Published last October in Nature, the study from scientists at Albert Einstein College of Medicine in New York was the eleventh most talked-about piece of research in 2016, according to Altmetric. The paper is not yet indexed in Clarivate Analytics’ Web of Science.
But now, multiple research teams have described what they see as flaws in either the statistical methods or underlying reasoning of the study. Today, Nature published five peer-reviewed rebuttals, in response to the study. Another scientist described his concerns about the paper in April in F1000 Research.
The five papers in Nature are published as Brief Communications Arising, the journal’s way of flagging an important debate over a paper. The short papers provide new data to challenge a central part of a paper’s conclusions. The study’s authors, however, have responded to all five, defending their methods, especially their controversial decision to rely in part upon a visual inspection of mortality data in concluding there is a limit to human lifespan. Senior author Jan Vijg, a geneticist, told Retraction Watch:
What else do we say? It boggles my mind how people can come up with these stupid arguments. You see there’s a plateau; mortality [for supercentenarians, people older than 110 years] is not going down and we’re not seeing any new [age] records. After the 1990s, there’s no longer an increase in the maximum age of death. Your eyes don’t lie.
The study and the responses highlight a fault line separating different types of aging researchers: those who think there’s a limit to how long humans can expect to live —one that’s close to what we’re seeing now with the oldest of the old — and those who think it could be much higher, or that it’s possible the limit does not exist.
Jim Vaupel, a gerontologist at the Max Planck Institute for Demographic Research in Rostock, Germany and one of the authors of one of the BCAs, was one of the harshest critics. He told Retraction Watch:
The “findings” stem from their naive use of inappropriate data…The balance of evidence suggests that if there is a limit, it is above 120, perhaps much above and perhaps there is not a limit at all. Whether or not there is a looming limit is an important scientific question. Because they used weak statistical methods to analyze inappropriate data, they contribute nothing useful to deeper understanding of human life expectancy and human lifespan.
Instead of publishing the criticisms to the paper as BCAs — which only appear online — Vaupel said:
…Nature should have promptly retracted [it].
But Vaupel is in the extreme. Jay Olshansky, of the University of Illinois at Chicago — who both reviewed the original article and wrote a “News and Views” piece accompanying it, told Retraction Watch that the authors of the rebuttals are “missing the point:”
The rebuttals are mostly focused on slightly different ways of looking at the same limited data; basically, if you tilt your head a little to the left or right and look at the same old age mortality/survival statistics for all humans, you might come to slightly different conclusions.
They quibble about how to deal with the mathematics of small numbers at extreme old age, and they fail to realize the obvious, staring them right in the face: the number of people surviving to extreme old age is so small because there is a biologically based limit to life operating on our species, and what they’re quibbling about is the byproduct of the very phenomenon they think does not exist.
Where have all the record breakers gone?
Vijg told Retraction Watch that the paper started with an observation kicked around his lab by a couple students:
The oldest person who ever lived died in 1997 [Frenchwoman Jeanne Calment, who was 122]. Have you seen anybody even close to her? The answer is no.
For two decades, the closest anyone has come is 117 years old, he told us. Prior to Calment’s death, the maximum age at death showed an uncontestable upward trend, but since then it appears to have plateaued. Vijg said he and his students wanted to know if this was a real “trend break,” or if it was something else.
As first reported by Dutch newspaper NRC in December 2016, the first version of the paper was rejected, but received an “enormous” amount of constructive criticism. Vijg told Retraction Watch that, going off the reviewer’s reports, his students soon had a “whole new paper,” which they resubmitted. The Dutch story reported this was at Nature’s invitation, characterizing it as a play for a story they knew would titillate on social media, but both Vijg and Olshanksy pushed back against many of the claims in the piece, saying their quotes were wildly misconstrued throughout it. Vijg told us:
The newspaper suggests that the editors took control of the decision over the reviewers. This is not true. They told us, “ If only one reviewer doesn’t want it, it’s over. We will not take it.”
“Evidence for a limit to human lifespan” looked on either side of 1995, just before Calment died, noting the ages at which the oldest people died, and found the numbers tended to peter out.
Siegfried Hekimi, a biologist at McGill University who studies aging, told Retraction Watch he thought that Vijg’s team chose 1995 “arbitrarily” because after that year, their graph “doesn’t seem to increase anymore.”
I think they jumped to conclusions. They saw that if you picked that date, [the plateau] seemed plausible. There is no justification to pick that date. There are many ways to pick dates.
He’s co-authored a BCA questioning that decision and he’s not alone — two other BCAs focused on that aspect of the study. In their responses, Vijg and his coauthors said their choice of where to partition the data was itself “data-driven.” But Hekimi told us:
However it’s picked, you can take the data set and arrange the data in a variety of ways which seem to be just as valid.
Hekimi and his co-author wrote that the age-at-death data were sufficiently noisy that they could be the start of several trends — one showing the 115 year limit, but also one showing a higher limit, and one showing no limit at all.
Saul Newman, a geneticist at Australia’s Commonwealth Scientific and Industrial Research Organisation who wrote up his criticisms for F1000 Research, told Retraction Watch that he didn’t have a problem with splitting the data, per se:
You could say maybe, there’s an inflection point at 1995. There’s nothing particularly wrong with it — if these were comprehensively surveyed data. But they’re not.
Both the U.S. and France, home to many of the world’s oldest people, were removed from the dataset the authors used in 2003, he said. If a supercentenarian had died at, say 118, that would have not been recorded.
It’s not because of a trend in those populations, it’s because you’re no longer measuring those populations.
Newman told us he tried to get Nature to publish his own rebuttal, but was told they’d received enough submissions already.
Vaupel also mentioned these missing data in his BCA. He told Retraction Watch that maximum age at death was even the wrong statistic — instead, the authors should have been looking at the oldest person alive in a given year:
In many years the maximum lifespan attained in that year is greater than the maximum age at death–because someone is still alive at an age greater than the maximum age at death. An analysis of maximum lifespans should focus on the oldest age attained over time. As the graph we cite in our note shows, maximum lifespans have tended to steadily rise over time with no looming limit in sight.
“There is no reason to retract it”
Vijg and his co-authors responded to all five BCAs, defending their statistical techniques, noting he had the data reviewed at his institution’s statistics core:
Do you really think I’m stupid enough to send papers out without getting my papers checked by professional statisticians?
There is no reason to retract it. You’re not talking about experiments we did in the lab. We’re talking about transparent calculations we did on totally open databases.
Nature told us:
The publication of one or more BCAs about a paper does not indicate that the paper is being considered for retraction or correction. The BCA format offers a way to highlight important discussions around a paper to the scientific community and facilitate scientific debate.
It is not unusual for us to publish multiple BCAs on a paper if the independent peer-reviewers consider the comments valuable additions to the scientific discussion at the time of consideration.
They are open about what they did. Then starts the scientific discussion about whether they made the right choices or not. They played it following the rules, they didn’t fabricate data.
I think these five BCAs, that is a kind of correction. As it should be.
Like Retraction Watch? Consider making a tax-deductible contribution to support our growth. You can also follow us on Twitter, like us on Facebook, add us to your RSS reader, sign up on our homepage for an email every time there’s a new post, or subscribe to our daily digest. Click here to review our Comments Policy. For a sneak peek at what we’re working on, click here.