Abstract Data: Population and Deaths

1 Background

In Canada, as in many other industrialised countries, half of all female deaths and a third of male deaths occur at advanced ages (80 years and over) [35]. The evolution of this phenomenon thus plays a significant role with respect to increases in life expectancy and the intensity of population aging. Our paper stems from the wider debate regarding the possible limits of human longevity and the evolution of mortality among the oldest-old [36] [37].

For a number of reasons, certain researchers are doubtful that survival probabilities at advanced ages can go on rising and expect life expectancy at birth to reach a ceiling at 85 years. They consider death over the age of eighty to be an inevitable, and natural, consequence of the aging process [4] [13] [14] [31] [32]. More optimistic writers have demonstrated, using simulation techniques, that life expectancy at birth could reach much higher levels; a mortality reduction of 2% at all ages over a long period of time would lead to a life expectancy at birth of 100 years for women and of 95 years for men [1]. This optimism is shared by other researchers who speculate that risk factors could gradually be eliminated by adopting more healthy lifestyles and by advances in medical biotechnology [28] [40].

Their hypotheses were not necessarily based on observed data due to imperfections in mortality statistics at advanced ages. These shortcomings have been partly overcome by researchers at the University of Odense (Denmark), who created a large database of mortality statistics at advanced ages from thirty-one countries known to have reliable data (Kannisto-Thatcher Oldest-Old Database); nonetheless, even in this sample of developed countries with fairly low levels of general mortality, data quality was relatively variable. In their study, still quoted in recent papers (such as [39]), Canadian and American data were classified as being of inferior quality, both in terms of their availability, and because of false declarations of age – in death registration and in census data [20] [21]. We must add that for Canada, the data used in the European study were limited to those published (data with an open age group at 90+, for example), which did not give the real picture in terms of quality of data and measurement of mortality. For this paper, we have obtained more detailed data that will permit a better assessment of data quality.

Other groups of researchers closely examined data on mortality at advanced ages from countries with reliable data sources, though with particular reference to the United States [7] [8] [9] [23] [24] [42]. These researchers established the following results:

  • Mortality at advanced ages has decreased since the beginning of the century, and most notably since 1950 [21].

For the majority of countries investigated, i.e. countries with high quality data, mortality rates at 80-99 years fell dramatically between 1960 and 1980. The annual improvement in mortality levels was more pronounced for the female population than for males, between 1 and 2%, and 0.5 and 1.5%, respectively.

  • The rate of increase of mortality above age 80 declines with age [7] [18] [19].
  • Mortality rates among the oldest-old are often understated in countries where death registration is imprecise and, in general, these age-at-death errors explain a large part of the mortality crossover [7].

In other respects, studies of North American data show that the United States has lower mortality levels than those of most European nations. The reason for this is not obvious, though three possible sources are worthy of mention: the inferior quality of the data [7] [8] [12]; the greater homogeneity of the elderly population in the USA with respect to health status [7] [17]; and a relative advantage for the elderly in terms of income, education and health-care [2] [17].

These studies have led to the introduction of the idea of a "North American profile", specific to the mortality patterns of the oldest-old. Canada has been assimilated into this profile without any real knowledge of the Canadian situation. However, a study has demonstrated that Canadian data had errors of age declaration similar to those of other English-speaking nations (United Kingdom, Australia and New-Zealand) [9].

In fact, the Canadian life tables are the source of almost all we know about Canadian mortality at advanced ages [3]. But do these life tables reflect accurately the mortality of the oldest-old in Canada? Figure 1 shows the probabilities of dying (qx) for men and women between age 80 until the end of the life table for each census year from 1951 to 1991 (Canadian Life Tables). First, it is clear that probabilities of dying at advanced ages rise smoothly with age; this is not surprising since they are produced by graduation methods that generate very smooth curves. According to Canadian life tables, mortality seems to have decreased since 1951, although there are some crossovers explained by changes in the method of estimation employed (as in 1980-1982) which may produce fairly surprising results. These probabilities of dying, however, cannot be compared to data from other countries since the methods are not the same. We propose, therefore, to apply the method of extinct generations to Canadian data in order to reconstitute a population that will provide us with a more precise denominator for calculating probabilities of death at old and extreme ages.


In this paper, our aim is to assess the Canadian situation, by evaluating the quality of mortality and population statistics, by estimating mortality levels with the aid of methods employed by Kannisto [21] (Kannisto-Thatcher Oldest-Old Database) and by Wilmoth [43] (Berkeley Mortality Database), and by clarifying the debate on the existence or otherwise of a North-American mortality profile [Note 1].

More precisely, we will try to answer the following questions: Are Canadian data really bad in measuring oldest-old mortality? And is Kannisto [20] right about the quality of Canadian data? We will use some indicators suggested by Coale and Kisker [8] and by Kannisto [22] to evaluate the quality of data in Canada, using more detailed data. If the various indicators show the same problems as previously found in other studies, it will confirm the bad quality of Canadian data. If not, we will get a better estimation of Canadian mortality among the oldest-old.


Abstract Data: Population and Deaths

logo70.gif (2450 bytes)

Mortality Statistics for the Oldest-Old: An Evaluation of Canadian Data
Robert Bourbeau and André Lebel
© 2000 Max-Planck-Gesellschaft ISSN 1435-9871