Part I: Statistical issues in WHI Part II: Further analyses of prior hormone therapy on breast cance

1
Part I Statistical issues in WHIPart II
Further analyses of prior hormone therapy on
breast cancer The WHI Estrogen Progestin Trial

• Garnet Anderson
• WHI Clinical Coordinating Center
• Fred Hutchinson Cancer Research Center
• October 7, 2003

2
Statistical methods

• Design and primary analyses based on weighted
  logrank statistic
• T ? w (O-E)
• Weights
• Specified for each disease endpoint
• Defined as time since randomization
• Used to increase efficiency

3
Weights by time since randomization for the
logrank statistic
4
Trial monitoring Early stopping for benefit
required

• Evidence of CHD benefit
• Statistical rules based on OBrien-Fleming (OBF)
  procedures using a 0.025-level, one-sided test
• AND
• Global index supportive of benefit
• Statistical rules based on OBF procedures using a
  0.05-level, one-sided test

Freedman, et al. Control Clin Trials.
199617509-525.
OBrien PC, Fleming TR. Biometrics.
197935549-556.
5
Trial monitoring Early stopping for adverse
effects required

• Evidence of increase in breast cancer
• OBF procedure using a 0.05-level one-sided
  weighted logrank test.
• OR
• Evidence of increase in CHD, stroke, PE, hip
  fracture, colorectal cancer, endometrial cancer,
  or death from other causes
• OBF procedure using a 0.05-level one-sided
  (weighted/unweighted) logrank test, with
  Bonferroni correction.
• AND
• Global index supportive of overall harm (Z

Freedman, et al. Control Clin Trials.
199617509-525.
6
Statistical approaches in reporting

• Intention-to-treat
• Unweighted hazard ratios
• Nominal and adjusted 95 confidence intervals
  from unweighted analyses
• Primary and designated safety endpoints adjusted
  for multiple interim analyses
• Designated secondary endpoints adjusted for
  multiple endpoints and interim analyses
• P-values may be shown from weighted analyses

7
Statistical approaches in reporting subgroup
analyses

• Inference based primarily on tests for
  interaction
• No adjustment for multiple testing required
• Considered hypothesis generating
• Must state
• Number of interactions tested
• Number expected to be significant by chance alone

8
Prior hormone use and breast cancer

• More detailed analysis of prior exposure
• Type of prior hormone use
• Duration of prior hormone use
• Recency of prior hormone use
• Extent of disease comparisons by prior hormone
  use
• Mammography performance

9
Breast cancer risk with EP by prior hormone use
Unweighted p for interaction .04 Weighted p
.10
10
Breast cancer incidence by prior PHT use

Unweighted p .04, Weighted p .10
Not to be reproduced or distributed without
permission from WHI
11
Breast cancer risk by type of prior PHT use
P-values from test for interaction with type of
prior use
Not to be reproduced or distributed without
permission from WHI
12
Breast cancer incidence among prior users of
E-alone and combined hormones
Unweighted p .08 Weighted p .17
Not to be reproduced or distributed without
permission from WHI
13
Breast cancer risk by duration of prior hormone
use
P-values from test for interaction with duration
of prior use
Not to be reproduced or distributed without
permission from WHI
14
Breast cancer incidence by duration of prior
hormone use
Unweighted p .09 Weighted p .15
Not to be reproduced or distributed without
permission from WHI
15
Breast cancer risk by recency of prior hormone use
P-values from test for interaction with recency
of use
Not to be reproduced or distributed without
permission from WHI
16
Breast cancer incidence by recency of prior
hormone use
PHT use at enrollment
PHT use within last 5 years
Unweighted p .59, Weighted p.65
Not to be reproduced or distributed without
permission from WHI
17
Breast cancer incidence by recency of prior
hormone use, continued
Last PHT use 5-10 years ago
Last PHT use 10 years ago
Unweighted p .59, Weighted p.65
Not to be reproduced or distributed without
permission from WHI
18
Breast cancer risk by prior hormone use and body
mass index
P-values from test for interaction with prior use
and BMI
Not to be reproduced or distributed without
permission from WHI
19
Breast cancer
BMI BMI 30
No prior hormone use
Unweighted p .11 Weighted p .12
Prior hormone use
Not to be reproduced or distributed without
permission from WHI
20
Characteristics of prior hormone users

• Younger, leaner
• More frequently
• White
• Have at least a high school education
• Former smokers
• Drink alcohol
• Exercise
• Have used oral contraceptives
• Have vasomotor symptoms
• Have parental history of fracture after age 40
• Had a mammogram in previous 2 years
• Have different pregnancy history

21
Duration of prior use by type
Not to be reproduced or distributed without
permission from WHI
22
Recency of prior use by type
Not to be reproduced or distributed without
permission from WHI
23
Estimated effects of EP in multivariate models
Unweighted p .03 Weighted p Controlling for differential characteristics of
prior hormone users and other breast cancer risk
factors and main effect of prior hormone use
Not to be reproduced or distributed without
permission from WHI
24
Estimated effects of EP in multivariate models
Unweighted p .05 Weighted p .64
Not to be reproduced or distributed without
permission from WHI
25
Estimated effects of EP in multivariate models
Unweighted p .08 Weighted p Not to be reproduced or distributed without
permission from WHI
26
Estimated effects of EP in multivariate models
Unweighted p .17 Weighted p Not to be reproduced or distributed without
permission from WHI
27
Estimated effects of EP in multivariate models
Unweighted p .05 Weighted p Not to be reproduced or distributed without
permission from WHI
28
Extent of disease by prior use and randomization
assignment
Not to be reproduced or distributed without
permission from WHI
29
Breast cancer incidence post-intervention(diagnos
es from 7/8/02 8/31/03)
Not to be reproduced or distributed without
permission from WHI
30
Cumulative breast cancer incidence, including
post-intervention diagnoses
Data through August 31, 2003
Not to be reproduced or distributed without
permission from WHI
31
SummaryBreast cancer

• Suggestion of greater EP hazard ratios in women
  with prior hormone exposure
• Potential confounding by differential
  characteristics of prior users
• Potential delay in diagnosis not taken into
  account
• More extensive modeling suggests
• Prior combined hormone use has a stronger effect
  than prior E-alone use
• Increasing duration of prior exposure is
  associated with greater risk
• Recency of exposure an unclear factor
• Modest evidence for interaction of EP with prior
  use and BMI

32
SummaryBreast cancer, continued

• Data too sparse to jointly examine
• Type
• Duration
• Recency
• Difference in extent of disease by randomization
  status consistent across prior use groups
• Some hints of differential effects by prior use
• Data on post-intervention comparisons are limited

33
Frequency of abnormal mammograms
Not to be reproduced or distributed without
permission from WHI
34
SummaryMammography

• EP increases rates of abnormal mammograms
• Among women who never had a breast cancer
  diagnosis, the proportions with abnormal
  mammograms were
• 32 for EP
• 22 for placebo
• Role of prior hormone use on mammography
  performance is small

35
(No Transcript)