The effects of regional characteristics on alcohol-related mortality—a register-based multilevel analysis of 1.1 million men
Introduction
Interest in investigating area effects on morbidity and mortality has increased considerably in recent years. There is growing consensus in the field of public health that characteristics of communities and areas may be important in understanding health outcomes. The effects of area context on the health of individuals have policy relevance, as it has been suggested that interventions to promote health should also focus on efforts to improve neighbourhoods and communities in socially disadvantaged areas (Robert, 1999; Diez-Roux, 2001). Most studies have concentrated on the neighbourhood level, although some research has been carried out at the US state level and at the metropolitan and county levels.
Most multilevel studies on area characteristics affecting health have focused on the effects of socioeconomic structure or deprivation and average income level of areas. These studies show somewhat conflicting results: some have found area effects after adjusting for individual characteristics (Haan, Kaplan, & Camacho, 1987; Davey Smith, Hart, Watt, Hole, & Hawthorne, 1998; Yen & Kaplan, 1999), while others show only modest independent effects after adjustment for individual confounders (Sloggett & Joshi, 1994; Anderson, Sorlie, Backlund, Johnson, & Kaplan, 1997; Sloggett & Joshi, 1998; Waitzman & Smith, 1998).
There is a large body of literature on the effects of income inequality on health and mortality. Some multilevel studies in this field have found an association between income inequality and health outcomes at the US state level (Kennedy, Kawachi, Glass, & Prothrow-Stith, 1998; Lochner, Pamuk, Makuc, Kennedy, & Kawachi, 2001), at the metropolitan level (Blakely, Lochner, & Kawachi, 2002) and at the county and census-tract levels (Soobader & LeClere, 1999), even after adjusting for individual-level income. However, methodological criticism has been levelled at the income inequality hypothesis. It has been suggested that the relation between income inequality and health outcomes is merely a statistical artefact due to the curvilinear association of individual income and health (Ellison, 2002; Gravelle, Wildman, & Sutton, 2002). Furthermore, some studies have found no independent effect of income inequality. For example Fiscella and Franks (1997) reported no association between income inequality and mortality on the county level in the United States after controlling for individual-level variables in a multilevel study. Moreover, Osler et al. (2002) found no association between parish-level income inequality and all-cause mortality in Danish adults after adjustment for individual income.
Social cohesion and social capital have attracted growing attention in research on community factors affecting health (Kawachi & Berkman, 2000; Kunitz, 2001), although these concepts have not been very clearly defined. However, social capital, defined as the density of membership of associations, levels of interpersonal trust and strengths of mutual aid and reciprocity (Putnam, 1993), has been associated with mortality and self-rated health, at least in the United States (Kawachi, Kennedy, Lochner, & Prothrow-Stith, 1997; Kawachi, Kennedy, & Glass, 1999). Baum (1999) emphasizes the need to distinguish between different types of social capital, for instance that generated by family and kinship compared with that arising from associational life or links that connect different groups within society. In our analysis, we do not explicitly distinguish between social cohesion and social capital, but we aim to take the different dimensions into account by separating family cohesion from societal cohesion.
Even though there is consensus on the importance of community- and area-level characteristics affecting health, the effects of socioeconomic position in the community are generally found to be modest compared to individual-level effects, and especially so after adjusting for confounding individual-level socioeconomic effects (Pickett & Pearl, 2001; Robert, 1999). According to the review by Pickett and Pearl (2001) on the effects of neighbourhood and local area characteristics on health outcomes (mortality, morbidity and health behaviours), most studies that have also adjusted for at least individual level socioeconomic status have found effects of one or more community-level variables, which however diminished or disappeared after adjustment for individual effects. Most of the mortality studies reviewed by Pickett and Pearl focused on all-cause mortality. However, all-cause mortality may be too general and heterogeneous as outcome to show strong independent effects of area characteristics.
We assume that contextual factors may have a stronger effect on alcohol-related mortality than on many other types of mortality, because this outcome is so strongly dependent on people's behaviour. Hence, alcohol-related mortality is a suitable outcome measure for testing the existence and nature of contextual effects. In a previous study on area differences in alcohol-related mortality in Finland (Mäkelä, Ripatti, & Valkonen, 2001), adjusting for the sociodemographic characteristics of individuals living in Finnish provinces reduced the differences between provinces by only 20%. Further analyses of the particular area characteristics that cause the geographical differences in alcohol-related mortality are thus needed.
According to Pickett and Pearl (2001), studies adjusting for more than one individual-level socioeconomic variable have found weaker associations between neighbourhood-level socioeconomic status and health. This suggests a compositional effect of the population structure in the areas, i.e. the observed differences in mortality or morbidity between areas arise because in some areas there are more residents with characteristics associated with poorer health outcomes (Curtis & Jones, 1998; Duncan, Jones, & Moon, 1998). The results also suggest that individual-level variables should be adequately adjusted for in studies of area effects on health outcomes. More analyses are also needed to investigate the effects of area characteristics other than socioeconomic structure on mortality, and to study the independent effects of various area characteristics when the effects of other area characteristics are simultaneously taken into account. Moreover, both average and spread variables should be included in the analyses (Pickett & Pearl, 2001).
In addition to the fact that alcohol-related mortality offers a better way of testing the effects of area-level characteristics, there are further reasons for selecting alcohol-related mortality as our outcome measure (for an exact definition, see Data and methods section). Alcohol-related deaths make a large contribution to premature mortality (Mäkelä, 1998), and to socioeconomic differences in mortality and their changes, particularly among young and middle-aged men (Mäkelä, Valkonen, & Martelin, 1997; Valkonen et al., 2000; Martikainen, Valkonen, & Martelin, 2001). Additionally, province-level differences in alcohol-related mortality in Finland are large, with the provinces in northern and eastern Finland showing higher mortality rates than those in western parts of the country (Mäkelä et al., 2001). It has also been shown that alcohol-related deaths make a rather large contribution to the province-level differences in all-cause mortality. An effect of alcohol-related deaths on geographical differences in all-cause mortality has also been shown at the neighbourhood level in the Helsinki metropolitan area (Martikainen, Kauppinen, & Valkonen, 2003). Furthermore, the emphasis on preventive measures against alcohol-related damage has moved from the national to the local level. This lends credence to the studies on the factors behind the large area differences in alcohol-related mortality.
Risky consumption of alcohol, in terms of acute intoxication or long-term heavy consumption, is a necessary—but not sufficient—cause of alcohol-related mortality (as it is defined in this study). We assume that the area-level factors used in the analysis affect alcohol-related mortality partly through the risky consumption of alcohol and partly through vulnerability to the effects of risky consumption. If valid area-level measures of risky consumption existed, it would be useful to add them to the analysis in order to separate the ‘risky consumption’ pathway from the ‘vulnerability’ pathway. However, there are no reliable statistics on regional alcohol consumption patterns but only on alcohol sales, which do not tell how alcohol is consumed and do not take into account that there are systematic regional differences in the extent to which alcohol is bought and consumed in the same place. Therefore, we do not attempt to explain geographical differences in alcohol-related mortality in terms of differences in alcohol sales.
The aim of this study is to assess to what extent selected characteristics of Finnish regions affect the level of alcohol-related mortality. We use the functional regions (n=84) as area units. We will first investigate the gross effects of the separate area characteristics, but the focus of the study is to assess the strength of the effects of each of the area characteristics on alcohol-related mortality while adjusting for individual-level and other area-level characteristics. This allows us to distinguish the contextual effects of the areas from the compositional effects of the individual characteristics. Individual-level variables are included in the analysis only to account for their possible confounding effects. We will also estimate to what extent the variation in alcohol-related mortality between the functional regions could be explained in terms of the individual-level and area-level variables considered.
The analysis is restricted to working-aged men (25–64 years during the period covered by the study). In 1991–1993, 80% of all alcohol-related deaths among men occurred in this age group (unpublished table from Mäkelä, 1998). In this paper, we decided to concentrate on men because 86% of all alcohol-related deaths in the period 1991–1993 occurred among men (Mäkelä, 1998). The proportion of alcohol-related deaths of all deaths in 1991–1993 was 28% among men aged 25–64 (unpublished data from Mäkelä, 1998).
Section snippets
Study population
The data in this study are on two levels, that of individuals and that of areas in which the individuals reside. The individual-level data were obtained from Statistics Finland (permission TK-53-8-02). The individual-level data are register data from the 1990 census linked to records from the death register for the years 1991–1996 for all men who were 25–64 years old during the study period. The linkage was carried out by Statistics Finland by means of personal identification codes. These
Results
In these analyses, individual-level variables were treated as confounders that reflect possible compositional differences in the population structures of the study areas. The detailed results for the individual-level variables are not presented, but the patterns are consistent with previous findings (Mäkelä (1998), Mäkelä (1999)). When all other individual- and area-level variables are adjusted for, alcohol-related mortality consistently increased with age, the mortality rate ratio (RR) of the
Discussion
Area effects on mortality have usually been studied in terms of the effects of area-level socioeconomic status or deprivation. Some multilevel studies on mortality, morbidity and health behaviours have found effects of area deprivation that are independent of individual-level variables, while others suggest that area effects are modest or non-existent, especially after controlling for individual-level variables (Pickett & Pearl, 2001). This variability in the findings is not surprising, as
Conclusion
In summary, the observed area differences in alcohol-related mortality were not accounted for by differences in the characteristics of the individuals in the areas. Moreover, the independent contextual effects of the area characteristics were substantial. The area-level variables representing social structure and cohesion in particular are significant in explaining area differences in alcohol-related mortality. Fuller understanding of the mechanisms underlying the effects of area measures of
Acknowledgements
This work is part of the European Science Foundation program on Social Variations in health Expectancy in Europe, in particular the working group on Macrosocial Determinants of Morbidity and Mortality. The study was supported by the Academy of Finland (grants 41498, 70631, 48600). We are grateful to Statistics Finland for granting permission (TK-53-8-02) to use the data.
References (40)
- et al.
Metropolitan area income inequality and self-rated health—a multi-level study
Social Science & Medicine
(2002) - et al.
Context, composition and heterogeneityUsing multilevel models in health research
Social Science & Medicine
(1998) Letting the Gini out of the bottle? Challenges facing the relative income hypothesis
Social Science & Medicine
(2002)- et al.
Income, income inequality and healthwhat can we learn from aggregate data?
Social Science & Medicine
(2002) - et al.
Aggregation and the measurement of income inequalityeffects on morbidity
Social Science & Medicine
(1999) - et al.
Mortality effects of community socioeconomic status
Epidemiology
(1997) Social capitalis it good for your health? Issues for a public health agenda
Journal of Epidemiology and Community Health
(1999)- et al.
Ecological effects in multi-level studies
Journal of Epidemiology and Community Health
(2000) - et al.
Is there a place for geography in the analysis for health inequality?
Sociology of Health and Illness
(1998) - et al.
Macro-to-micro links in the relation between income inequality and mortality
The Milbank Quarterly
(1998)