Artificial Consciousness Group 3 Mohit, Raghuvar, Avin, Aniruddha.
Adaptive Group Sequential Trials with Population ... · Acknowledgements • Joint research with...
Transcript of Adaptive Group Sequential Trials with Population ... · Acknowledgements • Joint research with...
![Page 1: Adaptive Group Sequential Trials with Population ... · Acknowledgements • Joint research with Ping Gao (The Medicines Company) and Jim Ware (Harvard University) • We thank Aniruddha](https://reader035.fdocuments.us/reader035/viewer/2022071217/604aa288f2e5a05d030fa175/html5/thumbnails/1.jpg)
Adaptive Group Sequential Trialswith Population Enrichment:Application to Cardiology
PhRMA Adaptive Working GroupKOL Series
July 11, 2008
Cyrus R. Mehta, Ph.DCytel Inc., Cambridge, MA
email: [email protected] – web: www.cytel.com – tel: 617-661-2011
1 PhRMA KOL Series. 11 July 2008
![Page 2: Adaptive Group Sequential Trials with Population ... · Acknowledgements • Joint research with Ping Gao (The Medicines Company) and Jim Ware (Harvard University) • We thank Aniruddha](https://reader035.fdocuments.us/reader035/viewer/2022071217/604aa288f2e5a05d030fa175/html5/thumbnails/2.jpg)
Acknowledgements
• Joint research with Ping Gao (The Medicines Company)and Jim Ware (Harvard University)
• We thank Aniruddha Deshmukh (Cytel Inc) foroutstanding programming support
• We thank Stuart Pocock and Misha Salganik for helpfuldiscussions
2 PhRMA KOL Series. 11 July 2008
![Page 3: Adaptive Group Sequential Trials with Population ... · Acknowledgements • Joint research with Ping Gao (The Medicines Company) and Jim Ware (Harvard University) • We thank Aniruddha](https://reader035.fdocuments.us/reader035/viewer/2022071217/604aa288f2e5a05d030fa175/html5/thumbnails/3.jpg)
Population Enrichment Design foran Acute Coronory Syndrome Trial
Major challenges to trial design for treatment of acutecoronary symptoms include:
• low event rates
• tiny effect sizes
• diverse patient population
Such trials require enormous sample size commitments.Therefore it is advisable to build in the possiblity for adaptivechanges to the design based on interim looks
3 PhRMA KOL Series. 11 July 2008
![Page 4: Adaptive Group Sequential Trials with Population ... · Acknowledgements • Joint research with Ping Gao (The Medicines Company) and Jim Ware (Harvard University) • We thank Aniruddha](https://reader035.fdocuments.us/reader035/viewer/2022071217/604aa288f2e5a05d030fa175/html5/thumbnails/4.jpg)
Case Study: Platelet Inhibitionduring Percutaneous CoronaryIntervention
• Composite primary endpoint – death, MI or ischemiadriven revascularization within 48 hours
• Placebo event rate between 8% and 10%
• New drug expected to reduce placebo event rate by 20%
• But actual risk reduction could be as low as 15%
4 PhRMA KOL Series. 11 July 2008
![Page 5: Adaptive Group Sequential Trials with Population ... · Acknowledgements • Joint research with Ping Gao (The Medicines Company) and Jim Ware (Harvard University) • We thank Aniruddha](https://reader035.fdocuments.us/reader035/viewer/2022071217/604aa288f2e5a05d030fa175/html5/thumbnails/5.jpg)
Design Options Evaluated
1. Fixed Sample Design
2. Group Sequential Design
3. Group Sequential Design with Sample Size Increase
4. Group Sequential Design with Sample Size Increase andPopulation Enrichment
5 PhRMA KOL Series. 11 July 2008
![Page 6: Adaptive Group Sequential Trials with Population ... · Acknowledgements • Joint research with Ping Gao (The Medicines Company) and Jim Ware (Harvard University) • We thank Aniruddha](https://reader035.fdocuments.us/reader035/viewer/2022071217/604aa288f2e5a05d030fa175/html5/thumbnails/6.jpg)
1. Fixed Sample Design
• Sponsor decides on an initial commitment of N = 8000
• This sample size provides more that 80% power if:
risk reduction ≥ 18% and placebo event rate ≥ 10%
• If these parameter estimates were be off by a fewpercentage points, study might be underpowered
6 PhRMA KOL Series. 11 July 2008
![Page 7: Adaptive Group Sequential Trials with Population ... · Acknowledgements • Joint research with Ping Gao (The Medicines Company) and Jim Ware (Harvard University) • We thank Aniruddha](https://reader035.fdocuments.us/reader035/viewer/2022071217/604aa288f2e5a05d030fa175/html5/thumbnails/7.jpg)
7 PhRMA KOL Series. 11 July 2008
![Page 8: Adaptive Group Sequential Trials with Population ... · Acknowledgements • Joint research with Ping Gao (The Medicines Company) and Jim Ware (Harvard University) • We thank Aniruddha](https://reader035.fdocuments.us/reader035/viewer/2022071217/604aa288f2e5a05d030fa175/html5/thumbnails/8.jpg)
8 PhRMA KOL Series. 11 July 2008
![Page 9: Adaptive Group Sequential Trials with Population ... · Acknowledgements • Joint research with Ping Gao (The Medicines Company) and Jim Ware (Harvard University) • We thank Aniruddha](https://reader035.fdocuments.us/reader035/viewer/2022071217/604aa288f2e5a05d030fa175/html5/thumbnails/9.jpg)
9 PhRMA KOL Series. 11 July 2008
![Page 10: Adaptive Group Sequential Trials with Population ... · Acknowledgements • Joint research with Ping Gao (The Medicines Company) and Jim Ware (Harvard University) • We thank Aniruddha](https://reader035.fdocuments.us/reader035/viewer/2022071217/604aa288f2e5a05d030fa175/html5/thumbnails/10.jpg)
Can we do better?
• 8000 patients is a very large fixed commitment
• We can do better with a group sequential design havingearly stopping boundaries for efficacy or futility
• Such a design would have a larger up-front commitmentbut lower expected sample size
10 PhRMA KOL Series. 11 July 2008
![Page 11: Adaptive Group Sequential Trials with Population ... · Acknowledgements • Joint research with Ping Gao (The Medicines Company) and Jim Ware (Harvard University) • We thank Aniruddha](https://reader035.fdocuments.us/reader035/viewer/2022071217/604aa288f2e5a05d030fa175/html5/thumbnails/11.jpg)
2. Group Sequential Design (GSD)
• Enroll up to Nmax = 8750 subjects
• First Interim Analysis (IA1): at 50% information
– Stop for efficacy if Z ≤ −2.9626 (OBF bdry)
– Stop for futility if 1% change in event rate in the wrongdirection
• Second Interim Analysis (IA2): at 70% information
– Stop for efficacy if Z ≤ −2.4623 (OBF bdry)
– Stop for futility if conditional power ≤ 20%
11 PhRMA KOL Series. 11 July 2008
![Page 12: Adaptive Group Sequential Trials with Population ... · Acknowledgements • Joint research with Ping Gao (The Medicines Company) and Jim Ware (Harvard University) • We thank Aniruddha](https://reader035.fdocuments.us/reader035/viewer/2022071217/604aa288f2e5a05d030fa175/html5/thumbnails/12.jpg)
12 PhRMA KOL Series. 11 July 2008
![Page 13: Adaptive Group Sequential Trials with Population ... · Acknowledgements • Joint research with Ping Gao (The Medicines Company) and Jim Ware (Harvard University) • We thank Aniruddha](https://reader035.fdocuments.us/reader035/viewer/2022071217/604aa288f2e5a05d030fa175/html5/thumbnails/13.jpg)
13 PhRMA KOL Series. 11 July 2008
![Page 14: Adaptive Group Sequential Trials with Population ... · Acknowledgements • Joint research with Ping Gao (The Medicines Company) and Jim Ware (Harvard University) • We thank Aniruddha](https://reader035.fdocuments.us/reader035/viewer/2022071217/604aa288f2e5a05d030fa175/html5/thumbnails/14.jpg)
14 PhRMA KOL Series. 11 July 2008
![Page 15: Adaptive Group Sequential Trials with Population ... · Acknowledgements • Joint research with Ping Gao (The Medicines Company) and Jim Ware (Harvard University) • We thank Aniruddha](https://reader035.fdocuments.us/reader035/viewer/2022071217/604aa288f2e5a05d030fa175/html5/thumbnails/15.jpg)
15 PhRMA KOL Series. 11 July 2008
![Page 16: Adaptive Group Sequential Trials with Population ... · Acknowledgements • Joint research with Ping Gao (The Medicines Company) and Jim Ware (Harvard University) • We thank Aniruddha](https://reader035.fdocuments.us/reader035/viewer/2022071217/604aa288f2e5a05d030fa175/html5/thumbnails/16.jpg)
3. GSD with Adaptive Sample SizeIncrease at IA2
• Suppose neither efficacy nor futility boundaries are crossedat IA2
• Is conditional power less than 80%?
• If so, increase sample size such that conditional powerequals 80%
• Subject to upper limit of 15,000 subjects
16 PhRMA KOL Series. 11 July 2008
![Page 17: Adaptive Group Sequential Trials with Population ... · Acknowledgements • Joint research with Ping Gao (The Medicines Company) and Jim Ware (Harvard University) • We thank Aniruddha](https://reader035.fdocuments.us/reader035/viewer/2022071217/604aa288f2e5a05d030fa175/html5/thumbnails/17.jpg)
Preservation of Type-1 Error afteran Adaptive Design Change
• Unblinded sample size increase can inflate α
• To control this you must preserve the conditional type-1error at the time of the adaptive change of design
• If you pre-specify that you will do this anytime you makean adaptive change, the overall type-1 error will still be α
(Muller and Schafer, 2001)
• Allowable adaptive changes include: sample size change,change in number of future looks, change in futurestopping boundaries, change in eligibility criteria
17 PhRMA KOL Series. 11 July 2008
![Page 18: Adaptive Group Sequential Trials with Population ... · Acknowledgements • Joint research with Ping Gao (The Medicines Company) and Jim Ware (Harvard University) • We thank Aniruddha](https://reader035.fdocuments.us/reader035/viewer/2022071217/604aa288f2e5a05d030fa175/html5/thumbnails/18.jpg)
18 PhRMA KOL Series. 11 July 2008
![Page 19: Adaptive Group Sequential Trials with Population ... · Acknowledgements • Joint research with Ping Gao (The Medicines Company) and Jim Ware (Harvard University) • We thank Aniruddha](https://reader035.fdocuments.us/reader035/viewer/2022071217/604aa288f2e5a05d030fa175/html5/thumbnails/19.jpg)
19 PhRMA KOL Series. 11 July 2008
![Page 20: Adaptive Group Sequential Trials with Population ... · Acknowledgements • Joint research with Ping Gao (The Medicines Company) and Jim Ware (Harvard University) • We thank Aniruddha](https://reader035.fdocuments.us/reader035/viewer/2022071217/604aa288f2e5a05d030fa175/html5/thumbnails/20.jpg)
20 PhRMA KOL Series. 11 July 2008
![Page 21: Adaptive Group Sequential Trials with Population ... · Acknowledgements • Joint research with Ping Gao (The Medicines Company) and Jim Ware (Harvard University) • We thank Aniruddha](https://reader035.fdocuments.us/reader035/viewer/2022071217/604aa288f2e5a05d030fa175/html5/thumbnails/21.jpg)
21 PhRMA KOL Series. 11 July 2008
![Page 22: Adaptive Group Sequential Trials with Population ... · Acknowledgements • Joint research with Ping Gao (The Medicines Company) and Jim Ware (Harvard University) • We thank Aniruddha](https://reader035.fdocuments.us/reader035/viewer/2022071217/604aa288f2e5a05d030fa175/html5/thumbnails/22.jpg)
4. GSD with Sample Size Increaseand Population Enrichment
• There is scientific evidence that some subgroups mightbenefit more than others from the new drug
• Sponsor wishes to take advantage of this possibility
• Design must allow drilling down into subgroups if overallbenefit cannot be established
22 PhRMA KOL Series. 11 July 2008
![Page 23: Adaptive Group Sequential Trials with Population ... · Acknowledgements • Joint research with Ping Gao (The Medicines Company) and Jim Ware (Harvard University) • We thank Aniruddha](https://reader035.fdocuments.us/reader035/viewer/2022071217/604aa288f2e5a05d030fa175/html5/thumbnails/23.jpg)
Qualifications for Subgroups
• Likely to influence the event rates
• More likely to influence experimental group than controlgroup
• Easily identifiable in a clinical setting
• Consistently collected in the trial
23 PhRMA KOL Series. 11 July 2008
![Page 24: Adaptive Group Sequential Trials with Population ... · Acknowledgements • Joint research with Ping Gao (The Medicines Company) and Jim Ware (Harvard University) • We thank Aniruddha](https://reader035.fdocuments.us/reader035/viewer/2022071217/604aa288f2e5a05d030fa175/html5/thumbnails/24.jpg)
Prior Selection of Subgroup
1. Using clinical judgement, create four non-overlappingpartitions of the overall population
Non-Overlapping Partition % of Population
high-risk and clopidogrel-naive 30%
high-risk and pre-treated with clopidogrel 30%
low-risk and clopiddogrel-naive 20%
low-risk and pre-treated 20%
high risk patients are those with diabetes and a troponin + marker
24 PhRMA KOL Series. 11 July 2008
![Page 25: Adaptive Group Sequential Trials with Population ... · Acknowledgements • Joint research with Ping Gao (The Medicines Company) and Jim Ware (Harvard University) • We thank Aniruddha](https://reader035.fdocuments.us/reader035/viewer/2022071217/604aa288f2e5a05d030fa175/html5/thumbnails/25.jpg)
2. Form three nested subgroups from these fournon-overlapping partitions:
• G0 = full population (100%)
• G1 = high-risk subgroup (60% of G0)
• G2 = high-risk, clopidogrel-naive subgroup (50% of G1)
3. The goal is to try and win with G0; if not, then with G1; ifnot, then with G2
25 PhRMA KOL Series. 11 July 2008
![Page 26: Adaptive Group Sequential Trials with Population ... · Acknowledgements • Joint research with Ping Gao (The Medicines Company) and Jim Ware (Harvard University) • We thank Aniruddha](https://reader035.fdocuments.us/reader035/viewer/2022071217/604aa288f2e5a05d030fa175/html5/thumbnails/26.jpg)
The Enrichment Strategy
Suppose the group sequential boundary was not crossed atIA1 or IA2. Then:
• If CP ≥ 80% at IA2, carry on with no change
• If CP < 80% at IA2:
– Try for 80% with G0 and sample size increase
– If unable, try for 80% CP by enriching with only G1
patients and sample size increase
– If unable, try for 80% CP by enriching with only G2
patients and sample size increase
• Terminate for futility if CP < 20% despite enrichment andsample size increase
26 PhRMA KOL Series. 11 July 2008
![Page 27: Adaptive Group Sequential Trials with Population ... · Acknowledgements • Joint research with Ping Gao (The Medicines Company) and Jim Ware (Harvard University) • We thank Aniruddha](https://reader035.fdocuments.us/reader035/viewer/2022071217/604aa288f2e5a05d030fa175/html5/thumbnails/27.jpg)
Hypothesis Testing Strategy
• For populations {G1, G2, G3}, let {H1, H2, H3} be correspondingnull hypotheses that treatment and control event rates are equal
• Consider three cases:
– Case 1 if population was not enriched, test H0 at final analysis
– Case 2: if population was enriched by G1, test H1 at final analysis
– Case 3: population was enriched by G2, test H2 at final analysis
• The testing strategy must ensure strong control of type-1 error. Thismeans that no matter which hypothesis is tested, the false positiverate must not exceed α
27 PhRMA KOL Series. 11 July 2008
![Page 28: Adaptive Group Sequential Trials with Population ... · Acknowledgements • Joint research with Ping Gao (The Medicines Company) and Jim Ware (Harvard University) • We thank Aniruddha](https://reader035.fdocuments.us/reader035/viewer/2022071217/604aa288f2e5a05d030fa175/html5/thumbnails/28.jpg)
28 PhRMA KOL Series. 11 July 2008
![Page 29: Adaptive Group Sequential Trials with Population ... · Acknowledgements • Joint research with Ping Gao (The Medicines Company) and Jim Ware (Harvard University) • We thank Aniruddha](https://reader035.fdocuments.us/reader035/viewer/2022071217/604aa288f2e5a05d030fa175/html5/thumbnails/29.jpg)
29 PhRMA KOL Series. 11 July 2008
![Page 30: Adaptive Group Sequential Trials with Population ... · Acknowledgements • Joint research with Ping Gao (The Medicines Company) and Jim Ware (Harvard University) • We thank Aniruddha](https://reader035.fdocuments.us/reader035/viewer/2022071217/604aa288f2e5a05d030fa175/html5/thumbnails/30.jpg)
Proof that Test Procedure is Closed
30 PhRMA KOL Series. 11 July 2008
![Page 31: Adaptive Group Sequential Trials with Population ... · Acknowledgements • Joint research with Ping Gao (The Medicines Company) and Jim Ware (Harvard University) • We thank Aniruddha](https://reader035.fdocuments.us/reader035/viewer/2022071217/604aa288f2e5a05d030fa175/html5/thumbnails/31.jpg)
31 PhRMA KOL Series. 11 July 2008
![Page 32: Adaptive Group Sequential Trials with Population ... · Acknowledgements • Joint research with Ping Gao (The Medicines Company) and Jim Ware (Harvard University) • We thank Aniruddha](https://reader035.fdocuments.us/reader035/viewer/2022071217/604aa288f2e5a05d030fa175/html5/thumbnails/32.jpg)
32 PhRMA KOL Series. 11 July 2008
![Page 33: Adaptive Group Sequential Trials with Population ... · Acknowledgements • Joint research with Ping Gao (The Medicines Company) and Jim Ware (Harvard University) • We thank Aniruddha](https://reader035.fdocuments.us/reader035/viewer/2022071217/604aa288f2e5a05d030fa175/html5/thumbnails/33.jpg)
33 PhRMA KOL Series. 11 July 2008
![Page 34: Adaptive Group Sequential Trials with Population ... · Acknowledgements • Joint research with Ping Gao (The Medicines Company) and Jim Ware (Harvard University) • We thank Aniruddha](https://reader035.fdocuments.us/reader035/viewer/2022071217/604aa288f2e5a05d030fa175/html5/thumbnails/34.jpg)
Properties of Testing Strategy
• Closed test!. Guarantees strong control of type-1 error;i.e., chance of false rejection of either H0 or H1 or H2
cannot exceed α
• Boosts power by 2-5% if new treatment benefits allsubgoups homogeneously
• Boosts power by 7-18% if new treatment benefits G1 orG2 differentially relative to G0
34 PhRMA KOL Series. 11 July 2008
![Page 35: Adaptive Group Sequential Trials with Population ... · Acknowledgements • Joint research with Ping Gao (The Medicines Company) and Jim Ware (Harvard University) • We thank Aniruddha](https://reader035.fdocuments.us/reader035/viewer/2022071217/604aa288f2e5a05d030fa175/html5/thumbnails/35.jpg)
Verify by Simulation
35 PhRMA KOL Series. 11 July 2008
![Page 36: Adaptive Group Sequential Trials with Population ... · Acknowledgements • Joint research with Ping Gao (The Medicines Company) and Jim Ware (Harvard University) • We thank Aniruddha](https://reader035.fdocuments.us/reader035/viewer/2022071217/604aa288f2e5a05d030fa175/html5/thumbnails/36.jpg)
36 PhRMA KOL Series. 11 July 2008
![Page 37: Adaptive Group Sequential Trials with Population ... · Acknowledgements • Joint research with Ping Gao (The Medicines Company) and Jim Ware (Harvard University) • We thank Aniruddha](https://reader035.fdocuments.us/reader035/viewer/2022071217/604aa288f2e5a05d030fa175/html5/thumbnails/37.jpg)
37 PhRMA KOL Series. 11 July 2008
![Page 38: Adaptive Group Sequential Trials with Population ... · Acknowledgements • Joint research with Ping Gao (The Medicines Company) and Jim Ware (Harvard University) • We thank Aniruddha](https://reader035.fdocuments.us/reader035/viewer/2022071217/604aa288f2e5a05d030fa175/html5/thumbnails/38.jpg)
38 PhRMA KOL Series. 11 July 2008
![Page 39: Adaptive Group Sequential Trials with Population ... · Acknowledgements • Joint research with Ping Gao (The Medicines Company) and Jim Ware (Harvard University) • We thank Aniruddha](https://reader035.fdocuments.us/reader035/viewer/2022071217/604aa288f2e5a05d030fa175/html5/thumbnails/39.jpg)
39 PhRMA KOL Series. 11 July 2008
![Page 40: Adaptive Group Sequential Trials with Population ... · Acknowledgements • Joint research with Ping Gao (The Medicines Company) and Jim Ware (Harvard University) • We thank Aniruddha](https://reader035.fdocuments.us/reader035/viewer/2022071217/604aa288f2e5a05d030fa175/html5/thumbnails/40.jpg)
40 PhRMA KOL Series. 11 July 2008
![Page 41: Adaptive Group Sequential Trials with Population ... · Acknowledgements • Joint research with Ping Gao (The Medicines Company) and Jim Ware (Harvard University) • We thank Aniruddha](https://reader035.fdocuments.us/reader035/viewer/2022071217/604aa288f2e5a05d030fa175/html5/thumbnails/41.jpg)
41 PhRMA KOL Series. 11 July 2008
![Page 42: Adaptive Group Sequential Trials with Population ... · Acknowledgements • Joint research with Ping Gao (The Medicines Company) and Jim Ware (Harvard University) • We thank Aniruddha](https://reader035.fdocuments.us/reader035/viewer/2022071217/604aa288f2e5a05d030fa175/html5/thumbnails/42.jpg)
Concluding Remarks• Why not run two independent trials; exploratory and confirmatory?
– exploratory trial won’t be adequately powered for subgroups
– if it fails, there won’t be any support for the confirmatory trial
– current design contains an insurance policy for handling subgroups
• Pre-specification of subgroups not a statistical requirement. Butimportant safeguard against chasing random noise
• Flexiblity to change sample size and enrich population greatlyincreases complexity of trial management
– Logistics of drug are complicated if sample size is increased at IA2
– Recruitment might slow down if population is enriched at IA2
• Sponsor cannot be involved in interim decision making. Must appointindependent committee with clinical and statistical expertise
• Simulate the design thoroughly before committing to it
42 PhRMA KOL Series. 11 July 2008