Background
The Asterix, IDeAl and InSPiRe projects
Regulatory guidance
Paper outline
Biostatistical Research by Asterix, IDeAl and InSPiRe
Levels of evidence
-
An approach will be developed based on Bayesian decision theory to obtain optimised clinical trial designs that account for prior belief regarding the benefits of treatment and the population size.
-
Methods to relax the traditionally applied levels of type 1 and type 2 errors based on the anticipated finite (and relatively small) future population to treat will be developed, essentially based on a mix of Bayesian and frequentist methods.
-
Value of information methods will be applied to assess appropriate sample sizes for clinical trials in small populations.
-
In small clinical trials properties of randomization procedure relying on approximate arguments fail to suffice. Recommendations for selection of the best practice randomization procedure maintaining the significance level in the presence of selection bias will be derived.
-
Determination of appropriate levels of evidence for decision-making in small population clinical trials will be investigated.
-
Evidence synthesis methods for small populations and rare diseases will be developed to support the planning, analysis and interpretation of a single randomised controlled trial. The feasibility and utility of the newly developed methods will be assessed in small population settings.
-
Generalized evidence synthesis approaches will be applied to paediatric studies and compounds developed for potentially multiple rare indications.
-
Evidence synthesis methods across trials (of similar or different design) that take into account the sequential nature of drug development will be developed.
Pharmacological considerations
-
Pharmacometric [14] approaches will be considered by modelling the disease – therapy relation using non-linear mixed effect models. These theoretical models are used for improved sample size estimation in small clinical trials allowing for uncertainty, defining appropriate outcome measures and investigating analysis methods.
-
Optimal design techniques, resulting in time points of observation based on minimization of the variance of the effects estimators in nonlinear mixed-effects models will be developed by means of using preliminary estimates from interim looks to improve the final estimator. New numerical techniques in case of small samples where asymptotic assumptions are no longer valid.
-
Innovative designs for early phase clinical trials taking into account safety, efficacy and pharmacokinetic/pharmacodynamic (PK/PD) measures will be developed to better estimate the dose level to be recommended based on limited sample sizes and subgroups with continuous and binary outcomes.
-
The performance of the novel methods will be evaluated in terms of information gain, number of subjects, efficiency, and robustness. Designs for within patient data modelling will be developed to allow as much information as possible to be used for benefit risk assessment.
-
Improved methods for identification of genetic prognostic factors will be derived, leading to efficient clinical trial design and analysis.
Choice of endpoints
-
A suitable framework for validation of biomarkers as surrogate endpoints in small clinical trials will be developed and optimal designs investigated. Hereby, the most appropriate approach, e.g., the causal inference and meta analytical paradigm will be evaluated within the context of small samples.
-
The impact of unbalanced small data sets on numerical accuracy for mixed effects models used for surrogate endpoint validation will be investigated.
-
Recommendations for models to describe the reliability of measurements in longitudinal studies, where the number and measurement times differs between subjects and the number of longitudinal measurements is large compared to the number of subjects of the trial will be established.
-
The possibility of using an individualized outcome measurement instrument, called Goal Attainment Scale [19], will be investigated. This may enable generation of evidence in diseases for which patient variability precludes the use of conventional variables to demonstrate efficacy.
Methodological and statistical considerations
-
Specific designs which are recommended in small population clinical trials will be evaluated concerning, whether it is possible to estimate the treatment effect without bias.
-
Once the treatment estimate could be estimated without bias, optimal designs minimizing the variation of the estimates, e.g., in non-linear mixed effects models to analyse data in small population groups will be derived.
-
The impact of selection and time trend bias on the validity of trial results depending on the selected randomisation procedure will be investigated and recommendations derived for different endpoints such as continuous or time to event.
-
A framework for scientific arguments to select the “best” randomisation procedure reflecting the practical clinical constraints will be derived.
-
A bias corrected randomisation test will be developed.
-
The advantage of covariate or stratified randomisation procedures in small population clinical trials will be examined.
-
The benefits and limitations of response adaptive randomisation in small population clinical trials will be evaluated.
-
Adaptive designs such as multi-arm multi-stage trials suitable for small populations will be developed.
-
Modelling and extrapolation tools will be incorporated in adaptive designs for confirmatory studies.
-
Adaptive designs that take into account multiple endpoints, biomarkers or surrogate endpoints efficiently will be developed.
-
The use of adaptive designs within a trial will be compared with the use of adaptive strategies across trials in a drug development programme.
-
The use of adaptive estimation in non-linear mixed effects models to gain in precision and sample size will be explored.
-
The extent of bias due to interim analysis will be investigated.
-
Pharmacogenetic information for tailored therapeutics including its application in crossover trials, n-of-1 trials and enrichment trials will be investigated.
-
Randomisation and bias arguments for the validity of n-of-1 trials will be elaborated and randomisation based inference will be developed.
-
The use of randomisation-based inference in small clinical trials, where asymptotic behaviour of parametric tests is questionable will be evaluated. In particular, randomisation tests for complex statistical models and response adaptive randomisation procedures will be developed
-
The gain in efficiency of stratification in small clinical trials will be investigated.
-
The use of non-linear and linear mixed effects models for longitudinal, hierarchical, and clustered data in small data sets and for unbalanced data will be considered. In particular, non- or semi-parametric methods that do not rely, or rely less, on asymptotic arguments will be developed.
-
More complex models that include genetic factors that influence the response to therapy in small population groups will be studied.
-
Longitudinal data analysis methods will be improved with respect to non-parametric methods to overcome potential weaknesses caused by the asymptotic theory.
-
Frequentist and decision theoretic methods will be developed to predict patients’ responses to targeted treatments based on genetic features or other biomarkers enabling subgroups of patients for which the benefit risk balance of a treatment is positive to be identified and confirmed.
-
The decision analysis framework will be expanded by explicitly studying the inter-relationship between the decision analyses made by health technology assessors, regulators, patients and trial sponsors.
-
The way in which decision analysis affects the design of a clinical trial will be evaluated.
Extrapolation
-
The uncertainty in dose–response information when knowledge of disease and drug is used to extrapolate from a large population to a much smaller target population will be quantified.
-
The use of extrapolation within a Bayesian framework will be explored.
-
Adaptive design strategies for extrapolation studies will be considered.
-
Guidance on the data to include in registries to support extrapolation and control group data for trial design will be developed.
Patient involvement and ethical considerations
-
A “Patient Think Tank” is established within one of the projects to directly involve the patient voice in a project and to work together on models for patient involvement in design stages of a trial. Patient organizations are represented in this Think Tank.
-
Involvement of patients at the design stage is also aimed to assess impact of design features (such as assessment schedules, informed consent procedures, multiple treatment arms) on willingness and ability to participate.
-
To what extent does our ethical framework need to be adjusted to accommodate using innovative trial designs in the context of research on orphan diseases?
-
Could we ethically accept less stringent standards of evidence in the context of rare diseases in order to benefit people with orphan diseases sooner?