1Demographers have long been aware that death rates calculated using statistics derived from vital records (the deceased person’s status reported at the time of death by the proxy informant) as numerator and from census reports as denominator do not always give a reliable measurement of sociocultural mortality differences, notably on account of frequent discrepancies between these two sources. The solution adopted in France in the 1960s and since used in many developed countries involves linking individual data from both sources, in such a way that the content of the sociocultural categories is established from one and the same type of information. In this article, Domantas Jasilionis, Vladimir M. Shkolnikov, Evgueni Andreev, Dmitri A. Jdanov, Dalia Ambrozaitiene, Vlada Stankuniene, France Meslé and Jacques Vallin apply the method for the first time to Lithuania, for which the rare studies that existed until now were limited to small samples. The results show the existence of sharp mortality differentials between social classes and constitute a timely addition to our understanding of social inequalities in Europe essential for monitoring the effects of public health policies.
2Social class differences in mortality have persisted and in many cases even accentuated over recent decades in all the industrialized countries for which studies are available (Valkonen, 2001; Mackenbach et al., 1997). In the FSU (Former Soviet Union) countries, the stagnation in life expectancy that characterized the region for so long seems to have been associated with marked social inequality in mortality. This at least is the conclusion drawn from a number of studies on mortality variations as a function of factors such as educational level, marital status, or ethnicity (Andreev and Dobrovolskaya, 1993; Shkolnikov, Leon et al., 1999; Shkolnikov, Deev et al., 2004; Kalediene and Petrauskiene, 2000, 2005; Shkolnikov, Andreev and Maleva, 2000; Valkonen, 2001; Leinsalu, Vagero and Kunst, 2003, 2004).
3Concerning the Baltic countries in particular, a systematic review of published works dealing with this question was carried out by Vlada Stankuniene, Domantas Jasilionis and Juris Krumins (1999), as part of a study of trends in mortality and causes of death since the 1950s. For Lithuania, four key studies have been conducted on marital status, showing that married people (men and women) enjoy a mortality advantage over those in the never-married, widowed or divorced states, notably for mortality from accidents and suicide (Petrauskiene et al., 1995; Kalediene et al., 1997; Kalediene, 1999; Kalediene and Petrauskiene et al., 1999); and four others on educational level (Kalediene, 1996; Petrauskiene et al., 1996; Kalediene and Petrauskiene, 2000, 2005). We should also mention three comparative studies in which the three Baltic countries are compared on ethnic differences (Zvidrine and Krumins, 1993; Krumins, 1995).
4The findings of such studies have major implications for public health policy, but all are based on rates calculated using numerators and denominators from different sources (records of deaths by age and sociocultural category on the one hand, census data for populations by age and sociocultural category on the other) between which there may be discrepancies. The rates thus risk being heavily biased if the information given by the deceased person’s family or next-of-kin when registering the death diverges substantially from that supplied by the individual himself during the previous census (Vallin, 1980; Levy and Vallin, 1981; Valkonen, 1993, 2002).
5To avoid bias of this kind, the death records of individuals must be linked to the corresponding census data, so that the content of the sociocultural categories on both sides is formed from exactly the same source of information. INSEE was in fact the first to do this on a national scale with a population sample drawn from the 1954 French census, where each individual’s survival was systematically followed using the Registre national d’identification des personnes physiques (national register of personal identities, RNIPP) (Calot and Febvay, 1965). The method has since been adopted, with improvements, in many developed countries, notably the United Kingdom (Fox, 1979; Fox et al., 1985; Goldblatt, 1990) and the northern European countries (Valkonen et al., 1993; Borgan, 1996; Diderichsen and Hallqvist, 1997; Keskimäki et al., 1997; Koskinen and Martelin, 1994; Vägerö and Lundberg, 1995; Valkonen and Jalovaara, 1996). Some studies have even linked individual death records with data in several successive censuses or with other registers, measuring the influence on mortality not only of social status but also of changes in this status (Cambois, 2004). In the early 1980s, the Council of Europe commissioned a Europe-wide synthesis of the results obtained on this question (Lynge, 1984). Recently, the European Union funded a project for a new international synthesis to determine the extent to which social mortality differentials in Europe have widened (Kunst et al., 2001, 2004).
6In the FSU countries, however, studies of social mortality differentials based on linkage between death records and individual characteristics were until recently extremely rare and restricted to small and nationally unrepresentative samples (Plavinski et al., 2003; Shkolnikov et al., 2004). In Lithuania, Juozas Kurtinaitis (2003) studied survival among cancer victims by linking hospital records to the corresponding death records. The same type of procedure has also been applied to Lithuanian data in large-scale international epidemiological surveys like the Kaunas-Rotterdam Intervention Study, MONICA [1] and CINDI [2] (Domarkiene et al., 2003; Radisauskas et al., 2003; Tamosiunas et al., 1999).
7In fact, representative record linkage at a national level has, to date, been undertaken in only three former communist countries of central or eastern Europe. These are Hungary, on deaths in 1980 linked to the same year’s census (KSH, 1987, quoted by Carlson, 1989); Bulgaria, on deaths registered between 5 December 1992 and 31 December 1993 linked to the census of 5 December 1992; and Lithuania, on deaths in the period 2001-2004 linked to the 2001 census. Some results from the Hungarian and Bulgarian studies have been published (KSH, 1987; Kohler, 2001). The more recent Lithuanian study has so far been covered by only two methodological publications (Shkolnikov et al., 2007; Jasilionis et al., 2006). Here we present the first substantive results.
8After briefly reviewing the data and method and comparing the findings with those that would have been reached with aggregate data (section I), we examine the differences in life expectancy obtained as a function of four sociocultural characteristics (section II), and the influence of age- and cause- specific mortality in these differences (section III). We also control for the interaction between characteristics to determine the contribution made by each. Lastly we identify the groups with the highest and lowest mortality risk more precisely when we calculate life expectancies for 24 cross-tabulated categories of the four variables under consideration (section IV).
I – Data and the advantages of linkage
1 – Data and method
9The Lithuania database compiled by Statistics Lithuania that we use here gives exhaustive coverage of population and deaths. The population data come from the individual records of the 6 April 2001 population census, the first to be conducted in Lithuania after the country regained its independence. The data on deaths cover all of the 138,409 deaths of persons over age 30 registered in Lithuania between 1 July 2001 and 31 December 2004. Data linkage was made possible by the personal identification number (PIN) that, in theory, is attributed to all permanent Lithuanian residents. This number was introduced in 1992 to accompany the creation of the population register and the new Lithuanian passport and is used by various government departments. A full 95% of the deaths registered in vital statistics could thus be linked to their corresponding census forms [3]. Unfortunately, for the cases not linked via the identification number, it was not possible to improve on this result by continuing the linkage procedure using other variables: first, the names and addresses of individuals in the census were removed from the Statistics Lithuania files on confidentiality grounds, and second, it was shown that linkage using other criteria (sex, date and place of birth, etc.) would very likely be unworkable. Statistics Lithuania performed the linkage between the individual census records and the vital records (deaths and international migration) in compliance with rules on confidentiality.
10Usual practice in studies of this kind is simply to ignore linkage failures (Kohler, 2001; Doblhammer et al., 2005). However, we observed that when unlinked deaths were excluded, life expectancy at age 30 was overestimated by 1 year for men and by 0.5 year for women (all sociocultural categories combined) (Shkolnikov et al. 2007). It was particularly important to include these deaths in the analysis since the available death registration data showed a strong relation between the variables we wished to examine and an absence of linkage. A problem arises because it is difficult to include unlinked deaths in the analysis without knowing the deceased individuals’ status at the time of the census. For unlinked deaths, therefore, we assigned values for the variables used that took into account the reported values in vital records and a correction model. The estimation/correction procedure is designed to supply information on the distribution of the unlinked deaths by educational level, marital status and ethnicity, comparable to that for the linked deaths. For linked deaths, a double set of information on educational level, marital status and ethnicity – from the census and vital records – is available. From this we derived the probabilities of belonging to a census category depending on the socio-demographic criteria recorded on the death certificate. These probabilities were estimated using a multinomial logistic regression. For the unlinked deaths, the final values for educational level, marital status and ethnicity were determined by means of a random choice method. For further details, the reader is referred to the methodological publication (Shkolnikov et al., 2007). The second source of bias in studies of this kind comes from migration. The problem arising in Lithuania during this period concerns emigration far more than immigration, which is much more limited. The person-years of exposure to the risk include only emigrants reported by the population register. We know that reporting of emigrants to the register is incomplete or delayed (Stankuniene et al., 2002), and that this shortcoming is likely to produce a slight under-estimation of mortality. Conversely, however, we could not easily include immigrants (whose characteristics are not known). Let us hope that these two sources of bias tend to cancel each other out.
11Using the population census data as a basis, the numbers of person-years used for the denominators of the mortality rates were calculated as a function of the length of time elapsed for each individual between 1 July 2001 and 31 December 2004, before emigration or death if that event occurred. The analysis involves a total of 7.3 million person-years.
2 – The advantage of linkage
12Although in certain exceptional cases the differences between the characteristics of an individual reported at the time of death and the corresponding census data are small enough to have no major consequences for the study of differential mortality (Rosamond et al., 1997; Goldblatt, 1989; Marmot and McDowel, 1986), it is more usual to find large differences that create serious bias in rates calculated without individual record linkage. Two US studies (Shai and Rosenwaike, 1989; Sorlie and Johnson, 1996) showed that around 30-40% of individuals were liable to change category during the interval between the census and registration of death.
13The database for Lithuania confirms the existence of large disparities for the variables used here – educational level, marital status, and ethnicity – supplied by the census and by vital records. This has serious implications for the measurement of mortality differences. Concerning educational level, for example, between men with higher education and those with only lower than secondary education, the difference in life expectancy at age 30 is 11.3 years when measurement uses linked records and when the deaths, like the person-years of exposure to the risk, are categorized according to the census data, whereas the difference would be 14.7 years if the mortality rates were calculated by determining the ratio of deaths categorized according to the vital records to the person-years as categorized by the census data. This result is the opposite of that observed for France in the cases where this kind of comparison has been made (Levy and Vallin, 1981; Kunst et al., 1998) but is not an isolated instance since a shift in the same direction seems to have occurred in England-Wales and perhaps in Italy (Kunst and Groenhof, 1996; Kunst et al., 1998).
14Similarly, the difference in life expectancy at age 30 between married and widowed men is only 11.2 years when based on linked records, whereas without linkage it would be estimated at 12.4 years (Shkolnikov et al., 2007). On the ethnicity variable, the difference between Russians and Lithuanians is even reversed since the result with linkage is a disparity of 1.8 years in favour of Lithuanians whereas without linkage the estimated disparity would be 1.9 years in favour of Russians. It is difficult to know exactly how much the differences between census observations and reported information in vital records can be attributed to the attitudes of declarants, and how much to the definition of the categories and the formulation of the questions used for collecting information in the census and in vital records. Whatever the nature of the differences, however, the essential point is that individual data linkage guarantees consistency between the denominator and numerator of the rates whereas without linkage calculation of rates is heavily biased. The downside is that the deceased individual may have changed status between the census and the time of death (notably, in our case, with regard to marital status and place of residence). We accept that our measurement of differential mortality is based on the differences observed at the time of the census and not at the time of death. This shortcoming is of limited significance here, however, since the length of time between the census and the death is only a few months (3-45 months).
II – Difference in life expectancy by selected sociocultural characteristics
15The mortality differences analysed here concern adult mortality only. Since the main aim of the study is to measure the influence of educational level on mortality, it appeared reasonable to focus on mortality after the age when education is completed. The age of 30 years was selected for this purpose.
16On confidentiality grounds, it was not possible to have simultaneous access to all the individual information collected in the census. For this study we were able to work on five sociocultural variables: sex, educational level, marital status, ethnicity, and place of residence [4].
17Educational level was summarized by three categories: higher [5], secondary [6], and "lower than secondary" [7], with the latter category also including the cases where educational level was unknown.
18Marital status comprised four categories: married, widowed, divorced, never-married. Individuals of unknown marital status were very rare (0.1%) and were excluded from the study.
19Ethnicity encompasses the concept of "ethnic origin" as distinct from that of "nationality". Thus a Lithuanian citizen may equally well have a Russian or other non-Lithuanian ethnic origin. A Russian who is naturalized Lithuanian becomes a Lithuanian national but remains of Russian origin. To avoid confusion over these two terms, we use the term ethnicity in preference to nationality. We distinguish between four ethnic groups: Lithuanians, Russians, Poles and "others". Individuals of unknown ethnicity, being very rare (0.4%), were excluded from the study.
20Place of residence is summarized by the distinction between urban and rural, in accordance with the definition given by Statistics Lithuania [8].
21Of the five variables, the most discriminating in relation to mortality is sex. On the other four variables, women in the least advantaged categories have a life expectancy at age 30 equal to and often higher than that of men in the most advantaged categories (Table 1 and Figure 1).
Life expectancy at age 30 and 95% confidence interval by five sociocultural variables, Lithuania, 2001-2004 (in years)
Life expectancy at age 30 and 95% confidence interval by five sociocultural variables, Lithuania, 2001-2004 (in years)
Male and female life expectancy at age 30, by educational level, marital status, ethnicity, and place of residence, Lithuania, 2001-2004 (in years)
Male and female life expectancy at age 30, by educational level, marital status, ethnicity, and place of residence, Lithuania, 2001-2004 (in years)
22Thus, women educated to lower than secondary level have a life expectancy at age 30 of 45.5 years, which is equal to that of men with higher education. The difference by ethnic group is large: life expectancy at age 30 for Lithuanian men, the most advantaged, is only 39.0 years while for Polish women, the least advantaged, it is 46.8 years. The difference by place of residence is even larger (from 39.9 years for urban men to 47.8 for rural women). We note, however, that place of residence is also a weak discriminator for mortality, and that the difference between urban and rural women is particularly small. The difference in life expectancy between urban men and rural women (7.8 years) alone makes up a large proportion of the total difference (10.2 years) between men and women.
23In fact, as all studies show, differential mortality is strongly contrasted on gender lines and analysis needs to treat men and women separately from the outset.
24Figure 1 illustrates for each variable the gradient of life expectancy at age 30 for men and women. Each value for life expectancy is given with its 95% confidence interval. As the study is very exhaustive, the majority of these intervals are very small.
25The observed differences are all much larger for men than for women. For men and women alike, however, some variables appear far more discriminating than others.
26By far the most discriminating variables for men are educational level and marital status, with a difference of over 11 years in life expectancy at age 30 between the most and least advantaged. On marital status there is an extremely sharp contrast between married men and all others: at age 30, a married man has a life expectancy of 41.5 years, whereas a never-married, widowed or divorced man has at best 31.3 years (divorced) and at worst 30.4 (widower). The total difference is roughly the same for educational level – 45.5 years of life expectancy for the highest category and 34.2 years for the lowest – but in this instance the intermediate category (secondary education) is indeed in an intermediate position, albeit closer to the lowest educational category than to the highest.
27On the other hand, and as noted earlier, place of residence is a much less discriminating variable, with a difference in male life expectancy of only 3.5 years, to the advantage of urban residents. The ethnicity variable is only marginally more discriminating, but a clear contrast is observed between Lithuanians (39.0 years of life expectancy at age 30) and Poles (35.9), with Russians in an intermediate position (37.2).
28As noted above, the differences for women are all much smaller than for men. Proportionally, educational level is clearly the most discriminating variable, with a difference of 6.9 years of life expectancy at age 30 between the highest and lowest educational levels while, as in the case of men, women educated to secondary level occupy a roughly midway position. Marital status, on the other hand, appears a less discriminating variable (with a maximum difference of 4.9 years between married and never-married women), though in this instance life expectancy at age 30 is significantly lower for never-married women than for widows (46.5 years) and, especially, divorced women (47.5 years).
29For women, like for men, mortality by ethnic group shows a contrast between Lithuanians and the other categories, with a relatively small overall difference (2.3 years), while the gap between women in urban and rural areas is even narrower (1.5).
III – The contribution of cause- and age-specific mortality to the differences in life expectancy
30The differences in life expectancy between sociocultural categories are of course accompanied by differences in the pattern of mortality by age and cause of death [9]. Various authors have proposed decomposition methods, all based on more or less the same approach, to analyse the role of age- and cause-specific mortality differentials in the difference between two life expectancies (Vallin and Caselli, 2001). Here we use the most recent version of the method developed by Evgueni Andreev et al (2002). The cause of death data were supplied in the form of 29 initial groups that we reorganized into 8 large groups (Appendix 1) to simplify graphical representation and interpretation of the main findings. The results from the decomposition of the differences in life expectancies are reported in Appendix 2.
31Although the number of variables analysed is not large, we tested the results using a multivariate analysis (Poisson regression) to verify that cross-classification of variables was limited and that each contributed specifically to the suggested differences. The table in Appendix 3 gives the excess mortality relative to the reference modality, for each variable of interest, according to two models. Model 1 controls for age only, while model 2 incorporates all the variables (including age). In fact, model 1 serves as a reference, reproducing the results from the bivariate analysis conducted on the basis of life expectancy.
32Despite the very different nature of the indicators, the all-cause excess mortality ratios from model 1 give closely comparable results to those from the analysis of differences in life expectancy. The hierarchy of the modalities is broadly the same for each variable. By contrast, because the excess mortality ratio is an indicator of relative difference, it gives a pattern of cause-specific mortality unlike that obtained by decomposing the disparities in life expectancy into absolute values. The latter measure the contribution from individual causes to the difference in mortality, while the excess mortality ratios allow us to identify the most discriminating causes associated with each variable.
33The most interesting result from this multivariate analysis is the finding that the four variables analysed each make a specific contribution that is largely independent of the other three. The results from model 2 for all-cause mortality are actually quite similar to those from model 1. Even when the two models are compared for each group of causes of death, the specific influence of each variable never diverges greatly from that already indicated by model 1. More specific results on the contribution by each group of causes to the observed mortality differences will be reported in the course of the article.
1 – Educational level
34Life expectancy at age 30 differs by 11.3 years between men with the highest and those with the lowest educational levels, mainly due to the excess mortality of the latter between 40 and 60 years of age (Graph 1, Figure 2). This excess male mortality at ages 40-60 alone accounts for a loss of 5.9 years among the least educated relative to the most educated, i.e. more than half of the total difference, despite the fact that mortality at these ages is much lower than at older ages.
Contribution of age- and cause-specific mortality to differences in life expectancy at age 30 by educational level, Lithuania, 2001-2004
Contribution of age- and cause-specific mortality to differences in life expectancy at age 30 by educational level, Lithuania, 2001-2004
35In terms of the causes of death, the life expectancy difference of 11.3 years is the result of cardiovascular mortality (3.5 years), violent deaths (3.3 years), tobacco-related cancers (1.1 year), respiratory diseases (0.9 years) and alcohol-related causes (0.9 years). Note that cardiovascular disease and tobacco-related cancers contribute more at older ages, while violent deaths and alcohol-related diseases contribute more at younger ages.
36Turning to the distribution of this overall difference between men with the highest and lowest educational levels, by comparing them separately with men educated to secondary level (Graphs 2 and 3, Figure 2), we observe that the same causes at the same ages produce broadly similar effects. For each of these causes, men educated to secondary level are situated midway between men with higher education and men with lower than secondary education.
37Multivariate analysis was used to explore these results in finer detail. Overall, the all-cause excess mortality of men educated to lower than secondary level relative to men with a higher-level qualification (Appendix 3) is not much higher than that of men who completed their secondary education, since the ratio stands at 1.97 rather than 1.53. In other words, after controlling for the influence of the other variables, the transition from secondary to higher education has a substantially larger impact on the all-cause mortality risk than the transition from lower than secondary to secondary level. The contribution of specific causes of death, on the other hand, is sharply contrasted by educational level. The excess mortality of men educated to secondary level does vary between causes but only moderately, whereas for men educated to lower than secondary level it varies widely. Among men with secondary education, the excess mortality ratio is between 1.2 and 1.5 for diseases of the circulatory system and non tobacco-related cancers (and for the "other diseases" category). It stands at around 2 for infectious diseases, diseases of the respiratory system, tobacco-related cancers and alcohol-related diseases, and is close to 1.8 for violent deaths. But a much more spectacular differentiation between causes of death is observed for men educated only to lower than secondary level. In this group, excess mortality from cardiovascular diseases, non tobacco-related cancers and "other diseases" is still close to 1.5 albeit generally above rather than below (between 1.4 and 1.8), but it is considerably higher for infectious diseases (3.7), diseases of the respiratory system (3.8), tobacco-related cancers (3.3), alcohol-related diseases (2.7) and violent deaths (2.3).
38All-cause excess mortality is not markedly higher for men at the lowest educational level than for those who completed secondary education. This is because all-cause mortality is dominated heavily by two main groups of causes of death (cardiovascular diseases and non tobacco-related cancers) whose social class gradient is not steep and which are as discriminating, if not more so, for men of the secondary educational level relative to the highest as they are for men of the lowest level relative to the secondary level. The fact remains, however, that between the latter two categories, the diseases that are most discriminating by social class (infectious and respiratory diseases, tobacco-related cancers) and those that are somewhat less so (alcohol-related diseases and violent deaths) affect men in the lowest educational group much more than men with a secondary education.
39It is worth stressing that the two most discriminating disease groups are infectious diseases and diseases of the respiratory system, rather than the alcohol-related diseases and violent deaths that might have been expected. The latter two groups actually arrive considerably behind tobacco-related cancers, which occupy third place. Also, given that alcohol contributes heavily to excess mortality in Lithuania compared with western countries, we see here that, compared with others, this cause of death is more evenly distributed between the social classes. Tobacco consumption, by contrast, appears to be much more differentiated socially than it was in the western European countries at the time when tobacco-related mortality was increasing rapidly.
40These results support the hypotheses based on surveys on tobacco use revealing higher consumption among men of the lowest educational level (Klumbiene and Petkeviciene, 2002). On the other hand, they contradict survey results showing higher alcohol consumption among men of the highest educational level (McKee et al., 2000) although it is known that responses to questions about alcohol consumption are frequently biased, due to the greater willingness of light drinkers to report their consumption. Some surveys find no association (Klumbiene and Petkeviciene, 2002). Here we see that alcohol consumption, although less discriminating than tobacco consumption, is clearly a factor of excess mortality among men of the lowest educational level.
41The situation for women as depicted in Figure 2 appears markedly different from that for men. On the one hand, the overall difference between the lowest and highest educational level is more influenced for women than for men by the mortality difference at older ages (Graph 4, Figure 2), and it owes more to cardiovascular disease. On the other hand, however, there is wide variation in the age- and cause-specific components of this overall difference when the difference between women of secondary and higher educational levels is compared with that between women of secondary and lower than secondary levels (Graphs 5 and 6, Figure 2). In the former case, the difference in life expectancy is due overwhelmingly to excess mortality at older ages, which stems almost exclusively from cardiovascular disease. By contrast, the central factor in the difference in life expectancy between women with the lowest and with secondary levels of education is excess mortality at young adult ages from non-cardiovascular causes.
42The multivariate analysis confirms this particularity of women by revealing an unexpected result. While the social gradient in female mortality in most western countries is generally much more moderate than that for men, in the case of Lithuania it is on the whole approximately comparable, with an all- cause excess mortality ratio, relative to women with higher education qualifications, of 1.53 for women with secondary qualifications and 1.73 for those with lower than secondary education (Appendix 3). And the differences by cause observed for men are in every case more marked for women. To start with, while the difference in general excess mortality between the secondary and lower than secondary levels is even smaller than for men – respectively 1.53 against 1.73 for women, compared with 1.53 against 1.97 for men – the pattern by individual causes is much more sharply contrasted, both for women with secondary qualifications and for those of the lowest educational level. Among women with secondary level qualifications, excess mortality relative to women of higher educational level is practically non-existent (1.1) for non tobacco-related cancers, which contrasts with values of 3.5 for infectious disease and 2.5 for respiratory diseases. These two cause-groups are thus much more discriminating for women than for men. At the other extreme, among women with secondary education, tobacco-related cancers are responsible for an excess mortality of only 1.3 as against nearly 2 among men. However, their excess mortality from alcohol-related diseases and violent deaths is identical to that for men.
43The range of cause-specific excess mortality observed for women of the lowest educational level, like for men, is more sharply contrasted than that for women with secondary education, indicating that certain causes are socially more discriminating for women than for men. This is true for infectious diseases, for which the excess mortality ratio between the lowest and highest educational levels is 4.8 (against 3.7 for men). But it is true also for alcohol-related diseases (3.4 against 2.7) and violent deaths (2.5 against 2.3). On the other hand, some diseases that are strongly discriminating for men are less so for women. This is the case for diseases of the respiratory system (3.2 against 3.8) but the contrast is even more spectacular for tobacco-related cancers (1.5 compared with 3.3). This last result confirms findings from surveys on tobacco consumption showing a social contrast that is large for men while statistically non-significant for women (Klumbiene and Petkeviciene, 2002).
44Considered with regard to the specific effect of educational level, therefore, the social class differences in female mortality in Lithuania are surprising by their scale and even more by the highly discriminating character of certain cause groups.
2 – Marital status
45Even more so than for educational levels, it is differences in mortality at young adult ages that account for most of the life expectancy differentials at age 30 between married and never-married men (Figure 3). More specifically, the excess mortality of never-married men aged 40-64 alone accounts for 7.4 years of the total difference of 10.8 years in life expectancy at age 30. The excess mortality at the oldest ages accounts for only 1.8 years. With respect to causes, the largest contributor is violent deaths and alcohol-related causes (4.3 years), closely followed by cardiovascular diseases (3.9 years). Respiratory diseases (0.7 years) and infectious diseases (0.6 years) account for most of the remainder, with only 0.4 years from tobacco-related cancers. Here too, of course, the influence of violence and alcohol-related causes is greatest at the young adult ages, while cardiovascular diseases predominate at older ages.
Contribution of age- and cause-specific mortalities to the difference in life expectancy at age 30 between married and never-married men, Lithuania, 2001-2004
Contribution of age- and cause-specific mortalities to the difference in life expectancy at age 30 between married and never-married men, Lithuania, 2001-2004
46The multivariate analysis confirms that certain causes are much more discriminating than others. With an excess mortality ratio of 4.8, infectious diseases form by far the most discriminating cause-group for never-married men. Alcohol-related diseases (3.3) come second, followed by respiratory diseases and violent deaths (2.4) in joint third place. The ratio is just below 2 for cardiovascular diseases which, because of their place in total mortality, are thus a major cause of overall excess mortality. On the other hand, cancer mortality, whether tobacco-related or not, has a limited effect. Indeed, for non tobacco-related cancer mortality, never-married men even show a slight advantage over married men, although this difference is not statistically significant.
47As noted earlier, the differences in life expectancy between never-married, widowed and divorced men are only minor, and little will be gained here by giving the detailed breakdown by causes. We simply point out that the most discriminating causes seem to vary by marital status: infectious diseases for never-married men, alcohol-related diseases for divorced men.
48Distinguishing between the four marital states is more interesting in the case of women. Mortality at the oldest ages has a proportionally larger effect in the overall difference between married and never-married women – the two extreme categories – than it does for men (Graph 1, Figure 4). The total divergence observed is 4.9 years, of which 2.0 are attributable to excess mortality of never-married women aged 60 or over. And, quite logically, excess cardiovascular mortality, dominant in old age, is alone responsible for more than half of the total difference (2.6 years). Note, however, that many other causes play a role at young adult ages.
Contribution of age- and cause-specific mortality to differences in female life expectancy at age 30 between pairs of marital states, Lithuania, 2001-2004
Contribution of age- and cause-specific mortality to differences in female life expectancy at age 30 between pairs of marital states, Lithuania, 2001-2004
49Another major difference compared with men is the existence of substantial disparities between the three categories of non-married women, with divorced and widowed women occupying quite distinct intermediate positions between married and never-married women. Note that the differences between these intermediate positions reflect mortality patterns by age and, most important, by cause of death that are also quite distinct.
50Mortality across almost all age groups contributes in roughly equal proportions to the difference in life expectancy between married and divorced women (Graph 2, Figure 4). The proportion of this difference due to cardiovascular diseases is considerably smaller than for that between married and never-married women, while other causes play a substantial role up to relatively advanced ages.
51By contrast, excess mortality at younger ages, principally from violent deaths or alcohol-related mortality, is responsible for the difference between divorced and widowed women (Graph 3, Figure 4).
52Last, for widows and never-married women, the excess mortality levels are reversed between young adult ages and older ages. Before age 50, mortality is higher among widows than among never-married women, and is generated mostly by violent deaths and alcohol-related causes. At ages over 60, by contrast, mortality is distinctly higher among never-married women than among widows, and is due mostly to cardiovascular diseases.
53Thus it is seen that the divergence in female life expectancy between marital states results from a wide variety of causes. The mortality of widows, in particular, probably differs from that of other women (i.e. married, divorced or never-married) partly because the conditions of their husband’s death may be a factor linked to their own death. The classic example is that of the traffic accident in which the wife dies some time after her husband from longer-term consequences. More generally, however, the same behavioural factors that led to a husband’s early death may also affect the wife. This is true notably for alcoholism.
54The multivariate analysis completes the picture in two respects. Firstly, it highlights the more discriminating role of cancers, and notably tobacco-related cancers, among women than among men. Secondly, all the other causes appear less discriminating for women than for men, since for none of them does excess mortality exceed 2.5 (Appendix 3).
3 – Ethnicity
55The causes of death that contribute to divergences in life expectancy at age 30 differ slightly between, firstly, Lithuanians and Poles, and secondly between Lithuanians and Russians (Figure 5). Whereas violent deaths and alcohol-related causes play an important role at young adult ages in the difference between Lithuanians and Poles, their role in that between Lithuanians and Russians is much more modest. Given what is also known about the role of alcoholism in violent deaths, especially at young adult ages, it can be inferred that the differences in life expectancy between Lithuanians and Poles are due largely to alcoholism.
Contribution of age- and cause-specific mortality to differences in male life expectancy at age 30 between Lithuanians and Poles or Russians, Lithuania, 2001-2004
Contribution of age- and cause-specific mortality to differences in male life expectancy at age 30 between Lithuanians and Poles or Russians, Lithuania, 2001-2004
56This pattern is less clear-cut among women, however, for whom cardiovascular diseases play the principal role at older ages (Appendix 2).
57The large differences by ethnicity might be considered surprising. To some degree, this is in fact related to differences in distribution by educational level and urban or rural place residence. In particular, the majority of Poles live in the rural regions of eastern Lithuania and in Vilnius (administered by Poland in the interwar period), while the Russian population resides mainly in the cities (Vilnius, Klaipeda). At the same time, the Russians have a higher educational level than the Lithuanians who, in turn, have a higher level than the Poles. However, the multivariate analysis (Appendix 3) shows that even when ethnicity is measured independently of educational level and place of residence, Russians and Poles have excess mortality with respect to Lithuanians of, respectively, 1.23 and 1.19 for the men and 1.23 and 1.17 for the women. This means that the difference in life expectancy between Russians and Lithuanians is narrowed by the higher educational level and more frequent urban residence of the Russians compared with the Lithuanians. Conversely, between Poles and Lithuanians the difference is increased. Other factors more specifically related to ethnicity are involved. For example, the Baltic Nutrition and Health Survey shows that whereas Russian men smoke more than Lithuanians and the others (chiefly Poles), alcohol consumption is higher among the "others" (mostly Poles) for men and among the Russians for women (Pudule et al., 1999; McKee et al., 2000).
4 – Place of residence
58The difference in life expectancy differentials between urban and rural populations is much smaller than those examined so far, but the excess mortality responsible for this differential has a most interesting age and cause structure (Figure 6). For men it is summed up in a simple pattern: a dominant proportion of violent deaths that alone produce a difference in favour of urban dwellers of 1.6 years out of a total of 3.5 years, although this includes no visible role for alcoholism. It seems that excess mortality from violence is, for once, largely unrelated to alcoholism, which appears to affect urban and rural dwellers without distinction. The external causes involved are probably more those associated with road accidents or agricultural activities as well as suicides. Alongside the effect of excess violent mortality concentrated chiefly in the young adult ages, we observe an effect of excess cardiovascular mortality spread fairly evenly across all age groups and the influence of excess mortality from respiratory diseases, also nearly uniform across ages.
Contribution of age- and cause-specific mortality to differences in life expectancy at age 30 between urban and rural dwellers, Lithuania, 2001-2004
Contribution of age- and cause-specific mortality to differences in life expectancy at age 30 between urban and rural dwellers, Lithuania, 2001-2004
59In a much less pronounced form, the same pattern is observed among women.
60Overall, the multivariate analysis confirms that the excess mortality of rural over urban populations (1.12 for men, 1.06 for women), although small, is statistically significant (Appendix 3). It is produced by the contrasting effects of causes responsible for appreciable excess mortality – such as diseases of the respiratory system (1.6 among men, 1.4 among women) and violent deaths (1.4 and 1.3) – and of other causes that produce larger mortality advantages than in the rare cases observed for the other variables ("other causes" and alcohol-related diseases).
61Men and women differ on one major point, however. Male excess mortality from tobacco-related cancers contrasts with a female mortality advantage that is larger than those observed for "other diseases" and alcohol-related diseases mentioned already. This contrast, rather than the higher male excess mortality from respiratory diseases and violent deaths, is responsible for the overall difference between men and women.
62For women, the latter result confirms the findings of the Baltic Health and Nutrition Survey, which shows that women smoke less in rural areas than in urban areas (Pudule et al., 1999; Puska et al., 2003).
IV – The categories most exposed to risk
63An alternative way to determine the effect of each variable involves cross-tabulating all the variables and calculating the life expectancies at age 30 for each group of cross-tabulated variables. In view of the results obtained already, however, we conduct this analysis using a simplified version of our database. We reduced the number of modalities for the four variables to keep only the most influential. Thus, marital status was reduced to two modalities (married and non-married), as too was ethnicity (Lithuanians and non-Lithuanians). A total of 3 (educational levels) X 2 (ethnic groups) X 2 (places of residence) = 24 cross-tabulated categories were selected, for which life expectancy was calculated separately by sex (Table 2).
Social gradient of life expectancy at age 30 according to cross-tabulations of four variables, Lithuania, 2001-2004
Social gradient of life expectancy at age 30 according to cross-tabulations of four variables, Lithuania, 2001-2004
64Figure 7 presents the different cross-tabulated categories in rising order of life expectancy for each sex. Each category has a length along the horizontal axis proportional to its share in the population. For each category, therefore, the line representing life expectancy is proportional in length to the relative size of the category, and its position is determined by the category’s level and rank. Finally, on the trajectory thus plotted, we have provided a few benchmarks for comparison by indicating the mean life expectancy levels at age 30 observed in Lithuania and in three other industrialized countries – Russia, the Czech Republic, and Japan – that lie along the range of life expectancies.
Social gradient in life expectancy at age 30 according to four cross-tabulated variables, Lithuania, 2001-2004 (in years)
Social gradient in life expectancy at age 30 according to four cross-tabulated variables, Lithuania, 2001-2004 (in years)
1 – Men
65The men with the highest life expectancy at age 30 (47.7 years) are in the highest educational group, married, Lithuanians and urban dwellers. The gap relative to the most disadvantaged category – men educated to lower than secondary level, non-married, non-Lithuanians and rural dwellers (27.3 years) – is 20.4 years. Note, however, that while the former category is numerically large (9% of the population), the latter is marginal (0.7%). To obtain a share of population at the bottom of the scale equal to that in the most advantaged category, it is necessary to join categories 24, 21, 22 and 16 to category 23, which together then contain 9% of the population. Life expectancy at age 30 does not exceed 30 years for any of these categories individually.
66Expressed as a weighted average, life expectancy at age 30 for this group of the most disadvantaged categories is 28.6 years. While one might wish to exclude the extreme cases where the sub-populations are too small, the gradient of over 19 years in Lithuanian male life expectancy between the upper and lower deciles is undeniable. This is a very steep gradient. Interestingly, however, it is not much greater than the distance between average national life expectancy (16.5 years) in the two industrialized countries that are furthest apart in this respect, i.e. Japan (49.3 years) and Russia (32.8 years). Does this illustrate the importance of downplaying the sociocultural inequality in mortality among Lithuanians, or on the contrary of taking the full measure of the distance that separates the Japanese from the Russians? That is not the subject of the present article, but these points of comparison offer material for reflection. We observe here that the range of Lithuanian inequalities roughly covers that of international differences, thus confirming the intermediate position of Lithuania in the industrial world’s league table of adult life expectancy. But we also note that the most advantaged category is still substantially below the average for Japan. In other words, for an adult man, it is better to be an average Japanese than to belong to the most advantaged group in Lithuania (on the four criteria selected here). Conversely, the lowest decile (for which life expectancy as a weighted average is 28.6 years) is quite a long way (over 4 years) below the average for Russia. The extreme group, for its part, is almost 6 years below the average for Russia, with a life expectancy at age 30 of 27.3 years.
67But let us return to the differences within the population of Lithuania. It is interesting to note that the categories forming the bottom decile of the scale are alike in being composed of non-married men, usually non-Lithuanians, none with a higher education qualification but of equally urban or rural residence. Marital status clearly emerges as the most discriminating criterion among men: non-married men are absent from practically all of the ten categories where life expectancy at age 30 exceeds 39 years [10]. A life expectancy of 39 years is roughly midway between the two extreme deciles, and is also a life expectancy level that divides the population into two large groups (65% above, 35% below). Conversely, there are no married men in any of the eight most disadvantaged categories where life expectancies are all shorter or equal to 33.1 years.
68Educational level seems to be a less discriminating factor. At the top of the hierarchy, having a higher level qualification is particular to only four categories, while at the bottom of the scale, the lack of a secondary level qualification distinguishes only four categories (compared with eight for marriage). In large part, however, this is because we kept three levels for education whereas for the other variables the number of modalities was reduced to two. If we had used only two levels for education also, it is highly likely that this variable would have been nearly as discriminating as marital status, especially if a division had been established inside the secondary category, whose composite structure seems to cause some clouding of the picture here. It must also be emphasized that of the four variables used, marital status is probably by far the most influenced by selection mechanisms, since marriage (or remarriage) is more difficult for persons with health problems than for others. Selection effects may also operate with respect to education, but certainly to a lesser extent.
2 – Women
69The situation for women is rather different. The range of inequalities is considerably narrower, since life expectancy at age 30 climbs from 40.0 years among women with no secondary qualification who are non-married, non-Lithuanian and in rural locations, up to 55.8 years among those with higher education who are married, non-Lithuanian and in urban locations (compared with 27.3 and 47.7 years among men). Between the extremes, therefore, the difference is only 16 years, compared with over 20 years for men. The smaller difference among women is unrelated to the fact that the two extreme categories contain only marginal fractions of the population (1.5% and 1%), since the interdecile range is only 12.1 years (54.7 – 42.6 years) for women compared with 19 for men. We note also that life expectancy in the most disadvantaged female decile is only 5 years lower than in the most advantaged male decile, whereas in the most advantaged female decile it is 7 years higher. More than two-thirds of the female population are thus in sociocultural categories where life expectancy at age 30 is higher than for the most advantaged male category.
70As the difference between men and women is largely sociocultural, the range of sociocultural mortality differences in Lithuania, on five variables (sex included) in combination may be said to extend from 27.3 years to 55.8 years, i.e. the extremes of life expectancy at age 30 are nearly 29 years apart. Even if the comparison is restricted to that between deciles – or more exactly to between the deciles for each sex separately, which is practically equivalent to an analysis by vigiciles (5% of the total population) – the difference is still 26 years (from 28.7 years to 54.7 years). In other words, life expectancy at age 30 for the most advantaged 5% is nearly double that for the least advantaged 5%. And of all the criteria, sex is of course the most discriminating.
71But let us again consider women in isolation. Like for men, a comparison of the Lithuanian sociocultural gradient with the Russian, Czech and Japanese national averages is eloquent and the same commentaries apply as those made earlier with respect to men.
72On the other hand, the roles of the different variables in the sociocultural spectrum of female mortality diverge appreciably from what was said for men. Marital status is less important for women. The "married" modality distinguishes only four categories at the top of the hierarchy (compared with five for men) and five cross-tabulated categories at the bottom of the scale (compared with eight for men). Educational level is at the forefront this time, although far less so than marital status is for men. Of the nine most advantaged categories, seven concern women with a higher education qualification and the other two have women with secondary level qualifications. Conversely, of the seven least advantaged categories, five have not obtained a secondary qualification and the other two have no higher qualification. However, because educational level is a variable with three modalities, the division is not clear-cut. Here again, the solution would probably have been to construct a dichotomous variable based on a distinction between secondary level qualifications.
Conclusions
73Despite the small number of variables available, this approach to social mortality differentials in Lithuania based on linkage of death certificates to census data reveals with great accuracy the extent of the mortality differentials observed in the country. This result also reflects the exhaustive nature of the operation, given the small size of the population.
74The multivariate analysis confirms the strongly significant nature of all the differences revealed by the bivariate analysis. It does however enable us to qualify certain results. For instance, where the bivariate analysis yields a gradient of life expectancies that – unsurprisingly – follows educational level to the advantage of better-educated persons, the multivariate analysis shows the effect of education to be much stronger between the higher and the secondary levels than between the secondary and the lower than secondary levels.
75Also, a bivariate analysis based on decomposition of the differences in life expectancy according to the contribution from each group of causes, combined with a multivariate analysis of relative mortality differences, shows clearly that while some causes contribute heavily to the differences through their role in total mortality, they are not necessarily the most discriminating. Thus, cardiovascular diseases and violent deaths contribute most to absolute differences in life expectancy, yet whereas excess mortality from the former is low, from the latter it is high. Excess mortality is higher still, however, for the less common causes, such as infectious and respiratory diseases, whose effect in absolute terms is slight.
76In sum, the sociocultural mortality differences observed here are generated mostly by cardiovascular diseases on the one hand, and by social pathologies (alcoholism, smoking, and violent deaths) on the other.
77Last, the observed differences are already large when the variables analysed are considered separately, notably for educational level and marital status though also for sex, but they assume gigantic proportions when combinations of variables are compared, so much so that 26 years separate the life expectancies at age 30 between the two extreme deciles.
78Acknowledgements: This study was conducted as part of the Vanguard Project of the Max Planck Institute for Demographic Research in Rostock, Germany. We are very grateful to Olga Trofimova of Statistics Lithuania, for her work on the initial version of the database of deaths linked with the census. We would also like to thank Danguole Svidleriene (head of the Demographic Statistics Division of Statistics Lithuania) for providing us with her expert knowledge of Lithuanian mortality statistics.
Appendix 1 – Groups of causes used for the decomposition of life expectancy differences
Appendix 2 – Contributions of the 8 groups of causes of death to differences in life expectancy at age 30, Lithuania, 2001-2004
A. Difference by educational level
A. Difference by educational level
B. Difference by marital status
B. Difference by marital status
C. Difference by ethnicity
C. Difference by ethnicity
D. Difference urban area – rural area
D. Difference urban area – rural area
Appendix 3 – Results of Poisson regression on four variables (excess mortality ratios for population aged 30+)
Notes
-
[*]
Max Planck Institute for Demographic Research, Rostock.
-
[**]
Statistics Lithuania, Vilnius.
-
[***]
Demographic Research Center, Institute for Social Research, Vilnius.
-
[****]
Institut national d’études démographiques, Paris.
Translated by Godfrey Rogers. -
[1]
Multinational MONItoring of trends and determinants in CArdiovascular disease.
-
[2]
Countrywide Integrated Noncommunicable Disease Intervention.
-
[3]
The 5% of unlinked deaths can be broken down into two categories: first, deaths for which the personal identification number (PIN) was missing from the death certificate; second, deaths for which no census form corresponding to the PIN given on the death certificate could be found in the census files, with no way of knowing whether this resulted from the lack of a PIN on the corresponding census form or the absence of a census form, the latter case itself being due to an omission by the census or to immigration.
-
[4]
The data source does not allow us to use a larger number of variables. However, in Lithuania, educational level appears to be the most robust indicator of socioeconomic status (Katsiaouni, Gorniak and Lazutka, 2000). Moreover, only three educational levels could be used for the study (higher, secondary and lower than secondary) due to changes in the schooling system in the twentieth century, from the Polish or Lithuanian system in the interwar period, to the Soviet system and finally the current Lithuanian system (Shkolnikov et al., 2007). This is how the problem was resolved for Finland (Valkonen et al., 1993). The other socioeconomic indicators are problematic in Lithuania, where people are reluctant to report their income, their occupation or their employment status, and where even data distinguishing simply between manual and non-manual workers were not available to us.
-
[5]
Graduation from university or equivalent higher education after at least 14 years of schooling from the first year of primary education.
-
[6]
Graduation from general or vocational secondary education (high school, middle school, technical school) after at least 10 years of schooling from the first year of primary education.
-
[7]
Less than 9 years of education (incomplete secondary, primary, basic vocational, etc.), or illiterate or level of education unknown.
-
[8]
The urban population comprises, first, the population of built-up localities with over 3,000 inhabitants where over two thirds of the economically active population work in the industrial, commercial, craft or service sectors, and second, the population of localities with under 3,000 inhabitants which acquired official "town" status before 19 July 1994, when the law on administrative units and their boundaries came into force (Statistics Lithuania, 2004).
-
[9]
In Lithuania, as elsewhere, the quality of the diagnosis of cause of death may vary between sociocultural categories. However, the WHO considers the Lithuanian cause-of-death statistics to be of good quality (Mathers et al., 2005). We believe in addition that the centralized coding system for causes of death and the relatively high proportion of autopsies (30%) reduces the risk of inconsistency in the quality of observation. This is all the more true given the large groups of causes of death that we use here.
-
[10]
With the never-married exception of non-married, rural, non-Lithuanian men with higher education, who form an extremely marginal group (22 deaths among 1,300 persons) for which the confidence interval is very large (37-47 years).
