Transcript Surrogate Endpoints and Non

Surrogate Endpoints and
Non-randomized Trials
Roseann White
Humble Biostatistician
© Guidant 2005
Type of non-randomized trials
1. Diagnostic/Natural History trial
2. Single arm trial that shows superiority/noninferiority in clinical endpoint
Example: Safety concerns for placebo or current
practice
© Guidant 2003
Motivation
• Chiron Corporation developed a method to
measure the amount of HIV-1 virus in the blood
• To obtain approval from FDA for the device,
Chiron needed to demonstrate clinical utility
• Investigators also saw the potential for Viral
Load to be a surrogate for HIV-1 disease
progression
© Guidant 2003
Prentice Criteria for a surrogate endpoint
Re-statement of Prentice Criteria for a surrogate
endpoint
1. Baseline measurements are predictive of
outcome
2. Changes in the measurement over time is
predictive of outcome
3. Changes in the measurement to external
forces (therapy) is predictive of outcome
© Guidant 2003
Non-randomized trial – a prospective
analysis of a retrospective cohort
• Description of Cohorts
 180 seropositive men studied for more than 10 years from
the Pittsburgh portion of Multicenter AIDS Cohort Study
(MACS)– Mellors, J.W., et. al. Science, 272
 1604 men infected with HIV-1 from four university-based
clinical centers participating in MACS – Mellors, J.W. et. al.
Annals of Internal Medicine, 126
 ~250 patients from New York Blood Center as part of a PMA
submission for the bDNA diagnostic
• Analysis
 Logistic regression using baseline values to predict survival
 Cox proportional hazards model with HIV-1 viral load as a
time dependant covariate
 Treatment effect?
© Guidant 2003
Predictive in stratified populations
Reprinted from Plasma Viral Load and CD4+ Lymphocytes as Prognostic Markers of HIV-1 Infection
John W. Mellors, et.al.
Annals 1997 126: 946-954.
© Guidant 2003
Rest of the story
• Viral load was used along with CD4 counts as
evidence of efficacy for accelerated approval of
the protease inhibitors
• Many efficacy trials measured viral load along
with CD4 count
• FDA Guidance to the industry (2002)
recommended the use of viral load for efficacy
in accelerated approvals
• “The Evaluation of Surrogate Endpoints in
Practice: Experience in HIV”* by Michael
Hughes
Uses several different methods to “validate” HIV viral
load and CD4 counts as surrogate endpoints
*Chapter 17 in The Evaluation of Surrogate Endpoints edited by T. Burzykowski, et. al. Springer, 2005
© Guidant 2003
Types of non-randomized trials
1. Quantitative Diagnostic
2. Single arm trial to show superiority or noninferiority in clinical endpoint
© Guidant 2003
Motivation
• Randomization is difficult
Cost prohibitive
Concerns for the safety of the patient
Limited population available for recruitment
• Potential surrogate endpoints available
© Guidant 2003
Design Considerations
• Evaluate the risk associated with the surrogate
for the product in question
If it’s a second generation product, will the surrogate
reflect the improvements in the product AND
Will the surrogate reflect potential problems?
Example: Using angiographic binary restenosis
as a surrogate at 6 months for drug eluting stent
whose drug has not completely eluted at six
months
© Guidant 2003
© Guidant 2003
Design Considerations (con’t)
• Choice of comparison – Historical Control
versus Objective Performance Criteria
Historical Control provides more of an opportunity to
demonstrate that the current trial population is similar
to the historical population in which the surrogate was
based.
When using an objective performance criteria,
develop a detailed method in which you will “validate”
the current population is reflective of the population
that surrogate was based.
 Subgroup analysis where the surrogate shows
difference, e.g. diabetics versus non-diabetics
© Guidant 2003
Design Considerations (con’t)
• Consider a co-primary clinical endpoint where
you demonstrate a trend in the same direction
as your surrogate
Less stringent alpha for superiority
Wider delta for non-inferiority
© Guidant 2003
Conclusion
• Validation of Surrogates endpoint using nonrandomized trials is challenging
More work needs to be done to develop techniques
that do not necessarily require a very effective
treatment
• Use of surrogates in single arm trials requires:
Careful consideration as to whether the surrogate will
reflect the true performance of the product
Use of a historical control or a detail plan of how to
assure the current population reflect the population on
which the surrogate was based.
© Guidant 2003