back(7.12.2003/11.02.07)
DrTim homepage link

Explanation and Critique of the WHI Paper on Hormone Replacement I

www.DrTimDelivers.com

by
Timothy D Bilash, MD
December 02, 2002 Bishop, CA
(first presented August 02, 2002 Cortland, NY)

original paper:
Risks and Benefits of Estrogen Plus Progestin in Healthy
Postmenopausal Women: Principle Results from the Women's
Health Initiative (WHI) Randomized Controlled Study
Writing Group for the Women's Health Initiative (WHI) Investigators
JAMA July 17, 2002 288(3):p321-349
  1. Purpose/ Conclusions/ Dosing
  2. Outcomes
  3. Discontinuation/ Dropouts
  4. Breast Cancer
  5. Statistical Problems I
  6. Error Table
  7. Statistical Problems II
  8. Grouping and Outcome Inaccuracies
  9. Life Table Analysis Limitations
  10. Hormones and Menopause
  11. Breast Cancer Biology
  12. Other Studies/ Epidemiology/ Animal Studies
1) PURPOSE/ CONCLUSIONS/ DOSING
  1. PURPOSE: DOES PREMPRO LOWER HEART DISEASE?
    1. "whether estrogen plus progestin has a favorable or unfavorable effect on CHD incidence and on overall risks and benefits in predominantly healthy women"
  2. CONCLUSIONS OF THE PAPER:
    1. Prempro .625/2.5 regimen "should not be initiated or continued for primary prevention of CHD"
  3. ONLY ONE DRUG REGIMEN WAS TESTED
    1. Prempro .625/2.5 daily with a uterus
      1. .625mg conjugated equine estrogen
      2. 2.5mg medroxyprogesterone acetate
      3. progestin dosing is continuous, not cyclic

2) OUTCOMES



3) DISCONTINUATIONS/OUTCOMES
  1. STUDY WAS DISCONTINUED EARLY (5.2 vs 8.5 year followup)
    1. because breast cancer statistic crossed monitoring parameters (26% increase)
      1. but, this was NOT statistically significant increase however (lost too many patients)
    2. heart attack findings found after further analysis of the data
      1. borderline statistically significant with uncorrected data only
  2. HIGH DROPOUT RATE ( > 90%)
    1. Only 8% of enrolled patients were left in the study after 7 years (1,300 of 16,000))
    2. 800 / 8000 on prempro, 500 / 8000 on sugar pills left
  3. ALL subgroup outcomes (except for DVT) NOT statitistically significant after correct data for multiple testing
    (CI-HR ranges include 0, non-significant, see col2)


4) BREAST CANCER

  1. Incidence (Diagnosis) vs Mortality (Death)
  2. Breast cancer diagnosis/deaths are difficult to interpret:
    1. long development time for breast tumors (10-20 years)
    2. 3 vs 2 deaths average per year for prempro vs placebo group
    3. invasive breast cancer includes the confidence interval CI = 1.0
      1. non-significant at 95% confidence
    4. there is NO significant difference in in-situ breast cancer seen, only invasive
      1. contradicts other data
  3. Estrogen does not cause deaths from breast cancer
    1. Patients may find more breast cancers, especially initially (incidence), but die less often (mortality) from all causes, including from breast cancer.
  4. Best evidence is that Estrogen does not cause breast cancer
    1. The WHI study of esrogen-only use in women who had previously undergone hysterectomy is continuing, with no reported increase risk of breast cancer.
    2. E2 levels >1000pg/ml in pregnancy, high estrogen doses [glinoer 1997 top p405]
      1. pregnancy decreases breast ca risk after, so must be another factor
    3. Reported rates of diagnosing breast cancer differed with prior hormone use:
      1. HRT with .....never prior use.....114 vs 102, HR = 1.06 [0.81-1.28]
        HRT with ........ <5yrs prior use...... 32 vs.15, HR = 2.13 [1.15-3.94]
        HRT with >5-10 yrs prior use........11vs....2, HR = 4.61 [1.01-21.02]
      2. note that as the statistical error decreases, so does the effect seen,
        supporting no statistical significance (ie, never-use has more patients)
    4. The overall rate of cancer was not different with HRT
      1. Breast Cancer HR = 1.26, Total Cancers HR = 1.03
      2. consistent with a subgroup outcome analysis bias

5) STATISTICAL PROBLEMS I

  1. inaccuracies in diagnosis and error ranges
    1. uncertainty of outcomes between local and central diagnoses adds to inaccuracy:
      1. [see Table 2 column(3)]
      2. breast, colorectal and endometrial cancers are diagnosed by hard data (98% by tissue pathology)
      3. Stroke, MI, DVT diagnosed more by clinical symptoms, thus less accurate
        1. severity of event not indicated
        2. multiple episiodes in a patient versus new episodes in different patients not distinguished
      4. this adds 5-18% to error ranges
      5. which was used for this paper, central or local diagnoses?
    2. ALL outcomes in this study (except for DVT) include the confidence interval of 1.0 (1-HR = 0) when statistically corrected
      1. [see Table 2, column (2)]
      2. this implys NO statistical significance (except for DVT) even assuming absolutely perfect date
      3. consider the further correction for disagreement in diagnoses and rarity of DVT

6) ERROR TABLE




7) STATISTICAL PROBLEMS II
  1. dont know what percentage of adverse events (ie., PE/DVT) were confirmed by imaging/ lab studies as opposed to clinical diagnoses
  2. "Overall CHD rates were low" in the study compared to the general population.
    1. most excess was in nonfatal MI
    2. this supports a diagnosis bias or a population bias (affects CHD conclusions)
  3. No trends were seen
    1. there are no statistical trends over the seven years for incidence of disease:
    2. there was no difference in mortality or cause of death between the groups
      1. [see Table 3]
    3. however, there is a trend after 5 years showing decreased death rate, just as meaningful as the incidence data
      1. [see figure 4]

8) Grouping and Outcome Inaccuracies

Summary of inaccuracies inherent to the study:
  1. 04% of prempro arm had received estrogen only in the beginning of the study
  2. 01% switched groups post hysterectomy
  3. 42% discontinued active treatment
  4. 35% additional patients were unblinded, additional 11% crossover to active treatment
  5. 05-16% discrepancy in diagnoses depending on diagnosis
  6. 03% died
  7. 03% lost to followup
  8. Only 8% of enrolled patients were left in the study after 7 years (1,300 of 16,000)a*

9) Life Table Analysis Limitations


"Enron" Accounting ???
  1. Experimental Results (compares numbers)
  2. Life Table Analysis (compares rates)
  3. Signal vs Noise
  4. This publication was published in JAMA with no author, prior to expected publication


10) Hormones and Menopause

  1. Hormones
    1. FSH
    2. LH
    3. Estrogen (E2)
    4. Estrone (E1)
    5. Progestins
      1. Progesterone
      2. Provera (medroxyprogesterone acetate)
      3. Norethindrone
      4. Newer ones
    6. Steroid interconversion diagram
    7. Free hormone is the active form, not total hormone

  2. Hormone changes in menopause
    1. run out of eggs in ovary, so make less estrogen (E2)
      1. E2 starts decreasing at 38 years
    2. estrogen (E2) source shifts to estrone (E1)
      1. converted from adrenal androgens (primarily androstenedione) in adipose
      2. not from ovary
    3. gradual rise in FSH in response to low estrogen and low inhibin
    4. lower sex hormone-binding globulin (causes more free estrogen and free testosterone)
    5. HRT Side Effects
      1. distinguish Nuisance vs Harm

11) Breast Cancer Biology
  1. Does it make sense?
    1. medroxyprogesterone acetate slows breast cancer lines positive for receptors in humans
      1. progesterone reduces cell divisions in breast
      2. progesterone decreases estrogen (progesterone) receptors
      3. progesterone stimulates 17beta-dehydrogenase (converts to estrone)
    2. estrogen induces progsterone receptors
    3. control of cellular response of breast epithelium appears unrelated to direct effects of estrogen and progesterone
    4. progesterone causes down-regulation of estrogen receptors in luteal phase (2nd E2 receptor pool shows no effect)
    5. in vivo and in vitro studies conflic
      1. there are two kinds of progesterone receptors giving varied responses depending on the tissue
      2. model: progesterone promotes start gene or stop gene depending on target tissue and timing of presence

12) Other Studies/ Epidemiology/ Animal Studies
  1. EPIDEMIOLOGY
    1. no increased risk with increased progesterone dose
    2. increased incidence confined to in-situ tumors (reverse of WHI)
    3. breast feeding protection may be due to bias- needs further study
    4. increased screening, increased diagnosis, increased survival (?)
      1. Iceland- incidence has increased, mortality rates stable
      2. Norway- increase in incidence has slowed now
      3. Sweden- cohorts by age showed no increase risk from OC's

  2. ANIMAL STUDIES
    1. in rats
      1. estrogen promotes tumors, progesterone inhibits tumors
      2. pretreatment with progesterone prevents or decreases growth promotor effect of estradiol

  3. OTHER (OBSERVATIONAL) STUDIES
    1. results are mixed, three studies indicate increased risk
      1. increased breast cancer risk from ERT runs 1.3 - 2 fold
      2. primarily in subgroup with alcohol consumption
      3. late natural menopause increases risk
      4. see Table C
      5. increased risk of breast cancer diagnosis, not risk of breast cancer mortality
    2. London Nurses Health Study (increased risk current users, but not with duration of use)
      1. Risk-benefit of HRT (model calculations, England)
        1. Greatest impact found on ischemic heart disease
          1. prevention of death (see figure D), gain in life span
          2. assumes breast cancer RR=1.3
          3. gains wiped out if RR=2 (but not worse)
        1. increased hospitalization?
    3. CASH data shows increased risk < 35 yrs, lower risk 45-54 yrs, but lower mortality
    4. HERS (Heart and Estrogen/Progestin Replacement Study)
      1. 2763 all had CHD
      2. early increased risk with no overall increase after 5 years

a*(Added 02.08.2005) The 2002 paper for Kaplan Meier Analysis indicates approximately 1300 patients remaining after 7 Outcome years [~800 Estrogen-Progestin/~500 Placebo depending on Outcome, Fig3&4 p328&329]. Censoring criteria are not the same for Cox Proportional Hazard Analysis, which indicates 9000 patients left [5129 Estrogen-Progestin/4243 Placebo, Table 4 p329]. The 2002 paper (as well as subsequent reports) have made no effort to clarify such issues, often placing critical information only in the small print subtext of figures, or not including any comment at all.

back

DrTim homesite


page views since Sept2007