Adaptive Clinical Trial Case Studies

Cytel has designed more adaptive trials for industry sponsors than any other service provider, helping pharmaceutical and biotech companies raise trial success rates while reducing development time and costs.

Here is a sampling of our recent adaptive trials that briefly summarizes the designs of early phase and FDA-approved confirmatory adaptive studies in virtually every therapeutic area, including medical devices.

Cytel’s experts have mastered the statistical methodologies of flexible trials. The methods developed at Cytel are validated and published, with their authors regularly invited speakers on a range of adaptive topics including:

- advances in Bayesian and Frequentist methods
- adaptive trial modeling and simulation
- key operational areas such as flexible randomization
- drug supply forecast and management

Cytel experts also periodically provide training to FDA biostatisticians on the use of its software.

Due to client confidentiality, we omit sponsor and product information.

Adaptive Trial Example #1
Population selection/enrichment and sample size re-estimation in phase 3 confirmatory trial

Disease area: cardiology

Situation: Developing new short-acting cardiovascular diseases treatments is especially challenging, particularly for Acute Coronary Syndromes (ACS).

This disease area’s specific attributes, including relatively low event rates and diverse patient populations, make phase 3 clinical development particularly difficult.

Companies developing ACS products with traditional clinical methods risk sponsoring huge confirmatory trials to discover only upon conclusion that efficacy could have been demonstrated with fewer patients, that the wrong patient population was targeted, or worse still, the treatment was ineffective all along.

In response to the client’s request for a better phase 3 strategy, Cytel recently compiled an innovative adaptive group sequential design that provided the sponsors with the ability to essentially combine both exploratory and confirmatory phases into one thereby avoiding the inherent trial risks.

To our knowledge, this particular adaptive design is the first of its kind approved by FDA/CDER cardiovascular/renal reviewers.

Procedural Milestones
• Cytel's adaptive clinical trial experts worked in close collaboration with the client company's statisticians to develop an innovative trial design solution.
• The most promising resulting design allowed for adaptive changes in both patient population enrichment and an increase of the study’s sample size based on the information available at an interim analysis.
• Cytel composed a detailed description of the new methodology for the confirmatory study’s statistical analysis plan (SAP).
• The design’s statistical procedures were adjusted to ensure strong control of the type-1 error in the face of multiple hypothesis tests and sub-group selection.

Outcomes
• The sponsor based their trial design and business case decisions using the predictive capabilities of an adaptive trial simulator created by Cytel.
• Cytel and the sponsor company’s representative presented and successfully defended the innovative trial design at an FDA statistical review meeting.
• The sponsor formed an Independent Interim Analysis Committee, including a Cytel biostatistical expert, to recommend the adaptive changes based on the unblinded interim analysis data.

Adaptive Trial Example #2
Sample size re-estimation in Phase 3 confirmatory trial

Adaptive Trial Example #3
Seamless Phase 2/3 design

Disease area: CNS

Situation: Start study with several doses. Make pairwise comparisons to placebo at interim. Drop poorer performing doses and increase sample size for remaining doses.

Procedure Milestones
• Developed criteria based on conditional power for dropping doses and increasing sample size
• Wrote detailed statistical methods, for inclusion in protocol, demonstrating validity of the adaptive procedure
• Created a simulation worksheet that could be used by sponsor or FDA to verify the operating characteristics (type-1 error, power, expected sample size, number of arms carried forward after interim analysis)
• Went to the FDA with sponsor to discuss trial detail

Adaptive Trial Example #4
Revised Phase 2/3 design

Disease area: Neurological Disease

Situation: An Orphan Drug indication for a HIV-associated neurological disease. The sponsor’s phase 2 study identified three possible effective doses.

The sponsor company submitted a combined phase 2/3 “seamless” design of their own, but it was rejected by the FDA. At this point the company engaged Cytel to analyze and redesign the trial.

Procedure Milestones
• Investigated several options for the study design:
• Single Four-Arm Trial: Run a single phase 3 four-arm trial; three dose groups, plus placebo
• Two Separate Trials: Run a four-arm phase 2 trial to select the best dose; then run a second independent phase 3 two-arm trial
• Seamless Phase 2/3 Trial: Start with four arms; select best dose at interim; continue with two-arm trial
• Created application-specific simulation software and evaluated operating characteristics of different designs.
• Recommended an implementation of the seamless phase 2/3 design based on the recent publication (Posch et al, 2005).
• The recommended seamless design was accepted by the FDA.

Adaptive Trial Example #5
Monitoring accruals and events in a Phase 3 survival endpoint trial

Adaptive Trial Example #6
Longitudinal trial with several interim looks to drop doses and possibly activate a second study

Disease area: Osteo arthritis of the knee

Situation: Radiologic measurements of joint space narrowing at baseline, month 12 and month 24. Up to six interim analyses.

Procedure Milestones
• Second pair of eyes for trial design. (Evaluate criteria for dropping a dose, terminating for futility, adding one additional measurement at 36 months, and starting a second trial based on interim results)
• Serve as member of DSMB with responsibility for making adaptive change recommendations to sponsor
• Cytel will serve as independent statistical center and produce the unblinded interim reports for the DSMB

Adaptive Trial Example #7
Pain relief study

Disease area: Rheumatoid and osteo arthritis

Situation: Endpoint is improvement at week 12 relative to baseline, measured on a visual analog scale. Large number of drop-outs are expected. Drop-out rates might be related to treatment. FDA insists on BOCF method for handling drop-outs.

Procedure Milestones
• Developed numerous simulations with various alternative patterns of informative drop-outs for the two arms
• BOCF implied large number of ties.
• Compared t-test and Smirnov test, adjusted for ties, in the simulations (only available in StatXact!)
• Sponsor decided to use Smirnov test and amended the protocol

Adaptive Trial Example #8
Rare disease

Disease area: Nephrology and metabolic disease

Situation: Two studies involving orphan drugs in a rare disease population. Sample size determined by availability of patients. Robust results can only be produced by increasing duration of follow-up.

Procedure Milestones
• Designed both trials; one with survival endpoint and other with longitudinal endpoint (6MWT).
• Survival: based on number of events observed at interim, specify when the trial should be terminated
• Longitudinal: based on between and within subject variance estimates at interim, specify how many additional repeated measurement should be taken
• Addressed FDA questions on trial design
• Addressed questions from medical, regulatory members of study team at the sponsor’s site
• Produced unblinded interim report for DMC for longitudinal trial
• Served on DMC for the longitudinal trial

Adaptive Trial Example #9
Group sequential and adaptive study designs for a cardiac ultrasound device

Disease area: Cardiac arrhythmia

Situation: Dual uncertainty in the endpoint: treatment effect δ = 30, 45 or 60 secs and std dev σ = 90 or 100 secs.

Procedure Milestones
• Discussed and reviewed protocol with sponsor over the phone (California based company)
• Verified and reproduced the sponsor’s sample size calculations for the fixed single-look designs with all six combinations of (δ, σ)
• Obtained sample sizes for a different randomization ratio (1:1 rather than 3:1)
• Total sample sizes were roughly 350, 160 or 80 depending on δ = 30, 45 or 60 secs
• Designed group sequential trials with early stopping for efficacy
• Designed for maximum sample size of roughly 350 with looks at 80 and 160 patients
• Investigated the user of three possible spending functions
• Produced tables to illustrate sample size inflation and final testing p-value penalty
• Simulated the trial under various (mis)specifications of (δ, σ)
• Reported the trial’s probability of success (concluding in favor of the treatment) at each interim look and also overall
• Also reported the distribution of sample sizes that stopped the trial
• Designed an adaptive two-stage designs using the Cui, Hung and Wang (1999) method to control the type I error rate
• Designed for minimum sample size of roughly 80 patients with option to increase the sample size at an interim look using an O’Brien Fleming spending function
• Investigated two filters for the estimated treatment effect at the interim look (Only adapt if the estimate is within the interval defined by the filter)
• Simulated the trial under various (mis)specifications of (δ, σ)
• Reported the trial’s probability of success (concluding in favor of the treatment) at each interim look and also overall
• Also reported the expected sample size of the trial and the probability of going through with an adaptation

Conclusion
Produced and submitted report demonstrating that group sequential design would be simpler to implement, however for financial reasons the adaptive design was most attractive to the sponsor

Adaptive Trial Example #10
Trial simulations reveal best design option; subsequent trial interim analysis

Disease area: CNS rare disease

Situation: Outcomes: normal, binomial, longitudinal data
Two arms: control and experimental therapy

Procedure Milestones
• Used extensive simulations to compare operating characteristics of potentially attractive adaptive designs with the simpler information-based, group-sequential, and single-look designs of these trials
• Showed that in this case, simpler designs are substantially more efficient and that more sophisticated designs may face unanticipated difficulties
• Recommendation to use well-established versions of flexible designs was accepted by the sponsor
• Analyzed data at interim look

Adaptive Trial Example #11
Adaptive Phase 2 trial design employing Bayesian methods for a cardiovascular drug

Disease area: Cardiovascular

Procedure Milestones
Original design:
• Double blind, randomized, parallel group
• Placebo + four doses (2.5, 5, 10, 20 mg)
• Fixed sample size : 400 subjects, 80/arm
• Accrual time = 24 weeks
• Primary endpoint (binary) at week 12
• Target dose is ED80

Alternative adaptive design:
• Randomize equally to each dose up to interim look.
• Interim look at week 18 (100 observations of end-point)
• Compute the Bayesian predictive probability of at least 80% response rate for each dose,
• Drop arms using predetermined rules

Illustrative rules:
• Drop dose if predictive probability is less than 0.4
• If two doses have predictive probabilities greater than 0.8 of at least 80% response rate drop the higher dose

Statistical analysis:
• Standard frequentist hypothesis testing methods used to determine which doses differ from placebo (FDA • guidance)
Specifically:
• The p-values from the two stages can be combined (using Bauer and Keiser method)
• Adjustments for multiple comparisons are available (Hochberg and Tamhane)
• Simulations with different scenarios were used to obtain power and number of subjects on effective doses
• Results showed superiority of adaptive design over the original design

Drug supply considerations:
• Adaptive designs lead to uncertainty in the quantity of doses required
• Calculated the units of each dose required under different scenarios, determining only a modest increase in drug requirement would be required

Adaptive Trial Example #12
Bayesian-based adaptive Phase 2 dose-finding trial design simulation

Disease area: Dental pain

Adaptive Design Parameters
• Primary endpoint is average pain relief from 0-8 hours after single dose treatment after surgical removal of molars
• Placebo + seven dose candidates
• Adaptive allocation of doses to subjects based on responses of earlier subjects
• Allocation rules designed to assign doses in the rising part of the dose response curve to improve response estimation at “interesting” doses

Procedure Milestones
• Simulations used to investigate operating characteristics of different sample sizes for several dose-response scenarios
• Data from previous dental pain trials used to develop simulation scenarios.
• Simulations showed that one adaptive trial can replace two standard PoC and dose-finding trials and has greater power than PoC while requiring 50% smaller sample size for the same Mean Square Error in response estimation at each dose
• A response-adaptive randomizer program developed for real-time deployment during trial