The 2002 Misreporting, Forensically

The most consequential number in the history of hormone therapy was not 26% — it was 8. On July 9, 2002, the world learned that estrogen-plus-progestin raised breast cancer risk by "26%," and within three years the treatment's use had collapsed. Both numbers appeared in the same journal article; only one made the news. This page reconstructs what the trial actually reported, what the press conference transmitted, and the arithmetic lesson the episode left behind — because the same reporting failure recurs in health news every single week. It explains evidence; it never prescribes.

The Press Conference and What It Transmitted

The sequence that created the scare is a matter of record. On May 31, 2002, the data and safety monitoring board stopped the estrogen-plus-progestin arm of the Women's Health Initiative after 5.2 years of a planned 8.5, because breast cancer had crossed a pre-specified boundary. The results were published in JAMA on July 17, 2002, with the National Institutes of Health holding a press conference days earlier. What the conference transmitted was a handful of relative risks: breast cancer up 26%, stroke up 41%, coronary events up 29%. "26% increased risk" became the sentence that traveled around the world — it survives today in countless retellings, most of which never mention the denominator.

The trial itself was extraordinary science — the largest randomized test of hormone therapy ever run, and the stopping decision was correct. What went wrong was downstream: the translation of a nuanced result into a single alarming percentage, applied without regard to who the participants were. The full corrected story lives on the parent topic page; this page focuses on the transmission failure itself, because that failure — not the trial — is what changed millions of medical decisions.

Relative vs Absolute: The One-Paragraph Arithmetic Lesson

In the WHI estrogen-plus-progestin arm, breast cancer appeared in 30 of every 10,000 women per year on placebo and 38 of every 10,000 on the hormone. The relative increase: 8 divided by 30, which is 26.7% — the famous "26%." The absolute increase: 8 cases per 10,000 women per year, or 0.08 percentage points. Both describe the same data; they differ by an order of magnitude in emotional force. Relative risk measures the ratio of change; absolute risk measures the size of change. A 100% increase in a one-in-a-million event is still a one-in-500,000 event. The rule worth internalizing: a relative risk without a base rate is a number without a scale.

The complementary tool is the number needed to harm (NNH): how many women would need a year of treatment for one extra harmful event to occur. For breast cancer on the WHI regimen, that number is roughly 1,250 woman-years per extra case (10,000 divided by 8). The chart below converts each of the trial's harms into this frame — the longer the bar, the smaller the per-person annual risk.

The Four Harms, One Number Each

Three readings matter. First, the harms are real: the trial stopped early for a reason. Second, they are small in absolute terms — the largest, blood clots, is still under two extra events per 1,000 women per year, concentrated in the first year of use. Third, the same table contains benefits the headlines compressed away: about a third fewer hip fractures and a third fewer colorectal cancers. Reporting that shows only the "+26%" column is not reporting; it is selection.

What the Paper Said vs What the Headlines Said

The striking forensic detail is that the JAMA article contained the absolute numbers from day one — the same 38-versus-30 breast cancer rates this page uses. The authors' own summary characterized the increases as small in absolute terms, and accompanying commentaries cautioned against overreaction. The press release and the resulting coverage, however, were built around relative risks and hazard-ratio shorthand, and the caution did not survive the trip through the news cycle. Media researchers later documented the pattern: coverage of the WHI overwhelmingly used relative-risk framing for harms, rarely gave base rates, and almost never mentioned that the trial's participants averaged 63 years old — a population far removed from the symptomatic early-fifties women who actually use this therapy (Brown, Climacteric, 2012).

The lesson generalizes beyond hormones. Any time a news item reports "X raises your risk of Y by Z%," the first forensic question is always the same: a percentage of what? If the story does not give the base rate, it has not given you a number you can act on. The site's methodology page applies this rule to every statistic on these pages.

The Exodus and What It Cost

Behavior changed faster than the science could correct it. US prescriptions for menopausal hormone therapy fell by roughly two-thirds between 2002 and 2005 (Hersh et al., JAMA, 2004), and the decline persisted for years; millions of women stopped therapy abruptly, often without a tapering conversation. Two consequences followed. The first is documented: a generation of women re-entered severe vasomotor symptoms and lost the fracture protection the therapy provided. The second is contested: estimates of excess mortality from the mass discontinuation exist — one widely discussed analysis of hysterectomized women in their fifties put the toll in the tens of thousands (Sarrel et al., AJPH, 2013) — but such models rest on assumptions that reasonable researchers dispute. The honest summary: the scare plausibly caused real harm at population scale, and the exact size of that harm is not knowable.

The Footnote the Headline Skipped

The second reporting failure was the population. The WHI enrolled women with a mean age of 63, most more than a decade past their final period, many with cardiovascular risk factors — and, because symptoms would have unblinded the study, very few of the severely symptomatic women who seek this treatment in real life. The headline applied the result to all women; the paper described a specific population. That mismatch is not a footnote: it is the basis of the timing hypothesis, which holds that results differ by how far a woman is past menopause. A trial finding reported without its population is a finding reported without its meaning.

The Reporting Standards This Taught

• 🧮 Always ask for the base rate. "Up 26%" is uninterpretable until you know 26% of what. The answer, when given, is usually a number per 1,000 or 10,000 people per year.
• 📏 Convert to NNH or NNT. "1,250 woman-years per extra breast cancer" is a number you can weigh against your own symptoms. "26%" is not.
• 👥 Check who was studied. A mean age of 63 versus 52 is not a detail; it can reverse a conclusion, as the arm-by-arm page and the timing page both show.
• ⚖️ Demand both columns. Harms without benefits — and vice versa — are marketing. The WHI paper reported both; the press conference effectively reported one.

Questions, Answered Briefly

• 😟 Was the 2002 headline wrong? The number was arithmetically correct; the framing was catastrophically incomplete. "26% increased risk" without the base rate is a true statement that misleads.
• 🧪 Did the trial itself hide the absolute numbers? No — they were in the original JAMA paper. The failure happened in transmission: press materials, then coverage, selected the relative-risk column.
• 📐 Which harm was largest in absolute terms? Blood clots: roughly 18 extra events per 10,000 women per year, front-loaded into the first year of use. Breast cancer and stroke each added about 8.
• 🩺 Does any of this mean hormone therapy is safe? It means the risk is quantifiable and modest in scale — a weighing decision with a clinician, laid out on the decision framework page. It is not a blanket endorsement, and nothing here prescribes.

Related Topics