On Surrogate Endpoints in HIV Vaccine Efficacy Trials

FDA/Industry Statistics Workshop, Sept 18-19, 2003 “Statistics: From Theory to Regulatory Acceptance” Steven Self, Peter Gilbert, Michael Hudgens FHCRC/UW

Outline

1.

2.

HIV Vaccine Trials: Current Status Clinical Endpoints in Vaccine Trials 3.

Endpoints in HIV Vaccine Trials 4.

A Simulation Approach 1. Goal 2. Approach 3. Example 5.

Conclusions/Discussion

HIV Vaccine Trials: Current Activity

• Phase I and II trials (ongoing) • 27 trials involving ~1,800 participants • 16 different vaccine candidates • 10 sponsors • Phase III trials • 1 completed • 1 to be completed in Q4 ’03 • 1 planned to start in ’04 • HVTN trials (Ph I, II) starting by Q3 ’04 • 9 different vaccine candidates • 1,453 participants

HIV Vaccine Trials: Current Results

• Immune Responses • Measurable cellular response in ~50% • No broadly neutralizing Ab in sera, mucosa • Non-human Primate Trials • Amelioration of disease course but no protection from infection upon challenge • Body of NHP literature difficult to assess • Efficacy • No overall efficacy in 1 completed efficacy trial

HIV Vaccine Trials: Summary

• Immune correlate of protection unknown • Many candidate vaccines but full range of desired immune responses poorly covered • Multiple efficacy trials will be required* • Plan for long-term, iterative development program* * Klausner et. al. (2003) Science

Classical Measure of Vaccine Efficacy

VE = % reduction in population incidence or morbidity/mortality rate

Classical Endpoint for Vaccine Efficacy

• Clinically significant morbidity and mortality • Pathogen specificity • Standard of care • For treatable infections: – Prevent/delay constellation of signs/symptoms sufficient to trigger treatment initiation (save cost/toxicity assoc with treatment) – Interact w/ treatment to improve risk/benefit profile of vaccine/tmt vs tmt alone

Measures of Vaccine Efficacy*

VE = % reduction in population incidence or morbidity/mortality rate VE S = % reduction in population infection rate VE P = % reduction in rate of morbidity/mortality VE I = % reduction in rate of 2 o transmission * Halloran, Longini, Struchiner

VE: % Reduction in Transition Intensities

VE Uninfected/ Seronegative VE S Infected/ Seropositive VE I VE P 2 o Transmission Morbidity/ Mortality

Endpoints in HIV Vaccine Efficacy Trials

– “Infection” Endpoint (A biomarker-based surrogate) • Operationally: presence of Ab and detectable HIV RNA • Aligned with one primary objective of HIV vaccine • Acceptable by all • However captures only one aspect of potential vaccine effects on clinical outcomes

Endpoints in HIV Vaccine Efficacy Trials

“Post-infection” Endpoints: Some Issues • Long-term FU required for morbidity/mortality endpoints esp with ARV treatment • Complicated dynamical process likely dominated by treatment effects • Uncertainty of optimal treatment initiation triggers • Variability in treatment initiation • Analytics – Key biomarker trajectories “dependently censored” by treatment initiation – Conditional vs unconditional analyses – Combination of analyses

 

Post-Infection Endpoints: Current Approach

 Provide treatment within trial – standardized treatment initiation guidelines (e.g. DHHS, UNAIDS) – standardized treatment monitoring/management Develop complementary array of endpoints to cover key aspects of post-infection outcomes – Early Endpoints - pre-ART – Mid-term Endpoints - peri-ART – Long-term Endpoints - post-ART “Reasonable conservatism” for interpretation of vaccine effects on surrogates

HIV Vaccine Efficacy Trial Endpoints

Short-term Endpoints: - Pre-ART VL Long-term Endpoints: - vaccine/tmt effects - CD4 - Morbidity/Mortality Uninfected/ Seronegative Infected/ Seropositive Treatment Initiation Morbidity/ Mortality Infection Mid-term Endpoints: - Composite (VL, tmt init) - Biomarker trajectories (VL, CD4)

A Composite Endpoint

• Definition: First event of ART initiation or virologic failure (VL > X cps/ml) • Composite endpoint directly tied to clinical events • virologic failure places a subject at risk for progression/transmission • starting ART exposes a subject to drug toxicities, resistance, loss of future drug options • Assess with standard statistical methods (Kaplan-Meier, Cox regression)

A Composite Endpoint

• Surrogate vaccine efficacy parameter: VEVL C (T;X) = percent reduction (vaccine vs. placebo) in the risk of the composite endpoint by T months post infection diagnosis • X calibrates the

magnitude

of virologic control (e.g., X = 1,500 copies/ml) • T calibrates the

durability

(e.g., use T  18 months) of virologic control

Example Analysis of VEVL

C

(18;X)

A Numerical Study*: Goal

 Provide an approach to facilitate the discussion of

how

to use surrogate endpoints – specific to trial design – specific to particular surrogate endpoints – accommodate statistical uncertainties – accommodate model uncertainties with desired degree of conservatism * Gilbert et al (2003) JID

A Numerical Study: Approach

 adopt empirically-based joint model of biomarker process and clinical outcomes as “true” prediction model*  modify model to incorporate degrees of “reasonable conservatism” – proportion vaccine effect explained** (attenuate log RR relating surrogate to clinical outcome by f percent ) – selection bias for conditional analyses*** (attenuate observed vaccine effect on surrogate outcome)  RCT simulation to identify minimum observed effects on specific surrogate endpoints that would generate 95% prediction intervals for VE parameters exceeding 40%, say * Albert et al (1998) Stat in Med ** Freedman et al (1992) Stat in Med *** Hudgens et al (2003) Stat in Med; Gilbert et al (2003) Biometrics

Numerical Study: An Example

• Question: • What inference on VEVL C (18;X) “reasonably” predicts a clinically significant VE P ?

• Numerical study based on the following predictions: • from the MACS * : Predicted(VE P ) = VEVL C (18;X) for X  5,000-10,000 cps/ml * Albert et al (1998) Stat in Med

Hypothetical Efficacy Trial

The numerical study is based on the following hypothetical trial:

Prediction of VE

P

• A lower 95% confidence bound for VEVL C (18;X) > 50% predicts VE P > 40% with f = 0.375

Summary/Conclusions

 Use of surrogate endpoints in HIV vaccine efficacy trials is question of

how

not

whether

 A framework is proposed to help interpret observed effects on surrogate endpoints that is – specific to particular trial designs/endpoints – captures relevant aspects of magnitude and durability of effect on surrogates – uses available empirical information relating biomarkers to clinical outcomes – is tunable with respect to degree of conservatism w/r/t use of empirical information – flexible to evolve with development program

Summary/Conclusions

 HIV vaccines showing strong and durable effects on post-infection endpoints should be licensed – use of standardized ART guidelines important – use simulation studies to assist in building agreement about defining “sufficiently strong” and “sufficiently durable” – design trials to detect significant levels of either VE S VEVL C (T;X) – use supporting data on other endpoints or  Long-term follow-up needed – for assessing VE and VE P directly – better understanding of surrogate endpoints