Rau et al 2006 — 10 Years After Kannisto Further Evidence for Mortality Decline at Advanced Ages in Developed Countries

Summary

Using the Kannisto-Thatcher Database (KTDB) — population and death counts for ages 80+ across 27 developed countries, 1950–2000 — Rau et al. update and extend Kannisto et al.'s (1994) landmark findings by a full decade. Old-age mortality has not only continued to decline but has done so at an accelerating rate in virtually every country studied. Population counts of the 80+ and 100+ age groups grew in every single country in the study period; annual improvement rates for ages 80–89 roughly doubled from the 1950s to the 1990s for both sexes. No evidence of a biological minimum emerges: the countries with the lowest current death rates show the fastest improvement, directly contradicting a floor hypothesis. The US is a prominent outlier whose stagnation since the mid-1980s "is not yet understood."

Key Claims

1. Population counts (Table 1): Over the study period, the 80+ population across 27 countries grew from 8.9 million to 29.7 million (+3.02%/year average). The centenarian population grew from 8,181 to 93,307 — an 11-fold increase at +6.09%/year. Japan was exceptional: its centenarian population grew at +10.76%/year. Every single country in the study showed an increase in both the 80+ and the 100+ population. This directly refutes Fries's (1980) prediction that "the number of very old persons will not increase."
2. Annual improvement rates accelerating (Table 2, 19 countries): Using a standardized annual improvement metric $\rho$ (computed from successive decade averages of age-standardized death rates, Swedish 1950–1990 population as standard), improvement was monotonically accelerating across all four successive decade pairs for both sexes:
   • Males 80–89: 0.81% → 0.89% → 1.20% → 1.88%/year
   • Females 80–89: 0.91% → 1.35% → 2.00% → 2.45%/year No country in the 19-country Table 2 sample showed deceleration at ages 80–89 across the full period.
3. No biological minimum: Countries with the currently lowest old-age death rates show the fastest improvement rates — the exact opposite of what a biological floor hypothesis predicts. A scatter of 1980s age-standardized death rates (x-axis) against 1990s improvement rates (y-axis) shows no positive slope (which would be expected if countries near a floor were running out of room for improvement). Japan exemplifies this: it has the lowest old-age death rates of any country and simultaneously among the fastest improvement rates.
4. Japan "in a class of its own": Confirming Kannisto et al.'s (1994) phrase, Japan shows the fastest improvement rates at nearly all age groups. Japanese female centenarians grew fastest of all national-age-sex groups.
5. US as outlier: The US had the lowest old-age mortality rates of any country through approximately the mid-1980s. Since then, the US has stagnated while virtually all other high-income countries continued improving. The authors explicitly note this stagnation is "not yet understood." Of the 27 countries in Table 1, 22 showed acceleration in the 1990s vs. 1980s; the US was among the 5 that did not. This finding corroborates and extends Wilmoth's (2005) Human Mortality Database (HMD)-based finding that 11 of 15 high-income countries showed accelerating improvement while the US was one of only 4 deviants.
6. East Germany catch-up: East Germany showed the single largest country-level improvement jump following German reunification in 1990. Women's death rates fell at more than 3%/year in the 1980s-to-1990s decade comparison; men's improvement also surged. This is consistent with rapid convergence of living standards and health system access.
7. Age gradient in improvement: Improvement rates are highest at age 80 and decline monotonically toward age 100. However, this gradient does not indicate a biological minimum — it mirrors the $b_x$ loading structure in the Lee-Carter model, which is highest at younger ages and lower at oldest ages. Crucially, improvement rates at every age group rose across successive decades; there is no crossover where older ages cease improving. For the best-performing four countries, women show acceleration at all ages from 80 to 100+; men show no crossover (i.e., the highest-age improvement does not lag behind lower ages relative to earlier decades).
8. Universality of the pattern: The acceleration of old-age mortality decline is not a statistical artifact of pooling; it appears in almost every individual country. The few exceptions (US, Denmark in some periods, former Eastern Bloc countries before reunification) are interpretable as anomalies, not as evidence for a general ceiling.

Concepts Introduced or Extended

• Period Mortality — primary update; Kannisto extension through 2000 with Table 2 improvement data
• Demographic Transition — Phase 1 accelerating old-age decline confirmed and extended
• Lee-Carter Model — age gradient in improvement rates (highest at 80, declining toward 100) is consistent with $b_x$ structure; does not indicate a biological floor
• Compression of Morbidity — Fries (1980) population count prediction directly refuted: every country showed 80+ and 100+ growth
• SSA Mortality Forecasting — US stagnation documented cross-nationally and over a longer window than Wilmoth (2005)

Entities Mentioned

• Roland Rau
• Domantas Jasilionis
• James W. Vaupel
• Jim Oeppen

Quotes

"Japan is in a class of its own" (confirming Kannisto's original 1994 characterization; Japanese centenarian growth at +10.76%/year is the fastest of any national group)

"the observed stagnation [in the United States] since the middle of the 1980s is not yet understood"

"The countries with the lowest current death rates are not necessarily approaching a floor; they often show the fastest rates of improvement"

My Take

This paper is the strongest available evidence against biological-limit arguments for old-age mortality, precisely because it combines three types of evidence: (1) population count growth in every country (refuting Fries's quantity prediction), (2) monotonically accelerating improvement rates in virtually every country over four successive decades (refuting the "running out of room" argument), and (3) the Japan paradox (lowest rates, fastest improvement — the direct falsification of a floor hypothesis). The US stagnation finding is important for Social Security Administration (SSA) forecasting debates: it shows that the stagnation the Office of the Chief Actuary (OACT) anchors its long-run projections to is a national anomaly, not a biological regularity. The $b_x$ connection (improvement age gradient mirrors Lee-Carter structure) is underemphasized by the authors but is conceptually significant: it explains why the gradient doesn't imply a floor while also confirming the LC model's empirical structure.