Biodemographic Trajectories of Longevity

James W. Vaupel, * James R. Carey, Kaare Christensen, Thomas E. Johnson,
Anatoli I. Yashin, Niels V. Holm, Ivan A. Iachine, Väinö Kannisto, Aziz A.
Khazaeli, Pablo Liedo, Valter D. Longo, Yi Zeng, Kenneth G. Manton, James
W. Curtsinger

Old-age survival has increased substantially since 1950. Death rates
decelerate with age for insects, worms, and yeast, as well as humans. This
evidence of extended postreproductive survival is puzzling. Three
biodemographic insights--concerning the correlation of death rates across
age, individual differences in survival chances, and induced alterations in
age patterns of fertility and mortality--offer clues and suggest research
on the failure of complicated systems, on new demographic equations for
evolutionary theory, and on fertility-longevity interactions. Nongenetic
changes account for increases in human life-spans to date. Explication of
these causes and the genetic license for extended survival, as well as
discovery of genes and other survival attributes affecting longevity, will
lead to even longer lives.

J. W. Vaupel is at the Max Planck Institute for Demographic Research,
D-18057 Rostock, Germany; Odense University Medical School, DK-5000 Odense
C, Denmark; the Center for Demographic Studies, Duke University, Durham, NC
27706, USA; and Andrus Gerontology Center, the University of Southern
California, Los Angeles, CA 90089-0191, USA.
*   To whom correspondence should be addressed at the Max Planck Institute
for Demographic Research, Doberaner Strasse 114, D-18057 Rostock, Germany.
E-mail: jwv@demogr.mpg.de
 

------------------------------------------------------------------------
Humanity is aging. The social, economic, and health-care consequences of
the new demography (Table 1) will drive public policy worldwide in coming
decades (1). Growth of the older population is fueled by three factors.
Baby-boom generations are growing older. The chance of surviving to old age
is increasing. And the elderly are living longer--because of remarkable,
largely unexplained reductions in mortality at older ages since 1950 (2-4).
Biodemography, the mating of biology and demography, is, we argue, spawning
insights into the enigma of lengthening longevity (5).

Increases in Old-Age Survival

For Sweden, accurate statistics on mortality are available going back for
more than a century. Female death rates at older ages have fallen since
1950, with large absolute reductions at advanced ages (Fig. 1). The pattern
is similar for males, although from conception to old age males suffer
higher death rates than females, and progress in reducing male mortality
has generally been slower than for females. Consequently, most older people
in Sweden--and nearly all other countries--are women. <Picture>
------------------------------------------------------------------------
Fig. 1. Shaded contour maps (47) of death rates (48) for Swedish females
from age 0 to 112 and years 1875 to 1995 (49), with contours on a ratio
scale of mortality doublings (A) and on an arithmetic scale (B). The color
of each small rectangle denotes the level of the death rate at that age and
year. White rectangles indicate ages and years when no female deaths were
recorded. Dark red rectangles at the highest ages mark the deaths of the
last survivor of a cohort. The vertical black line marks the year 1950,
when increases in old-age survival accelerated. The horizontal black line
is at age 85. The large relative reductions in mortality at younger ages,
especially before 1950, are apparent when a ratio scale is used to set
contours (A). The vertical light line at 1919 in (A) is a consequence of
deaths from the Spanish flu epidemic. The low level of mortality at ages
below age 70 and the large absolute reductions in mortality at advanced
ages are highlighted when an arithmetic scale is used (B). [View Larger
Version of this Image (44K GIF file)]

------------------------------------------------------------------------

For other developed countries, trends in mortality since 1900 have been
roughly similar to those in Sweden. For example, old-age survival has also
increased since 1950 for female octogenarians in England, France, Iceland,
Japan, and the United States (Fig. 2). If there were an impending limit to
further declines in death rates at older ages, countries with low levels of
mortality would tend to show slow rates of reduction. There is, however, no
correlation between levels of mortality and rates of reduction (2). In most
developed countries the rate of reduction has accelerated, especially since
1970 (2, 4). Japan, which enjoys the world's longest life expectancy and
lowest levels of mortality at older ages, has been a leader in the
quickening pace of increase in old-age survival (Fig. 2). Since the early
1970s female death rates in Japan have declined at annual rates of about 3%
for octogenarians and 2% for nonagenarians. Mortality among octogenarians
and nonagenarians has been low in the United States (Fig. 2). The reasons
for the U.S. advantage and the recent loss of this advantage to Japan and
France are not well understood (4, 6). <Picture>
------------------------------------------------------------------------
Fig. 2. Deaths per 1000 women at ages 80 to 89 from 1950 to 1995 for Japan
(dashed black line), France (blue line), Sweden (green line), England and
Wales (red line), Iceland (gray line), the United States (light blue line),
and U.S. whites (brown line). The U.S. data (light blue line) may be
unreliable, especially in the 1960s. Source: (49, 50). [View Larger Version
of this Image (26K GIF file)]

-----------------------------------------------------------------------

The reduction in death rates at older ages has increased the size of the
elderly population considerably (2, 4, 7). In developed countries in 1990
there were about twice as many nonagenarians and four to five times as many
centenarians as there would have been if mortality after age 80 had stayed
at 1960 levels. Reliable data for various developed countries indicate that
the population of centenarians has doubled every decade since 1960, mostly
as a result of increases in survival after age 80 (7).

The decline in old-age mortality is perplexing. What biological charter
permits us (or any other species) to live long postreproductive lives (8)?
A canonical gerontological belief posits genetically determined maximum
life-spans. Most sexually reproducing species show signs of senescence with
age (9), and evolutionary biologists have developed theories to account for
this (10). The postreproductive span of life should be short because there
is no selection against mutations that are not expressed until reproductive
activity has ceased (11-13).

The logic of this theory and the absence of compelling countertheories (14)
have led many to discount the evidence of substantial declines in old-age
mortality. Often it is assumed that the reductions are anomalous and that
progress will stagnate (15). Only time can silence claims about the future.
And empirical observations are not fully acceptable until they are
explicable. We have therefore focused on testing hypotheses and developing
new concepts.

Mortality Deceleration

A key testable hypothesis is that mortality accelerates with age as
reproduction declines. We estimated age trajectories of death rates (Fig.
3) for Homo sapiens, Ceratitis capitata (the Mediterranean fruit fly),
Anastrepha ludens, Anastrepha obliqua, and Anastrepha serpentina (three
other species of true fruit fly), Diachasmimorpha longiacaudtis (a
parasitoid wasp), Drosophila melanogaster, Caenorhabditis elegans (a
nematode worm), and Saccharomyces cerevisiae (baker's yeast). To peer into
the remote realms of exceptional longevity we studied very large cohorts.
<Picture>
------------------------------------------------------------------------
Fig. 3. Age trajectories of death rates (48). (A) Death rates from age 80
to 122 for human females. The red line is for an aggregation of 14
countries (Japan and 13 Western European countries) with reliable data,
over the period from 1950 to 1990 for ages 80 to 109 and to 1997 for ages
110 and over (49). The last observation is a death at age 122, but data are
so sparse at the highest ages that the trajectory of mortality is too
erratic to plot. Although the graph is based on massive data, some 287
million person-years-at-risk, reliable data were available on only 82
people who survived past age 110. The exponential (Gompertz) curve that
best fits the data at ages 80 to 84 is shown in black. The logistic curve
that best fits the entire data set is shown in blue (16). A quadratic curve
(that is, the logarithm of death rate as a quadratic function of age) was
fit to the data at ages 105 and higher; it is shown in green. (B) Death
rates for a cohort of 1,203,646 medflies, Ceratitis capitata (17). The red
curve is for females and the blue curve for males. The prominent shoulder
of mortality, marked with an arrow, is associated with the death of
protein-deprived females attempting to produce eggs (51). Until day 30,
daily death rates are plotted; afterward, the death rates are averages for
the 10-day period centered on the age at which the value is plotted. The
fluctuations at the highest ages may be due to random noise; only 44
females and 18 males survived to day 100. (C) Death rates for three species
of true fruit flies, Anastrepha serpentina in red (for a cohort of 341,314
flies), A. obliqua in green (for 297,087 flies), and A. ludens in light
blue (for 851,100 flies), as well as 27,542 parasitoid wasps,
Diachasmimorpha longiacaudtis, shown by the thinner dark blue curve. As for
medflies, daily death rates are plotted until day 30; afterward, the death
rates are for 10-day periods. (D) Death rates for a genetically homogeneous
line of Drosophila melanogaster, from an experiment by A.A.K. and J.W.C.
The thick red line is for a cohort of 6338 flies reared under usual
procedures in J.W.C.'s laboratory. The other lines are for 17 smaller
cohorts with a total of 7482 flies. To reduce heterogeneity, eggs were
collected over a period of only 7 hours, first instar larvae over a period
of only 3 hours, and enclosed flies over a period of only 3 hours. Each
cohort was maintained under conditions that were as standardized as
feasible. Death rates were smoothed by use of a locally weighted procedure
with a window of 8 days (52). (E) Death rates, determined from survival
data from population samples, for genetically homogeneous lines of nematode
worms, Caenorhabditis elegans, raised under experimental conditions similar
to (53) but with density controlled (21). Age trajectories for the
wild-type worm are shown as a solid red line (on a logarithmic scale given
to the left) and as a dashed red line (on an arithmetic scale given to the
right); the experiment included about 550,000 worms. Trajectories for the
age-1 mutant are shown as a solid blue line (on the logarithmic scale) and
as a dashed blue line (on the arithmetic scale), from an experiment with
about 100,000 worms. (F) Death rates for about 10 billion yeast in two
haploid strains: D27310b, which is a wild-type strain, shown in red; and
EG103 (DBY746), which is a highly studied laboratory strain, shown in blue
(34). Surviving population size was estimated daily from samples of known
volume containing about 200 viable individuals. Death rates were calculated
from the estimated population sizes and then smoothed by use of a 20-day
window for the EG103 strain and a 25-day window for the D27310b strain.
Because the standard errors of the death-rate estimates are about one-tenth
of the estimates, the pattern of rise, fall, and rise is highly
statistically significant. (G) Death rates for automobiles in the United
States, estimated from annual automobile registration data. An automobile
"dies" if it is not re-registered (26, 54). The blue and dashed blue lines
are for Chevrolets from the 1970 and 1980 model years; the red and dashed
red lines are for Toyotas from the same years. [View Larger Version of this
Image (29K GIF file)]

------------------------------------------------------------------------

For humans (Fig. 3A), death rates increase at a slowing rate after age 80.
A logistic curve that fits the data well from age 80 to 105 indicates that
death rates may reach a plateau (16). A quadratic curve fit to the data at
ages 105+ suggests a decline in mortality after age 110.

For four species of true fruit flies in two genera and for a parasitoid
wasp (Fig. 3, B and C), death rates rise and then fall. The data on
medflies (Fig. 3B) generated considerable controversy when published
because it was generally believed that for almost all species mortality
inexorably increases at ages after maturity (9, 17). Previously unpublished
data on three species from a different genus and a species from a different
order (Fig. 3C) demonstrate that mortality decline is not unique to
medflies. Theories of aging will have to confront the vexing observation of
mortality decline.

Mortality deceleration can be an artifact of compositional change in
heterogeneous populations (18). Previously unpublished Drosophila data
(Fig. 3D) demonstrate that a leveling off of death rates can occur even
when heterogeneity is minimized by rearing genetically homogeneous cohorts
under very similar conditions.

The mortality trajectories for C. elegans (Fig. 3E) are based on data from
experiments more extensive than earlier ones. The trajectory for the
wild-type strain decelerates when about a quarter of the cohort is still
alive, similar to observations for Drosophila. For age-1 mutants mortality
remains low throughout life, which demonstrates that simple genetic changes
can alter mortality schedules dramatically.

Data from about 10 billion individuals in two strains of S. cerevisiae were
used to estimate mortality trajectories (Fig. 3F). Because the yeast were
kept under conditions thought to preclude reproduction, death rates were
calculated from changes in the size of the surviving cohort. Although they
need to be confirmed, the observed trajectories suggest that for enormous
cohorts of yeast, death rates may rise and fall and rise again.

The trajectories in Fig. 3 differ greatly. For instance, human mortality at
advanced ages rises to heights that preclude the longevity outliers found
in medflies (3, 16, 17). Such differences demand explanation. But the
trajectories also share a key characteristic. For all species for which
large cohorts have been followed to extinction (Fig. 3), mortality
decelerates and, for the biggest populations studied, even declines at
older ages. A few smaller studies have found deceleration in additional
species (19). For humans, the insects, and the worms, the deceleration
occurs at ages well past normal reproductive ages.

If older individuals contribute to the reproductive success of younger,
related individuals, then they promote the propagation of their genes.
Hence, in social species, the effective end of reproduction may be much
later than indicated by fertility schedules (20). The deceleration of human
mortality, however, occurs after age 80 and the leveling off or decline
after age 110, ages that were rarely if ever reached in the course of human
evolution (8) and ages at which any reproductive contribution is small.

In our early experiments, flies and worms were held in containers, with the
density of living individuals declining with age. To check whether
mortality deceleration could be an artifact of such changes in crowding, we
held density constant--and still observed deceleration (21).

Biodemographic Explanations

It is not clear how to reconcile our two key findings--that mortality
decelerates and that human mortality at older ages has declined
substantially--with theory about aging. The proximate and ultimate causes
of postreproductive survival are not understood (12, 22). Theories that
leave "non-zero late survival ... unexplained" are unsatisfactory (13).
Three biodemographic concepts--mortality correlation, heterogeneity in
frailty, and induced demographic schedules--point to promising directions
for developing theory.

Mortality correlation. Demographers have long known that death rates at
different ages are highly correlated across populations and over time (23).
In addition to environmental correlation, there may be genetic correlation:
Mutations that raise mortality at older ages may do so at younger ages as
well, decreasing evolutionary fitness (12). A pioneering Drosophila
experiment found mortality correlation and no evidence of mutations with
effects only at late ages (24). Postreproductive life-spans might be
compared with postwarranty survival of equipment (25). Although living
organisms are vastly more complex than manufactured products, they too are
bound by mechanical constraints that may impose mortality correlations. The
trajectory of mortality for automobiles (Fig. 3G) decelerates, suggesting
the possibility that both deceleration and mortality correlation are
general properties of complicated systems (26).

Heterogeneity in frailty. All populations are heterogeneous. Even
genetically identical populations display phenotypic differences. Some
individuals are frailer than others, innately or because of acquired
weaknesses. The frail tend to suffer high mortality, leaving a select
subset of survivors. This creates a fundamental problem for analyses of
aging and mortality: As a result of compositional change, death rates
increase more slowly with age than they would in a homogeneous population
(18).

The leveling off and even decline of mortality can be entirely accounted
for by models in which the chance of death for all individuals in the
population rises at a constant or increasing rate with age (18). A frailty
model applied to data on the life-spans of Danish twins suggests that
mortality for individuals of the same genotype and with the same nongenetic
attributes (such as educational achievement and smoking behavior) at some
specified age may increase even faster than exponentially after that age
(27). On the other hand, mortality deceleration could result from
behavioral and physiological changes with age.

Verification of the heterogeneity hypothesis hinges on empirical estimation
of the variation in frailty within a population. If at specified ages
cohorts of Drosophila (or some other species) could be subjected to a
stress that killed the frail and left the survivors neither weaker nor
stronger, then comparison of the trajectories of mortality for the stressed
cohorts with the trajectories for control cohorts would reveal the degree
of heterogeneity (28). In practice, however, stresses generally weaken some
survivors and strengthen others. Experiments with multiple intensities of
stress, including nonlethal levels, may permit experimental estimates of
heterogeneity in frailty.

Induced demographic schedules. A key construct underlying evolutionary
theory is the Lotka equation, which determines the growth rate of a
population (or the spread of an advantageous mutation) given age schedules
of fertility and survival (29). The simplistic assumption in the Lotka
equation that fertility and survival schedules are fixed is surely wrong
for most species in the wild: Environments in nature are uncertain and
changing (30). Many species have evolved alternative physiological modes
for coping with fluctuating conditions, including dauer states (C.
elegans), stationary phase (yeast), diapause (certain insects), and
hibernation. In social insects the same genome can be programmed to produce
short-lived workers or long-lived queens (9). That is, alternative
demographic schedules of fertility and survival can be induced by
environmental conditions.

To reproduce medflies need protein--and this is only occasionally available
in the wild. Medflies fed sugar and water can survive to advanced ages and
still reproduce when fed protein. Regardless of when medflies begin
reproducing, their subsequent mortality starts low and rises rapidly. This
is a striking example of how, depending on the environment, organisms can
manipulate their age-specific fertility and survival (31).

In nematodes, exposure to nonlethal heat shock or other stresses early in
life induces increases in both stress resistance and longevity (32). In
Drosophila, stress can also produce increases in subsequent longevity,
attributable in part to the induction of molecular chaperones (33).
Deletion of the RAS2 gene in S. cerevisiae doubles the mean chronological
life-span of yeast in stationary phase (34). RAS2 mutants exhibit striking
similarities to long-lived nematode mutants, including increases in stress
resistance (32, 34). Rodents raised on restricted diets have extended
life-spans and increased resistance to environmental carcinogens, heat, and
reactive oxidants (35). These findings suggest that stress-related genes
and mechanisms may affect longevity across a broad range of species
(32-35).

In sum, induced physiological change can lower mortality substantially.
There is also evidence for physiological remolding to cope with damage in
organisms (9, 36). An individual does not face fixed fertility and survival
schedules, but dynamically adopts alternative schedules as the environment
and the individual's capabilities change. For this and other reasons (30,
37), Lotka-based evolutionary theory needs rethinking. Post-Lotka equations
should incorporate "grandparental and multigenerational terms, ...
homeostatic feedback and fluctuating environments" (37), as well as induced
demographic schedules.

Although simplistic, the Lotka equation captures a fundamental insight: It
is reproductive success that is optimized, not longevity. Deeper
understanding of survival at older ages thus hinges on intensified research
into the interactions between fertility and longevity (19, 31, 38).

Survival Attributes

The concepts of mortality correlation, heterogeneity in frailty, and
induced demographic schedules can be tied together by a general question:
How important are an individual's survival attributes (that is, persistent
characteristics, innate or acquired, that affect survival chances) as
opposed to current conditions in determining the chance of death? For
humans, nutrition and infections in utero and during childhood may program
the development of risk factors for several important diseases of middle
and old age (39). Conflicting evidence suggests that current conditions may
affect old-age survival chances much more than conditions early in life (2,
40).

A frailty model applied to Danish twin data sheds some even-handed light on
this controversy. The model suggests that about 50% of the variation in
human life-spans after age 30 can be attributed to survival attributes that
are fixed for individuals by the time they are 30; a third to a half of
this effect is due to genetic factors and half to two-thirds to nongenetic
survival attributes (related to, for example, socioeconomic status or
nutritional and disease history). The model suggests that the importance of
survival attributes may increase with a person's life expectancy. For
persons who at age 30 can expect to survive into their 90s, more than 80%
of the variation in life-span may be due to factors that are fixed by this
age (41).

How many survival attributes account for most of the variation in
life-spans? The number required to "survive ad extrema" may be "hundreds,
not tens-of-thousands" (37); research over the next decade may resolve this
question. For nematode worms and yeast, the mutation of a single gene can
result in a qualitative change in the age trajectory of mortality (Fig. 3E)
(34). For other species, including Drosophila and humans, no genes with
such radical demographic effects have yet been discovered, but some
polymorphisms, such as ApoE alleles in humans, alter substantially the
chance of surviving to an advanced age (42). The emerging field of
molecular biodemography seeks to uncover how variation at the microscopic
level of genetic polymorphisms alters mortality trajectories at the
macroscopic level of entire populations.

Analyses of data on Danish twins and other populations of related
individuals indicate that 20 to 25% of the variation in adult life-spans
can be attributed to genetic variation among individuals; heritability of
life-span is also modest for a variety of other species (43). The
possibility that genetic polymorphisms may play an increasing role with age
is supported by evidence of increases with age in the genetic component of
variation in both cognitive and physical ability (44).

Although genes and other survival attributes are fixed for individuals,
their distribution in a cohort changes with age as the frail die. Hence, it
is possible to develop survival attribute assays based on demographic
analysis of changes with age in the frequency of fixed attributes. In
longitudinal research in progress, we are gathering information on
lifestyle and environmental conditions as well as DNA from 7000 Chinese
octogenarians and nonagenarians, 3000 Chinese centenarians, and 14,000
elderly Danes. Survival-attribute assays applied to these data may uncover
a suite of genetic and nongenetic determinants of longevity.

Experiments with insects, worms, yeast, and other organisms permit
alternative approaches for discovering survival attributes; the diet and
stress experiments sketched above provide examples. That genes can alter
mortality trajectories is now certain; research on the mechanisms will shed
new light on aging and longevity (45). The importance of diet, stress, and
reproduction in inducing alternative mortality schedules has been
demonstrated, but the potential of such studies to clarify causal
relationships is just beginning to be tapped. The emerging dialogue between
biologists and demographers (5) is changing the terms of discourse and
opening new vantage points for research on aging.

REFERENCES AND NOTES

1.In China the 60+ population in 2050 may number around half a billion
people, about the number worldwide today (Table 1). To keep the proportion
over 60 to a quarter that size, population size would have to rise from
1.25 to 2 billion [J. W. Vaupel and Y. Zeng, Policy Sci. 24, 389 (1991); Y.
Zeng, J. W. Vaupel, Z. Wang, Math. Pop. Stud. 6, 187 (1997)]. More
generally, see J. W. Vaupel and A. E. Gowan, Am. J. Public Health 76, 430
(1986) [Medline] and D. A. Wise, Ed., Advances in the Economics of Aging
(Univ. of Chicago Press, Chicago, IL, 1996). 2.V. Kannisto, J. Lauritsen,
A. R. Thatcher, J. W. Vaupel, Pop. Dev. Rev. 20, 987 (1994); V. Kannisto,
Development of Oldest-Old Mortality 1950-1990 (Odense Univ. Press, Odense,
Denmark, 1994); V. Kannisto, The Advancing Frontier of Survival (Odense
Univ. Press, Odense, Denmark, 1996); J. R. Wilmoth in (5), p. 38. 3.B.
Jeune and J. W. Vaupel, Eds., Exceptional Longevity: From Prehistory to the
Present (Odense Univ. Press, Odense, Denmark, 1995). 4.J. W. Vaupel,
Philos. Trans. R. Soc. London Ser. B 352, 1 (1997) . 5.K. W. Wachter and C.
E. Finch, Eds., Between Zeus and the Salmon: The Biodemography of Longevity
(National Academy Press, Washington, DC, 1997). 6.K. G. Manton and J. W.
Vaupel, N. Engl. J. Med. 333, 1232 (1995) [Medline]. 7.J. W. Vaupel and B.
Jeune, in (3), p. 109.  8.Remaining life expectancy at age 65 for
Paleolithic populations may have been about 7 years [R. E. Lee, in (5), p.
212]. For Swedish females in 1900, 1950, and 1995 it was 12.9, 14.3, and
19.8 years, and for Japanese females in 1995 it was over 20.8 years, triple
the Paleolithic level. Remaining life expectancy at age 50 from the stone
age through the middle ages may have varied from 10 to 16 years [J. R.
Wilmoth in (3), p. 125], compared with values of 23.8, 26.4, and 33.0 for
Swedish females in 1900, 1950, and 1995. 9.C. E. Finch, Longevity,
Senescence, and the Genome (Univ. of Chicago Press, Chicago, IL, 1990).
10.P. B. Medawar, An Unsolved Problem in Biology (Lewis, London, 1952); G.
C. Williams, Evolution 11, 398 (1957) ; W. D. Hamilton, J. Theor. Biol. 12,
12 (1966) [Medline]; B. Charlesworth, Evolution in Age-Structured
Populations (Cambridge Univ. Press, New York, 1994); P. Abrams and D.
Ludwig, Evolution 49, 1055 (1995) ; L. Partridge in (5), p. 78. For
discussion of the mixed empirical support for this theory, see (13); J. W.
Curtsinger, P. M. Service, T. Prout, Am. Nat. 144, 210 (1994); D. E. L.
Promislow, M. Tatar, A. A. Khazaeli, J. W. Curtsinger, Genetics 143, 839
(1996) [Abstract]. 11.J. W. Curtsinger, Genetica 96, 187 (1995) ; S.
Tuljapurkar, in (5), p. 65. 12.B. Charlesworth and L. Partridge, Curr.
Biol. 7, R440 (1997) [Medline]. 13.The quote, from (12, p. R441), pertains
to L. D. Mueller and M. R. Rose, Proc. Natl. Acad. Sci. U.S.A. 93, 15249
(1996) [Abstract/Full Text]; also see S. D. Pletcher and J. W. Curtsinger,
Evolution, in press. 14.L. Keller and M. Genoud, Nature 389, 958 (1997) .
15.N. Keilman, J. Off. Stat. 13, 245 (1997); J. F. Fries, N. Engl. J. Med.
303, 130 (1980) [Medline]; S. J. Olshansky, B. A. Carnes, C. Cassel,
Science 250, 634 (1990) [Medline]. 16.A. R. Thatcher, V. Kannisto, J. W.
Vaupel, The Trajectory of Mortality from Age 80 to 120 (Odense Univ. Press,
Odense, Denmark, 1998). 17.J. R. Carey, P. Liedo, D. Orozco, J. W. Vaupel,
Science 258, 457 (1992) [Medline]; J. R. Carey, Demography 34, 17 (1997)
[Medline]. 18.J. W. Vaupel, K. G. Manton, E. Stallard, ibid. 16, 439 (1979)
[Medline]; J. W. Curtsinger, H. H. Fukui, D. R. Townsend, J. W. Vaupel,
Science 258, 461 (1992) [Medline]; J. W. Vaupel and J. R. Carey, ibid. 260,
1666 (1993) [Medline]; A. I. Yashin, J. W. Vaupel, I. A. Iachine, Mech.
Aging Dev. 74, 1 (1994). 19.M. Tatar, J. R. Carey, J. W. Vaupel, Evolution
47, 1302 (1993) ; D. L. Wilson, Mech. Aging Dev. 74, 15 (1994). But most
smaller studies have not found deceleration (9). 20.J. R. Carey and C.
Gruenfelder, in (5), p. 127; S. N. Austad, ibid., p. 161. 21.J. R. Carey,
P. Liedo, J. W. Vaupel, Exp. Gerontol. 30, 605 (1995) [Medline]; A.A.
Khazaeli, L. Xiu, J. W. Curtsinger, J. Gerontol. 52, 48 (1995) ; A. A.
Khazaeli, L. Xiu, J. W. Curtsinger, Genetica 98, 21 (1996) [Medline]. In
our nematode experiments, the volume of the container was reduced as worms
died, to keep density constant. 22.K. Christensen and J. W. Vaupel, J. Int.
Med. 240, 333 (1996). 23.A. Coale and P. Demeny, Regional Model Life Tables
and Stable Populations (Academic Press, New York, 1983); R. D. Lee and L.
R. Carter, J. Am. Stat. Assoc. 87, 659 (1992) . 24.S. D. Pletcher, D.
Houle, J. W. Curtsinger, Genetics 148, 287 (1998) [Abstract/Full Text].
25.L. Hayflick, How and Why We Age (Ballantine Books, New York, 1994); L.
S. Gavrilov and N. S. Gavrilova, The Biology of Life Span (Harwood, Chur,
Switzerland, 1991). Contrary to J. F. Fries and L. M. Crapo [Vitality and
Aging (Freeman, San Francisco, 1981)] and R. Dawkins [Sci. Am. 273, 80
(November 1995)], reliability engineering constraints make it virtually
impossible for organisms to approximate the "one-hoss shay" of Oliver
Wendell Holmes, which ran perfectly until one day when all of its pieces
fell apart simultaneously. 26.J. W. Vaupel, in (5), p. 17. 27.A. I. Yashin
and I. A. Iachine, Demography 34, 31 (1997) [Medline]. 28.J. W. Vaupel, A.
I. Yashin, K. G. Manton Math. Pop. Studies 1, 21 (1988); J. W. Curtsinger
and A. A. Khazaeli, Exp. Gerontol., in press. 29.A.J. Lotka, Theorie
Analytique des Associations Biologiques (Hermann, Paris, 1939). The
equation is 1 = <Picture: int >e<Picture: ->rxl(x)m(x)dx, where r is the
intrinsic rate of growth of the population, l(x) is the proportion of
females surviving to age x, and m(x) is the average number of female
offspring to females at age x. 30.S. Orzack and S. Tuljapurkar, Am. Nat.
133, 901 (1989); M. Mangel and C. W. Clark, Dynamic Modeling in Behavioral
Ecology (Princeton Univ. Press, Princeton, NJ, 1988). 31.J. R. Carey, P.
Liedo, H.-G. Müller, J.-L. Wang, J. W. Vaupel, in preparation. 32.G.
Lithgow, T. M. White, S. Melov, T. E. Johnson, Proc. Natl. Acad. Sci.
U.S.A. 92, 7540 (1995) [Medline]; S. Murakami and T. E. Johnson, Genetics
143, 1207 (1996) [Abstract]. 33.M. Tatar, A. A. Khazaeli, J. W. Curtsinger,
Nature 390, 30 (1997) [Medline]. 34.V. Longo et al., in preparation. 35.E.
J. Masoro and S. N. Austad, J. Gerontol. 51A, B387 (1996) ; S. M.
Jazwinski, Science 273, 54 (1996) [Abstract]. 36.C. Franceschi, et al.,
Int. Rev. Immunol. 12, 57 (1995) [Medline]; E. G. Lakatta, Aging 6, 213
(1994) . 37.K. W. Wachter in (5), p. 1. 38.L. Partridge and M. Farquhar,
Nature 294, 580 (1981) ; L. Partridge and P. Harvey, ibid. 316, 20 (1985) .
39.W. Kermack, A. McKendrick, P. McKinlay, Lancet 1, 698 (1934) ; D. J. P.
Barker, Fetal and Infant Origins of Adult Disease (British Medical Journal,
London, 1992); I. T. Elo and S. H. Preston, Pop. Index 58, 186 (1992); R.W.
Fogel and D. R. Costa, Demography 34, 49 (1997) [Medline]. 40.K.
Christensen, J. W. Vaupel, N. V. Holm, A. I. Yashin, Br. Med. J. 310, 432
(1995) ; V. Kannisto, K. Christensen, J. W. Vaupel, Am. J. Epidemiol. 145,
987 (1997) [Medline]. 41.Calculation by I. A. Iachine based on frailty
model described in (27). 42.F. Schächter, et al., Nature Genet. 6, 29
(1994) [Medline]; G. De Benedictis, et al., Hum. Genet. 99, 312 (1997)
[Medline]; J. Maynard Smith, Nature 181, 496 (1958) . 43.A. M. Herskind, et
al., Hum. Genet. 97, 319 (1996) [Medline]; J.W. Curtsinger, et al., Annu.
Rev. Genet. 29, 553 (1995) [Abstract]; C. E. Finch and R. E. Tanzi, Science
278, 407 (1997) [Abstract/Full Text]. 44.G. E. McClearn, et al., Science
276, 1560 (1997) [Abstract/Full Text]; K. Christensen et al., in
preparation. 45.A promising line of inquiry we are pursuing focuses on
lines of medflies (31) and yeast (34) that survive to and reproduce at
advanced ages. 46.Census Bureau International Data Base (updated 10 October
1997), available at http://www.census.gov/ipc/www/idbnew.html; United
Nations Population Division, World Population Prospects: The 1996 Revision,
Annex II and III (United Nations, New York, 1997). 47.J. W. Vaupel, Z.
Wang, K. Andreev, A. I. Yashin, Population Data at a Glance: Shaded Contour
Maps of Demographic Surfaces (Odense Univ. Press, Odense, Denmark, 1998).
48.Death rates are the so-called central death rates calculated by dividing
the number of deaths at the specified age by the years or days of exposure
for the population at risk. 49.Own calculations from data in the
Kannisto-Thatcher Oldest-Old Database and other databases in the Archive of
Population Data on Aging maintained by Odense University Medical School,
Denmark [see (2)], as well as from data in the Berkeley Mortality Database
(http://demog.berkeley.edu/wilmoth/mortality). 50.U.S. data are from the
Social Security Administration. Data on U.S. whites are based on Social
Security data supplied to J.W.V. by the Health Care Financing
Administration. Concerning reliability and calculation methods, see (4, 6);
B. Kestenbaum, Demography 29, 565 (1992) [Medline]; L. B. Shrestha and S.
H. Preston, Survey Method. 21, 167 (1995). 51.H.-G. Müller, J.-L. Wang, W.
B. Capra, P. Liedo, J. R. Carey, Proc. Natl. Acad. Sci. U.S.A. 94, 2762
(1997) [Abstract/Full Text]. 52.T. J. Hastie and R. J. Tibshirani,
Generalized Additive Models (Chapman & Hall, New York, 1990). 53.A. Brooks,
G. J. Lithgow, T. E. Johnson, Science 263, 668 (1994) [Medline]; J. W.
Vaupel, T. E. Johnson, G. J. Lithgow, ibid. 266, 826 (1994) [Medline].
54.Calculations by J.W.V. and C. R. Owens in manuscript on "Automobile
Demography." 55.Our research was supported by the U.S. National Institutes
of Health (grant AG08761), Danish Research Council, Max Planck Society,
Alfred P. Sloan Foundation, and Wellcome Trust. We thank K. Andreev, K.
Brehmer, C. E. Finch, L. G. Harshman, B. Jeune, P. Laslett, H. Lundström,
M. K. McGue, H.-G. Müller, D. Orozco, C. R. Owens, L. Partridge, S. D.
Pletcher, S. H. Preston, D. Roach, R. Suzman, M. Tatar, A. R. Thatcher, S.
Tuljapurkar, N. G. Vaupel, K. W. Wachter, J.-L. Wang, J. R. Wilmoth, and
the Moscamed Program in Metapa, Mexico.
 

Volume 280, Number 5365 Issue of 8 May 1998, pp. 855 - 860
©1998 by The American Association for the Advancement of Science.