Cyrus Mehta Ph.D. and Lingyun Liu Ph.D Cytel Inc., Cambridge MA

Adaptive Designs to Demonstrate Risk Reduction in Cardiovascular Outcome Trials

A Case Study of the EXAMINE Trial

Cyrus Mehta Ph.D. and Lingyun Liu Ph.DCytel Inc., Cambridge MA

FDA and Industry: September 23, 2014 2

• Multicenter randomized double blind placebo controlled study of Alogliptin a DPP4 inhibitor

• Designed to rule out excess risk of CV outcomes in very high risk T2DM patients

• Primary Endpoint: composite of first occurrence of CV death, MI or stroke (MACE)

• FDA Guidance: estimate upper bound of 95% RCI for the hazard ratio of Alogliptin/Placebo• if < 1.8: submit pre-marketing NDA for alogliptin• if < 1.3: CV safety demonstrated

The EXAMINE Trial


EXAMINE: Group Sequential Design

1.8

1.3

1

80 100 125 150

550 650

550 650

• O’Brien-Fleming spending function

• Sample size: 5400

• Events: 150 for 1.8 and 650 for 1.3

• Enrolment 2 years

• Trial duration 4.25 years

• 94% power for 1.8 for true HR=1

• 91% power for 1.3 for true HR=1

Only 53% power to claim superiority if true HR=0.85


EXAMINE: Study Results

Non-

inferiority HR 1.8

Non-

inferiority HR 1.3

Superior

ity HR 1

83 100 125 150

550 650

550 650

• Total enroll: 5383

• IA1: 83 events• Upper bound of RCI: 1.51• Submitted NDA

• IA2: 550 events• HR: 0.96• Upper bound RCI: 1.17• Stopped to claim non-inferiority

• White et. al. published in NEJM

Trial could have continued to 650 events in hopes of showing superiorityWhy stop at 550 and accept a non-inferiority claim?


Trial was underpowered for Superiority

Hazard Ratio 80% Power 90% Power

Events Sample Size Events Sample Size

0.8 631 5253 845 70340.85 1189 9644 1592 12913

0.9 2829 22377 3787 29954

0.95 11933 92116 15975 123317

Deterrents to an up-front superiority design• Don’t know what HR to target• HR = 0.8 unlikely• HR > 0.85 leads to very large trial with high risk of failure

Sample size requirements for a trial powered to show superiority(assumes 2 years accrual and 3 additional years of follow-up)


• SAVOR-TIMI Trial (NEJM 2013)• Designed up front for superiority• Planned for 1040 events and 16,500 patients• 85% power to detect HR=0.83• O’Brien-Fleming boundary at 50% information

• After enrolling 16,492 patients with 2.1 years median follow-up and 1222 MACE events, the trial failed to show superiority

• Could have designed a smarter trial!

The Risks of Powering Up-Front for Superiority


• Power for non-inferiority with 650 events and 5400 patients

• At interim analysis (IA-550), re-power for superiority by increasing the number of events and sample size provided:1. NI criterion is met (RCI bound < 1.3)2. CP(Sup) is in a “promising zone”

Less risky to allow switching from NI to Superiority objective at interim analysis


Decision Rules at Interim Analysis

IA at 550 events

NI criterion met (RCI <1.3)

Compute CP(Sup)

CP(Sup) < 20%

Terminate and claim NI

CP(Sup) ≥20%

Switch to Adaptive Design

NI criterion not met

Final Analysis at 650

Test for NI and Sup


The Path to an Adaptive Switch

IA at 550 events

NI criterion met (RCI <1.3)

Compute CP(Sup)

CP(Sup) < 20%

Terminate and claim NI

CP(Sup) ≥20%


NI criterion not met

Final Analysis at 650

Test for NI and Sup



Unfavorable Zone 20%<CP(Sup)<50%

Go to 650 events

Promising Zone50%≤CP(Sup)<90%

Increase events to 1300Incr sample size to 10,800

Go to 650 events

Criteria for adaptive increase of events and sample size


Zones for Decision Making to Claim Superiority after NI Boundary is Crossed

Stop and claim NI: CP(sup)<20%

Unfavorable Zone: 20% ≤CP(sup)<50%Carry on with no change

Promising Zone: 50% <CP(sup)≤90%Double the events and sample size

Favorable Zone: CP(sup)>90%Carry on with no change


• Since promising zone starts at 50% conditional power, no special adjustment needed for the adaptive increase in events (Chen, DeMets and Lan, 2004)

• Thus one can use the classical RCI for a group sequential superiority trial

where D is the number of MACE events at the final analysis (here 650)

No adjustment necessary for adaptation

Dα 2

ln(HR) c


Results: If IA-550 is in Promising Zone

0.8 0.85 0.9 0.95 10

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Hazard Ratio

Pow

er


Operating Characteristics if HR=0.85

• NI claim is assured• Sup claim has good prospects too

• 40% chance at IA with 550 events• 0.12x0.37= 4% chance at 650 events (no adaptation)• 0.11*0.96=11% chance at 1300 events (adaptation)

• If adaptation occurs, 96% chance of showing Sup


Operating Characteristics if HR=1

• NI claim has at least 90% power (per design) with 76% chance of early stop• Very little (<2%) chance of showing superiority• Only 2% chance of expanding the trial by doubling sample size and events


• All design details are included in DMC charter• DMC buys into design at the kick-off meeting, but

reserves right to exercise clinical judgment• Although the trial is expanded only if the IA at 550

events shows promise of superiority:• Actual interim decision should only be conveyed on

need to know basis (to drug supply and IVRS teams)• Investigators may be told only that this adaptive design

has a maximum sample size of 13,000 patients and possibility of showing superiority

• Double blind design also important to avoid bias

Operational Considerations


• No anti-glycemic agent has shown protection to CV risk

• Huge up-side to being the first on the market• Traditional superiority design requires:

• large up-front commitment of resources• considerable optimism about underlying HR

• SAVOR-TIMI example underscores the risks• Start with modest expectations and expand

only if interim results are promising

Concluding Remarks

Documents

Cyrus Mehta Ph.D. and Lingyun Liu Ph.D Cytel Inc., Cambridge MA