My Test Presentation

Adaptive Clinical Trials

From Basics to Bayesian

Alex Kaizer, PhD

June 5, 2019

ACCORDS Emerging Research Designs Conference

Outline

• What are adaptive designs?

• Adaptive design elements and examples

• Bayesian methods in adaptive trials

Adaptive Trial Designs

What are adaptive designs?

• The key to all my research woes!

• Designs where I can do whatever I want, whenever I want to

(ethically) answer my research questions.

• The “good” designs that statisticians have been selfishly keeping to

themselves all this time!

• “An adaptive design is defined as a clinical trial design that allows for

prospectively planned

modifications to one or more aspects of the

design based on accumulating data from subjects in the trial.” (FDA

2018 Adaptive Designs for Clinical Trials Guidance Document)

FDA Adaptive Elements

• Group sequential designs (i.e., interim analyses)

• Adaptations to sample size (i.e., sample size re-estimation based on interim

results to preserve power)

• Adaptations to the patient population (i.e., adaptive enrichment)

• Adaptations to treatment arm selection (i.e., adding or terminating arms)

• Adaptations to patient allocation (i.e., adaptive randomization)

• Adaptations to endpoint selection

• Adaptations to multiple design features (combining multiple features

above)

Sample Size Re-Estimation

Adaptations to Sample Size

Why Adapt the Sample Size?

Hypothetical Scenario:

• You design and power a study on a research

topic with limited prior information (i.e., there

is uncertainty in your sample size calculation

assumptions)

• As the study is being conducted, the observed

treatment effect is smaller than expected, but

still clinically meaningful

• If we maintain the planned sample size, we may

be underpowered to detect this difference

Image Source: Everyday Health

Sample Size Re-Estimation

• Using interim estimates we can address the prior uncertainty about

the treatment effect size

• These can be blinded or unblinded, however they involve different

statistical approaches and the FDA Guidance focuses primarily on the

unblinded context

• FDA recommends steps should be taken to limit personnel with

detailed knowledge to maintain trial integrity

• It can be challenging if the re-estimation suggests the need for a

much larger sample size

Trial Example

• Study powered for composite

event rates of 5.1% vs. 3.9% in

study arms  10,900 patients

for 85% power and two-sided

α=0.05

• Unblinded sample size re-

estimation planned after 70%

enrolled

• At the interim analysis, an early

stopping efficacy boundary was

crossed but DSMB decided to

continue the trial as planned

(i.e., no sample size increase)

Adapting the Patient Population

Adaptive Enrichment

Why “Enrich” the Patient Population?

Hypothetical Scenario:

• You expect the treatment effect to be

greater in a certain targeted subset of the

trial population: >

• Do we enroll only the targeted

subpopulation?

Reasons for Population Enrichment

• Want information about both the targeted and non-targeted

subpopulations

• Uncertain about treatment effect in non-targeted subpopulation (i.e.,

perhaps the treatment is as effective or less effective but still clinically

meaningful)

• Can provide greater power relative to a fixed sample design without

enrichment (i.e., if we restrict enrollment we have more

subpopulation observations versus having equivalent power)

Simple Enrichment Example

1. Enroll both groups at start

2. At interim analysis,

determine if you

continue enrollment in

the overall population

or restrict future

enrollment to the

targeted

subpopulation.

3a. Continue enrollment of both

3b. Restrict enrollment to subpop.

Seamless Designs

Adaptations to Treatment Arm Selection

Seamless Designs

• Seamless study designs combine multiple phases of a study into one

trial

• e.g., Phase II and Phase III combined to include both treatment selection and

confirmation in one trial

• Interim analyses used to determine what continues from Phase II

portion of the study to Phase III

• Advantages include reducing overall study size, shorter development

time, more long term safety information

• Disadvantages include logistical challenges and issues maintaining

statistical properties

One Version of Seamless Phase II/III Designs

• Compare Treatment 1 and 2 after Phase II and drop least effective arm.

• Then compare efficacy after Phase III between the SOC and continued

treatment using all data from Phases II and III.

Multi-Arm Multi-Stage

MAMS can drop ineffective arms

early on at an interim analysis.

Promising arms seamlessly

continue to a (confirmatory)

Phase III trial.

One disadvantage is that you can

only compare “Novel” arms to

the Control arm (to maintain the

type I error and power).

Adaptive Randomization

Adaptations to Patient Allocation

Baseline (Covariate) Adaptive Randomization

• The probability of the next treatment assignment is altered on the

basis of the previous assignments in order to achieve better balance

(i.e., biased coin, minimization procedures).

• Considerations:

• How to implement (central entity vs. local entities)

• Multiple treatments

• What is considered a lack of balance

• What covariates to use for balance

• Main advantage: opportunity to balance composition of treatment

groups on several characteristics without stratification

Response/Outcome Adaptive Randomization

• Assignment probabilities are modified based on observed responses

or outcomes

• The motivation is to allocate as many patients as possible to the

“best” treatment arm

• Recent research has identified that outcome adaptive randomization

may result in randomization to the inferior arm, concerns about

sample size imbalance (leading to reduced power), and challenges

where time effects are present

Response Adaptive Randomization Example

Zelen’s 1969 Play the Winner Design (2 arm study):

1. Assign 1

participant to either arm with equal probability

2. Observe success/failure in arm

3. Depending on outcome…

1. Observed success leads to use increasing the probability of the successful

treatment being assigned for the next participant

2. Observed failure leads to a decreased probability

Disadvantages are that sample size/power is challenging to calculate a

priori and you need to know the previous response before randomizing

the next individual (although you could update in blocks)

Master Protocol Designs

Umbrellas, Baskets, and Platforms

Master Protocols

• Traditionally we have conducted separate standalone studies for at

most a few interventions in targeted populations, however these are

becoming increasingly expensive and prohibitive

• Precision medicine and the need for flexible designs to consider

multiple drugs, diseases, populations, or combinations of these are

needed

• Master protocols provide a unifying framework that use one master

protocol for a study that is designed to answer multiple questions

MP Innovation

• Woodcock and LaVange

describe the many areas of

innovation that can be found

in master protocols

General Types of Master Protocols

Umbrellas and Baskets

Umbrella trials identify a single

(broad) disease, but then

further classified by subtypes

and treated accordingly

Basket trials identify a common

mutation (or trait) across sites

and then treat all with a

common intervention

Another example

figure of basket

and umbrella

designs (Figure 1

from Woodcock

and LaVange)

Platform Trials

Platform designs

can be very

flexible and

potentially

complex (Figure 2

from Woodcock

and LaVange)

Master Protocols

• Designs can be noncomparative or comparative

• If comparative, you may have a common control group or multiple control

groups depending on design

• Designs can include adaptive elements or not

• Designs can be exploratory or confirmatory

• LOTS of flexibility

Umbrella Trial Example

Umbrella Example: Study Design (BATTLE-1)

• Outcome was complete or partial response, stable disease,

progression free survival, overall survival, toxicity

• Phase II, single-center, comparative trial with (response) adaptive

randomization

• Four therapies (three mono and one combination)

• Study enrolled advanced NSCLC with specific mutations

• 255 adults who had at least 1 failed chemotherapy regimen

Umbrella Example cont.

Umbrella Example: Conclusion

• Demonstrated the feasibility of the umbrella design to advance

personalized treatment of NSCLC

• Different responses by mutation type and status:

Platform Trial Example

West Africa Ebola Virus Disease Outbreak

Platform Example: PREVAIL II

• No known candidate therapies or vaccines for Ebola

• Outcome was 28-day mortality

• Design sequentially considered multiple treatments within a single

trial to most effectively identify beneficial therapeutics

• Used a Bayesian design with frequent interim monitoring (starting

after 12 participant outcomes observed, 6 per arm)

• Used Haybittle-Peto style boundaries for interim monitoring based on

the posterior probability of the experimental treatment being better

than the current standard of care

PREVAIL II Example Design

Due to concerns

with time effects,

only concurrent

controls were

used in analyses

(i.e., only

information within

each segment).

PREVAIL II Conclusion

• Terminated early due to success of public health measures, which

prevented desired enrollment of 100 per arm in first segment

• Patients in treatment arm had lower 28-day mortality rate (22% vs.

37%), but it did not meet the prespecified statistical threshold for

efficacy

• Did demonstrate minimal safety concerns with the intervention

Extension to Incorporate Past Information

Use methods to

incorporate past

segments when

“exchangeable” (i.e.,

potentially use non-

concurrent data)

Adaptively randomize

to maintain

information balance

between oSOC and

Experimental arms

Bayesian Methods in Adaptive

Designs

A Brief Introduction

A Brief Note on Frequentist vs. Bayesian Designs

“Everyone is Bayesian in the design phase” (i.e., power, type I error,

effect size, etc. are usually based on prior studies or evidence)

Bayesian Adaptive Design

Essentially, any designs that use Bayesian approaches for statistical

reasoning and/or calculations, with some examples being:

• Use of predictive statistical modeling

• Use of assumed (prior) dose-response relationships to govern dose

escalation and selection

• Borrowing information from external sources (e.g., previous trials,

natural history studies, registries) via informative prior distributions

• Use of posterior probability distributions to form trial success criteria

(as opposed to frequentist p-values)

Design Considerations

• The FDA requires that all trials maintain desired frequentist operating

characteristics, including Bayesian trials:

• Power

• Type I error rate

• Evaluating trial operating characteristics generally involves extensive

simulation studies (i.e., this is how you calculate power, the target

sample size, etc.)

• Prior specification can be challenging (e.g., conjugate priors,

informative priors, vague priors, etc.)

Source: xkcd.com #1132

Should I consider adaptive designs?

Advantages

• Improved flexibility

• More efficient use of resources

(financial and administrative)

• Greater statistical power

possible

• Ability to answer broader

questions that may be refined as

the trial progresses

Challenges

• Advanced analytic methods

needed to avoid type I errors

• Gains in efficiency have trade-

offs with other trial components

• Logistics to maintain trial

conduct and integrity

• Adaptations may be limited by

clinical/scientific constraints

Thank you!