Latency
الخميس، ١٢ نوفمبر ٢٠٢٦
Dear reader, between 1966 and 1969, a hospital in Boston treated seven young women for a cancer of the vagina. That sentence should not be possible. Clear cell adenocarcinoma of the vagina is so rare that a gynecologist may practice forty years without seeing one, and when it appears, it appears in women past middle age. These patients were between 15 and 22 years old. An eighth case followed. All had been born in New England hospitals between 1946 and 1951.
Arthur Herbst, with Howard Ulfelder and David Poskanzer, set out to find the common factor. They matched each patient with four healthy women born in the same hospital at about the same time, thirty-two controls in all, then did the part that cannot be automated: they interviewed the mothers in detail about pregnancies that had ended two decades earlier. Medications, illnesses, bleeding, everything.
The answer, published in the New England Journal of Medicine in April 1971, was almost unbelievable. Seven of the eight mothers of the cancer patients had been given diethylstilbestrol, a synthetic estrogen, beginning in the first trimester. Not one of the thirty-two control mothers had. The probability of that distribution arising by chance was less than one in a hundred thousand. The disease had originated not in anything these young women had done or eaten or been exposed to, but in a tablet their mothers swallowed before they were born.
Diethylstilbestrol was synthesized in 1938 and deliberately left unpatented, so anyone could make it cheaply. In 1947, it was approved in the United States to support high-risk pregnancies, on an elegant theory: pregnancy loss is associated with low estrogen, so supplying estrogen should prevent it. Between two and four million American women received it during pregnancy.
Here is the detail that should trouble every prescriber. In 1953, eighteen years before Herbst, William Dieckmann and colleagues at the University of Chicago tested the drug against a placebo in more than 1,600 pregnancies, randomized and double-blind, and found no benefit. It did not prevent miscarriage. It did not improve outcomes. The trial was published, and prescribing continued for nearly two more decades, on the strength of a plausible mechanism and enthusiastic marketing. The cancer was the second failure. The first was that we kept giving a drug that had already been shown not to work.
The Food and Drug Administration contraindicated it in pregnancy in November 1971, months after Herbst's paper. A registry was established, and the eight cases became dozens, then hundreds. The registry eventually recorded some 700 cases, with a 20-year survival rate of around 69%, thanks to early detection and aggressive surgery. Later work found more: structural abnormalities of the reproductive tract in exposed daughters, raised breast cancer risk in the mothers and in the daughters after forty, and in 2018, a report of elevated mortality across the whole lifespan of affected women. Questions about the effects of reaching a third generation remain open.
What should we take from it, half a century on? Three things, none of them historical curiosities.
The first is latency. We are trained to look for harm near the exposure, and the whole apparatus of pharmacovigilance is built on that assumption: side effects during treatment, adverse events in the trial window. This drug caused cancer twenty years later in a different person than the one who took it. Nothing in our routine surveillance would have caught that pattern today, and there’s no reason to think it was the only compound capable of it.
The second is that a plausible mechanism is not evidence. Estrogen supports pregnancy; therefore, supplementing it prevents miscarriage: the reasoning is clean, physiological, and wrong. Medicine is littered with interventions that made perfect sense and killed people, and the only defense is the trial, followed by the discipline to act on it when it comes back negative.
The third is pregnancy itself, and it has not been fixed. Pregnant women are excluded from almost all clinical trials, for reasons that begin as protective. The result is thin evidence for most drugs in pregnancy, which we then prescribe anyway, because pregnant women get ill like everyone else. We did not learn to study pregnancy after this. We learned to avoid studying it, which reproduces the evidence vacuum that allowed the drug in the first place.
In the end, eight patients and a list of careful questions overturned a drug given to millions. That is a small, unglamorous, deeply serious piece of work, of the kind our institutions could do and rarely fund. The instruments today are far better than Herbst's. What is scarcer is the willingness to sit with a mother and ask what happened twenty years ago.
Arthur Herbst, with Howard Ulfelder and David Poskanzer, set out to find the common factor. They matched each patient with four healthy women born in the same hospital at about the same time, thirty-two controls in all, then did the part that cannot be automated: they interviewed the mothers in detail about pregnancies that had ended two decades earlier. Medications, illnesses, bleeding, everything.
The answer, published in the New England Journal of Medicine in April 1971, was almost unbelievable. Seven of the eight mothers of the cancer patients had been given diethylstilbestrol, a synthetic estrogen, beginning in the first trimester. Not one of the thirty-two control mothers had. The probability of that distribution arising by chance was less than one in a hundred thousand. The disease had originated not in anything these young women had done or eaten or been exposed to, but in a tablet their mothers swallowed before they were born.
Diethylstilbestrol was synthesized in 1938 and deliberately left unpatented, so anyone could make it cheaply. In 1947, it was approved in the United States to support high-risk pregnancies, on an elegant theory: pregnancy loss is associated with low estrogen, so supplying estrogen should prevent it. Between two and four million American women received it during pregnancy.
Here is the detail that should trouble every prescriber. In 1953, eighteen years before Herbst, William Dieckmann and colleagues at the University of Chicago tested the drug against a placebo in more than 1,600 pregnancies, randomized and double-blind, and found no benefit. It did not prevent miscarriage. It did not improve outcomes. The trial was published, and prescribing continued for nearly two more decades, on the strength of a plausible mechanism and enthusiastic marketing. The cancer was the second failure. The first was that we kept giving a drug that had already been shown not to work.
The Food and Drug Administration contraindicated it in pregnancy in November 1971, months after Herbst's paper. A registry was established, and the eight cases became dozens, then hundreds. The registry eventually recorded some 700 cases, with a 20-year survival rate of around 69%, thanks to early detection and aggressive surgery. Later work found more: structural abnormalities of the reproductive tract in exposed daughters, raised breast cancer risk in the mothers and in the daughters after forty, and in 2018, a report of elevated mortality across the whole lifespan of affected women. Questions about the effects of reaching a third generation remain open.
What should we take from it, half a century on? Three things, none of them historical curiosities.
The first is latency. We are trained to look for harm near the exposure, and the whole apparatus of pharmacovigilance is built on that assumption: side effects during treatment, adverse events in the trial window. This drug caused cancer twenty years later in a different person than the one who took it. Nothing in our routine surveillance would have caught that pattern today, and there’s no reason to think it was the only compound capable of it.
The second is that a plausible mechanism is not evidence. Estrogen supports pregnancy; therefore, supplementing it prevents miscarriage: the reasoning is clean, physiological, and wrong. Medicine is littered with interventions that made perfect sense and killed people, and the only defense is the trial, followed by the discipline to act on it when it comes back negative.
The third is pregnancy itself, and it has not been fixed. Pregnant women are excluded from almost all clinical trials, for reasons that begin as protective. The result is thin evidence for most drugs in pregnancy, which we then prescribe anyway, because pregnant women get ill like everyone else. We did not learn to study pregnancy after this. We learned to avoid studying it, which reproduces the evidence vacuum that allowed the drug in the first place.
In the end, eight patients and a list of careful questions overturned a drug given to millions. That is a small, unglamorous, deeply serious piece of work, of the kind our institutions could do and rarely fund. The instruments today are far better than Herbst's. What is scarcer is the willingness to sit with a mother and ask what happened twenty years ago.