IJE Advance Access originally published online on February 1, 2006
International Journal of Epidemiology 2006 35(3):633-643; doi:10.1093/ije/dyl009
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Article |
Disentangling contextual effects on cause-specific mortality in a longitudinal 23-year follow-up study: impact of population density or socioeconomic environment?
1 Community Medicine and Public Health, Department of Clinical Sciences in Malmö, Malmö University Hospital, Faculty of Medicine, Lund University, Malmö, Sweden
2 Research Unit in Epidemiology, Information Systems, and Modelling (INSERM U707), National Institute of Health and Medical Research, Paris, France
3 Department of Epidemiology, Biostatistics and Occupational Health, McGill University, Montreal, Canada
* Corresponding author. Department of Clinical Sciences, Malmö University Hospital, S-205 02 Malmö, Sweden. E-mail: chaix{at}u707.jussieu.fr
 |
Abstract |
|---|
 Top Abstract MethodsResultsDiscussionReferences |
|---|
Background Various studies have investigated urban/rural differences in cause-specific mortality. A separate body of literature has analysed effects of socioeconomic environment on mortality. Almost no studies have attempted to disentangle effects of population density and socioeconomic environment on mortality, beyond the effects of individual characteristics.
Methods Considering all individuals living in the region of Scania, Sweden, from 1970–93, we performed 10 year mortality follow-ups on (i) individuals aged 55, (ii) individuals aged 65, and (iii) individuals aged 75 years at baseline. Cox multilevel models adjusted for individual factors allowed us to investigate the independent effects of population density and median income in the parish of residence on mortality from ischaemic heart disease (IHD), lung cancer, and chronic obstructive pulmonary disease (COPD) among individuals who had lived in the same parish for at least 10 years prior to mortality follow-up.
Results In females, as in males, after adjustment for individual and contextual socioeconomic status, we found a dose–response association between population density and mortality from lung cancer and COPD in all age groups investigated, and from IHD especially in the youngest age group. Overall, the population density effect was the strongest on lung cancer mortality. Median income had an additional impact only in 2 out of 16 subgroups of age x gender x cause of death.
Conclusions In our region-wide study conducted at the parish level, contextual disparities in mortality were dominated by the population density effect. However, it may be unwise to conclude that truly contextual effects exist on mortality, before identification of plausible mediating processes through which urbanicity may influence mortality risk.
Keywords Residential context, socioeconomic environment, population density, spatial analysis, ischaemic heart disease, lung cancer, chronic obstructive pulmonary disease
Accepted 10 January 2006
Socioeconomic disparities in cause-specific mortality have been documented in previous research.1 However, recent literature has emphasized that, beyond individual factors, characteristics of the residential context may contribute to disparities in mortality.2 On one hand, some studies have investigated urban/rural differences in mortality.3–6 Certain authors have found that the rate of cancer mortality increases linearly with population density.7 They hypothesized that such patterns are attributable to the behavioural habits and environmental hazards of urban life. However, almost all these studies were ecological and failed to control for socioeconomic characteristics of individuals.3–5,7,8 On the other hand, many studies conducted either within urban territories or on a larger scale have investigated effects of the socioeconomic environment on mortality.9–19 Most of them reported an increased mortality risk in deprived areas, after adjustment for individual factors.9–17
To our knowledge, no studies have simultaneously investigated the independent effects of these two contextual factors—population density and socioeconomic environment—on mortality after adjustment for individual factors, either within urban territories or on a broader scale. However, it is possible to do so because the two contextual dimensions constitute theoretically distinct constructs with independent definitions (population density referring to the spatial concentration of inhabitants, irrespective of the distribution of socioeconomic resources). Second, because of the complex pattern of association that may exist between the two factors, it is necessary to adjust their effect for each other in order to avoid incorrect inferences on the contextual influences at play. Third, disentangling the impact of population density and socioeconomic environment is relevant in a public health perspective, to assess which of the two factors should be primarily used as a risk marker of high mortality areas.
Our study was not meant to investigate the mediating mechanisms between the residential context and mortality, but makes a substantive contribution at an earlier step of the process of understanding of contextual influences on health. Based on previous literature,20 the mediating mechanisms of contextual effects investigated probably pertain to the physical environment (e.g. built environment, air pollution, noise exposure)21–23 and social environment (which is characterized by cultural milieus and social interaction patterns that may shape individual health-damaging and healthcare-seeking behaviour).20,24,25 Presumably, similar mediating factors may be at play in the population density and socioeconomic environment effects, but may have a different social meaning in those two distinct causation processes. For example, the increased consumption of tobacco or alcohol observed in urban settings may result from the consumerist attitudes prevailing there.26 Conversely, similar health-damaging behaviour in deprived neighbourhoods may be attributed to feelings of hopelessness or a lack of health knowledge. In any case, before investigating mediating processes, it is necessary to describe contextual effects on mortality, a task for which the large register database used in our study was particularly appropriate.
We used longitudinal data on all individuals living in the region of Scania in southern Sweden, aged 55, 65, or 75 years at baseline in the 14 successive years from 1980 to 1993, and followed-up over 10 years for mortality. Individuals were georeferenced at the level of the 370 parishes of Scania. Those parishes contain a median number of inhabitants inferior to 1000, but have a larger area size in rural than in urban territories. Therefore, such geographical units were appropriate to capture spatial variations in mortality, both within urban territories at a rather local scale and between rural and urban territories. Beyond utilization of exhaustive data, our longitudinal framework allowed us to conduct analyses among individuals who had resided at least 10 years in the same parish prior to mortality follow-up (most of them residing in the parish for much longer), ensuring that individuals had been exposed to a similar residential background over a relatively long period of time and thereby minimizing selective migration biases.27 Such a longitudinal design allowed us to compare the magnitude of population density and socioeconomic environment effects adjusted for individual socioeconomic characteristics on mortality from different causes, including ischaemic heart disease (IHD), lung cancer, and chronic obstructive pulmonary disease (COPD).
|
Methods |
|---|
|
TopAbstractMethodsResultsDiscussionReferences |
|---|
Data and measures
With the approval and assistance of Statistics Sweden and the Swedish Centre for Epidemiology, a 34 year (1969–2002) longitudinal database (LOMAS) including all inhabitants in the region of Scania, Sweden (
1 000 000), was assembled. A personal identification number assigned to each person in Sweden was used to link different registers: (i) yearly information on individual income and parish of residence over the 34 year period from the Swedish Population Register; (ii) data on educational attainment from the 1970 population census; and (iii) information on age and causes of death from the National Mortality Register. For each of the 14 successive calendar years from 1980 to 1993 we considered all individuals aged 55, 65, or 75 years, and followed-up on them over a 10 year period. In order to stratify analyses by age at baseline, we constructed three separate databases, each including 14 birth cohorts of individuals with a definite baseline age (i.e. 55, 65, or 75 years). The strength of this design is that in each separate database (wherein the calendar year at baseline varies from 1980 to 1993) all individuals have the same baseline age, enabling us to directly assess contextual effects at those ages.
We aimed to exclude the possibility that contextual effects could result from the migration of individuals who were ill into specific parishes.27 Therefore, we conducted our analyses among individuals who had been living in the same parish for at least 10 years prior to mortality follow-up (individuals who moved during the previous 10 years were excluded). As shown below, most of the stable 10 year residents actually lived in the same parish for a much longer period of time.
Analyses were conducted separately among females and males, resulting in six different databases (see headlines of Table 1 for sample sizes). In each database, we examined whether deaths from IHD, lung cancer, or COPD occurred over the 10 year follow-up period. We used International Classification of Diseases diagnosis codes for the underlying and contributing causes of death to define these outcome variables (see Table 1 for detailed codes).
|
At the individual level, we took birth year, gender, education, and income into account. Birth year was coded into five categories of successive cohorts (1925–27, 1928–30, 1931–33, 1934–36, 1937–38 for 55-year-old individuals, 1915–17, 1918–20, 1921–23, 1924–26, 1927–28 for 65-year-old individuals, and 1905–07, 1908–10, 1911–13, 1914–16, 1917–18 for 75-year-old individuals). Education was divided into three classes (7 years or less, 8–9 years, >9 years). To take into account the past and current socioeconomic position of individuals, we considered personal income at 1 and 10 years before baseline. It was necessary to obtain income values comparable across years, because baseline year was not the same for all individuals. To do so, income each year was expressed as a rank between 1 and 100 among all individuals aged 25 years or over in Scania. Income variables were divided into quartiles (with specific cut-offs in each age x gender stratum).
Individuals were geocoded to the 370 parishes of Scania. The median number of inhabitants in a parish in 1980 was 836 (interquartile range: 385–2579). Two types of contextual factors were considered: population density and socioeconomic environment. For each individual, we defined contextual factors at 1 year prior to baseline. Population density28,29 was defined as the total number of inhabitants per square kilometre. For example, the median parish population density for individuals aged 55 years at baseline was 311 inhabitants per km2 (interquartile range 70–1463). We defined the socioeconomic environment with median income of individuals (males and females together) aged 25 years or over in each parish. To obtain a factor comparable across years, the socioeconomic index was expressed as the rank of each parish among the 370 parishes with respect to median income. The two contextual factors were divided into four different classes containing an equal number of individuals (specific cut-offs were used in each age x gender stratum).
Statistical methods
In a descriptive perspective, we used geoadditive models to derive continuous maps showing variations in mortality risk independent of administrative boundaries.30 The models did not include any effect of individual or contextual characteristics, but captured spatial variations with a two-dimensional (latitude/longitude) smooth function of the spatial coordinates of individuals (geocoded to the centroid of their parish). The fully parametric smooth term was defined as a thin plate regression spline. Those models were estimated with R.31
To examine whether there were between-parish variations in mortality risk after adjustment for individual socioeconomic factors, we used the R software32 to estimate Cox multilevel models with a parish-level random intercept and individuals nested within parishes. Observations were right-censored in case of death by another cause. A gamma distribution was postulated for the parish-level random effect, to allow for a non-normal distribution. In order to facilitate assessment of between-parish variations, we expressed these differences on the commonly employed hazard ratio scale,33,34 using the interquartile hazard ratio.35 Such a hazard ratio quantifies the difference in mortality risk between the 25% of all individuals in parishes with the lowest mortality risk and the 25% of all individuals in parishes with the highest mortality risk.35 Let u be the parish-level random effect. More specifically, u12.5 and u87.5 correspond to the 12.5th and 87.5th quantiles in the distribution of the parish-level random effect in the sample of individuals (i.e. to the median value of the random effect among the 25% of all individuals in parishes with the lowest and highest mortality risk). Accordingly, the interquartile hazard ratio was computed as exp(u87.5 – u12.5). It is a very conservative estimate of between-parish variations, since it is based on the parish-level random effect obtained after shrinkage.
We then included all individual variables into the models, and in a second step the two contextual variables (separately or simultaneously). We used the Akaike Information Criterion (AIC)36 to examine whether including each contextual variable led to a significant improvement in model fit (the lower the AIC, the better the fit of the model). Since it may be desirable to find the least complex model that best fits to the data, the AIC combines a measure of complexity and a measure of fit to identify the model that describes the data in the most efficient way. Finally, we examined the extent to which the introduction of population density and median income explained the observed geographic variability in mortality risk.
|
Results |
|---|
|
TopAbstractMethodsResultsDiscussionReferences |
|---|
The present study focused on individuals who had been exposed to the same residential environment for at least 10 years prior to mortality follow-up. In fact, most of those individuals resided in the same parish for even longer. For example, for individuals aged 55 years for whom mortality follow-up began in 1990, 89% of the 10 year stable residents resided for at least 15 years and 66% for at least 20 years in the same parish (the corresponding figures were 91 and 80% for individuals aged 75 years at baseline).
As shown in Table 1, the risk of mortality from each of the causes investigated increased with age at baseline (with the exception of lung cancer among females). Among individuals aged 55 years at baseline, death by COPD occurred in <0.5% of all individuals. Owing to an insufficient number of events, mortality from COPD was not further investigated in the youngest group.
Figure 1 reports continuous maps of risk of mortality from specific causes among individuals aged 65 years at baseline, as obtained from the geoadditive models. A visual inspection of these descriptive maps indicates similarities in the spatial distributions of lung cancer mortality and COPD mortality, with higher mortality rates on the west coast (where the two largest cities of Scania, Malmö and Helsinborg, are located).
|
Among males of all ages, Cox multilevel models including individual factors indicated that both a low income at 1 year and a low income at 10 years prior to baseline were independently associated with an increased risk of mortality from IHD and COPD (results not shown). Among 55-year-old females, IHD mortality increased with decreasing income at 1 year prior to baseline, but there was evidence of a significantly higher lung cancer mortality among affluent females.
In those Cox multilevel models adjusted for individual factors, the between-parish variance and interquartile hazard ratio35 allowed us to assess the extent to which mortality depended on the residential context (Table 2). Regarding IHD mortality, between-parish variations were substantial only among the youngest individuals and decreased with increasing age. After individual-level adjustment, between-parish variations in lung cancer mortality were not statistically significant among females but markedly increased with age among males. There were also statistically significant between-parish variations in COPD mortality, except among the eldest females.
|
We then considered contextual variables. As shown in Figure 2, even if there was a great variability in median income at each level of population density, a non-monotonic association was found between the two factors, with socioeconomic level first increasing with population density (as one moves from rural to urban parishes) then decreasing with further increase in population density (as one goes from affluent to impoverished urban parishes). This inverted U-shape pattern was confirmed with correlation statistics, since we found a positive correlation between population density and median income when considering all parishes except those of Malmö, the largest city of Scania (Pearson correlation for 351 parishes = 0.31; P < 0.001), but a negative correlation among the 19 parishes of Malmö (Pearson correlation = –0.50; P = 0.03). Even if it was substantial, the correlation between the two contextual factors was not high enough to prevent us from investigating the independent effects of population density and median income on cause-specific mortality.
|
We first computed percentages of deaths in each quartile of population density and median income. The most obvious pattern of association was a strong graded increase in lung cancer and COPD mortality with population density, especially among males. We then introduced population density and median income into the models. Figures 3 and 4 report associations between contextual factors and mortality risk, with contextual effects mutually adjusted for each other and adjusted for individual factors. Effects of population density were found for the three causes of death investigated in females and males of all ages, except for IHD in the oldest females where the effect was non-existent and for lung cancer in the youngest females where the observed dose–response effect was not statistically significant. The association between population density and IHD mortality weakened with increasing baseline age. The strongest population density effect was found for lung cancer mortality among males aged 65 and 75 years. In addition to the effect of population density, a dose–response relationship between median income and mortality was noted only for IHD among 55-year-old females and for COPD among 65-year-old males.
|
|
As shown in Table 3, inclusion of population density in models already containing median income led to a decrease in AIC ranging from 16 to 155, indicating a substantial improvement in model fit. In all cases, but one, introduction of population density resulted in a far larger decrease in AIC than inclusion of median income. In 7 cases out of 16, inclusion of median income increased the AIC, indicating absolutely no improvement in model fit.
|
As shown in Table 2, inclusion of population density and median income into the models led to a 73–98% decrease in between-parish variance (except in the model for IHD among oldest females, in which there was no variance before inclusion of contextual characteristics). Between-parish variance was reduced to non-significance in all the models (P-value between 0.08 and 0.90).
We verified that associations between population density and mortality did not result from the exclusion of individuals who moved from one parish to another during the 10 years prior to mortality follow-up—something that might have occurred if individuals with a high propensity to die who were residing in low population density parishes had significant odds of moving into high density parishes in order to be closer to healthcare services. We found no evidence of such a selection bias, since associations remained statistically strong even when the analyses included individuals who had moved.
|
Discussion |
|---|
|
TopAbstractMethodsResultsDiscussionReferences |
|---|
We sought to determine whether the occurrence of cause-specific mortality in the region of Scania depended on the residential context (apart from any effects associated with individual characteristics) and whether parish of residence influenced mortality through processes related to population density and/or median income (as a marker of overall socioeconomic environment).
One strength of our study is that our database included all individuals born between 1905 and 1938 residing in the region of Scania, thus, providing good power to assess the mortality risk in each parish, and thereby between-parish variations. Furthermore, we selected individuals exposed to the same residential environment for at least 10 years prior to mortality follow-up (most of those individuals residing in the same parish for even longer). On one hand, such a design ensured a more correct measurement of exposure to contextual factors than if residents who moved had also been included. However, owing to the long induction time of the chronic diseases investigated, this period of exposure to a similar residential environment may not have been long enough to entirely resolve the measurement issue. On the other hand, contrary to most previous studies, selecting stable, long-term residents certainly allowed us to rule out the possibility that associations between contextual factors and mortality result from the selective migration of individuals who were ill in the direction of densely populated or deprived parishes.
The most important limitation of our study was that our large-scale database did not include information on biomedical and behavioural risk factors. However, before studies investigate the exact mediating mechanisms of contextual effects, it is necessary to describe those effects properly, a task to which our study contributes. Even if our findings do not have a direct aetiological relevance, they provide useful guidance for future investigation of the detailed causal pathways. Another limitation was that we used individual income, rather than household income. This may have resulted in a worst adjustment of the models among females. Moreover, income data were on pre-tax income that includes income from employment and business but excludes capital income or income transfers. However, since those data allowed us to capture individual-level socioeconomic effects on mortality, it is very unlikely that the absence of a systematic effect of the socioeconomic environment could result from our inability to measure it adequately.
In previous research, the investigation of effects of population density7,28 and socioeconomic environment9–16,18,19 has been conducted independently, with population density effects generally investigated in regional or national territories, and the impact of the socioeconomic environment analysed within urban areas or in broader territories. Moreover, most previous studies of the population density effects on mortality were ecological.3–5,7,8 In our region-wide study, we found population density effects on mortality from IHD, lung cancer, and COPD in all age x gender groups investigated, but one, after adjustment for individual characteristics. Median income showed an additional effect on mortality only for IHD among 55-year-old females and for COPD among 65-year-old males. Therefore, parish-level contextual influences on cause-specific mortality were dominated by population density effects.
It is important to note that the relative strength of population density and median income effects observed in our analysis is relative to (i) the study territory (region-wide area) and (ii) geographical scale of analysis (parishes). Different results may have been found in exclusively urban territories (e.g. the city of Malmö) or if more or less local geographical units had been considered. Our approach that investigates simultaneously population density and socioeconomic environment effects clearly needs to be replicated in those circumstances.
At the parish level, we found an inverted U-shape association between population density and median income. Urbanicity may be associated with higher socioeconomic circumstances, but within urban territories higher population densities may be found in deprived parishes. Such a complex pattern of association raises the possibility of confounding biases playing in different directions depending on the territory and geographical scale of analysis. Therefore, in studies interested in the effects of one of those two contextual dimensions, it may be recommended to consider both of them simultaneously, to avoid making misleading inferences on contextual effects.
In our study, contextual influences at the parish level appeared to be dominated by population density effects. However, it may be unwise to conclude that truly contextual effects exist on mortality, before identification of plausible mediating processes that are clearly associated with context and the specific outcome. Following previous literature,20 future research may consider that the facets of urban living that negatively affect health are related to the (i) physical environment and (i) social environment. Relevant features of the physical environment include the built environment, automotive and industrial air pollution, and noise pollution.37,38 For example, urban air pollution is associated with respiratory morbidity and mortality,21,22 and noise exposure may increase the risk of hypertension and IHD.23 The social environment, characterized by specific cultural milieus, social norms, and social interaction patterns,24 may affect the risk of mortality through individual health-damaging and healthcare-seeking behaviour.20,25 For example, more consumerist attitudes may be found in urban territories, leading to heavier consumptions of high-fat food, tobacco, and alcohol.7,26,37,39 Moreover, a stronger propensity to individualism in the urban environment combined with effects of overcrowding on social behaviour and mistrust between residents may result in patterns of interaction between individuals that provide less social support and material assistance in case of need.40–42 However, analyses will also have to consider the positive aspects of urban living, such as the far wider availability of healthcare services than that in non-urban territories.20
Future investigation of the underlying causal mechanisms, which are specific to each cause of death, will allow one to assess whether population density may be seen or not seen as an important upstream determinant of mortality operating through intermediate processes such as those enumerated above. A clear understanding of the mediating mechanisms will be necessary to propose public health interventions targeted at the exact contextual processes at play.43 In the meanwhile, the population density of an area may be considered as a risk marker in itself, allowing one to identify places of higher mortality from IHD, lung cancer, and COPD.
KEY MESSAGES
|
|
Acknowledgments |
|---|
We want to express our gratitude to Statistics Sweden, the Centre for Epidemiology (National Board of Health and Welfare), and to Region Skåne. This investigation was supported by a grant awarded to Basile Chaix by the French Foundation for Medical Research; the Swedish Council for Working Life and Social Research (PI Juan Merlo, Dnr: 2003-05809); and the Swedish Research Council (PI Juan Merlo, Dnr: 2004-6155).
|
References |
|---|
|
TopAbstractMethodsResultsDiscussionReferences |
|---|
1 Wong MD, Shapiro MF, Boscardin WJ, Ettner SL. Contribution of major diseases to disparities in mortality. N Engl J Med 2002;347:1585–92.
2 Kawachi I, Berkman LF (eds). Neighborhoods and Health. New York, NY: Oxford University Press, 2003.
3 Barnett E, Halverson J. Disparities in premature coronary heart disease mortality by region and urbanicity among black and white adults ages 35–64, 1985–1995. Public Health Rep 2000;115:52–64.[ISI][Medline]
4 Barnett E, Halverson JA, Elmes GA, Braham VE. Metropolitan and non-metropolitan trends in coronary heart disease mortality within Appalachia, 1980–1997. Ann Epidemiol 2000;10:370–79.[CrossRef][ISI][Medline]
5 Ingram DD, Gillum RF. Regional and urbanization differentials in coronary heart disease mortality in the United States, 1968–85. J Clin Epidemiol 1989;42:857–68.[CrossRef][ISI][Medline]
6 Smith MH, Anderson RT, Bradham DD, Longino CF Jr. Rural and urban differences in mortality among Americans 55 years and older: analysis of the National Longitudinal Mortality Study. J Rural Health 1995;11:274–85.[Medline]
7 Mahoney MC, LaBrie DS, Nasca PC, Wolfgang PE, Burnett WS. Population density and cancer mortality differentials in New York State, 1978–1982. Int J Epidemiol 1990;19:483–90.
8 Reijneveld SA, Verheij RA, de Bakker DH. Relative importance of urbanicity, ethnicity and socioeconomic factors regarding area mortality differences. J Epidemiol Community Health 1999;53:444–45.[ISI][Medline]
9 Haan M, Kaplan GA, Camacho T. Poverty and health. Prospective evidence from the Alameda County Study. Am J Epidemiol 1987;125:989–98.
10 Yen IH, Kaplan GA. Neighborhood social environment and risk of death: multilevel evidence from the Alameda County Study. Am J Epidemiol 1999;149:898–907.
11 Martikainen P, Kauppinen TM, Valkonen T. Effects of the characteristics of neighbourhoods and the characteristics of people on cause specific mortality: a register based follow up study of 252,000 men. J Epidemiol Community Health 2003;57:210–17.
12 Bosma H, van de Mheen HD, Borsboom GJ, Mackenbach JP. Neighborhood socioeconomic status and all-cause mortality. Am J Epidemiol 2001;153:363–71.
13 Marinacci C, Spadea T, Biggeri A, Demaria M, Caiazzo A, Costa G. The role of individual and contextual socioeconomic circumstances on mortality: analysis of time variations in a city of north west Italy. J Epidemiol Community Health 2004;58:199–207.
14 Jaffe DH, Eisenbach Z, Neumark YD, Manor O. Individual, household and neighborhood socioeconomic status and mortality: a study of absolute and relative deprivation. Soc Sci Med 2005;60:989–97.[CrossRef][ISI][Medline]
15 Subramanian SV, Chen JT, Rehkopf DH, Waterman PD, Krieger N. Racial disparities in context: a multilevel analysis of neighborhood variations in poverty and excess mortality among black populations in Massachusetts. Am J Public Health 2005;95:260–65.
16 Anderson RT, Sorlie P, Backlund E, Johnson N, Kaplan GA. Mortality effects of community socioeconomic status. Epidemiology 1997;8:42–47.[ISI][Medline]
17 van Lenthe FJ, Borrell LN, Costa G, Diez Roux AV, Kauppinen TM, Marinacci C et al. Neighbourhood unemployment and all cause mortality: a comparison of six countries. J Epidemiol Community Health 2005;59:231–37.
18 Sloggett A, Joshi H. Higher mortality in deprived areas: community or personal disadvantage? BMJ 1994;309:1470–74.
19 Roos LL, Magoon J, Gupta S, Chateau D, Veugelers PJ. Socioeconomic determinants of mortality in two Canadian provinces: multilevel modelling and neighborhood context. Soc Sci Med 2004;59:1435–47.[CrossRef][ISI][Medline]
20 Vlahov D, Galea S. Urbanization, urbanicity, and health. J Urban Health 2002;79:S1–12.[Medline]
21 Nafstad P, Haheim LL, Wisloff T, Gram F, Oftedal B, Holme I et al. Urban air pollution and mortality in a cohort of Norwegian men. Environ Health Perspect 2004;112:610–15.[ISI][Medline]
22 Sunyer J, Schwartz J, Tobias A, Macfarlane D, Garcia J, Anto JM. Patients with chronic obstructive pulmonary disease are at increased risk of death associated with urban particle air pollution: a case-crossover analysis. Am J Epidemiol 2000;151:50–56.
23 van Kempen EE, Kruize H, Boshuizen HC, Ameling CB, Staatsen BA, de Hollander AE. The association between noise exposure and blood pressure and ischemic heart disease: a meta-analysis. Environ Health Perspect 2002;110:307–17.[Medline]
24 Stafford M, Bartley M, Wilkinson R, et al. Measuring the social environment: social cohesion and material deprivation in English and Scottish neighbourhoods. Environ Planning A 2003;35:1459–75.[CrossRef]
25 Veenstra G, Luginaah I, Wakefield S, Birch S, Eyles J, Elliott S. Who you know, where you live: social capital, neighbourhood and health. Soc Sci Med 2005;60:2799–818.[CrossRef][Medline]
26 Chaix B, Chauvin P. Tobacco and alcohol consumption, sedentary lifestyle and overweightness in France: a multilevel analysis of individual and area-level determinants. Eur J Epidemiol 2003;18:531–38.[CrossRef][ISI][Medline]
27 Verheij RA, van de Mheen HD, de Bakker DH, Groenewegen PP, Mackenbach JP. Urban-rural variations in health in The Netherlands: does selective migration play a part? J Epidemiol Community Health 1998;52:487–93.[Abstract]
28 Howe HL, Keller JE, Lehnherr M. Relation between population density and cancer incidence, Illinois, 1986–1990. Am J Epidemiol 1993;138:29–36.
29 Nasca PC, Burnett WS, Greenwald P, Brennan K, Wolfgang P, Carlton K. Population density as an indicator of urban-rural differences in cancer incidence, upstate New York, 1968–1972. Am J Epidemiol 1980;112:362–75.
30 Wood SN. Thin plate regression splines. J R Stat Soc Ser B Stat Methodol 2003;65:95–114.[CrossRef]
31 Wood SN. Multiple smoothing parameter estimation and GAMs by GCV. Available at: http://www.r-project.org/ (20 January 2006, date last accessed).
32 Therneau T. Survival R package, version 2.20. Available at: http://www.r-project.org/ (20 January 2006, date last accessed).
33 Merlo J, Chaix B, Ohlsson H, et al. A brief conceptual tutorial on multilevel analysis in social epidemiology—using measures of clustering in multilevel logistic regression to investigate contextual phenomena. J Epidemiol Community Health 2005; (In press).
34 Larsen K, Merlo J. Appropriate assessment of neighborhood effects on individual health: integrating random and fixed effects in multilevel logistic regression. Am J Epidemiol 2005;161:81–88.
35 Chaix B, Merlo J, Subramanian SV, Lynch J, Chauvin P. Comparison of a spatial perspective with the multilevel analytic approach in neighborhood studies: the case of mental and behavioral disorders due to psychoactive substance use in Malmö, Sweden, 2001. Am J Epidemiol 2005;162:171–82.
36 Burnham KP, Anderson DR. Model Selection and Inference. New York: Springer, 1998.
37 Doll R. Urban and rural factors in the aetiology of cancer. Int J Cancer 1991;47:803–10.[Medline]
38 Pope CA 3rd, Burnett RT, Thun MJ, Calle EE, Krewski D, Ito K et al. Lung cancer, cardiopulmonary mortality, and long-term exposure to fine particulate air pollution. JAMA 2002;287:1132–41.
39 Fukuda Y, Nakamura K, Takano T. Accumulation of health risk behaviours is associated with lower socioeconomic status and women's urban residence: a multilevel analysis in Japan. BMC Public Health 2005;5:53–62.[CrossRef][Medline]
40 Korte C. Helpfulness in the urban environment. In: Baum A, Singer JE, Valins S (eds). Advances in Environmental Psychology: Vol I - The Urban Environment. New York: John Wiley & Sons. 1978, pp. 85–109.
41 Baum A, Valins S. Architectural mediation of residential density and control: Crowding and the regulation of social contact. In: Berkowitz L (ed.). Advances in Experimental Social Psychology. Vol. 12. New York: Academic Press, 1979, pp. 131–75.
42 Boots BN. Population density, crowding and human behaviour. Prog Hum Geogr 1979;3:13–63.[Medline]
43 Diez Roux AV. Residential environments and cardiovascular risk. J Urban Health 2003;80:569–89.[ISI][Medline]
CiteULike 
Connotea 
Del.icio.us What's this?
This article has been cited by other articles:
|
|
 |
|
  B Chaix, M Lindstrom, M Rosvall, and J Merlo Neighbourhood social interactions and risk of acute myocardial infarction J. Epidemiol. Community Health, January 1, 2008; 62(1): 62 - 68. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
M. Riva, L. Gauvin, and T. A Barnett Toward the next generation of research into small area effects on health: a synthesis of multilevel investigations published since July 1998 J. Epidemiol. Community Health, October 1, 2007; 61(10): 853 - 861. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 M Rosvall, G Engstrom, B Hedblad, L Janzon, and G Berglund Area social characteristics and carotid atherosclerosis Eur J Public Health, August 1, 2007; 17(4): 333 - 339. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
B. Chaix, M. Rosvall, and J. Merlo Assessment of the magnitude of geographical variations and socioeconomic contextual effects on ischaemic heart disease mortality: a multilevel survival analysis of a large Swedish cohort J. Epidemiol. Community Health, April 1, 2007; 61(4): 349 - 355. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 B. Chaix, M. Rosvall, and J. Merlo Recent Increase of Neighborhood Socioeconomic Effects on Ischemic Heart Disease Mortality: A Multilevel Survival Analysis of Two Large Swedish Cohorts Am. J. Epidemiol., January 1, 2007; 165(1): 22 - 26. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 J. MERLO and B. CHAIX Neighbourhood effects and the real world beyond randomized community trials: a reply to Michael J Oakes Int. J. Epidemiol., October 1, 2006; 35(5): 1361 - 1363. [Full Text] [PDF]
|
|
|
|
|
|
J M. Oakes Commentary: Advancing neighbourhood-effects research--selection, inferential support, and structural confounding Int. J. Epidemiol., June 1, 2006; 35(3): 643 - 647. [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||