Visualization of sex-specific differences in survival Acknowledgements

6. Conclusions

The present report demonstrates how intensity regression techniques can be applied to study age- and period-specific differences in survival between males and females in Canada. In the previous section I have also shown how the same phenomenon can be studied with the help of visualization techniques. A brief illustration of the character of the results obtained by both methods is presented as well.

In this particular case the visualization techniques perform exceptionally well. The trends in mortality are apparent and the sex differences are clearly revealed. There are two important factors which contribute equally to the success of this technique:

  • The data set includes only three covariates and it can be easily separated into male and female mortality maps.
  • The data represent the entire population of Canada. This makes it possible to compute reliable, direct estimates of the death rates, so the plots will not be disturbed by a high degree of stochastic noise.

If one or another of these conditions is not fulfilled the application of the visualization methods will be less appealing.

The intensity regression techniques allow us to conduct a more detailed analysis and to test the statistical significance of the estimates. Some important details can be revealed by intensity regression, which would be virtually invisible on the Lexis maps. For example, it is evident from Figures 2a and 2b that infant and childhood mortality increased significantly in 1937, but on the Lexis maps (Figures 4 and 5) this fact hardly attracts attention. In sum, intensity regression is a means of summarizing information in the data set in a concise, numeric manner whereas the Lexis maps provide a visualization of the entire data array.

The analysis conducted here shows that the choice of analytical method should be driven not only by the information present in the data set (occurrences and exposures) but by the structure of the data as well. If the structure is very simple, as was the case for what we have analyzed here, the visualization techniques might prove to be as useful as regression techniques.

Also I would like to note that the age-specific differences in Canadian survival (Fig. 6) are strikingly similar between those found for other countries [Andreev 1999]. There are still some differences between the Lexis maps but there are far more similarities to be observed. It might be the case that there exist certain factors like smoking prevalence that affect mortality in some uniform way, thereby maintaining a more or less fixed pattern of male-female differences over various countries.

Excess of male mortality over the last decades is a well-known phenomenon and a number of hypotheses have been put forward as possible explanations [Nault 1997, Waldron 1983, Waldron 1993]. These hypotheses are based mainly on analysis of cause-specific mortality, social and behavioral differences, risk factors prevailing in male and female populations, and inherent biological and genetics differences. Complete analysis of observed differences in survival between sexes in Canada is well beyond the scope of this paper, so I limit myself to a brief overview of possible explanations.

At adult ages significantly higher male death rates are to be observed from cardiovascular diseases (CVDs) and cancers. Death rates from CVDs declined over time both for males and females but the pattern of the decline was different between sexes. If we return to Figure 6, the maximal relative differences are found in the years around 1980 and at ages 50-75. Emergence of this gap in death rates between males and females is due in a large part to developments in cardiovascular mortality, especially ischaemic heart disease. For example, at ages 60-64, male death rates stayed at a level of about 1300 deaths per 100,000 over the period 1950-70 and only after that period started to decline [Note 6]. For females, however, the death rates declined almost in linear fashion over the years 1950-97: the death rates fell from the level 800 deaths per 100,000 at the beginning of 1950s to 200 in the middle of 1990s. As a result, the sex ratio of mortality from this cause of death significantly increased over 1950-80, then stagnated at a level about 2.7, and finally dropped slightly in the 1990s.

On the other hand, trends in cancer mortality were very different from those observed for CVDs. For females, the level of mortality was more or less constant over the years 1950-97 while male death rates were increasing up to the year 1990 and then sharply declining, thereby reducing the gap between male and female mortality. The relative difference between sexes is less pronounced compared to CVDs. Over the 1980s male death rates were only 1.5 times higher than females. This analysis of cancer mortality indicates that a change in this cause of death is the main driving force behind the convergence of male and female death rates during the 1990s. Especially unfavorable trends are to be observed in the female death rates due to lung cancer. The death rates have been persistently increasing since the late 1960s while male death rates have been declining since the late 1980s. As a result, the huge excess of male mortality (over the years 1950-70 male death rates were about 8 times higher than female at ages 55-75) has been significantly reduced and now male death rates are approximately only two times higher than female death rates.

At young and young adult ages (15-35) the excess of male mortality during recent decades is mostly due to accidental causes. Death rates from suicide and traffic accidents were considerably higher for males than for females. On the contrary, prior to the 1950s female death rates were in fact higher than male, which is probably mostly attributable to maternal mortality and complications in childbearing. Improvements in health care system significantly reduced this component of female mortality. At the beginning of the 1950s an excess of female mortality completely disappeared.

We can see that analysis of trends in cause-specific mortality can highlight certain differences in survival between males and females but unfortunately it cannot provide a complete explanation of the observed differences in survival. More analysis will require investigation of risk factors prevailing in the male and female populations and their relation to mortality. Most researchers agree that smoking provides one of the main contributions to the observed mortality differences between both sexes and countries. However, the relation of this risk factor to trends in death rates is less clear. In Canada, for example, smoking rates have declined over the period 1977-95 for both males and females, but the level has always been higher for males than for females [Millar 1996]. Trends in lung cancer mortality, however, were quite different. For males they rose until the late 1980s and then started to decline for virtually all ages from 50 to 75. For females, however, the rates have been increasing persistently since 1970. It could be that smoking affects females more than males so the relation between this risk factor and trends in lung cancer is not straightforward.

On the other hand, females have always had lower mortality from CVDs and trends in this cause of death were similar to those observed in the male population. This contradicts the fact that smoking is also a risk factor for CVDs. Lower female rates from CVDs can be partially explained by the protective effect of female sex hormones and better utilization of the health care system: genetic and social factors, respectively [Waldron 1983, Waldron 1993]. This demonstrates that the observed sex differences in survival are affected by a variety of causal factors. Their influence on mortality still calls for explanation.

 

Visualization of sex-specific differences in survival Acknowledgements

Sex differentials in survival in the Canadian population, 1921-1997: A descriptive analysis with focus on age-specific structure
Kirill Andreev
© 2000 Max-Planck-Gesellschaft ISSN 1435-9871
http://www.demographic-research.org/Volumes/Vol3/12