back
(08.08.03/10.31.07*)
DrTim homepage link
Comments on the Million Woman Study of Breast Cancer
by Timothy D. Bilash MD, MS
August 18, 2003
www.DrTimDelivers.com
about the paper:
Breast cancer and hormone-replacement therapy
in the Million Women Study
Million Woman Study Collaborators
The Lancet Aug 09, 2003, 362:419-427
1) this is a 5 year cohort study with mammograms taken at baseline and 3 year intervals. patients already on HRT that had a mammogram prior to the study period used that as the baseline.
2) study period was short (makes this like a prevalence study, since screening was at 3 year intervals)
.......
2.6 years for cancer incidence
.......
4.1 years for cancer mortality
3) 10-20% of patients (estimated) became non-compliant with their study med at 2.8 years, which adds dosing uncertainty due to compliance bias
|
22% of current HRT users had stopped HRT
19% of
..
past HRT users
..
had started HRT
11% of
.... ....
never users
.
had started HRT
|
|
|
|
|
|
|
4) Cox's Proportional Model is subject to startpoint artifact in the first years because it assumes a constant chance of having and constant chance of detecting disease
- since screened in three year intervals, there is only one reporting period (the first)
- if find tumors earlier in HRT group (such as from doing more self-exams while on HRT) rather than more tumors, would expect to inflate the number found in first period after baseline mammogram
5) There is no control group in the study (all selected from response to survey)
6) MSW study finds an unexpectedly low postmenopausal breast cancer rate of 3.2/1000 (3.2/1000 per year over 2.6 years at average age of 55.9 years), compared to population estimated rate of 32/1000 per year from the ages of 50 to 65 years (see page 424)
|
|
#cancers/total#
|
|
cancers/year
|
|
study postmenopausal never users
|
2894 / 392757
|
=
|
2.8/1000
|
|
study postmenopausal past users
|
1044 / 150179
|
=
|
2.7/1000
|
|
study postmenopausal current Est
|
991 / 115383
|
=
|
3.3/1000
|
|
study postmenopausal current Est+Prog..
|
1934 / 142870
|
=
|
5.2/1000
|
|
|
|
|
|
|
study average all postmenopausal pts
|
6863 / 828923
|
=
|
3.2/1000
|
|
typical, average age 50-65
|
.
|
=
|
32/1000
|
|
|
|
|
|
7) The risk of finding invasive breast cancer according to the study was lower in postmenopausal users than in premenopausal never users (RR=0.63, CI=0.58-0.68)
- contrary to clinical experience
- premenopausal women with baseline and 3 year followup mammogram would tend to have a (1st vs 2nd mammogram) rather than (2nd vs 3rd) or (3rd vs 4th) mammogram
- premenopausal women generally have denser breasts at baseline, and this would increase the sensitivity for detection by followup mammogram or if start HRT during the study (see #8 below)
- these factors would favor finding new tumors in the younger women
- premenopausal women had a 10% rate of HRT everuse, while postmenopausal women had >50% HRT everuse. this might help explain the inverted post/pre findings, if past exposure to HRT decreases the risk of breast cancer.
- may indicate outcome detection bias or subject selection bias
8) if the total never-user group has the same or lower incidence of breast cancer as the general population, then finding a higher-than-expected-rate in the premenopausal-never-user group, would require a lower-than-expected-rate in the postmenopausal-never-user group than the general population
- study does find a lower postmenopausal-never-user rate compared to premenopausal-never-user rate
9) re mammograms, if HRT makes it harder to find breast cancers as stated in the paper, then:
- -women already on HRT at baseline would find new breast lesions easier since would have a good baseline mammogram comparison (already dense at baseline), inflating the incidence >5 year duration
- -women not on HRT at baseline would have an interval increase in their breast density that would make it harder to find new tumors, deflating the incidence <5 year duration (since they started HRT during the study and density increase is seen in the first years in other studies)
- -this might contribute to the reported duration-of-use-effect on breast cancer (current users fig3 page 422, fig5 page 423), and the reported current-user-effect (more current users were on HRT >4 years, 80585/139569 = 58%,
see fig2 page 421
10) there was no dose-amount-effect seen (higher dose, estrogen alone, estrogen+progestin all same fig4, fig5). although a dose-time(duration)-effect is reported for combined HRT, this is problematic in a short-term life-table analysis study
11) note that "the relative risk of breast cancer showed little variation according to the progestogen constituent or whether sequential or continuous regimens were used" (fig 5)
12) there was NO statistically valid increase in breast cancer deaths for current-users, contrary to newspaper reports (RR=1.22, confidence interval of 1.0-1.48 includes 1.0 means NOT significant, Findings page 419)
13) study RR for incidence of breast cancer was affected by body-mass index:
|
|
|
|
group
|
RR for BMI > 25
|
RR for BMI < 25
|
|
estrogen only
|
1.53
|
1.17
|
|
estrogen-progestin
|
2.31
|
1.78
|
|
|
|
|
|
[ ]switched < >
- 40%-50% increase in RR of breast cancer was due to high BMI (is a large effect compared to a 40-100% increase RR study effect)
- higher BMI tend to have higher total estrogen levels
- effect of BMI for progestin only was not reported
- BMI demographics for study population was not reported
14) size of corrections due to other stratification variables applied to RR for breast cancer appear to be on the order of several percent each. these could cumulatively account for an additional 5-20% of noted effects depending on the exact corrections (not reported)
- ages
- time since menopause
- parity and age at first birth
- family history of breast cancer
- region
- deprivation index
15) total cancer incidence rate was not reported
16) total cancer death rate was not reported
17) total death rate was not reported or compared to general population
18) causation generally cannot be proved with a cohort study, because they are observational and not experimental, but they can provide strong evidence for causation because they possess the correct time sequence for cause and effect
19) this study is not consistent with other observational data
[an error occurred while processing this directive]page views since Sept2007