Analyzing Trials with Active Control Arms

1
Analyzing Trials with Active Control Arms

• Non-Inferiority Analyses
• David Harrington
• Dana-Farber Cancer Institute

2
Disclaimers

• No financial support from sponsor
• Expenses paid by FDA as a member of ODAC
• Will not discuss in detail or evaluate the
  analyses in the application, but will use some
  data for context

3
Trials in Application
Trial Number of Participants Population Postmenopausal women Primary Endpoint
STAR 19,747 High risk of breast cancer Invasive Breast Ca
RUTH 10,101 With or at risk of coronary events Coronary events, inv. Breast Ca
MORE 7,705 With osteoporosis Vertebral fracture
CORE 4,011 With osteoporosis Invasive Breast Ca
4
Main Statistical Issue in Analysis of STAR

• Application contains non-inferiority (NI)
  analysis of Raloxifene vs Tamoxifen in STAR
  trial.
• Primary endpoint is invasive breast cancer
• Raloxifene is test agent
• Tamoxifen is an active control
• Analysis of active control trials uses
  information outside of current trial to infer
  effect of a study drug (raloxifene) vs placebo in
  absence of a direct comparison
• See Rothmann et al. (2003), Temple Ellenberg
  (2000), many others
• What questions should be asked about NI analysis?

5
Tamoxifen vs Placebo NSABP P1 Trial Subset of
7,998 Women ? 50 years old
Favors Tamoxifen
Favors placebo
RR 0.47 (0.33 - 0.66)
Interpretation T reduces the rate of invasive
breast cancer incidence on placebo by 53
(confidence interval 34 to 67)
1.4
1.2
1.0
0.8
0.6
0.4
1.6
Relative Risk for Invasive Br Ca Tamoxifen /
Placebo
6
Tamoxifen vs Placebo NSABP P1 Trial Subset of
Women ? 50 years old
RR 2.12 (1.52 - 3.03)
Interpretation P increases the rate of invasive
breast cancer incidence compared to Tam by 112
(confidence interval 52 to 303)
Relative Risk for Invasive Br Ca Placebo /
Tamoxifen
7
Potential Designs for Evaluating Raloxifene
Raloxifene
R A N D O M I Z E
STAR designed as superiority trial R vs T
85 power if RR (R/T) lt 0.67 95 power if RR
(R/T) gt 1.56
Tamoxifen
Placebo
Observed RR 1.02 Application has
non-inferiority analysis of STAR
8
Tamoxifen vs Placebo NSABP P1 Trial Subset of
Women ? 50 years old
Favors Tamoxifen
Favors Placebo
RR 2.12 (1.52 - 3.03)
NI margin 50 of Active control effect retained.
56 increased risk on P
RR 1.56
1.84
1.28
2.2
2.4
2.6
2.0
1.8
1.6
1.4
1.2
1.0
0.8
0.6
Relative Risk for Invasive Br Ca Placebo /
Tamoxifen
9
Possible Outcomes of Active Control Trials
Favors Active Control
Favors Test Tx
NI margin 50 of Active control effect retained.
56 increased risk on P
Placebo vs Active Control RR 2.12, with C.I.
Relative Risk Test Treatment / Active Control
10
Goal of Active Control (NI) Analysis

• How would test treatment compare to placebo, had
  placebo been present in the trial?
• Estimate effect of T compared to active control
  (C) in current trial (T vs C)
• Use data from previous trials to estimate effect
  of P vs C, along with a margin of error for that
  effect
• Combine these estimates to evaluate putative
  effect of T vs P
• Sometimes done by estimating the range of percent
  retention of the P vs C effect consistent with
  data (confidence interval for NI margin)
• If T has fewer side effects than active control,
  T may be useful even if not as effective as C

11
Most important issue when evaluating NI
analysesextrapolation

• NI analyses are based on one or more previous
  trials and a current trial and use information
  gained in potentially different settings.
• More careful labeling is helpful
• Current trial T test treatment, C2 active
  control, P a putative (unobserved) Placebo
• Previous trial(s) C1 the same control, P1 the
  placebo

12
Assumptions in NI Analysis

• T vs C2 well conducted.
• P vs C2 P1 vs C1 if placebo had been present
  in current trial (assay sensitivity)
• P1 vs C1 effect has not changed since prior
  trials, or any change can be modeled
• Uncertainty in P1 vs C1 effect can be estimated
• Both within and between trial variability
  relevant
• P1 vs C1 sometimes estimated from meta-analysis
  of prior trials.
• Clinically relevant non-inferiority margin was
  specified before analysis
• Assumptions all used in inference

13
Some Questions to ask about NI analyses

• Is the claim of NI supported by a biological
  rationale?
• Might the effect of Active Control (vs placebo)
  have been different in current trial?
• Changes in administration of agent
• Differences in populations using the drug or in
  endpoint determination
• Has long term follow-up changed the thinking of
  the value of the active control?
• Does the analysis use the best available
  historical data on active control to estimate
  both treatment effects and uncertainty in the
  estimate?
• Justification for omitted trials

14
Questions

• Is an estimated NI margin clinically relevant?
  Was it specified in advance of the analysis?
• Is a reduced therapeutic effect for the test
  agent balanced by other benefits?
• What is the margin of error (confidence interval)
  in the estimate of possible loss of efficacy?
• Are results consistent across related endpoints?
• As in all trials, treatment effects measured in
  NI analyses are estimates of population effects,
  not predictions of efficacy for individuals
• Is there a clear signal to the treating clinician
  on when to use the active control vs the new
  treatment?