INTRODUCING BAYESIAN DOSE ESCALATION IN AN … › wp-content › uploads › ...INTRODUCING...
Transcript of INTRODUCING BAYESIAN DOSE ESCALATION IN AN … › wp-content › uploads › ...INTRODUCING...
![Page 1: INTRODUCING BAYESIAN DOSE ESCALATION IN AN … › wp-content › uploads › ...INTRODUCING BAYESIAN DOSE ESCALATION IN AN INDUSTRIAL ENVIRONMENT: CHALLENGES AND OPPORTUNITIES BEYOND](https://reader033.fdocuments.in/reader033/viewer/2022060401/5f0e40487e708231d43e565b/html5/thumbnails/1.jpg)
INTRODUCING BAYESIAN DOSE ESCALATION IN AN INDUSTRIAL ENVIRONMENT: CHALLENGES AND OPPORTUNITIES BEYOND STATISTICS
Daniela Fischer, Boehringer Ingelheim Pharma GmbH & Co. KG
Oliver Schönborn-Kellerberger, Cogitars GmbH
cogitars supporting innovation
![Page 2: INTRODUCING BAYESIAN DOSE ESCALATION IN AN … › wp-content › uploads › ...INTRODUCING BAYESIAN DOSE ESCALATION IN AN INDUSTRIAL ENVIRONMENT: CHALLENGES AND OPPORTUNITIES BEYOND](https://reader033.fdocuments.in/reader033/viewer/2022060401/5f0e40487e708231d43e565b/html5/thumbnails/2.jpg)
Outline
• Available algorithmic designs
• Bayesian approaches for dose escalation
• Communicating Bayesian Designs
• Implementing Bayesian Designs in Practice – A
company perspective
cogitars supporting innovation
![Page 3: INTRODUCING BAYESIAN DOSE ESCALATION IN AN … › wp-content › uploads › ...INTRODUCING BAYESIAN DOSE ESCALATION IN AN INDUSTRIAL ENVIRONMENT: CHALLENGES AND OPPORTUNITIES BEYOND](https://reader033.fdocuments.in/reader033/viewer/2022060401/5f0e40487e708231d43e565b/html5/thumbnails/3.jpg)
Available Algorithmic Designs (3+3)
cogitars supporting innovation
Next Cohort 3 new patients
Same dose
Current Cohort Dose 1
3 Patients
Next Cohort 3 new patients
Next higher dose
Next cohort 3 new patients at
previous dose level OR
Termination of study OR
Declaring previous dose level as MTD
No DLT One DLT Two ore more DLTs
Advantages: • Easy to implement
• Used for over 50
years
• No need of a
statistician
Disadvantages: • Fixed cohort size
• Algorithm makes
the decision
• Only in-cohort
information
• Target rate 35%-
40% [1]
![Page 4: INTRODUCING BAYESIAN DOSE ESCALATION IN AN … › wp-content › uploads › ...INTRODUCING BAYESIAN DOSE ESCALATION IN AN INDUSTRIAL ENVIRONMENT: CHALLENGES AND OPPORTUNITIES BEYOND](https://reader033.fdocuments.in/reader033/viewer/2022060401/5f0e40487e708231d43e565b/html5/thumbnails/4.jpg)
Bayesian Approaches for Dose Escalation
Logit {πθ(d)}=log α + β log (d/d*), α, β >0 where θ =(log α, log β)
and d* is the reference dose [2]
cogitars supporting innovation
●
●
●
●
●
●
Median (95% CrI)
Pro
babili
ty o
f D
LT
1 2 3.3 5 6.7 9
00.2
0.4
0.6
0.8
1
Dose
0/1Data
Interval probabilitiesP
roba
bility
1 2 3.3 5 6.7 9
00.5
00
.50
0.5
1
Dose
0/1Data
Advantages: • Flexibility
• Cost-effectiveness
• Risk prevention
• Joint decision making
• Target rate 60%-70%[2]
Disadvantages: • you need a
statistician
• Longer protocol
development
• Analysis tools
![Page 5: INTRODUCING BAYESIAN DOSE ESCALATION IN AN … › wp-content › uploads › ...INTRODUCING BAYESIAN DOSE ESCALATION IN AN INDUSTRIAL ENVIRONMENT: CHALLENGES AND OPPORTUNITIES BEYOND](https://reader033.fdocuments.in/reader033/viewer/2022060401/5f0e40487e708231d43e565b/html5/thumbnails/5.jpg)
Communication of Bayesian Designs
Communication essentials
• F2F with full team / take the time
• Limit use of statistical terminology
• Understand the perspective of the
stakeholders (especially non-statisticians)
• Receive feedback
cogitars supporting innovation
![Page 6: INTRODUCING BAYESIAN DOSE ESCALATION IN AN … › wp-content › uploads › ...INTRODUCING BAYESIAN DOSE ESCALATION IN AN INDUSTRIAL ENVIRONMENT: CHALLENGES AND OPPORTUNITIES BEYOND](https://reader033.fdocuments.in/reader033/viewer/2022060401/5f0e40487e708231d43e565b/html5/thumbnails/6.jpg)
Communication of Bayesian Designs
Flexibility
• Cohort size
• Dose selection
• Regimen / population
• Avoids amendments
cogitars supporting innovation
![Page 7: INTRODUCING BAYESIAN DOSE ESCALATION IN AN … › wp-content › uploads › ...INTRODUCING BAYESIAN DOSE ESCALATION IN AN INDUSTRIAL ENVIRONMENT: CHALLENGES AND OPPORTUNITIES BEYOND](https://reader033.fdocuments.in/reader033/viewer/2022060401/5f0e40487e708231d43e565b/html5/thumbnails/7.jpg)
Communication of Bayesian Designs
Cost-effectiveness
• Accelerated escalation
• Optimal patient number at different doses
• Prior information: replace on-study
cohorts
cogitars supporting innovation
![Page 8: INTRODUCING BAYESIAN DOSE ESCALATION IN AN … › wp-content › uploads › ...INTRODUCING BAYESIAN DOSE ESCALATION IN AN INDUSTRIAL ENVIRONMENT: CHALLENGES AND OPPORTUNITIES BEYOND](https://reader033.fdocuments.in/reader033/viewer/2022060401/5f0e40487e708231d43e565b/html5/thumbnails/8.jpg)
Communication of Bayesian Designs
Risk prevention
• Avoid closing doses by pure chance
findings
• Evaluate intermediate doses
• Incorporate additional information in
decision making
cogitars supporting innovation
![Page 9: INTRODUCING BAYESIAN DOSE ESCALATION IN AN … › wp-content › uploads › ...INTRODUCING BAYESIAN DOSE ESCALATION IN AN INDUSTRIAL ENVIRONMENT: CHALLENGES AND OPPORTUNITIES BEYOND](https://reader033.fdocuments.in/reader033/viewer/2022060401/5f0e40487e708231d43e565b/html5/thumbnails/9.jpg)
Common misunderstandings of Bayesian
Designs
„If the results are the same than a 3+3 why make it
complicated? As we do not expect any tox anyway“
„Bayesian escalation requires more patients and time“
„The Bayesian approach is less safe than a 3+3“
„The model selects the dose“
![Page 10: INTRODUCING BAYESIAN DOSE ESCALATION IN AN … › wp-content › uploads › ...INTRODUCING BAYESIAN DOSE ESCALATION IN AN INDUSTRIAL ENVIRONMENT: CHALLENGES AND OPPORTUNITIES BEYOND](https://reader033.fdocuments.in/reader033/viewer/2022060401/5f0e40487e708231d43e565b/html5/thumbnails/10.jpg)
Implementing a Bayesian Dose Finding Design at
Boehringer Ingelheim
• Some past studies where characteristics of the 3+3
design were of disadvantage
Observation of early DLTs
Study where the 3rd patient of a cohort needs to be
replaced close to end of observation period,
therefore long study duration
Need for a more flexible design
Why transition from 3+3 design to a Bayesian dose finding design?
cogitars supporting innovation
![Page 11: INTRODUCING BAYESIAN DOSE ESCALATION IN AN … › wp-content › uploads › ...INTRODUCING BAYESIAN DOSE ESCALATION IN AN INDUSTRIAL ENVIRONMENT: CHALLENGES AND OPPORTUNITIES BEYOND](https://reader033.fdocuments.in/reader033/viewer/2022060401/5f0e40487e708231d43e565b/html5/thumbnails/11.jpg)
Implementing a Bayesian Dose Finding Design at
Boehringer Ingelheim
• Centrally:
Trial statisticians contact „expert team“
Expert team supports:
understanding general design and protocol
development
(Optional) escalation board meeting preparation, TSAP
writing, trial implementation, etc.
Statistics involvement in the trial process at the moment:
cogitars supporting innovation
![Page 12: INTRODUCING BAYESIAN DOSE ESCALATION IN AN … › wp-content › uploads › ...INTRODUCING BAYESIAN DOSE ESCALATION IN AN INDUSTRIAL ENVIRONMENT: CHALLENGES AND OPPORTUNITIES BEYOND](https://reader033.fdocuments.in/reader033/viewer/2022060401/5f0e40487e708231d43e565b/html5/thumbnails/12.jpg)
Implementing a Bayesian Dose Finding Design at
Boehringer Ingelheim
• Locally:
Each trial statistician is familiar with the design
„Expert team“ only involved in
Exceptional cases
Specific questions
Adaptations of the standard model / development of
new models
To achieve this, a training for all statisticians is
planned to be given in autumn
Statistics involvement in the trial process in future:
cogitars supporting innovation
![Page 13: INTRODUCING BAYESIAN DOSE ESCALATION IN AN … › wp-content › uploads › ...INTRODUCING BAYESIAN DOSE ESCALATION IN AN INDUSTRIAL ENVIRONMENT: CHALLENGES AND OPPORTUNITIES BEYOND](https://reader033.fdocuments.in/reader033/viewer/2022060401/5f0e40487e708231d43e565b/html5/thumbnails/13.jpg)
Implementing a Bayesian Dose Finding Design at
Boehringer Ingelheim
• More trials using Bayesian dose finding design
• Enhancement of the current design
• Using Bayes in Phase II regarding
– Confirmation of dose
– Bayesian decision criteria for Phase II (PoC), cmp
Gsponer et al.
Future planning with Bayesian statistics:
cogitars supporting innovation
![Page 14: INTRODUCING BAYESIAN DOSE ESCALATION IN AN … › wp-content › uploads › ...INTRODUCING BAYESIAN DOSE ESCALATION IN AN INDUSTRIAL ENVIRONMENT: CHALLENGES AND OPPORTUNITIES BEYOND](https://reader033.fdocuments.in/reader033/viewer/2022060401/5f0e40487e708231d43e565b/html5/thumbnails/14.jpg)
Biometrisches Kolloquium (10.12.2008, Julia
Hocke)
Thank you.
cogitars supporting innovation
![Page 15: INTRODUCING BAYESIAN DOSE ESCALATION IN AN … › wp-content › uploads › ...INTRODUCING BAYESIAN DOSE ESCALATION IN AN INDUSTRIAL ENVIRONMENT: CHALLENGES AND OPPORTUNITIES BEYOND](https://reader033.fdocuments.in/reader033/viewer/2022060401/5f0e40487e708231d43e565b/html5/thumbnails/15.jpg)
References
cogitars supporting innovation
[1] LIN, Y., AND SHIH, W.
Statistical properties of the traditional algorithm-based designs for phase I cancer clinical trials.
Biostatistics 2 (2001), 203–215.
[2] NEUENSCHWANDER, B., BRANSON, M., AND GSPONER, T.
Critical aspects of the Bayesian approach to phase I cancer trials.
Stat Med 27 (2008), 2420–2439.
[3] NEUENSCHWANDER, B., MATANO, A., TANG, S., ROYCHOUDHURY S. WANDEL, S. BAILEY,
S.
A Bayesian Industry Approach to Phase I Combination Trials in Oncology
Statistical Methods in Drug Combination Studies, CRC Press (2015). Chapter 6