinical endpoint ? Ignored between-study variability validation usually within similar settings no “long-term validation”
12/20 N Benda: M&S in Drug Approval Federal Institute for Drugs and Medical Devices Simulations Simulation = numerical tool Complex models / methods require unfeasible high dimensional numerical integration e.g. type-1 error / power calculation under complex assumptions (drop- outs, adaptive designs, etc) or model deviations Simulation = visualization Focus on statistical distributions between subjects / within subjects considering complex variance structures / non-linear mixed models Visualize resulting distribution for specific settings (treatments, fixed covariates)
13/20 N Benda: M&S in Drug Approval Federal Institute for Drugs and Medical Devices Simulations Advantages: visualization on distributions / populations allow for an population based assessment Disadvantages often (unconsciously ?) misinterpreted as “new” data inference from simulation impossible depend on (unverifiable) model assumptions incorrect variance modelling may be misleading
14/20 N Benda: M&S in Drug Approval Federal Institute for Drugs and Medical Devices Longitudinal model-based inference Repeated Scientific Advice question: Pivotal confirmatory Phase III study Longitudinal measurements at time t 1, t 2,..., t n relevant endpoint at t n (end of treatment) primary analysis based on t n only or on a longitudinal model ? different possibilities time dependency functional or categorical ? covariance structured or unstructured ? Robustness (t n ) vs more informative analysis “borrowing strength” or “relying on assumptions difficult to verify” ?
15/20 N Benda: M&S in Drug Approval Federal Institute for Drugs and Medical Devices Longitudinal model-based inference Case-by-case decision Relevant missing data issue and non-inferiority: consider assay sensitivity longitudinal analysis / MMRM (Mixed-Effect Model Repeated Measure) preferred justify model (by M&S ?) Non-compliance and superiority vs placebo: use of measurements under non-compliance / after discontinuation (retrieved data):“effectiveness” longitudinal analysis under compliance: “efficacy”
16/20 N Benda: M&S in Drug Approval Federal Institute for Drugs and Medical Devices Small population dilemma Independent confirmation vs historical information Population concerned vs extrapolation from other population Modelling approaches to bridge historical information extrapolate from other population Trade-off Robustness and independent confirmation vs presumably more informative analysis Less data available – more assumptions needed
17/20 N Benda: M&S in Drug Approval Federal Institute for Drugs and Medical Devices Small population proposals M&S approaches to extrapolate Surrogate endpoints (PD) + adult evidence Meta-analytic approaches using historical data Bayesian: Evidence synthesis (Paediatric) subgroup analyses rely on transferability of (some) model components Increase type-1 error Relying on more assumptions False positives - false negatives missed drug worse than ineffective drug ?
18/20 N Benda: M&S in Drug Approval Federal Institute for Drugs and Medical Devices Conclusions (1) Differentiate M&S to optimise study design M&S to explore and optimise development program M&S to predict efficacy and safety Differentiate M&S / Model building and exploration Model-based inference
19/20 N Benda: M&S in Drug Approval Federal Institute for Drugs and Medical Devices Conclusions (2) Be honest with simulations Numerical tool Visualizing tool Be honest with modelling confirmatory inference independent of model building inference is always model-based amount and quality of assumptions to be assessed simplicity preferred if robustness is of concern trade-off between precision vs robustness false positives vs false negatives
20/20 N Benda: M&S in Drug Approval Federal Institute for Drugs and Medical Devices Conclusions (3) Virtues of M&S increased understanding of underlying process may facilitate focus on distributions may optimise development program design Independent confirmation still required in Phase III in most applications low amount of assumptions / simplicity to ensure robustness possible exceptions where false positive decisions are worse than false negatives