Statistical Issues in NDA 21239 Laura Lu, Ph.D FDA/CDER

1
Statistical Issues in NDA 21239 Laura Lu,
Ph.DFDA/CDER

2
Outline

• ITT vs. Per-Protocol Analyses in Study 95-02
• Definitions for A Responder in Study 95-02
• Original Definition
• Window Definition Proposed by Sponsor
• Window Sensitivity Analysis
• Subgroup Analysis in Patients with Baseline
  SLEDAIgt2

3
ITT vs. Per-Protocol Population in Study 95-02

• ITT Population
• Specified in the original protocol
• All randomized patients
• Per-Protocol Population
• Proposed in a later submitted statistical plan
  (most patients had finished study)
• Excluded drop-outs within the first 60 days

4
ITT Analysis

• Preserves randomization---the base for valid
  statistical inference.
• In general, avoids over estimation of treatment
  effect.

5
Sponsors Argument for Per-Protocol Analysis

• Treatment needs at least 60 days to take into
  effect

6
Patient Disposition in Patients Excluded from ITT
Population
7
Validity of Per-Protocol Analysis?

• Excluding early dropouts in the per-protocol
  analysis may bias conclusion--there are treatment
  related dropouts.

8
Original Definition for A Responder in Study
95-02

• Improvement or stabilization in SLAM, SLEDAI,
  KFSS, Patient VAS (post-baseline weighted average
  of each score no worsening than the baseline
  score)
• No clinical deterioration

9
Result for Responder Rate (ITT)
10
Later Proposed Window for Responder Definition

• Compared with baseline, post-baseline weighted
  average for
• SLAM no worse than 1
• SLEDAI no worse than 0.5
• KFSS no worse than 0.5
• Patient VAS no worse than 10
• No clinical deterioration

11
Window Definition by Percent of Change from
Baseline

• For example, a -5 window definition for a
  responder is
• 1) weighted average for each of SLAM, SLEDAI,
  KFSS and Patient VAS no worse than 5 from
  baseline
• 2) no clinical deterioration

12

13
Sensitivity of Responder Rate to Window
Definition

• The numerical trend of responder rates in
  treatment groups is sensitive to whether
  worsening is allowed in the responder definition.

14
Subgroup Analysis in Patients with Baseline
SLEDAIgt2

• Study 94-01 Hypothesis Generating
• Study 95-02 Specified in Protocol Amendment

15
Results in Study 94-01
Responder Rate
16
Results in Study 94-01 (cont.)
Percent Change from Baseline in Prednisone Dose
in Baseline SLEDAIgt2 Group
17
Results in Study 95-02
Responder Rate (original definition, ITT
population)
18
Results in Baseline SLEDAIgt2 Group

• In Study 94-01, the results of primary
  endpoints were not consistent in baseline
  SLEDAIgt2 group.
• Numerical advantage in responder rate
• no advantage in mean percent change in prednisone
  dose

19
Results in Baseline SLEDAIgt2 Group

• In Study 95-02, DHEA showed numerical
  advantage over placebo in responder rate in
  patients with baseline SLEDAIgt2. Statistical
  significance was not demonstrated by ITT analysis
  with original definition (p0.17). A small
  p-value (0.005) was found by per-protocol
  analysis with a window definition.

20
Question

• Are additional studies needed for the baseline
  SLEDAIgt2 group to support an efficacy claim?

21
(No Transcript)
22
(No Transcript)
23
(No Transcript)