Sato's fraud was one of the biggest in scientific history.
The impact of his fabricated reports—many of them on how to reduce the risk of
bone fractures—rippled far and wide. Meta-analyses that included his trials
came to the wrong conclusion; professional societies based medical guidelines
on his papers. To follow up on studies they did not know were faked,
researchers carried out new trials that enrolled thousands of real patients.
Exposing Sato's lies and correcting the literature had been a bruising struggle
for Avenell and her colleagues.
Yet they could not understand why Sato faked so many
studies, or how he got away with it for so long. They puzzled over the role of
his co-authors, some of whom had their names on dozens of his papers. (“Do we
honestly believe they knew nothing at all about what was going on?” Avenell
asked.) They wondered whether other doctors at his hospital read Sato's
work—and whether the Japanese scientific community ever questioned how he
managed to publish more than 200 papers, many of them ambitious studies that
would have taken most researchers years to complete.
The tools of science that the group had used—analyzing
studies, calculating statistics, writing papers—could reveal fraud. But they
could not expose the personal and cultural factors that drove it, or assess its
emotional toll. So I set off on a quest that would eventually lead me to the
Mitate Hospital in Tagawa, a small town on the island of Kyushu, where Sato had
worked in the last 13 years of his life…
Avenell's own quest began in 2006, when she was combing
through dozens of papers for a review evaluating whether vitamin D reduces the
risk of bone fractures. In two papers by Sato, she stumbled on a weird coincidence.
They described different trials—one in stroke victims, the other in Parkinson's
disease patients—but the control and study groups in both studies had the exact
same mean body mass index. Looking further, she quickly found several other
anomalies. She decided not to include Sato's studies in her analysis.
She wasn't the first to notice something was off. In a 2005
Neurology paper, Sato claimed that a drug named risedronate reduces the risk of
hip fractures in women who have had a stroke by a stunning 86%. In a polite
letter to the journal, three researchers from the University of Cambridge in
the United Kingdom noted that the study was “potentially of great importance,”
but marveled that the authors had managed to recruit 374 patients in just 4
months.
Two years later, a letter in what was then the Archives of
Internal Medicine was less polite. A study of male stroke patients published by
Sato had managed to enroll 280 patients in just 2 months; another one, of women
with Alzheimer's disease, recruited a staggering 500 in an equally short
period. Sato claimed to have diagnosed all of the Alzheimer's patients himself
and done follow-up assessments of all 780 patients every 4 weeks for 18 months.
Both studies had very few dropouts, and both showed risedronate, again, to be a
resounding success. “We are deeply concerned whether the data provided by Sato
et al are valid,” Jutta Halbekath of Arznei-Telegramm, a Berlin-based bulletin
about the drug industry, and her co-authors wrote. Sato apologized in a published
response and claimed the study had been conducted at three hospitals, not one.
“The authors did not describe this fact, the reason being that these hospitals
were reluctant to have their names in the article,” he wrote. He didn't name
the other hospitals or explain why they wanted to remain anonymous. The journal
apparently accepted the explanation.
The letter's authors also spotted a troubling pattern. In
addition to the two papers in the Archives of Internal Medicine, they found 11
further studies by Sato, published elsewhere, that tested whether sunlight,
vitamin D, vitamin K, folate, and other drugs could reduce the risk of hip
fractures. All but two reported “extremely large effects with significant
results,” they noted. But the Archives of Internal Medicine didn't want to
point fingers at other journals. “You may allude to your concern that other
papers have similar concerns,” its editors warned Halbekath, “but we cannot
allow you to mention those other papers by journal name.”
By now, several researchers had raised red flags and waved
them for everyone to see—and then everybody moved on. “The trail just went
cold,” Avenell says.
Mark Bolland had never heard of Sato when Avenell first
mentioned him in late 2012. She and Bolland, a clinical epidemiologist at the
University of Auckland in New Zealand, have never met in person, but they
joined forces to write meta-analyses on calcium supplements in 2008, together
with Andrew Grey and Greg Gamble, both also at the University of Auckland. One
topic the quartet discussed frequently was why meta-analyses on the same topic
sometimes reach different conclusions. Avenell mentioned Sato's studies and
noted that the effects they reported were so strong that they might swing
meta-analyses if they were included.
Intrigued, Bolland looked up the papers. He, too, was
stunned by the large cohorts, the low number of dropouts, and the big effects
of almost any treatment tested. “There is nothing that I can think of that
produces a 70% to 80% reduction in hip fractures, yet Sato was able to do it
consistently in all his trials,” he says.
To follow up on his suspicions, Bolland turned to
statistics. When scientists compare a treatment and a control group, they
usually report “baseline characteristics” for each—things like age, weight, and
sex, or, in osteoporosis studies, bone density and calcium intake. From these
values, scientists can calculate p-values that are a measure of the similarity
of two groups for a given characteristic; the closer to one the value is, the
more the groups resemble each other. Because the groups are randomly selected,
the p-values should normally be “equally distributed”; the value for age or
weight is just as likely to be between 0 and 0.1 as between 0.9 and 1.0, for
example.
Bolland extracted the baseline characteristics from the 33
clinical trials Sato had published at the time, more than 500 variables all in
all, and calculated their p-values. More than half were above 0.8, he found.
“That just shouldn't happen,” he says. “The randomized groups were incredibly
similar.” There was just one plausible explanation, he says: Sato had
fabricated data for both groups and had made them more similar than they would
ever be in real life.
The team felt it had a damning indictment. “I thought: ‘This
is so convincing. Everybody is going to believe this,’” Avenell says. Still,
“It needed detailed statistical refereeing, and it needed to be published by a
journal so that other affected journals would take note,” she adds. So they
wrote their accusation as a scientific paper. All they had to do was publish it
and wait for researchers, journals, and institutions to react, investigate, and
retract. Or so they thought.
In March 2013, the team submitted the manuscript to The
Journal of the American Medical Association (JAMA), the highest profile journal
Sato had published in, and one it felt might have the resources for an in-depth
investigation. After reviewing the evidence, JAMA Editor-in-Chief Howard
Bauchner told the team the editors would ask Sato and, if necessary, his
institution to respond.
Two years later, in April 2015, JAMA told the researchers
the hospital had not responded, and it would publish an “expression of
concern”—a short note to flag Sato's JAMA paper as suspicious. It would not
publish the whistleblowers' paper, however; if the team had concerns about
other papers, it should contact the journals that had published them, Bauchner
said.
The four researchers were shocked. “To find out after
waiting 2 years that in fact nothing much had really happened and, other than
an expression of concern, was going to happen in JAMA, was quite frustrating,”
Bolland says. (Bauchner declined to answer Science's questions about the case.)…
Next, the paper was rejected by JAMA Internal Medicine,
which had also published Sato's work. The Journal of Bone and Mineral Research,
a highly rated journal in the osteoporosis field, said it would investigate
Sato's papers, but would not publish the manuscript either. The editors of
Trials, which had not published Sato's work, said it would not be appropriate
to get involved.
Bolland became demoralized. The other three persuaded him
not to give up. “If you ever embark on something like this, make sure you have
a good support team,” he says now. Avenell, too, was sometimes despondent.
Whereas the other three researchers at least saw each other in Auckland, she
was on her own, frustrated, in the dreary, gray town of Aberdeen. Sometimes,
she says, she would just sit in a corner of her open floor plan office and cry.
Then, in June 2015, came a small success: The Journal of
Bone and Mineral Research retracted one of the 33 trials the team had analyzed.
A few other journals followed suit in the months after. But some seemed
irritated by the group's persistence. “It is apparent that the responses to the
JAMA investigation by Dr. Sato and his institution have been either inadequate
or not forthcoming,” Grey wrote to Bauchner in December 2015. “At what point
will JAMA consider more decisive action, such as retraction?” “We will consider
your opinion about how you think it best we should conduct the investigation,”
Bauchner responded. “We often hear from people how they think we should perform
our responsibilities as editors.”
In what Bolland calls “really just the last throw of the
dice,” that same month the group submitted the paper to Neurology, where Sato
had published three papers about bone fractures in patients with neurological
disease. When it was accepted 8 months later, Avenell cried again. “I'm not one
usually given to showing such emotion, especially when all I have is a computer
screen and emails to look at,” she says…
By the time Neurology published the investigation in
December 2016, 10 of the 33 trials had been retracted, all but one by journals
the team had contacted. Three months later, Avenell received an email from an
editor with troubling news. Sato was dead…
Today, 21 of Sato's 33 trials have been retracted by the
journals or Sato himself; Avenell has crossed them off a list taped next to her
computer with a red marker. But now the team is following the ripples that the
studies caused, focusing, for the time being, on a dozen papers published in
the journals with the highest impact factors. Together, these studies reported
results for 3182 participants. They have been referenced more than 1000 times,
and 23 systematic reviews or meta-analyses have included one or more of the 12
trials…
The letter does not mention fraud, however. “I couldn't
force him to confess,” Ogawa says. “I think he had a mental illness.” His
emails were not logical, he says. “To tell the truth, I predicted that he would
commit suicide.”
Suicide. Is he sure that's what happened?
“I received the information from the lawyer of Mr. Sato,”
Ogawa says. Sato also left a note, he says, and he paraphrases it: “I am very
sorry for Mr. Iwamoto. I decided to commit suicide.”
When I call Avenell after my return from Japan and tell her
what I have learned, there is stunned silence at first. “That's what we were
dreading,” she says. “That's horrible, really horrible.” Exposing the
misconduct was important, she says. “Could we have done it without Sato
committing suicide? So that he felt less guilty? I just don't know.”
Later she follows up with an email, still astonished at “how
such a small piece of data analysis a long time ago can end up with someone
dying.” As a clinician and a researcher, Avenell wrote, she knows her work can
eventually make the difference between life and death. “But seldom is the
connection between a clinician and another human being's death so obvious.”
Courtesy of a colleague
