Why is Mortality Low among the Swedish-Speaking Minority in Finland?

Life expectancy has been substantially longer among the Swedish-speaking Finns than in the rest of the population. The relative mortality difference appears to be particularly marked among the middle-aged. This study examines the possible reasons for this mortality difference. The mortality advantage of the Swedish-speaking Finns is connected with their more favorable geographic location and socioeconomic position. For women these factors explain all of the mortality difference but among men two-thirds of the original difference, a 20% excess mortality of the Finnish-speaking majority, persists after adjusting for the structural differences. In men, the main part of the mortality difference results from factors responsible for excess mortality of the Finnish-speaking population from cardiovascular diseases and non-natural causes of death. A similar mortality contrast is seen in women as well, but it is compensated by other causes of death which are more common among Swedish-speaking than among Finnish-speaking women.

The mortality advantage of the Swedish-speaking Finns is connected with their more favorable geographic location and socioeconomic position.For women these factors explain all of the mortality difference but among men two-thirds of the original difference, a 20% excess mortality of the Finnish-speaking majority, persists after adjusting for the structural differences.In men, the main part of the mortality difference results from factors responsible for excess mortality of the Finnish-speaking population from cardiovascular diseases and non-natural causes of death.A similar mortality contrast is seen in women as well, but it is compensated by other causes of death which are more common among Swedish-speaking than among Finnish-speaking women.

lntroduction
In intemational comparison life expectancy is very high in the Nordic countries.Sweden, Norway and Iceland, in particular, have been among the small number ofleading countries throughout the 20th century.Finland constitutes an exception among the Nordic countries, although it has been drawing closer to the others during recent decades.The mortality of Finnish men, in particular, has been higher than the average level in the westem industrialized countries (Valkonen1991; Statistics Finland 200 la).
There is a significant ethnic minority,comprisingnearly six percent ofthe total Finnish population, however, with clearly lower-than-average mortality.Among the Swedishspeaking Finns, life expectancy has been substantially longer than in the rest of the population.The earliest data, presented by Fougstedt (1951), cover the period 1936-45, when the proportion of Swedish speakers was 10% of the whole population.At that time women's life expectancywas 3.5 years higher in the Swedish-speakingsection than in the population as a whole.Among men the corresponding difference was 2.5 years.Among men the mortality difference between the language groups has been quite constant, whereas among women mortality has declined more slowly in the Swedish-speaking minority than in the Finnish-speaking population.At the beginning of the l 990s the difference in life expectancy had diminished to one year among women but remained at 2.5 years among men (Finnäs 1995, 9).During the 1990s the relative mortality differencesbetween the two language groups remained rather stable (Statistics Finland 200lb,(90)(91).
In the early l 940s the relativemortality differencebetweenlanguagegroups was clearly largest among infants and small children.In the adult age groups the difference tended to be much smaller, particularly among men (Fougstedt 1951, 225-227;Central Statistical Office of Finland 1952, 48).By the late l 970s the excess mortality of Finnishspeaking boys had totally disappeared.The relative advantage of Swedish-speaking girls had remained at the earlier level (Finnäs 1986, 44), but its impact on life expectancy had practically disappeared due to the rapid decline of death rates among girls in both language groups.In the adult age groups the relative contrast has, sincethe l 940s, somewhat declined among women and increased among men (Finnäs 1986(Finnäs , 1995)).
The Swedish-speakingpopulation in Finland appears to be healthier than the Finnishspeaking majority also according to indicators other than mortality.For example, disability pensions (Hyyppä and Mäki 200la), occupational accidents (Salminen and Johansson 2000) and poor self-reported health among schoolchildren (Suominen et al. 2000) as well as among adults (Hyyppä and Mäki 200lb) are less frequent among Swedish-speaking than among Finnish-speaking Finns.
Significant linguistic minorities exist in most European countries (e.g. Stephens 1976;Allardt 1979), but in most countries mortality differences between language groups have not been studied extensively.Many examples of mortality contrasts between language groups within national populations have been reported, however, conceming both rather recent immigrants (e.g.Balarajan et al. 1984;Bollini and Siem 1995;Wild and Mckeigue 1997;Uitenbroek and Verhoeff 2002) and indigenous population groups (e.g.Komitzer et al. 1979).
The aim of this study is to delineate the reasons för the mortality difference between the language groups by means of analyzing a data set comprising the deaths during 1971-85 among the entire Finnish population, linked to the censuses of 1970, 1975 and 1980.The analysis is restricted to the age group 35-64 years which makes a particularly large contribution to the difference in life expectancy at birth between the language groups.We will start by reviewing the areas where the causes för the mortality contrast could lie.

Possible causes
It seems unlikely that the language spoken by an individual would directly affect her or his health status.Instead, we regard language as an indicator of membership of an ethnic group.Mortality differences between ethnic groups, as well as differences between subgroups of the population in general, may result from any factor or set of factors affecting the probability of death, providing that the prevalence of this determinant varies between the groups which are compared.The origin of mortality differences may thus lie in variations in the environment as well as in health-related behavior, genetic dissimilarities and differences in the nature and extent of health-related mobility to and from the population group.
Many altemative explanations may be relevant.First, the low mortality among the Swedish-speakingFinns may be due to a favorable population structure with regard to other sociodemographicfactors.Most Swedish speakers live in the southem and westem coastal parts ofthe country where mortality is known to be lowest.Furthermore, the proportion of persons who are married or belong to upper socioeconomic groups is higher in the Swedish-speaking sector than in the rest of the population (Central Statistical Office ofFinland 1987, 31-32;Allardt and Starck 1981;Sandlund 1985).In these groups, death rates are significantly lower than average (Valkonenet al. 1993;Koskinen and Martelin 1994).Whatever the ultima te reasons för these sociodemographic mortality differences are, it seems quite possible that a major part of the low mortality in the Swedish-speakinggroup could simply be attributed to their favorable population structure.
Contrary to the structural explanation, the cultural and genetic explanations share the basic assumption that membership in a linguistic minority has an independent effect on one's health status.The cultural hypothesis suggests that being a member of a linguistic or ethnic group affects several aspects of one's social environment and way of life.It may, för example, have implications on one's health-related behavior, such as pattems of drinking or eating.In addition, psychosocial factors, för example, such as social support and pattems of experiencing and dealing with stressful situations, may differ between the language groups.Hyyppä has postulated that the low mortality of the Swedish-speaking group results from their social capital, which gives them better support than in the Finnish-speaking population (Hyyppä 1994;Hyyppä and Mäki 200la).
Genetic characteristics can also produce mortality contrasts between the language groups if the beneficial traits are particularly common in the Swedish-speaking population.There seem to be some genetic differences between the language groups (Virtaranta- Knowles et al. 1991), owing to the history ofthe settlement of Finland.
Health-related mobility from one language group to another can hardly play a significant role in generating mortality differences between the groups.Changing one's language is rather infrequent (Finnäs 1986, 51-70) and there is no evidence or reason to suspect that changing one's language would be significantly influenced by one's health.

Data, strategy and methods
The data set of this study was constructed by linking individual records of the 1970, 1975 and 1980 Censuses with death records för the subsequent five-yearperiods 1971-75, 1976-80 and 1981-85, through the cooperative efforts of Statistics Finland 1 and the research group at the Department of Sociology, University of Helsinki (Valkonen et al. 1990;Valkonen et al. 1993).It includes all the 172,000 deaths in the age group 35-64 in 1971-85 in Finland.The coverage of the data can be regarded as excellent since less than 0.5% of the deaths in Finland during 1971-85 are missing from the files, mainly due to undercoverage ofthe census data.The compilation ofthe data set has been described in detail earlier (Valkonenet al. 1990;Koskinen 1994;Martelin 1994).The data set does not cover the most recent years, but this does not diminish its value, as the relative mortality differences have remained more or less similar för several decades.
Table 1 shows the classification of causes of death.The variables used in the analysis as well as their classifications are presented in Table 2, which also gives the numbers We are mdebted to Statistics Fmland för the perrmssron (TK 53-69-87) to use the data files.
of person-years and deaths in each class.In the data set, age is continuously updated during the föllow-up so that it refers to the exact age at any time during the föllow-up.
Other census variables, on the other hand, describe the situation at the time of the census.Possible changes in place of residence, occupation, education, marital status or language during the five-year föllow-up are not included in the data.With regard to the alternative explanations of the mortality difference, we are able to directly examine the hypothesis suggesting that the gap between the language groups is due to differences in their sociodemographic structure.If, on the other hand, mortality differentials persist even after controlling för sociodemographic structure, we may conclude that language has an independent effect on mortality.It is, however, a more difficult challenge to specify the mechanisms because behavioral and other cultural factors as well as genetic characteristics of individuals are not included in the data.Conclusions about the impact of these factors will thereföre rely on indirect conclusions based on findings from the analyses concerning causes of death, as well as subgroups ofthe population.
There are many behavioral factors which may contribute to the mortality difference.If smoking is the major cause of the mortality difference between the Swedish-and Finnish-speakinggroups, after havingcontrolledför the population structure, we should see the largest mortality differences in those causes of death which are strongly related with smoking.A particularly extensive contrast could thus be expected in lung cancer, a somewhat weaker one in respiratory diseases and a rather small one in other cancers and cardiovascular diseases, while in mortality from other diseases and nonnatural deaths the language groups may not differ from each other.Differences caused by smoking can be expected to be larger in men as the prevalence of smokinghas been much greater among men than among women (Martelin 1984).If alcohol consumption has a significant role, we may assume that the mortality differences are especially large in accidental and violent causes of death, as well as in the group "other diseases" which includes alcoholism, liver cirrhosis and diseases of the pancreas, för example.Mortality contrasts caused by drinking should be particularly apparent in men since men drink, on average, much more than women and their drinking habits tend to be more extreme (Simpura 1987).Conclusions about the role of dietary habits are difficult to draw because none of the causes of death can be interpreted as an indicator of an excess intake or a shortage of a certain substance.The possible effect of varying fat and salt consumption may, however, be seen in cardiovascular diseases.
If more or less permanent cultural traditions, such as social interaction pattems, are the main ultimate cause oflow mortality among the Swedish speakers, those Swedishspeaking population groups which have best retained their original cultural features might be anticipated to be in the most advantaged position.Such groups may include residents of rural areas, perhaps farmers in particular, in those regions where the Swedish-speaking group förms the majority of the population.
Finally, if genetic factors are important causes of the mortality differential between the language groups, the contrast ought to be largest in causes of death where the role of inherited factors is greatest.It is unförtunately not possible to classify causes of death reliably according to this dimensionbut some causes, such as lung cancer, obviously result mainly from environmental and behavioral factors.Secondly, if genetic factors play an important role, the mortality difference should be largest in those areas where the Swedish-speaking group has best retained its genetic characteristics.Such areas may be Åland and the coastal (rural) parts of the province of Vaasa.In the Helsinki Region, för example, the population groups are probably more mixed in terms of genetic properties.
The analysis is based on a multivariate table för which the data are cross-tabulated according to all the variables included in Table 2.Each cell in this table gives the number of deaths and the number of person-years during the study period.The table is analyzed employing Poisson regression analysis of the mortality rates (Aitkin and Clayton 1980).The analyses are perförmed using the GLIM package (Payne 1985).
The analysis proceeds in three steps.First, the role of differences in population structure is examined by means of controlling för the other sociodemographic factors (age, education, occupational class, marital status, region of residence and level of urbanization) with the model.Second, the mortality differencebetween the language groups is examinedseparately within sociodemographicsubgroups in order to find out whether the gap is particularly prominent in certain parts of the population.Cause-specific mortality differences are studied as the third step of the analysis.

The role of population structure
Age-standardized mortality was seven percent lower among Swedish-speakingwomen than among the other women during the period 1971-85 (relative mortality = 0.93; Table 3a).Adding other sociodemographicvariables to the explanatory model in Table 3a shows that this advantage disappears totally when differences in the population structure are controlled för.This is partly due to the higher-than-average education and occupational class ofthe Swedish-speaking women in comparison with the other women.Also the geographic location of the Swedish speakers is favorable, with respect to risk of death, even when the socioeconomic structure ofthe regional popula-tions has been adjusted för.Members of the upper socioeconomic strata and persons living in the southwest have lower-than-average mortality, and för women the advantageous socioeconomic and geographic location of the Swedish speakers seems to explain their low mortality totally.The contrast in age-standardized mortality is much starker in middle-agedmen: Swedish-speaking men have a 27% advantage over the others (relative mortality = 0.73; Table 3b).Approximately one-third of this variation is related to different socioeconomic and geographic distributions of the language groups.Nevertheless, the relative mortality of the Swedish-speaking men remains low, 83% of the level among the other men, even after controlling för all variables included in the data.In the subsequent analyses the main emphasis will thereföre be put on language-group differences amongmen.Is the mortality difference equally wide in ali population groups?
Table4 revealsthat the relative differencesin age-standardizedmortality are not equally wide in each subgroup ofthe population.Part of the diversity obviously results from random variation due to the small number of deaths in many subgroups.Among men mortality of the Swedish speakers is, however, lower than that of the Finnish-speaking majority in each subgroup.Even though controlling för the differences in the sociodemographic structure of the language groups appeared to abolish the mortality differential among women, a significant difference is found in some subgroups.In men the contrast is much wider than in women, as was already observed above.Furthermore, the relative advantage of the Swedish-speaking persons is particularly manifestin the youngest age groups and among persons residing in rural areas.In addition,the contrast in age-standardized mortality is slightlymore pronounced among farmers than in other occupational classes, particularly among women.This finding results, however, partly from the altogether different geographic distribution of the Swedish-speaking and Finnish-speaking agricultural population, which can be seen when the effect of region is adjusted för.Among white-collar men, the contrast is weaker than in the other classes, whether other variables are controlled för or not, and among white-collar women there is a statistically non-significant excess mortality in the Swedish-speaking group.
Controlling för other variables also somewhat changes the findings för the regional populations.In general, the relative advantage of the Swedish-speaking population declines or, in some instances, even reverses to a slight disadvantage, when their higher socioeconomic position is taken into account.Among both women and men, adjusting för the other sociodemographic variables produces a particularly small change in the results concerning the province of Vaasa as well as för Uusimaa, excluding the Helsinki Region.In these areas the socioeconomic distribution ofthe Swedish-speaking persons differs only slightly from that in the rest of the population of the respective region.Thereföre, when all other variables are controlled för, the relative mortality of the Swedish-speaking men, compared to the other men, is particularly low in Vaasa and "other Uusimaa", in addition to Åland, för which the confidence intervals are very large, however, due to the very small number of deaths in the Finnish-speaking group.Among women the adjustment för other variables leads to the result that it is only in Vaasa that the Swedish speakers have significantly lower mortality than the Finnish-speaking group.

Is the mortality difference similar in ali causes of death?
The advantage of the Swedish speakers is obvious in men's age-adjusted mortality from each cause of death, except neoplasms other than lung cancer.Controlling för all sociodemographic variables diminishes the mortality contrasts but they remain marked and statistically significant (Table 5).The mortality excess of the Finnish-speaking men is largest in cardiovascular and respiratory diseases as well as in non-natural causes of death.Also lung cancer and "other" diseases are more common in the Finnishspeaking group, but the difference is relatively small.Among women the situation is more complex.The language groups do not differ from each other in mortality from all causes combined when the sociodemographic structure of the groups is taken into account.Behind this similarity we find opposite pattems.Finnish-speaking women have statistically significant excess mortality from cardiovascular diseases and violent and accidental causes, excluding suicide.In ischemic heart disease the mortality contrast between the language groups is almost equally great among women and men.On the other hand, mortality from neoplasms is significantly higher among Swedish-speaking than among Finnish-speaking women and smaller differences, in the same direction, are seen also in other diseases and suicides.
Circulatory diseases are responsible för nearly two-thirds of the excess mortality of Finnish-speaking men compared with Swedish-speaking men (Table 6).Together with non-natural deaths, they cause 90% of the mortality difference in men.For women these two groups of causes of death are the only ones where the Swedish-speaking population has a mortality advantage.Table 6.Excess mortality (deaths / 100 000 / year) of the Finnish-speaking population aged 35-64, compared with the Swedish-speaking population by cause of death in 1971-1985.Age and all other variables" controlled för.

Cause of death Women Men
Ischemic The cause-specific analyses were carried out also in regional and other subgroups of the population.Due to small numbers of deaths, causes of death were combined into circulatory diseases, neoplasms, other diseases, and accidental and violent causes.In circulatory diseases, wide and statistically significant variation was föund in the extent of language-group differences.The findings replicated those in mortality from all causes, presented in Table 4.When all other factors were controlled för, a particularly large advantage of the Swedish-speaking population in mortality from circulatory diseases was observed in Åland and the province of Vaasa, in rural areas and in the agricultural population, and among the younger middle-aged men.
In other causes of death, the language-group contrast varied significantly between subgroups of the population only in some instances.In women's accidental and violent deaths the Swedish-speaker's advantage was particularly large in Åland, in the province ofVaasa, in Uusimaa excluding the Helsinki Region and among blue-collar employees.In mortality from neoplasms among men, blue-collar workers had a widerthan-average contrast between language groups.In male mortality from other diseases the differencewas especially evident in the provinces of Vaasa and Uusimaa, in rural areas and among farmers.

Conclusions
The mortality advantage of the Swedish-speaking Finns is partly due to their favorable geographic location and socioeconomic position.Owing to the history of the settlement of Finland, the Swedish-speakingpopulation resides in the western coastal areas where mortality is low in both language groups, and the proportion of higher socioeconomicgroups is larger than average in the Swedish-speakingpopulation.For women these factors appear to totally explain the difference in mortality from all causes but among men, two-thirds of the original difference, a 20% excess mortality of the Finnish-speakingmajority, persists after adjusting för the structural differences.
It turns out, however,that evenamong womena mortality differenceemergesin certain subgroups and in some causes of death.This implies that the structural explanation suggested by the first step of the analysis is partly biased and it is due to differences taking place in opposite directions.Moreover, it has been shown elsewhere (Martelin 1994)that among elderlywomenthe excess mortality of the Finnish-speakingmajority persists even after controlling för the sociodemographic structure.
Clues för possible causes of the difference, other than the structural explanation, have been searched för by means of separately analyzing various subgroups as well as causes of death.To summarize, the advantage of the Swedish-speaking group is particularly large in the younger age groups, especially among men, and in Åland, föllowed by Vaasa, and, with respect to men, 'other Uusimaa'.In addition, the mortality contrast appears to be slightly wider among manual workers than among non-manual workers, and in rural than in urban areas, even when the other sociodemographic variables are controlled för.
After adjusting för structural differences between the language groups, excess mortality among the Finnish speakers, relative to the Swedish-speaking persons, is seen för both genders only in cardiovascular diseases as well as in accidental and violent deaths.In ischemic heart disease the relative mortality difference between the language groups is equally great among both sexes, whereas in other cardiovascular diseases and non-natural causes the advantage is more evident för Swedish-speaking men than women.In neoplasms other than lung cancer, a slight mortality excess is seen among Swedish-speaking women and men.All the other causes of death show opposite language-group differences för women and för men.
The föllowing assessment of reasons, other than population structure, för the causespecific findings is divided in three parts.We shall first discuss the most important question,i.e. the reasons för the lowmortalityfrom circulatorydiseasesamong Swedishspeaking women and men.Next, possible causes för the difference in accidental and violent deaths will be considered.Finally,we shall attempt to evaluatewhy the mortality differences between language groups are opposite among women and men för most of the remaining causes of death.Swedish-speaking women and men seem to possess some feature(s) which protect(s) them from dying of circulatory diseases and non-natural causes.It seems likely that the causes of these mortality differencesbetween the language groups are rather similar för women and men.What could these features be?The results presented above fit best with the genetic and cultural hypotheses: the mortality contrast is seen in all population groups but it is starkest in those subgroups which have assumably maintained their original cultural and genetic characteristics, i.e. farmers, persons living in Åland and Vaasa province and those residing in rural areas.Social networks (Hyyppä and Mäki 200la,200lb) and dietary practices are among the cultural factors which might come into question as mechanisms mediating the effect.
In addition to cardiovascular diseases, accidental and violent deaths are also less common among Swedish-speaking men and women as compared to the Finnish-speaking population, the difference being greater among men.One of the causes för this phenomenon may lie in differences in pattems ofthe use of alcohol.In the early 1970s, mortality of Finnish-speaking men from selected alcohol-related diseases'' was about 20% higher compared with mortality among Swedish-speaking men, while deaths from alcohol poisoning were as much as föur times more common among Finnishspeaking than Swedish-speaking men (Valkonen1982).Valkonen's results implythat, among men, there were large differences between the language groups in drinking habits.This finding parallels the survey results reported by Simpura (1990).
Of the causes of death included in this study, excessive alcohol use is mainly related with non-natural deaths and the heterogeneous group 'other diseases' which together account för about 30% of the mortality difference between Swedish-and Finnishspeaking men.Among male deaths from non-natural causes, alcohol plays a role in less than half of the cases (Koskinen and Martelin 1994, 184).Among deaths from 'other diseases', the corresponding proportion can hardly be higher.We may thus approximate that the impact of excessive alcohol consumption on the mortality difference between the language groups probably lies somewhere between 10-30% among men.
The language group differences in alcohol-related causes of death were small among women.This corresponds with the survey findings suggesting that drinking habits among Swedish-speaking women are almost similar to those in the general female population (Simpura 1990).As a result, mortality differences among women can not be explained by differences in drinking habits -unless one is willing to propose that the way the Finnish-speaking men use alcohol is dangerous to Finnish-speaking women.
In addition to patterns of drinking, some other causal factors behind the mortality advantage ofthe Swedish speakers appearto be specific to men only.Swedish-speaking men are less vulnerable than Finnish-speaking men regarding all remaining causes of death except neoplasms other than those ofthe lung.Genetic and general environmental factors could hardly provide an adequate explanation because they would probably exert their effect similarly on both genders.Behavioral patterns, on the other hand, show considerable variation between women and men.Differences in lung cancer mortality suggest that Finnish-speaking men smoke more than Swedish-speaking men.Five percent of the excess deaths among Finnish-speaking men were due to lung cancer, as was shown in Table 6.As lung cancer has been estimated to cause about one-third of smoking-related deaths (Martelin and Valkonen 1981;Peto and Lopez 1992), we can conclude that, among men, about 15% ofthe excess mortality of the Finnish speakers is due to differences in smoking habits between the language groups.
The other causes of death where the effect of differences in smoking habits can be seen are mainly respiratory and cardiovascular diseases.Among men, deaths from circulatory diseases account för more than 60% of the mortality difference between the language groups.The advantage of the Swedish-speaking menin mortality from cardiovascular diseases is, therefore, mainly due to other etiological factors.In women, lung cancer and respiratory diseases were more common in the Swedish-speaking population.This finding indicates that smoking has been more common among Swedish-speaking than Finnish-speaking women.While differences in smoking habits between the language groups appear to run in opposite directions in women and men, the greater difference among men, compared with women, in cardiovascular mortality between the language groups may largely ensue from smoking patterns.In summary, there is no simple all-embracing answer to the question ofwhy mortality is low in the Swedish-speaking population of Finland.The contrast is partly due to their favorable geographic location and socioeconomic position, but differences in genetic factors as well as behavioral patterns or other cultural factors are likely to be important causes as well.While a comprehensive data set consisting of census variables and införmation from death certificates provides an excellent hasis för testing the hypothesis of the effect of structural differences, the possibilities för drawing conclusions about the role of genetic or cultural factors are limited.

Table 1 .
Deaths by cause of death, gender and language in 1971-85 in Finland in the 35-64-year-old population.

Table 2 .
Person years and numbers of deaths by language in subgroups of the 3 5-64-year-old population inFinland in 1971Finland in  -1985.   .
a The Helsinki region includes Helsinki, Espoo, Vantaa and Kauniainen.

Table 3a .
Mortality of Swedish-speaking Finns relative to other Finns (=1.00) according to different models" in women aged35-64 in 1971-85.Model 1 includes only two explanatoryvariables, namely language (LAN) and age.Model 2 includes the main effects of language, age and education.Similarly, each successive model includes one more explanatoryvariable, relative to the previousone, so that the main effects of ali variables includedin the data are incorporated in Model 6, which thus gives the relative mortality of the Swedish-speaking populationcontrollingför the effects of age (AGE), education(EDU), occupationalclass (OCC), marital status (MAR), region ofresidence (REG) and level ofurbanization (URB). a
a See footnote in table3a

Table 4 .
Mortality of Swedish-speaking persons as percentage of mortality ofFinnish-speaking persons in subgroups of the population aged35-64 in 1971-1985.(95%confidence intervals in parentheses.)Age and ali the other variables (region, level ofurbanization, education, occupational class and marital status), except the variable under study, adjusted för.

Table 5 .
Mortality of Swedish-speaking persons as percentage of mortality of Finnish-speaking persons by cause of death in the population aged 3 5-64 in1971- 1985.(95% confidence intervals in parentheses.) a Age, region, level of urbanization, education, occupational class and marital status adjusted för