This document discusses adaptive designs for phase 3 oncology trials. It uses the VALOR trial as a case study to illustrate a sponsor's dilemma in designing a trial with limited prior data. It proposes a promising zone design that allows staged investment - an initial modest sample size with the option to increase size and power if interim results are promising. Simulations show this two-stage investment approach increases power over a non-adaptive design while managing risks for sponsors. The document also discusses extensions to population enrichment designs.
A Non-Revisiting Genetic Algorithm for Optimizing Numeric Multi-Dimensional F...ijcsa
Genetic Algorithm (GA) is a robust and popular stochastic optimization algorithm for large and complex search spaces. The major shortcomings of Genetic Algorithms are premature convergence and revisits to individual solutions in the search space. In other words, Genetic algorithm is a revisiting algorithm that escorts to duplicate function evaluations which is a clear wastage of time and computational resources. In this paper, a non-revisiting genetic algorithm with adaptive mutation is proposed for the domain of MultiDimensional numeric function optimization. In this algorithm whenever a revisit occurs, the underlined search point is replaced with a mutated version of the best/random (chosen probabilistically) individual from the GA population. Furthermore, the recommended approach is not using any extra memory resources to avoid revisits. To analyze the influence of the method, the proposed non-revisiting algorithm is evaluated using nine benchmarks functions with two and four dimensions. The performance of the proposed genetic algorithm is superior as contrasted to simple genetic algorithm as confirmed by the experimental results.
We analyzed data that was recorded from a previous experiment done on the use of palm vein recognition that used two devices, cradle and non-cradle. Our goal was to see which device provided more benefits as a biometric system. We compared both devices by evaluating the data results that were provided . The main factors we looked at were capture times of each test subject when verifying using each device, and the amount of failure rates that occurred from both devices. From our conclusion, we believe the results we found will help lead to an improvement in the biometric system for future use. If the industry demands a clean environment, then non-cradle is recommended. If throughput time is of priority, then cradle is recommended.
Many resources discuss machine learning and data analytics from a technology deployment perspective. From the business standpoint, however, the real value of analytics is in the methodology for solving some systemic holistic problems, rather than a specific technology or platform.
In this presentation, the focus is shifted from the technology deployment to the analytics methodology for solving some holistic business problems. Two examples will be covered in detail:
(i) Analysis of the performance and the optimal staffing of a team of doctors, nurses, and technicians for a large local hospital unit using discrete event simulation with a live demonstration. This simulation methodology is not included in most Machine Learning algorithms libraries.
(ii) Identifying a few factors (or variables) that contribute most to the financial outcome of a local hospital using principal component decomposition (PCD) of the large observational dataset of population demographic and disease prevalence.
Presentation Description: Patient flow is a universal challenge in healthcare. Through a systematic evaluation of our cardiac catheterization lab operations (EMR assessment, metric development, capacity analysis using simulation), we found opportunities to reduce unnecessarily bedded cath lab patients by 8 per day on high-demand telemetry units at a large Midwestern hospital. Objective/Purpose: : Frequently, patients at Iowa Methodist Medical Center’s ED and ICU are waiting for a telemetry bed to become available. Patients who are either waiting for a cath procedure or have just had one, occupy nearly 50% of the available telemetry beds, many of whom do not require inpatient level of care. Inadequate, non-transparent cath lab operations and poor schedule adherence were signaled as reasons for why many patients are unnecessarily bedded. Objectives: Understand EMR usage; Scheduling / Reporting Capabilities Develop Sustainable Operations Metrics (KPI’s) Use Discrete Event Simulation to Model Various Improvement Scenarios Track impact to patient access and usage of inpatient telemetry beds Methods/Approach: : Methods/Approach Used: Understanding EMRs:24+ hours of observations of the practical use of our EMR for scheduling, documenting, and reporting. Participated in vendor meetings and webinars on current and alternative systems. Developing Metrics: pulled EMR data and developed 6 key metrics + a process for the ongoing delivery of charts and graphs for use by cath lab leadership. Discrete Event Simulation and Impact to Operations: Data was pulled and validated from EMRs, and Arena® DES was used to model over 20 scenarios for changes. Outputs were captured, compared, and selected for implementation. Results/Findings: : The following scenarios modeled using discrete event simulation had desired impacts: Limit outpatient elective cases to 3 on Mondays and Fridays Increase same day discharges of PCI patients Improve amount of cath proceduralists arriving at 7:00AM (or 7:30AM/8:00AM) Provide 8 hours of capacity to perform catheterizations for urgent patients and inpatients on the weekend Work to develop a process of sending same day elective outpatients to other underutilized cath lab within health system (<10 minutes away) Items 1-3 reduce inpatient bed usage by 5 per day; all 5 together result in a reduction of 8 occupied telemetry beds. Conclusion/Practical Implications: : Our analysis demonstrated the value of understanding workflow and information flow for making improvements. Information leads to transparency and ongoing reporting of performance is essential. Using simulation to model change allows for making large operational changes in a safe, virtual environment. We saw the downstream effects of our changes on the hospital overall with the potential to greatly impact telemetry bed usage and patient flow at UnityPoint Health – Des Moines.
A Non-Revisiting Genetic Algorithm for Optimizing Numeric Multi-Dimensional F...ijcsa
Genetic Algorithm (GA) is a robust and popular stochastic optimization algorithm for large and complex search spaces. The major shortcomings of Genetic Algorithms are premature convergence and revisits to individual solutions in the search space. In other words, Genetic algorithm is a revisiting algorithm that escorts to duplicate function evaluations which is a clear wastage of time and computational resources. In this paper, a non-revisiting genetic algorithm with adaptive mutation is proposed for the domain of MultiDimensional numeric function optimization. In this algorithm whenever a revisit occurs, the underlined search point is replaced with a mutated version of the best/random (chosen probabilistically) individual from the GA population. Furthermore, the recommended approach is not using any extra memory resources to avoid revisits. To analyze the influence of the method, the proposed non-revisiting algorithm is evaluated using nine benchmarks functions with two and four dimensions. The performance of the proposed genetic algorithm is superior as contrasted to simple genetic algorithm as confirmed by the experimental results.
We analyzed data that was recorded from a previous experiment done on the use of palm vein recognition that used two devices, cradle and non-cradle. Our goal was to see which device provided more benefits as a biometric system. We compared both devices by evaluating the data results that were provided . The main factors we looked at were capture times of each test subject when verifying using each device, and the amount of failure rates that occurred from both devices. From our conclusion, we believe the results we found will help lead to an improvement in the biometric system for future use. If the industry demands a clean environment, then non-cradle is recommended. If throughput time is of priority, then cradle is recommended.
Many resources discuss machine learning and data analytics from a technology deployment perspective. From the business standpoint, however, the real value of analytics is in the methodology for solving some systemic holistic problems, rather than a specific technology or platform.
In this presentation, the focus is shifted from the technology deployment to the analytics methodology for solving some holistic business problems. Two examples will be covered in detail:
(i) Analysis of the performance and the optimal staffing of a team of doctors, nurses, and technicians for a large local hospital unit using discrete event simulation with a live demonstration. This simulation methodology is not included in most Machine Learning algorithms libraries.
(ii) Identifying a few factors (or variables) that contribute most to the financial outcome of a local hospital using principal component decomposition (PCD) of the large observational dataset of population demographic and disease prevalence.
Presentation Description: Patient flow is a universal challenge in healthcare. Through a systematic evaluation of our cardiac catheterization lab operations (EMR assessment, metric development, capacity analysis using simulation), we found opportunities to reduce unnecessarily bedded cath lab patients by 8 per day on high-demand telemetry units at a large Midwestern hospital. Objective/Purpose: : Frequently, patients at Iowa Methodist Medical Center’s ED and ICU are waiting for a telemetry bed to become available. Patients who are either waiting for a cath procedure or have just had one, occupy nearly 50% of the available telemetry beds, many of whom do not require inpatient level of care. Inadequate, non-transparent cath lab operations and poor schedule adherence were signaled as reasons for why many patients are unnecessarily bedded. Objectives: Understand EMR usage; Scheduling / Reporting Capabilities Develop Sustainable Operations Metrics (KPI’s) Use Discrete Event Simulation to Model Various Improvement Scenarios Track impact to patient access and usage of inpatient telemetry beds Methods/Approach: : Methods/Approach Used: Understanding EMRs:24+ hours of observations of the practical use of our EMR for scheduling, documenting, and reporting. Participated in vendor meetings and webinars on current and alternative systems. Developing Metrics: pulled EMR data and developed 6 key metrics + a process for the ongoing delivery of charts and graphs for use by cath lab leadership. Discrete Event Simulation and Impact to Operations: Data was pulled and validated from EMRs, and Arena® DES was used to model over 20 scenarios for changes. Outputs were captured, compared, and selected for implementation. Results/Findings: : The following scenarios modeled using discrete event simulation had desired impacts: Limit outpatient elective cases to 3 on Mondays and Fridays Increase same day discharges of PCI patients Improve amount of cath proceduralists arriving at 7:00AM (or 7:30AM/8:00AM) Provide 8 hours of capacity to perform catheterizations for urgent patients and inpatients on the weekend Work to develop a process of sending same day elective outpatients to other underutilized cath lab within health system (<10 minutes away) Items 1-3 reduce inpatient bed usage by 5 per day; all 5 together result in a reduction of 8 occupied telemetry beds. Conclusion/Practical Implications: : Our analysis demonstrated the value of understanding workflow and information flow for making improvements. Information leads to transparency and ongoing reporting of performance is essential. Using simulation to model change allows for making large operational changes in a safe, virtual environment. We saw the downstream effects of our changes on the hospital overall with the potential to greatly impact telemetry bed usage and patient flow at UnityPoint Health – Des Moines.
This talk is presented at Bio-inspiring and evolutionary computation: Trends, applications and open issues workshop, 7 Nov. 2015 Faculty of Computers and Information, Cairo University
Conferencia: "Superando la brecha entre Investigación y Aplicación", a cargo del Dr. Ander Ramos, investigador de TECNALIA.
TECNALIA Perspectives 2015. “Industria y Tecnología: Investigación traslacional, de la ciencia al mercado” es el título del evento que contó con la participación del Dr. Niels Birbaumer, experto mundial en el desarrollo de interfaces cerebro-computador y del investigador Dr. Ander Ramos, Premio al Mejor Investigador Joven de Alemania en 2014 por la Academia Alemana para las Ciencias y las Letras.
También presentamos dos novedosas iniciativas de TECNALIA como son un dispositivo de estimulación eléctrica funcional para rehabilitación de pacientes de ictus y el robot quirúrgico con visión 3D y sensaciones táctiles.
Más información en http://www.tecnalia.com
Episode 18 : Research Methodology ( Part 8 )
Approach to de-synthesizing data, informational, and/or factual elements to answer research questions
Method of putting together facts and figures
to solve research problem
Systematic process of utilizing data to address research questions
Breaking down research issues through utilizing controlled data and factual information
SAJJAD KHUDHUR ABBAS
Chemical Engineering , Al-Muthanna University, Iraq
Oil & Gas Safety and Health Professional – OSHACADEMY
Trainer of Trainers (TOT) - Canadian Center of Human
Development
Innovative Strategies For Successful Trial Design - Webinar SlidesnQuery
Full webinar available here: https://www.statsols.com/webinar/innovative-strategies-for-successful-trial-design
[Webinar] Innovative Strategies For Successful Trial Design- In this free webinar, you will learn about:
- The challenges facing your trials
- How to calculate the correct sample size
- Worked examples including Mixed/Hierarchical Models
- Posterior Error
- Adaptive Designs For Survival
www.statsols.com
Underestimated Input of a Central Lab During the Clinical Trial Planning PhaseMichal Dysko
Clinical trials are growing in complexity, particularly with regards to laboratories.
There is a tendency to include unnecessary protocols and inappropriate laboratory procedures in studies that can lead to complicated study sample logistics and large courier costs as a result.
Early engagement of a central lab during a clinical trial planning phase can save a lot of money and time for the study sponsor as well as a lot of unnecessary stress to the project management and investigator teams. We would like to share with you few real life cases that we have experienced in the last couple of years.
We will present the consequences of the protocol's initial assumptions, our proposed solutions and the achieved results. In such a competitive market as Pharmaceuticals, drug developers cannot afford to waste money and must utilize expertise and experience of all clinical trials parties, especially central laboratories, at the earliest possible stage of a clinical trial.
Machine learning and Internet of Things, the future of medical preventionPierre Gutierrez
Title:
"Machine learning and Internet of Things, the future of medical prevention"
Abstract:
In this talk, Pierre Gutierrez, a data scientist at Dataiku, will discuss Dataiku's experiences using machine learning on IOT data. We will talk about the challenges processing and cleaning IoT data, and how to successfully train a model that can be deployed in production. We will illustrate our talk with two examples from our previous work. Creating algorithm for early epilepsy seizure detection based on wearable tech and Detecting people activity through sensor data.
Episode 12 : Research Methodology ( Part 2 )
Approach to de-synthesizing data, informational, and/or factual elements to answer research questions
Method of putting together facts and figures
to solve research problem
Systematic process of utilizing data to address research questions
Breaking down research issues through utilizing controlled data and factual information
SAJJAD KHUDHUR ABBAS
Chemical Engineering , Al-Muthanna University, Iraq
Oil & Gas Safety and Health Professional – OSHACADEMY
Trainer of Trainers (TOT) - Canadian Center of Human
Development
ReComp and P4@NU: Reproducible Data Science for HealthPaolo Missier
brief overview of the ReComp project (http://recomp.org.uk) on Selective recurring re-computation of complex analytics, and a brief outlook for the P4@NU project on seeking digital biomarkers for age-0related metabolic diseases
Propagating Data Policies - A User StudyEnrico Daga
When publishing data, data licences are used to specify the actions that are permitted or prohibited, and the duties that target data consumers must comply with. However, in com- plex environments such as a smart city data portal, multiple data sources are constantly being combined, processed and redistributed. In such a scenario, deciding which policies ap- ply to the output of a process based on the licences attached to its input data is a difficult, knowledge-intensive task. In this paper, we evaluate how automatic reasoning upon se- mantic representations of policies and of data flows could support decision making on policy propagation. We report on the results of a user study designed to assess both the accuracy and the utility of such a policy-propagation tool, in comparison to a manual approach.
Presentation of Hexoskin Validation for KHealth's Dementia Project
The paper is available at: http://www.knoesis.org/library/resource.php?id=2155
Citation for the paper: T. Banerjee, P. Anantharam, W. L. Romine, L. Lawhorne, A. Sheth, 'Evaluating a Potential Commercial Tool for Healthcare Application for People with Dementia' in Proc. of the Intl Conf on Health Informatics and Medical Systems (HIMS), Las Vegas, July 27-30, 2015.
This talk is presented at Bio-inspiring and evolutionary computation: Trends, applications and open issues workshop, 7 Nov. 2015 Faculty of Computers and Information, Cairo University
Conferencia: "Superando la brecha entre Investigación y Aplicación", a cargo del Dr. Ander Ramos, investigador de TECNALIA.
TECNALIA Perspectives 2015. “Industria y Tecnología: Investigación traslacional, de la ciencia al mercado” es el título del evento que contó con la participación del Dr. Niels Birbaumer, experto mundial en el desarrollo de interfaces cerebro-computador y del investigador Dr. Ander Ramos, Premio al Mejor Investigador Joven de Alemania en 2014 por la Academia Alemana para las Ciencias y las Letras.
También presentamos dos novedosas iniciativas de TECNALIA como son un dispositivo de estimulación eléctrica funcional para rehabilitación de pacientes de ictus y el robot quirúrgico con visión 3D y sensaciones táctiles.
Más información en http://www.tecnalia.com
Episode 18 : Research Methodology ( Part 8 )
Approach to de-synthesizing data, informational, and/or factual elements to answer research questions
Method of putting together facts and figures
to solve research problem
Systematic process of utilizing data to address research questions
Breaking down research issues through utilizing controlled data and factual information
SAJJAD KHUDHUR ABBAS
Chemical Engineering , Al-Muthanna University, Iraq
Oil & Gas Safety and Health Professional – OSHACADEMY
Trainer of Trainers (TOT) - Canadian Center of Human
Development
Innovative Strategies For Successful Trial Design - Webinar SlidesnQuery
Full webinar available here: https://www.statsols.com/webinar/innovative-strategies-for-successful-trial-design
[Webinar] Innovative Strategies For Successful Trial Design- In this free webinar, you will learn about:
- The challenges facing your trials
- How to calculate the correct sample size
- Worked examples including Mixed/Hierarchical Models
- Posterior Error
- Adaptive Designs For Survival
www.statsols.com
Underestimated Input of a Central Lab During the Clinical Trial Planning PhaseMichal Dysko
Clinical trials are growing in complexity, particularly with regards to laboratories.
There is a tendency to include unnecessary protocols and inappropriate laboratory procedures in studies that can lead to complicated study sample logistics and large courier costs as a result.
Early engagement of a central lab during a clinical trial planning phase can save a lot of money and time for the study sponsor as well as a lot of unnecessary stress to the project management and investigator teams. We would like to share with you few real life cases that we have experienced in the last couple of years.
We will present the consequences of the protocol's initial assumptions, our proposed solutions and the achieved results. In such a competitive market as Pharmaceuticals, drug developers cannot afford to waste money and must utilize expertise and experience of all clinical trials parties, especially central laboratories, at the earliest possible stage of a clinical trial.
Machine learning and Internet of Things, the future of medical preventionPierre Gutierrez
Title:
"Machine learning and Internet of Things, the future of medical prevention"
Abstract:
In this talk, Pierre Gutierrez, a data scientist at Dataiku, will discuss Dataiku's experiences using machine learning on IOT data. We will talk about the challenges processing and cleaning IoT data, and how to successfully train a model that can be deployed in production. We will illustrate our talk with two examples from our previous work. Creating algorithm for early epilepsy seizure detection based on wearable tech and Detecting people activity through sensor data.
Episode 12 : Research Methodology ( Part 2 )
Approach to de-synthesizing data, informational, and/or factual elements to answer research questions
Method of putting together facts and figures
to solve research problem
Systematic process of utilizing data to address research questions
Breaking down research issues through utilizing controlled data and factual information
SAJJAD KHUDHUR ABBAS
Chemical Engineering , Al-Muthanna University, Iraq
Oil & Gas Safety and Health Professional – OSHACADEMY
Trainer of Trainers (TOT) - Canadian Center of Human
Development
ReComp and P4@NU: Reproducible Data Science for HealthPaolo Missier
brief overview of the ReComp project (http://recomp.org.uk) on Selective recurring re-computation of complex analytics, and a brief outlook for the P4@NU project on seeking digital biomarkers for age-0related metabolic diseases
Propagating Data Policies - A User StudyEnrico Daga
When publishing data, data licences are used to specify the actions that are permitted or prohibited, and the duties that target data consumers must comply with. However, in com- plex environments such as a smart city data portal, multiple data sources are constantly being combined, processed and redistributed. In such a scenario, deciding which policies ap- ply to the output of a process based on the licences attached to its input data is a difficult, knowledge-intensive task. In this paper, we evaluate how automatic reasoning upon se- mantic representations of policies and of data flows could support decision making on policy propagation. We report on the results of a user study designed to assess both the accuracy and the utility of such a policy-propagation tool, in comparison to a manual approach.
Presentation of Hexoskin Validation for KHealth's Dementia Project
The paper is available at: http://www.knoesis.org/library/resource.php?id=2155
Citation for the paper: T. Banerjee, P. Anantharam, W. L. Romine, L. Lawhorne, A. Sheth, 'Evaluating a Potential Commercial Tool for Healthcare Application for People with Dementia' in Proc. of the Intl Conf on Health Informatics and Medical Systems (HIMS), Las Vegas, July 27-30, 2015.
1. Adaptive Designs for
Phase 3 Oncology Trials:
Case Study and Extensions
Cytel User Group Meeting, Paris
October 13, 2011
Cyrus Mehta, Ph.D.
President, Cytel Inc., Cambridge, MA
email: mehta@cytel.com – web: www.cytel.com – tel: 617-661-2011
1 Cytel User Group Meeting, Paris. October 13, 2011
2. Outline of Presentation
• Motivating Example: the VALOR trial
• Sponsor’s dilemma with conventional design
• Promising zone design an alternative approach
• Benefits: Staged investment versus large up-front
investment
• Extension to population enrichment designs
• Role of technology in design implementation
2 Cytel User Group Meeting, Paris. October 13, 2011
3. The VALOR Trial for AML
• Vosaroxin and Ara-C combination evaLuating Overall
survival in Relapsed/refractory AML
• Phase 3, double-blind, placebo-controlled, multinational
trial for first-relapsed or refractory Acute Myeloid
Leukemia (AML)
• Evaluate efficacy and safety of Vosaroxin plus Cytarabine
versus placebo plus Cytarabine
• Vosaroxin is a first-in-class anticancer quinolone derivative
under development by Sunesis pharmaceuticals
3 Cytel User Group Meeting, Paris. October 13, 2011
4. Design Objectives
• Primary endpoint is overall survival
• Design for 90% power at 5%
significance level
• Complete the trial in 30 months
– Enroll for 24 months
– Follow for 6 additional months
4 Cytel User Group Meeting, Paris. October 13, 2011
5. Prior Phase 2 Data
• Limited information on Vosaroxin from a single phase 2 trial of 69 patients
with no active comparator
• Median OS for Vosaroxin estimated to be 7 months from phase 2 trial
• Median OS for Cytarabine estimated to b 5 months, from meta-analysis of
prior studies and consultation with KOLs
• Hazard ratio estimated to be 0.71 amidst considerable uncertainty
5 Cytel User Group Meeting, Paris. October 13, 2011
6. Sponsor’s Dilemma
• Based on phase 2 data:
– Assume 5/7 month median on Ctrl/Trtm (HR=0.71)
– Require 375 events and 450 subjects @ 19/month
• But phase 2 estimates are subject to uncertainty:
– What if 5/6.5 month median on Ctrl/Trtm (HR=0.77)?
– HR = 0.77 is still clinically meaningful
– Require 616 events and 732 subjects @ 31/month
– Not a feasible option for sponsor
6 Cytel User Group Meeting, Paris. October 13, 2011
7. Sponsor is Resource and Time Constrained
Power if designed with Power if designed with
True base-case assumption: alternative assumption:
HR ( HR = 0.71) (HR=0.77)
0.71 91% 99%
0.74 83% 97%
0.77 71% 90%
Resources Needed 450 patients @ 19/month 732 patients @ 31/month
Why not design up-front for HR=0.77 (smallest clinically meaningful effect)?
• Unable to muster resources for large investment with limited phase 2 data
• Rule of thumb cost/patient is $50-80K for an oncology trial with OS
• Study would be extremely overpowered under base-case of HR=0.71
7 Cytel User Group Meeting, Paris. October 13, 2011
8. Sponsor Adopts a Strategy of Staged
Investment
• Design optimistically up-front. Power study to detect
HR=0.71 ( requires 375 events; 450 subjects @ 19/month)
• One interim analysis after 50% information (187 events)
– Stop early if overwhelming evidence of efficacy
– Stop early for futility if low conditional power
– Increase number of events, sample size and (if possible)
rate of recruitment at the interim if results are promising
Key Idea: Invest additional resources and re-power the
study to detect HR=0.77 only after seeing interim results
8 Cytel User Group Meeting, Paris. October 13, 2011
9. The Promising Zone Design
• Partition the interim outcome into three zones based on
the interim estimate of conditional power. For example:
Unfavorable: HR hat ≥ 0.86; no change to design
Promising: 0.74 ≤ HR hat < 0.86; increase resources
Favorable: HR hat ≤ 0.74; no change to design
• Control type-1 error by using Cui, Hung and Wang (1999)
weighted statistic modified for survival data
• Evaluate operating characteristics of design by simulation
9 Cytel User Group Meeting, Paris. October 13, 2011
10. Adaptive Decision Rule: Representation I
10 Cytel User Group Meeting, Paris. October 13, 2011
11. Adaptive Decision Rule: Representation II
11 Cytel User Group Meeting, Paris. October 13, 2011
12. Preserving the Type-1 Error
• Let D1 and D2 be the pre-specified total events at interim
and final analysis. (Here D1 = 187 and D2 = 375)
• Let LR1 and LR2 be the corresponding logrank statistics
• Suppose D2 is altered to D∗
2 > D2 at the interim
• Let LR∗
2 denote the corresponding altered logrank statistic
• Type-1 error is preserved if we use
Zchw =
D1
D2
× LR1 +
D2 − D1
D2
×
D∗
2LR∗
2 −
√
D1LR1
D∗
2 − D1
instead of LR∗
2 for the final analysis
12 Cytel User Group Meeting, Paris. October 13, 2011
13. Adaptation Principles
• Primary driver of power is number of events
• FDA guidance recommends increase only, not decrease
• Increase events by amount needed to achieve some target
conditional power, subject to a cap
• Compute sample size increase necessary to achieve the
desired increase in events without undue prolongation of
the trial
• Complex relationship exists between increase in events,
increase in sample size and study duration. Best evaluated
by simulation
13 Cytel User Group Meeting, Paris. October 13, 2011
15. Operating Characteristics
Under Pessimistic Scenario, HR = 0.77 (10,000 simulations)
Power Duration (months) SampSize
Zone P(Zone) NonAdpt Adapt NonAdpt Adapt NonAdpt Adapt
Unf 25% 33% 33% 28 28 436 439
Prom 34% 71% 90% 29 38 453 680
Fav 41% 95% 95% 26 26 414 413
Total — 71% 78% 28 31 432 509
• Two-stage investment
• Sponsor unable to invest resources needed for 90% unconditional power at
HR=0.77; too risky
• But, if stage-1 results from 172 events (375 subjects) are promising, sponsor
can invest needed resources to boost power to 90% at greatly reduced risk
15 Cytel User Group Meeting, Paris. October 13, 2011
16. Power Curves of Adaptive and
Non-adaptive Designs in Promising Zone
16 Cytel User Group Meeting, Paris. October 13, 2011
17. Attractiveness of Approach
• Up-front sample size investment can be modest
• Additional investment is only made if interim results are
promising
• If that happens, chances of success are dramatically
increased
17 Cytel User Group Meeting, Paris. October 13, 2011
18. Metrics for Evaluating an Adaptive Design
• Traditional View: Unconditional power and average sample
size evaluated before trial begins should be the main
criteria for evaluating risk versus benefit
• Modern View: Presence of an independent data
monitoring committee with a charter to alter the future
course of the trial is a game changer. It permits staged
investment based on a more accurate assessment of power
and lower risk to sponsor as well as to patients
18 Cytel User Group Meeting, Paris. October 13, 2011
19. 19 Cytel User Group Meeting, Paris. October 13, 2011
20. 20 Cytel User Group Meeting, Paris. October 13, 2011
21. 21 Cytel User Group Meeting, Paris. October 13, 2011
22. 22 Cytel User Group Meeting, Paris. October 13, 2011
23. Extension to Population Enrichment
• Goal: prospective strategy to identfy patient who would
respond to particular compound (Pfe)
• Assumptions
– Potential markers identified pre-clinically based on
biology and mechanism of action
– Phase I completed in all comers and phase 2 dose
established
23 Cytel User Group Meeting, Paris. October 13, 2011
24. Phase 2-3 Enrichment Strategy
24 Cytel User Group Meeting, Paris. October 13, 2011
25. Illustrative Example
• Phase 3 trial of Cetuximab vs. SOC for advanced colorectal cancer
showed statistically significant OS (Jonker et. al.,NEJM 2007)
• Retrospective analysis revealed benefit from Cetuximab strongly
correlated with mutation status in exon-2 of the K-ras gene
Gene Median OS by Treatment
Status Cetuximab SOC
Wild Type 9.5 months 4.8 months
Mutant 4.5 months 4.6 months
• A population enrichment design might have established the above
conclusion prospectively
25 Cytel User Group Meeting, Paris. October 13, 2011
26. Conclusions
• Difficult to launch studies with large up-front resource commitments
• Adaptive designs offer option to start small and ask for more if
interim results are promising
• Better suited to advanced and metastatic disease where sufficient
events are obtained before enrollement closes
• Careful attention must be paid to details of implementation such as:
– Patient arrival rates and endpoint arrival rates
– Auditable documentation that the sample size decision was strictly
based on the interim logrank statistic (see demo of Cytel’s ACES
solution later in the program)
– Preservation of confidentiality about interim results, especially
from investigators
26 Cytel User Group Meeting, Paris. October 13, 2011