THE VIRGINIAN-PILOT
Copyright (c) 1995, Landmark Communications, Inc.
DATE: Sunday, July 16, 1995 TAG: 9507140595
SECTION: COMMENTARY PAGE: J1 EDITION: FINAL
SOURCE: BY ANN G. SJOERDSMA
LENGTH: Long : 150 lines
``There are three kinds of lies - lies, damned lies and statistics.''
- Benjamin Disraeli
When the media trumpet statistics, especially statistics affecting women's health, with a fanfare befitting the Second Coming, I take up my own instrument - reasoned skepticism - and go to the source.
I understand probability, know my mode from my mean and my median, and appreciate the difference between the so-called ``rate of incidence'' and the actual number of cases. Percentages don't scare me. In fact, they invite a challenge. But I am not a statistician, emphatically not.
Inevitably, what appears in splashy, results-oriented statistics to be meaningful, as well as anxiety-producing, is much less so upon investigation. Even I can discern this much.
The latest case in point: the estrogen-linked-to-breast-cancer findings of Harvard University's ``Nurses' Health Study,'' published in the June 15 issue of The New England Journal of Medicine. In reporting the latest from this ongoing study, newspapers and magazines nationwide trumpeted, with bold statistics, the alarming news that estrogen taken by menopausal women to decrease the risk of osteoporosis (by slowing the loss of bone), as well as heart disease, might cause breast cancer.
I do not seek to minimize the risk of breast cancer or to suggest that there is no link between estrogen and breast cancer: Scientists recognize that there is one. Estrogen ``receptors'' exist in most breast cancers, and current therapy for menopausal women with breast cancer recommends against estrogen replacement.
But to make the leap from acknowledgment of an uncalculated link between estrogen and breast cancer to calculated proof of an increased risk of breast cancer associated with hormone replacement therapy (HRT) is problematic. When results from the more than 30 estrogen-breast cancer studies done are combined - including a study published just last week in the Journal of the American Medical Association that showed no increased risk - it appears that there is very little or no overall risk of breast cancer associated with HRT.
Am I saying that the Harvard study is flawed? It may be, but that is not my claim to make. What I am saying is that the study is not about medical reality; it is not about individual women - about you or me. It is about mathematical reality, about statistics. Number-crunching. I know. I've read the report, and reread it, and dissected it, with expert medical counsel. I would like to meet the gynecologist who can navigate its foggy, uncharted statistical sea. As for the media, they cannot begin to evaluate its significance.
The Associated Press reported that menopausal women who have taken estrogen for several years ``appear to raise their risk of breast cancer by nearly half.'' Half of what? I wondered.
The Washington Post concluded that any menopausal woman who takes estrogen ``runs a slightly higher chance of developing breast cancer,'' while also cautioning that the risks and benefits of estrogen replacement ``aren't fully known, and almost certainly differ from woman to woman.'' How confusing is that?
All media reports reached this bottom line: Overall, postmenopausal women who have taken estrogen, with or without progestin, for at least five years have a ``46 percent greater chance of developing breast cancer'' than women who have never taken the hormone. Thus, 46 percent - the AP's ``half'' - became the buzz statistic.
After reading what I could in the lay press, I was left with many questions, including: 1) How were the study data collected?; 2) What was the hormone dosage/regimen of the various women?; 3) How were risk factors such as age and familial cancer history evaluated?; and most important, 4) What were the actual numbers? How many of the women taking estrogen, with or without progestin, developed breast cancer and how many did not? And of those stricken, what were their cancer ``risk profiles''?
I sought details, substantiation, accountability, and so I went to the source. Bear with me, as I show how analysis of the study methodology quickly breaks down:
The Nurses' Health Study started in 1976 with 121,700 female nurses, ages 30 to 55, who answered a mailed questionnaire that asked them about ``risk factors'' for cancer and cardiovascular disease. The researchers sent follow-up questionnaires every two years, and by 1990, they had classified 69,586 women as postmenopausal (PM).
The number 69,586 - actual lives - can be readily understood. But in analyzing their data, the researchers converted lives to ``person-years of follow-up.'' From 725,550 person-years of follow-up, they identified 1935 cases of invasive breast cancer among PM women. How does the individual woman, with her individual hormone treatment, fit in here? She doesn't.
They then decided to use only those cancer cases diagnosed from 1988 to 1992, thus reducing the number of cases to 1,373, of which 923 took no hormone therapy. Of the 450 ``hormone''-cancer cases, 20 percent used conjugated estrogens (Premarin) alone.
To measure the ``relative risk'' of breast cancer - necessary for the calculation of percentages of increased risk - the researchers divided the incidence of breast cancer among PM women who had taken hormones by the incidence among PM women who had never used such therapy.
This is straightforward enough, but then the number-crunching begins. Consider these all-important kickers, which I (trust me) simplify: The researchers adjusted the relative risks for various hormone treatments by conducting ``stratified analyses to control for risk factors''; by using ``proportional-hazards models to adjust for multiple risk factors simultaneously''; and by applying ``the odds ratio from a logistic regression in which we controlled for the year of birth and the age at menopause,'' thereby accounting for changes in estrogen use over time.
Whew! Regression models and life-tables. Nothing but numbers. Translated, this means that important factors such as age, age at menopause, type of menopause, family history, history of benign breast disease, other ``risks'' and much more were ``controlled'' through mathematics.
And yet, if I read their graphics correctly, the researchers calculate incidence of breast cancer among women 60 to 64 who have used hormones for at least five years (the highest relative risk group) at 6.5 per 1,000, or 0.65 per 100, or less than 1 percent. For the same-age women who have not taken hormones, the incidence is 3.5 per 1,000, again less than 1 percent.
There is so much more here, but I have belabored enough.
If statistics are going to underlie health care reports and medical decision-making - and they certainly are - then it's important to strive for context, which, because of deficiencies in data reporting, is difficult to establish.
Allow me to demystify at least one misleading statistic: one in eight. For a long time, we heard that women had a one-in-nine lifetime risk of developing breast cancer; later, this rose to one-in-eight. Why? Because the earlier data were based on an 85-year lifetime, the latter on a 95-year lifetime. How many of us are going to live to 95? Or 85, for that matter? The ``average'' female lifespan is about 79 years.
According to the National Cancer Institute, only women 95 and older have a one-in-eight risk of developing breast cancer. The risks for younger women include: at age 25, one in 19,608; at 45, one in 93; at 65, one in 17. Of course, the pool of women shrinks with age, as mortality (death often by heart disease) takes its toll. The figures increase, but not alarmingly so. Breast cancer is associated with age.
Women who have a mother, sister or daughter with breast cancer, or who have ovarian, uterine or colon cancer in their families, comprise a known high-risk breast cancer group. But consider these ``risk factors'' identified by the National Cancer Institute and the Centers for Disease Control and Prevention's cancer division: early age at menarche, late age at menopause, late age at first live birth and higher education and socioeconomic status. Clearly, the reporting system - where and whom the data come from - skews the profile.
What we have here is the tip of a very big, hitherto neglected iceberg. Confusion and contradiction abound. Every informed woman must assess for herself the risks and benefits of HRT, in light of her own history and health, as well as the evolving (and often inconclusive) state of medical knowledge.
There is much to read between the lines of a sketchy daily-newspaper article. With statistics themselves becoming more and more often the ``news,'' it behooves us all to dust off our old textbooks, or buy new ones, and get our calculators charged. Our health may depend on it. MEMO: Ann G. Sjoerdsma is a lawyer and book editor of The Virginian-Pilot
and The Ledger-Star. ILLUSTRATION: Color graphic and chart by Adriana Libreros, Staff
by CNB