Model Based Adaptive Designs in Dose Finding, Including Case - PowerPoint PPT Presentation

Model Based Adaptive Designs in Dose Finding, Including Case Studies Alun Bedding, Director, Quantitative Sciences

Outline of the Seminar Introduction – Why statistical modelling? – Modelling and simulation in designing a trial Problems behind dose finding and adaptive designs Some examples

Why statistical modelling? Let’s take an simple example Finding the right dose Might use a study with three doses and placebo in a parallel group design Analysis could be a hypothesis test comparing each dose and placebo at the 5% significance level – Is the dose different from placebo? Where are the problems?

Dose Dose Response Curve Response 95%

Other issues Multiple testing - the more you test the more you increase your chance of finding a false positive (type I error) – 3 doses versus placebo – 3 tests – Need an adjustment for multiple comparisons many of which are conservative Using hypothesis tests does not give you a good idea of the dose response curve – it just tells you the dose is different to placebo May not give you the dose of interest

Answer is modelling Deriving a model for the shape of the dose response does not involve hypothesis testing No need to correct for multiple testing Knowledge of the shape and location of the dose response curve will allow the determination of doses of interest ICH E4 - “Assessment of dose-response should be an integral component of drug development with studies designed to assess dose-response an inherent part of establishing the safety and effectiveness of the drug.”

“What about in other situations?” Modelling together with simulation allows you to explore the design and better understand the drug’s properties For example in FTIH it might allow the skipping of doses to more efficiently reach the maximum tolerated dose – example later Give greater understanding of the inter-relationships between dose, exposure and response In a well designed and analysed dose response study allows you to find the minimally effective dose, dose which gives maximal response as well as ease of having to change dose post approval Look at effect of a study on drug supplies to prevent overage

Why Simulation? We want to be sure we have picked the best design under degrees of uncertainty for various aspects – Number of doses, number of subjects on each dose, analysis methods, dose selection criteria, decision criteria........ Reproduce the design under different scenarios – Treatment differences, recruitment rates, dropouts, stopping and decision rules Demonstrate the robustness of the design to customers – Internal governance, regulators, ethics committees

Operating Characteristics of the design If there is no effect (in reality), how often do we find an effect? Type I error Simulate data assuming no effect If there is an effect, how often do we pick it up? Power (1 – Type II error) Simulate assuming various effects How often do we make the correct decision?

Dose Response 20% of post approval changes are to dose Best guess at doses in a pre-determined randomisation - we do not always use the data as it is collected Do not know dose response until the end of the study Significant number of subjects randomised to ineffective or potentially toxic doses Inefficient - may miss dose response curve

Dose Dose Response Curve Response 95%

Dose Dose Response Curves Response

Increase Number of Doses (~15) Dose Response

Wasted Doses Increase Number of Doses (~15) Dose Wasted Doses Response

Wasted Doses Increase Number of Doses – and adapt Dose Wasted Doses Response

Definition Adaptive Design – any design which uses accumulating data to decide how to modify aspects of the study without undermining the validity and integrity of the trial. An adaptive design should be adaptive by "design" not an ad hoc change of the trial conduct and analysis. Adaptation is a design feature, not a remedy for poor planning. To maintain study validity means providing correct statistical inference (such as adjusted p-values, unbiased estimates and adjusted confidence intervals), assuring consistency between different stages of the study, minimizing operational bias. To maintain study integrity means providing convincing results to a broader scientific community, preplanning, as much as possible, based on intended adaptations, and maintenance of blinding.

General Structure of Adaptive Designs An adaptive design requires the trial to be conducted in several stages with unblinded access to the accumulated data. An adaptive design may have one or more rules: – Allocation Rule : how subjects will be allocated to available arms. – Sampling Rule : how many subjects will be sampled at the next stage. – Stopping Rule : when to stop the trial (for efficacy, for harm, for futility). – Decision Rule : interim decisions (to change endpoint, to modify initial plan, optimal dose to be randomized to).

Some Examples

Phase 1 Trial in Treatment for Type II Diabetes Tibaldi F.S, Beck B.H.L., Bedding A.W.- Implementation of a Phase 1 Adaptive Clinical Trial in a Treatment of Type 2 Diabetes, 2008, Drug Information Journal, Vol. 42, pp. 455–465

The Study Design Phase I multiple dose study for safety and tolerability of a compound for diabetes Involves six active doses and placebo. Treatment of type 2 diabetes mellitus (T2DM) and includes a total of 36 subjects. Data from a single dose study of the same compound is available

• Firstly, 4 patients per cohort with 2 dose levels (3 active and 1 placebo) • Escalation rule: Pr(non-tolerability < 30%) > 90% • Dose j to be used in cohort C is determined using info from previous doses • Planned doses : 0.05, 0.3, 1, 3, 5, and 8 mg based on SDSS

Model for Safety Logistic model – The dose-response model for non-tolerability – Let p ij be the probability that a patient i who receives dose j experiences non-tolerability ( ) ( ) = α + α logit p log dose ij 0 1 j – j = 1,…..,7 where dose 1 is used for placebo – i = 1,…..,36 ( ) ( ) α + α exp log dose 0 1 j = p ( ) ( ) ij + α + α 1 exp log dose 0 1 j

Results

Probability of NT

Adaptive Design for Severe Asthma

Study purpose A Phase IIb study in the treatment of moderate or severe asthmatics – Determine the minimum efficacious dose – Gain an understanding of the dose response Can we use a flexible design in this situation – Increase the chances of a successful trial – Reduce the chances of a failed study – Increase the probability of stopping a poor drug early

Study design Single dose Parallel group Primary endpoint of FEV1 – Objective to find a dose which gives an increase in FEV1 of 150 mL (SD – 430 mL) Maximum number of subjects to be recruited – 600 Time of endpoint – 7 days Recruitment rate – 5 days per week Study already conducted in COPD suggests that doses – 15.5, 31.25, 62.5, 125 mg should be sufficient to characterise the dose response curve Placebo and active comparator included Assume Emax model

Adaptive design process (one example) Placebo Final Analysis Interim Analysis Placebo n=50 (n=300) n=50 AC Futility AC • Pr(D max -P>150)<0.1 n=50 Drop arms n=50 • Pr(D-P>150)<0.1 15.5 mg Success n=50 • Pr(D max -P>150)>0.7 Stop for futility • Pr(D max -P>150)<0.1 31.25 mg n=50 Stop for success 62.5 mg 62.5 mg • Pr(D max -P>150)>0.95 n=50 n=50 125 mg 125 mg n=50 n=50

Simulation Results Adaptive design Non-adaptive design Design Total N Power for Superiority of Dose vs Placebo Pairwise Emax model Linear model Comparisons 6 Trt Parallel Group 1038 (173/arm) 0.78 0.90 0.94 6 Trt, 3 Period crossover 360 (3/sequence) 0.83 0.96 0.85 6 Trt, 4 Period crossover 360 (1/sequence) 0.94 0.98 0.96

On-Going Work Cross-over study is possible in this population Investigate an adaptive cross-over study How to adapt – patients already in the study when an arm is dropped Do you adapt by sequence rather than treatment ?????

Other Therapeutic Areas for Adaptive Designs Neuroscience – Migraine, MS, Alzheimer's Cardiovascular Ophthalmology Oncology

Final Remarks Adaptivemethods can help better characterise the dose response for a compound Simulations need to be carried out to investigate the study operational characteristics Model based designs are ideal but adaptive are not a panacea in all situations