2
Editorial Predicting Future Longevity: At Long Last, Extrapolation Is On the Way Out Aubrey D.N.J. de Grey ‘‘One of the problems with posing a ’bold new plan’ is that you can’t just extrapolate from previous plans.’’ – Nathan Myhrvold O ne of the first and most fundamental tenets of the scientific method, instilled in aspiring researchers before they even leave high school, is that when your hy- pothesis is contradicted by your data you need a new hy- pothesis. In reality, of course, this tenet is a bit fuzzy to work with, being influenced by such inconvenient truths as the unreliability of instrumentation or reagents, the divisi- bility of a hypothesis into variants with similar simplicity, and so on. But still, at the end of the day it is a dictum that one flouts at one’s peril. You would not know this, observing the history of expert attempts to predict future life expectancy. Over the past 50 years, actuaries and demographers have lined up to pro- nounce on the trajectory which they believe longevity will follow — and every single time, they have been wildly incorrect. By and large, the rate of increase of life expec- tancy has far exceeded what was predicted. Why? What’s so difficult about this? Surely we can just look back at past trends and extrapolate them into the fu- ture? Well, yes, that’s a great starting hypothesis — and indeed, it’s exactly what everyone has always done. Why doesn’t it work? There are, broadly, three reasons. The first is that the past trajectory is somewhat complex: for a start, when viewed over the very long term longevity has clearly been increasing at an accelerating rate, having hardly risen at all from the dawn of civilization until 1800 or so and roughly trebled since then. This means that the trend must be described by a function with more than two pa- rameters — and unfortunately there are many ways to do so that are pretty similar in terms of simplicity and plausibility. The second problem is that, whatever method you may choose to do your extrapolation, the answer you get depends a lot on the details of what question you ask. A classic case in point is Oeppen and Vaupel’s 1 celebrated report from a decade ago that ‘‘best practice’’ longevity, defined as the world record national life expectancy in any given year, has risen uncannily linearly since 1840 or so. Few were the commentators who looked under the hood at the supple- mentary material and noted that the strength of this corre- lation relies greatly on fine details, such as the use of just the right type of data for New Zealand in the period 1900–1950, or the fact that Japan transitioned to having negligible infant mortality a couple of decades later than the West. But ultimately, the third difficulty eclipses those two in terms of its severity. It is that those who perform these extrapolations almost invariably feel the need to come up with answers that coincide with their intuition, and the quality of their intuition is limited by their command of the relevant knowledge. Unfortunately, that knowledge is gen- erally technological, especially biomedical, and demogra- phers and actuaries don’t generally know much about medical research. Thus it is, again picking just one example for illustration, that Olshansky and colleagues 2 have found it possible to predict an imminent and permanent plateauing of life ex- pectancy, on the basis of the diminishing returns that would result from continuing to decrease mortality rates at each age at the proportional rate that they have declined in the past century. Few were the commentators who looked more closely at the statistics and noted that that decline has ac- celerated for the older ages where mortality is high, and that if the extrapolation is refined to take that into account then one instead derives the prediction that most people alive today will live to at least 200. Of course, this prediction is every bit as worthless as the plateau prediction, being based on the same flawed method, but the point is that the plateau prediction is the one that gets published and believed, simply because it is the answer that most people, including most demographers, were already making off the top of their heads before the study was even conceived. I have been gratified that those with a financial interest in the accuracy of long-term longevity predictions, such as pension funds and life insurance companies, have been in- viting me regularly for nearly a decade to offer my thoughts on how this parlous situation can be improved. Until quite recently, however, my pleasure at this was tempered by the SENS Research Foundation, Mountain View, California. REJUVENATION RESEARCH Volume 17, Number 3, 2014 ª Mary Ann Liebert, Inc. DOI: 10.1089/rej.2014.173.edit 247

Predicting Future Longevity: At Long Last, Extrapolation Is On the Way Out

Embed Size (px)

Citation preview

Editorial

Predicting Future Longevity:At Long Last, Extrapolation Is On the Way Out

Aubrey D.N.J. de Grey

‘‘One of the problems with posing a ’bold new plan’is that you can’t just extrapolate from previous plans.’’

– Nathan Myhrvold

One of the first and most fundamental tenets ofthe scientific method, instilled in aspiring researchers

before they even leave high school, is that when your hy-pothesis is contradicted by your data you need a new hy-pothesis. In reality, of course, this tenet is a bit fuzzy towork with, being influenced by such inconvenient truths asthe unreliability of instrumentation or reagents, the divisi-bility of a hypothesis into variants with similar simplicity,and so on. But still, at the end of the day it is a dictum thatone flouts at one’s peril.

You would not know this, observing the history of expertattempts to predict future life expectancy. Over the past 50years, actuaries and demographers have lined up to pro-nounce on the trajectory which they believe longevity willfollow — and every single time, they have been wildlyincorrect. By and large, the rate of increase of life expec-tancy has far exceeded what was predicted.

Why? What’s so difficult about this? Surely we can justlook back at past trends and extrapolate them into the fu-ture? Well, yes, that’s a great starting hypothesis — andindeed, it’s exactly what everyone has always done. Whydoesn’t it work? There are, broadly, three reasons.

The first is that the past trajectory is somewhat complex:for a start, when viewed over the very long term longevityhas clearly been increasing at an accelerating rate, havinghardly risen at all from the dawn of civilization until 1800 orso and roughly trebled since then. This means that the trendmust be described by a function with more than two pa-rameters — and unfortunately there are many ways to do sothat are pretty similar in terms of simplicity and plausibility.

The second problem is that, whatever method you maychoose to do your extrapolation, the answer you get dependsa lot on the details of what question you ask. A classic casein point is Oeppen and Vaupel’s1 celebrated report from adecade ago that ‘‘best practice’’ longevity, defined as theworld record national life expectancy in any given year, hasrisen uncannily linearly since 1840 or so. Few were thecommentators who looked under the hood at the supple-

mentary material and noted that the strength of this corre-lation relies greatly on fine details, such as the use of just theright type of data for New Zealand in the period 1900–1950,or the fact that Japan transitioned to having negligible infantmortality a couple of decades later than the West.

But ultimately, the third difficulty eclipses those two interms of its severity. It is that those who perform theseextrapolations almost invariably feel the need to come upwith answers that coincide with their intuition, and thequality of their intuition is limited by their command of therelevant knowledge. Unfortunately, that knowledge is gen-erally technological, especially biomedical, and demogra-phers and actuaries don’t generally know much aboutmedical research.

Thus it is, again picking just one example for illustration,that Olshansky and colleagues2 have found it possible topredict an imminent and permanent plateauing of life ex-pectancy, on the basis of the diminishing returns that wouldresult from continuing to decrease mortality rates at eachage at the proportional rate that they have declined in thepast century. Few were the commentators who looked moreclosely at the statistics and noted that that decline has ac-celerated for the older ages where mortality is high, and thatif the extrapolation is refined to take that into account thenone instead derives the prediction that most people alivetoday will live to at least 200. Of course, this prediction isevery bit as worthless as the plateau prediction, being basedon the same flawed method, but the point is that the plateauprediction is the one that gets published and believed,simply because it is the answer that most people, includingmost demographers, were already making off the top of theirheads before the study was even conceived.

I have been gratified that those with a financial interest inthe accuracy of long-term longevity predictions, such aspension funds and life insurance companies, have been in-viting me regularly for nearly a decade to offer my thoughtson how this parlous situation can be improved. Until quiterecently, however, my pleasure at this was tempered by the

SENS Research Foundation, Mountain View, California.

REJUVENATION RESEARCHVolume 17, Number 3, 2014ª Mary Ann Liebert, Inc.DOI: 10.1089/rej.2014.173.edit

247

niggling suspicion that the enthusiastic reception I alwaysreceived would transmute by the following day into a distantmemory of a bad dream, and my audience would continue asbefore. This was, of course, because my own predictions ofthe trajectory of longevity are dramatically more optimisticthan those on the basis of which such professionals havealready placed very, very large bets.

It is thus a source of unconfined joy to me that, within onlythe past year or so, one very respected forecasting companyhas altered the playing field entirely. Risk Management So-lutions (RMS) recently began to sell a system that they havedeveloped over the past five years, in which future life ex-pectancies are calculated on the basis of expert opinionconcerning the likely time frame for the development andsubsequent dissemination of new technologies that mightsubstantially disrupt prior trends. Regenerative medicine andmore general postponement of age-related ill-health featurefront and centre in their model; check it out at < http://rms.com/liferisks > . (For the avoidance of doubt, neither Inor SENS Research Foundation have any financial interestsin RMS.) By this means, RMS is already touting the possi-bility of considerable deviation from current expectations of

longevity. They can’t tell the whole truth quite yet, of course,because they’re in the business of making a profit and peopledon’t buy bad dreams. But the number one obstacle is nomore: there is now a genuine, robust conduit of informationfrom those who are creating the future of longevity to thosewho are predicting it. One of the most fundamental barriersto the defeat of aging — the credibility barrier — has taken abig step into becoming history.

References

1. Oeppen J, Vaupel JW. Demography. Broken limits to lifeexpectancy. Science. 2002;10;296(5570):1029–1031.

2. Olshansky SJ, Carnes BA, Desesquelles A. Prospects forhuman longevity. Science 2001;291(5508):1491–1492.

Address correspondence to:Aubrey D.N.J. de Grey

SENS Research FoundationMountain View, CA

E-mail: [email protected]

248 DE GREY