Transcript Endpoints

Lessons Learned From Recent
Safety Meta-Analyses
Mark Levenson, Ph.D.
Quantitative Safety and Pharmacoepidemiology Group
Office of Biostatistics
Center for Drug Evaluation and Research, FDA
v. 1 Oct. 2009
Disclaimer
The views expressed in this presentation
represent the opinions of the author, and do not
necessarily represent the views of the United
States Food and Drug Administration
Focus Today
•
•
•
•
•
•
Safety
Regulatory setting
Pre- and post-market
Clinical trials
Access to relevant data
Not: locating, accessing quality of, and
extracting data from studies
Outline
1.
2.
3.
4.
Study inclusion criteria
Endpoints
Methodology
Examples
– Suicidality meta-analyses
– Aprotinin
Meta-Analysis Steps
1. Define research goals
2. Research/understand relevant trials
3. Define analysis plan (Prespecify!)
–
–
–
–
–
–
Research questions
Study inclusion criteria
Analysis set and subgroups
Endpoints
Primary methods
Sensitivity methods
4. Make request to sponsor(s) / obtain data
5. Implement analysis plan
6. Report and interpret findings
Study Inclusion Criteria
Study Inclusion Criteria
• Valid comparison groups
• Similarity in
– Design
– Interventions
– Study population
– Studied indication
• Data availability
8
9
Duration: Example
• Assume event of interest takes some
time to develop (increasing hazard)
Survival
1
2
Years
Duration: Example (Cont.)
• Scenario 1:
– 10 trials
– 100 patients per trial
– 3 month duration
• 1,000 patients
• 250 person-years
• Scenario 2:
– 1 trial
– 100 patients per trial
– 2 year duration
• 100 patients
• 200 person-years
Endpoints
Endpoints
• First choice: prospectively collect and
adjudicate endpoints
• Second choice: use common post-hoc
adjudication procedure across trials
• Last choice: make do with existing
information from trials
Endpoints: Follow-Up Time
Time
Randomization
End of
Treatment
Event
End of Follow-up
Endpoints: Follow-Up Time
• Follow-up should be long enough to
capture event of interest
• Use common cut-off point across trials
when possible
• Need to balance lasting effect of drug
versus confounding with post-trial
therapy and dilution of drug effect
(see: NEJM Vioxx APPROVE
discussion, 2006)
Data Availability
Patient-level data allows more thorough
analysis
• Time-to-event
• Subgroup
• Treatment duration effects
• Internal validation
Methodology
Methodology
• Need to use appropriate methods for
problem
• Need to justify method
• Need to perform sensitivity analyses
along several fronts
Methodology Considerations
•
•
•
•
•
Number of trials
Number of subjects per trial
Rates of events
Zero-event trials
Heterogeneity of effect
Methods
•
•
•
•
•
Inverse variance weighting
Mantel-Haenszel odds ratio or risk ratio
Exact method for odds ratio
Mantel-Haenszel risk difference
Bayesian methods
– Encompass fixed- and random-effect
models and hierarchical models
Sensitivity Analysis
• Consequences of low event rate
• Consequences of zero-event trials
• Consequences of heterogeneity of trials
– Random effects models
– Trials with large influence
Sensitivity Strategy
• Primary method: Exact method for OR
• Sensitivity methods:
– Mantel-Haenszel RD
– GLMM, qualitatively compare results with
primary method
Suicidality Meta-Analyses
Suicidality Meta-Analyses
• Concern that drugs may be associated
with suicidality
• Requested all patient-level data from all
placebo-controlled trials from sponsors
• Patients retrospectively classified into
suicidality outcomes by blinded experts
Suicidal Behavior and Ideation
Psychiatric Indications
Odds Ratio
Trials
OR (95% CI)
bupropion
citalopram
duloxetine
escitalopram
fluoxetine
fluvoxamine
mirtazapine
nefazodone
paroxetine
sertraline
venlafaxine
1.41
2.21
0.81
1.57
0.65
1.37
1.04
0.61
0.96
0.63
0.68
OVERALL
(0.40,
(0.79,
(0.43,
(0.38,
(0.44,
(0.69,
(0.34,
(0.27,
(0.59,
(0.32,
(0.40,
5.58)
7.63)
1.56)
7.88)
0.96)
2.84)
3.35)
1.35)
1.58)
1.21)
1.16)
0.84 (0.69, 1.02)
0.1
0.3
1
Odds Ratio
3.2
10
25
Suicidal Behavior and Ideation
Psychiatric Indications
Odds Ratio
Age Class
OR (95% CI)
2.22 (1.40, 3.60)*
Pediatric Data
18 to 24
1.55 (0.91, 2.70)
25 to 30
1.00 (0.60, 1.69)
31 to 64
0.77 (0.60, 1.00)
65 and Up
0.39 (0.18, 0.78)
Adult Overall
0.84 (0.69, 1.02)
0.1
0.3
1
3.2
10
Odds Ratio
26
* Reanalysis of FDA/Hammad 2004 data
Suicidal Behavior or Ideation
Odds Ratio Estimates
Drug
OR (95% CI) [Sample Sizes]*
Carbamazepine
0.65 (0.08, 4.42) [2/252 3/250]
Divalproex
0.72 (0.29, 1.84) [11/1327 9/992]
Felbamate
ND. (ND., ND.) [0/170 0/170]
Gabapentin
1.57 (0.12, 47.66) [2/2903 1/2029]
Lamotrigine
2.08 (1.03, 4.40) [27/2865 11/2070]
Levetiracetam
2.75 (0.62, 19.36) [8/2554 2/1549]
Oxcarbazepine
1.91 (0.15, 56.33) [2/1342 1/827]
Pregabalin
1.88 (0.41, 13.58) [7/7201 2/3125]
Tiagabine
inf (0.20, inf) [2/835 0/608]
Topiramate
2.53 (1.21, 5.85) [40/7742 8/3971]
Zonisamide
2.52 (0.26, 67.94) [3/672 1/438]
Overall
1.80 (1.24, 2.66) [104/27863 38/16029]
0.1
0.3
1
3.2
10
27
Odds Ratio
*[Treat. Events/Treat. n Plac. Events/Placebo n]
Antiepileptic AC
Paraphrase
• Does committee agree with agency that
findings should apply to all 11 drugs?
Yes: 18, No: 3, Abstain: 0
• Does committee agree with agency that
findings should apply to all approved
antiepileptics?
• Yes: 15, No: 5, Abstain: 1
Aprotinin
The Aprotinin Story
• Aprotinin: used to reduce blood loss and
transfusion in patients undergoing
coronary artery bypass graft surgery
(CABG) with cardiopulmonary bypass
• 2006 NEJM Mangano paper raised
safety concerns
• FDA held 2 Advisory Committee
meetings on the safety of aprotinin
motivated by 3 observational studies
Disparate Findings
• Mangano in-hospital death: no effect relative
to no drug
• Mangano 5-year death: 1.37 HR, pvalue=0.008 relative to no drug
• Sponsor Global CABG RCT Database:
Death 2.9% aprotinin, 2.5% placebo (9/06 AC)
• Meta-analyses Henry et al. Cochran Review
2007: RR=0.90 (0.67, 1.20) relative to control
(no drug). No effects relative to other drugs.
BART
• Blood Conservation Using
Antifibrinolytics in a Randomized Trial
(BART)
• Compared aprotinin to two active drugs
• 30-day death secondary endpoint
Jan. 2007 Second Interim Analysis
• Aprotinin 5.0% vs. 3.9% and 4.3% for
comparator drugs
• DSMB: “did not consider the results of
[Mangano Study] convincing.”
• DSMB: [Systematic reviews] “less
biased than the observational studies”
• Four systematic reviews showed no
death effect
October 2007
• Aprotinin 6.5% vs. 4.2% and 4.3% for
comparator drugs (p-value near nominal
significance)
• DSBM recommends terminating trial
What Happened to the Meta-Analysis?
• Ray Editorial (NEJM)
• Trials not designed to collect follow-up
mortality information
• Limited data on head-to-head
comparisons with other drugs
• Trial and patient heterogeneity (surgical
procedure, patient risk) may hide signal
Conclusions
• Prospectively plan trials for pooled analyses,
e.g. endpoint definition and ascertainment
• Prespecify analysis plan
• Select trials with similar and appropriate
designs
• Consider methodology issues of sparse
events and perform sensitivity analyses
• Understand the limitations of meta-analysis
• Draft FDA Guidance for safety meta-analyses
December 2009
References
• Guidance for Industry Premarketing Risk Assessment
(FDA, 2005)
• Much ado about nothing: a comparison of the
performance of meta-analytic methods with rare
events (Statis. Med., Bradburn et al., 2007)
• The Aprotinin Story – Is BART the final chapter?
(NEJM, Ray, 2007)
• Time-to-Event Analysis for Long-Term Treatments –
The APPROVe trial (NEJM, Lagakos, 2006)
• Understanding the New Drug Safety Standards: The
Emerging Science of Meta-Analysis (The Pink Sheet,
2007)