Selected Issues in Oncology Trial Design
Grant Williams, M.D.
DODP, CDER, FDA
Outline of Presentation
• Challenges in oncology trial design
• Non-inferiority trials in oncology
• Time to Progression (TTP)– The TTP question in a regulatory
framework– TTP-like endpoints– Pros and Cons of TTP
Blinding Oncology Trials
• Problems– Unmasking of blind by side-effects– Need to adjust doses
• Opportunities:– Oral drugs with fewer side-effects
Use of Placebos in Oncology Trials
• Problem:– Placebo-alone control usually not feasible in advanced cancer
• Potential use of placebos– Settings: “prevention”, adjuvant, or early disease – Add-on designs (Drug A plus Drug B versus Drug A plus placebo)– May allow continuation of drug and placebo after failure of Drug A (e.g.,
bisphosphonates)
– practical orPlacebo-alone treatment is uIn advanced settings it Often may not be practical and/or ethical for cancer patientuse a placebo-alone treatment arm
No Blind or Placebo, Consequences:
• Limits choice of clinical-benefit endpoints• Limits trial designs:
– Control must be an active drug• Superiority design (preferred)
– requires new drug to be more effective– or use add-on design
• Non-inferiority design – requires large trials– Quality of historical data on active control limits NI design
• Result: It is difficult to approve drugs that are similar but less toxic
•
The Combination Drug Problem
• Drug approvals, drug labels, and drug marketing focus on effects from individual drugs.
• Many oncology regimens are combinations where the efficacy contribution of individual drugs may not be precisely defined.
Non-inferiority
• Superiority: – Determined with statistical confidence
• Equivalence: – Has no statistical meaning
• Non-inferiority– Definition: no worse by a specified margin– Proving non-inferiority does not necessarily prove
efficacy (next slides)
• Not statistically different:– has no meaning without details
Non-Equivalent Words
Regulatory Goal of NI Trial
• Demonstrate Drug B is effective– By referring to historical Drug A effect – By randomizing A versus B– By prospectively identifying a margin that includes
an acceptable fraction of Drug A efficacy– By proving that Drug B is no worse than Drug A
by that margin– By determining that the “constancy assumption” is
valid
Critical Assumption of NI Trial• “Constancy assumption”: The historically observed drug effect of the active control drug also
exists in the current NI trial and population
Potential differences– Population– Supportive care– Additional available therapies– Study design (observation frequency, etc.)
• Violating this assumption could lead to approval of “toxic placebo”
Sloppiness / Poor Quality Data
• Sloppiness obscures differences– Superiority trial designs: obscures efficacy– For NI trials: could lead to false efficacy
claim
Determining the Margin from Historical Cancer Drug Effects
• Step 1: Estimate effect size and confidence intervals of active control drug– Needed (Ideally):
• Multiple historical trials showing effect• Consistent large drug effect
– Oncology reality:• Small historical drug effect in one or two trials• Leads to very small margin• Leads to very large NI studies
• Drug combinations even more complicated
The Effectiveness Standard
• 1962 amendments: “claimed effect”
• Subsequent rulings: “Clinical meaning”
• “Clinical meaning” in oncology– 1970s: minimal activity– 1985 : survival or effect on “QOL”
(symptoms or function)– 1990s-2000s: use of some surrogates
Surrogates in Drug Approval
• Surrogate endpoint definition*:– Substitute for a clinically meaningful endpoint that
measures directly how a patient feels, functions or survives.
– Changes are expected to reflect changes in a clinically meaningful endpoint.
*Temple RJ, Clinical Measurement in Drug Evaluation. Nimmo and Tucker. John
Wiley & Sons Ltd, 1995.
Established Surrogates Supporting Regular Approval
• Blood pressure
• Blood sugar
• Blood cholesterol
Oncology Surrogates
• AA surrogate: reasonably likely
• “Validated” Surrogates– Few and far between
• Surrogates for CB supporting regular approval – Judged by FDA and experts in the field to
be reliable indicators of CB
The Ideal:Prentice’s Sufficient Conditions
The surrogate endpoint must be correlated with the clinical outcome
The surrogate endpoint must fully capture the net effect of treatment on the clinical outcome
• Meta-analyses of clinical trials data
• Comprehensive understanding of:
– The causal pathways of the disease
process
– The intervention’s intended and unintended mechanisms of action
Surrogate Endpoint Validation*
From Tom Fleming, Ph.D.
Is TTP a Clinical Benefit Measure?
• Does TTP have clinical meaning?– Cancer growth leads to suffering and death– Delaying cancer growth is good
Is TTP a Clinical Benefit Measure?
• The critical issues:– Can you measure TTP reliably?– How much progression delay is worth how
much toxicity?– What is the relative meaning of a TTP
benefit to other benefits such as survival?
Acceptance of Clinical Benefit Based on Tumor Effects (RR or TTP), Examples
• Hormonal drugs for metastatic breast cancer – Primary endpoint: response rate (RR)– Secondary endpoints: TTP and Survival– Regulatory acceptance
• long experience with tamoxifen• no proven survival benefit for drugs in this setting• low drug toxicity
TTP and Cytotoxic Drugs for First-line Treatment of Metastatic Breast Cancer
(ODAC, 1999) • Determination:
– Not for full approval– Yes for Accelerated Approval
• Acceptable effect size not stated
• Deliberations:– Possible survival benefit from chemotherapy?– Only small TTP benefits with current drugs– Poor correlation with survival?– Unreliable TTP measurements?– Reliability requires frequent measurement?
What is TTP?
• Complex: Check the protocol,case report form, & statistical analysis plan!
• Time from randomization to first evidence of progression. RECIST:– 20% increase in sum of marker lesions – New lesions– Unequivocal increase in non-marker
lesions
Which Events Count?Time to Tumor Progression (TTP)
• TTP event = progression– Measures tumor effects– Deaths are censored at last visit
• Non-informative censoring assumption
Which Events Count?Progression Free Survival (PFS)
• PFS events = progression + death
• Better surrogate for CB?
• Poor follow-up causes prolongation of progression time – Need careful follow-up– Need analysis rules for deaths after loss to
follow-up?
Which Events Count?Time to Treatment Failure (TTF)
• TTF events = death, progression, toxicity, etc. – Does not isolate efficacy– Not adequate as the primary regulatory
endpoint• Drug must be safe and effective• Demonstrating less toxicity is not adequate
• Measured in all patients
• Measures cytostatic activity
• Oncologists usually change therapy at progression
• Assessed before crossover
• Requires smaller studies
• Face validity?
TTP: Advantages
• Doesn’t always “correlate” with survival(vs. inadequate data to assess relationship?)
• Indirect measure of patient benefit• Unclear meaning of small difference• Reliability in unblinded setting?• Unknown reliability of small TTP difference
with usual trial monitoring• Expensive to measure, difficult to verify
TTP: Problems
• Data are usually inadequate to assess– Many different cancer settings – Large survival benefits are rare– Cited “lack of correlation” usually invalid
• Greater statistical power for TTP than survival• Studies cannot rule out survival effect• Significant TTP analysis and non-significant
survival analysis would be expected
• Crossover may obscure survival effect
The Relationship between TTP and Survival
Survival versus TTP
Survival TTP-100% Accurate Event -Less Accurate-100% Accurate Time -Less Accurate-Assessed Daily -Assessed every 2-6 mo-Importance Unquestioned -Uncertain-Both Safety & Efficacy -Only Efficacy-Takes Longer -Faster-Might be Obscured by Secondary Rx
-Not Obscured
Problem #2:TTP is Indirect measure of benefit
TTP would be more persuasive benefit measure when:– When symptoms frequently occur at or
soon after progression time – When TTP increment is large– When treatment toxicity is low – When benefit of available drugs is less
Incorporate symptoms into TTP: “time to symptomatic progression”
• Represents full clinical benefit
• Potential bias in symptom data
• Symptom data needed beyond tumor progression time
• Confounding effects of additional treatments
Visit 1 Visit 2Randomization
= Date of Death or actual tumor progression
Survival Event Date
Visit 1 Visit 2Randomization
TTP Event Date
Survival Analysis
TTP Analysis
Determining Event Dates
Verifying TTP: Difficulties for Sponsors and for FDA
• What if:– Not all lesions are followed?– Measurements occur at non-standard times?– Some measurements are missing from a visit?
• How do you:– Assure equal screening for new lesions?– Evaluate bias from lack of blinding?– Verify progression of “evaluable disease?”
Endpoint for Future Research: Single Time Progression Analysis
• Specify analysis point (e.g., 6 months)
• Requires only two data collections:– Document baseline data– Document either:
• Progression before time point• Stable disease at time point
Single Time Progression Analysis
• Advantages:– Less data collection– Minimize time-related bias
• Research questions:– Potential loss of statistical power– Uncertainty of predicting optimal ST– Potential for losing information in TTP curve
• Different early effects • Benefit in curve plateau
TTP Issues for Consideration
• TTP as a drug approval endpoint?– Factors determining acceptable settings?– Amount of evidence needed for TTP claim
(# trials, p value, effect size)
TTP Issues for Consideration
• Can we improve our approach?– Research on novel progression endpoints?– Research on validating TTP?– Standard approach to endpoint definition
and censoring methods?– Blinding investigators and patients?– Blinded review?– Including symptoms in endpoint?