adaptive clinical trials in the real world presentation to mbc 23 rd april 2008...
TRANSCRIPT
The questions
3 categories:
1. Should we use adaptive clinical trials or not?
2. What’s the impact of using them?
3. How do we use them?
To use or not to use
Adaptive Clinical Trials – to use or not to use?
1. When is it most appropriate to run an adaptive clinical trial?
2. When in a drug's development is the most appropriate time to conduct an adaptive clinical trial?
3. What indications particularly lend themselves to the use of adaptive clinical trials?
4. What is the value proposition in the use of Adaptive Clinical Trials?
5. What's driving the increasing use of adaptive clinical trials?
6. What are the key benefits for utilizing an adaptive clinical trial design?
1. When is it most appropriate to run adaptive clinical trial?
When you have a lot to learn about the drug and the disease in your target population
You do not have the time or money to simply recruit enough subjects in a simple way to answer you questions
And there are outcomes early enough in treatment to adapt to
2. When in a drug's development is the most appropriate time to conduct an
adaptive clinical trial?
Any phase where there is significant uncertainty over the drug behavior• But Phase 1 is adaptive anyway (could use better
methods and could look at efficacy as well as toxicity)
• Phase 2 (PoC) and Phase 2 (Dose Finding)• In Phase 3 there are regulatory issues – classical
(frequentist) but not Bayesian statistics? Need for safety data?
• Phase 4? A lot of scope – but less budget.
3. What indications particularly lend themselves to the use of adaptive clinical
trials?
Quick response (<25% of recruitment period) Range of doses available Subjects are expensive Don’t want to learn equally about every
treatment regardless of outcome.Example Cumulative Subjects and Responses
0
50
100
150
200
250
300
350
1 2 3 4 5 6 7 8 9 10 11 12 13 14
Months into trial
# S
ub
ject
s /
Res
po
nse
s
Monthly recruitment
Total Recruitment
2 Month Response
• Migraine, dental pain, post-operative pain, neuropathic pain
• Stroke, Alzheimer's, Schizophrenia• Diabetes, cholesterol lowering• Cancer• Orphan indications
3b. What indications lend themselves to Adaptive Clinical Trials?
3c. What indications don’t lend themselves to the use of adaptive clinical trials?
Very long time to final response Very swift recruitment Population change over duration of trial Subjects are cheap Want to learn equally about all treatment arms
• Save 25-30% over parallel group with interim for futility.
• Additional investment ~$500,000 but net saving of $1.5M (400 subject trial) due to early termination for futility
• Costs:• Extra supplies $200,000• Additional design $100,000• Response collection $100,000• Adaptive algorithm $100,000
4. What is the value proposition of using Adaptive Clinical Trials?
• If successful, better characterization of the efficacy and toxicity of the drug
• More data on the dose of interest• Less risk of an inconclusive outcome• Better model of drug effect and disease
progression – more persuasive• Faster/smarter overall development through
better targeted trials
4b. What is the value proposition of using Adaptive Clinical Trials?
5. What's driving the increasing use of adaptive clinical trials?
Level of failure in Phase 3 • Need better information before Phase 3• Need better killing of ineffective compounds before
Phase 3 Time spent in development
• Can we learn faster by combining phases in a cleverer trial?
• Phase 1 & 2a• Phase 2a & 2b• Phase 2b & 3
6. What are the key benefits of adaptive clinical trials?
Better Ethics• Fewer subjects allocated to ineffective or over-toxic
treatment arms• Fewer subjects used in studies that fail
Better Science• Can try more doses (Phase 1 & 2)• Can try more doses (Phase 3?)• Explore other dimensions – combinations,
indications, sub-populations Better Business
• Swifter curtailment of failing compound• Better information -> better decisions at the next
phase
6b.The key benefits
Better definition of trial goal Modeling of trial data:
• Borrow ‘strength’ from neighboring points• Borrow ‘strength’ from other outcomes (biomarkers,
longitudinal, prior data, etc.) Optimization of dose allocation:
• Put fewer subjects on treatment arms that are clearly not working
• Put more subjects on treatments arms that seem to be showing the desired target effect.
Result: better characterization of the dose behavior
Tom Parke [email protected] (44)(0) 1235-555511
©2007 Tessella Support Services plc
Example Adaptive Trial
Impact
Impact
1. How do adaptive clinical trials impact the whole development program?
2. What are the principle disadvantages (difficulties and costs) one faces when utilizing this approach?
1. How do adaptive clinical trials impact the development program?
More flexibility in design of whole program• Trials used to have very predefined task• Now – what are your questions, and lets design a
trial to answer them as efficiently as possible Consider trial in whole development program
• What will follow, what will be in parallel, what is the right order to answer the questions
Need to think about the next trial earlier and longer
Need to integrate the development team
2. What are the difficulties and costs of implementing adaptive clinical trials?
Longer Design Time• Need to identify candidate trial• Design less “off-the-shelf”• Design needs interaction with clinical team• Design needs simulation and optimization
More Integrated Trial Management System• Quick capture of key responses• Frequent modification of randomization
Drug Supply• Need to be able to deliver more doses• Need to be able to use central randomization
Implementation
Implementation
1. Can we capture the response data quickly and reliably2. Can we calculate and agree the adaptation quickly3. How do we implement the adaptation?4. How does one effectively manage the clinical drug
supply chain in an adaptive trial?5. How do we get all the stakeholders aligned so the trial is
a success?6. From a clinical operations perspective, what are the
challenges in managing a complex trial that could have from 300 to 800 subjects at multiple sites in different global locations?
7. How do we go about deploying adaptive clinical trials?8. How do you make it mainstream and industrialize the
process?
Example Adaptive Trial Infrastructure
Model
IVRS
ResponseData
Capture
Drug SupplyManagement
DataMonitoringCommittee
Relative %Randomization
TreatmentPack Data
RandomizationList
ResponseData
DMCReport
Randomization
EDCLite
Adaptive Trial Infrastructure
Model
IVRS
EDCDrug SupplyManagement
DataMonitoringCommittee
Relative %Randomization
TreatmentPack Data
RandomiszationList
ResponseData
DMCReport
Randomization
EDCLite
IVRS
• IVRS need modification to allow adaptation:• To be able to regularly replace the
randomization list• after interim to drop or add doses• after model update to adjust relative proportion of
randomization
• Randomize dynamically based on the currently available arms and/or proportions of randomization
• Randomize dynamically using a combined blocking and proportionate randomization
Partial BlockingRequired Randomization is:Placebo: 25%Dose1: 6%Dose2: 9%Dose3: 15%Dose4: 26%Dose5: 13%Dose6: 6%
Partiallyblocked
Random is now:Dose1: 8%Dose2: 12%Dose3: 20%Dose4: 35%Dose5: 17%Dose6: 8%
Ran
dom
Pla
cebo
Pla
cebo
Ran
dom
Ran
dom
Ran
dom
Ran
dom
Ran
dom
Partial blocking of placebo ensures % allocated to placebo and consistent allocation to placebo through time.
IVRS treatment allocation
• IVRS if not loading a randomization list needs to be able to supply a treatment allocation list:
Patient ID, Treatment Arm05041101, 205042301, 305041102, 105040701, 1
• From first patient first visit and weekly or fortnightly thereafter.
Adaptive Trial Infrastructure
Model
IVRS
EDCDrug SupplyManagement
DataMonitoringCommittee
Relative %Randomization
TreatmentPack Data
RandomiszationList
ResponseData
DMCReport
Randomization
EDCLite
EDC• EDC needs to be able to extract key response data:
Patient ID, Visit #, resp1, resp205041101, 1, 6.3, 005041101, 2, 5.2, 005041101, 3, 5.0, 005042301, 1, 4.3, 005042301, 2, 4.6, 105041102, 1, 5.9, 005040701, 1, 6.5, 0
• Within a 1-2 months of first patient first visit and weekly or fortnightly thereafter.
EDC-Lite
• If the main EDC cannot produce response data quickly, frequently and reliably
• A parallel EDC system can be used, just collecting headline response values (possibly just two values per patient visit)
• Can be made convenient to use• EDC-Lite data can be replaced by main EDC data
as it becomes available• Forward EDC-Lite data to EDC to aid data checking
Faxes in from centres:• Subject screened• Subject eligible• Subject mobile phone #
Faxes back to centres:• Subject randomised• Subject response overdue
Monitoring by study manager
Web access
Subjects phone in:• for randomisation• with response
Subjects receive:• text reminder
EDC-Lite
Adaptive Trial Infrastructure
Model
IVRS
EDCDrug SupplyManagement
DataMonitoringCommittee
Relative %Randomization
TreatmentPack Data
RandomiszationList
ResponseData
DMCReport
Randomization
EDCLite
Drug Supply
• Initial negotiation with supply as to what is possible number of different doses, quantity of API, etc. before design
• Trial design simulations provide estimates of max number of subjects allocates to any one treatment arm
• Trial supply simulations allow manufacturing estimate to be fine tuned, and supply / logistics trade-offs to be explored
Drug supply during
• Unblinded supply representative included on supply implications of DMC report. • supply proportionate to probability of randomization• total supply requirements implied by predictive
probabilities
Adaptive Trial Infrastructure
Model
IVRS
EDCDrug SupplyManagement
DataMonitoringCommittee
Relative %Randomization
TreatmentPack Data
RandomiszationList
ResponseData
DMCReport
Randomization
EDCLite
Data Monitoring
• A process change not infrastructure• DMC include someone competent to check:
• correctness of the data supplied to the model, • the design’s performance,• the implementation of the adaptation (is the randomization
adapting?)
• Phase 1 & 2 trial DMCs staffed internally unless external specialist required.
• Regular automated DMC report with 10 minute teleconferences to review.
• Small number of big review meetings. Timing flexible based on review of report
DMC report
• The current recommendation
• The data
• The model fit
• The decisions
• The likely outcome (predictive probability)
Example trial setup
• Phase 2 dose finding• Designs by Berry Consultants• Data weekly from EDC (in-house, 3rd party)• Possibly supplemented by direct fax of key
endpoint data• New randomizations sent to IVRS (in-house
or 3rd party)• DMC report• Secure file transfer
Example Weekly Update System
Weeklycompleteresponsedata
New randomization list, or randomization probabilities
DMC report
EDC IVRS
Stats
Fax
SMS
Reminders
Trial monitoring website
Main Clinical Operations challenges
• The high level data is collected and sent to the adaptive ‘back box’ reliably, accurately and frequently
• Efficiently supplying in a changing world
• But, you will be able to monitor your trial better
Adaptive Trial Infrastructure
Model
IVRS
EDCDrug SupplyManagement
DataMonitoringCommittee
Relative %Randomization
TreatmentPack Data
RandomiszationList
ResponseData
DMCReport
Randomization
EDCLite
Adaptive Design
• Need good tools (R, S-Plus, WinBUGS, Matlab) and good statisticians to generate designs, or very customizable implementations of designs.
• R, S-Plus, WinBUGS, Matlab – are very statistician friendly and good tools for researching designs – but slow for running large numbers of simulations.
• Can code them up (C++ / Fortran) once proven.• Berry Consultants with Tessella will be releasing
customizable implementations of Berry Consultants designs later this year.
Why simulate designs?
• For some trial designs we can no longer simply prescribe our desired probability of a false positive (alpha) and or false negative (beta).• Simulate with treatment arms no more efficacious than placebo
• Simulate with different arms (and different numbers of arms) being clinically effective
• But there are other properties of interest too:• How likely is the best treatment arm to be chosen?
• How likely are we to stop early and will it be correctly or incorrectly?
• What if we are studying more than one endpoint?
• Or more than one compound?
Why simulate designs? (2)
• We have more to decide:• Is it worth doing an adaptive design?• Which of these adaptive designs is better?• What is the impact of this protocol change (more visits,
more treatment arms, longer follow-up, change of endpoint)?
• For this design what values should I choose as design parameters:
• required confidence of futility/success to stop early• the earliest the trial is allowed to stop early• frequency of looking at data• thresholds for dropping arms, adding arms etc.
Simulation Functionality
Black Box
Set and validate designparameters and scenarios
to simulate
Orchestrate runningsimulations of all versions
of the designover all scenarios
Display, analyze and chart the results
Execute Trial with selected design
and parameters
Centralize storage ofdesigns and run simulations on
a computing grid
Compare designswith common design
constraints and scenarios
Server Simulationmanager
GUI
Trial Execution
Deployment at the company level
• Decide on type of adaptive trials you can and want to run.
• Establish cross functional adaptive review team (clinical, biostats, supply, IT, trial operations) to review candidate trials and assist teams to Go Adaptive.
• Development teams should be responsible for all compounds in an indication, not a single compound or• So they see benefit in early determination of futility• Can learn across a a number of trials
Aligning the stakeholders
• Involve them early• help them understand what adaptive clinical
trials are and why the company wants to use them
• in identifying the problems and solving them
• Ensure personal objectives are aligned with running adaptive clinical trials
• Top down & bottom up
Development teams
• Don’t design and evaluate design in isolation• Trials aren’t islands,• or steps on a single path• They are decision nodes in a complex tree of
investigation – looking at different endpoints, populations, indications, combinations.
• The more you can learn each trial and the more quickly you can learn, the more efficient you decision making and overall development.
• Can use interim data to start/stop other branches in the development
Example
Phase 1
A
B
Combined phase 2a/2b in A & B
Operationally seamlessphase 3 with best of A or B
Second confirmatorytrial
Poc complete start 2nd indication development
Sufficient confidence in efficacy to start manufacturing API for rest of development
Start planning 2nd confirmatory trial
Drop a dose and chose dose for 2nd trial
Implementation
1. Can we capture the response data quickly and reliably2. Can we calculate and agree the adaptation quickly3. How do we implement the adaptation?4. How does one effectively manage the clinical drug
supply chain in an adaptive trial?5. How do we get all the stakeholders aligned so the trial is
a success?6. From a clinical operations perspective, what are the
challenges in managing a complex trial that could have from 300 to 800 subjects at multiple sites in different global locations?
7. How do we go about deploying adaptive clinical trials?8. How do you make it mainstream and industrialize the
process?
You are not on your own
• Tessella and Berry Consultants can help you do this.
• Berry Consultants:• Designs & “Black Boxes”
• Tessella:• Simulation framework for black boxes• Systems and services to help execute the
trial
Summary
• Despite their differences from normal trials, Adaptive Clinical Trials can be implemented
• They are becoming increasingly easy to implement as we • learn the lessons from the early adopters • and build tools to support them
• As we integrate them fully into the development process, the benefits of cost savings and quicker and better informed decisions will continue to grow as the development process is redesigned