Variations in mortality, morbidity, and risk factors across neighbourhoods and larger areas can be explained by compositional characteristics of residents and by contextual characteristics of the area. Living in a deprived area may be associated with poorer health over and above individual risk factors.^1–12 The importance of area deprivation for health may vary by country for several reasons. Firstly, socioeconomic segregation may be greater in some countries, with people of different socioeconomic characteristics living in separate neighbourhoods. The extent of socioeconomic segregation may be influenced by provision of socially assisted housing and local planning policies as well as factors outside the remit of local government, such as local population demographics and the private housing market.¹³ In a country with greater socioeconomic segregation, we expect to see a wider range of deprivation and affluence and larger neighbourhood effects on health. Secondly, the effect of neighbourhood deprivation on health may be diminished in countries with universal access to high quality facilities and services.¹¹ Accordingly, Ross et al¹⁴ found an association between income inequality in metropolitan areas and mortality in the US but not in Canada. They suggested that differences in the provision of local services, which are market led in the US but universally available and publicly funded in Canada, may explain their findings.

Here we compare neighbourhood variations in health and the health effects of neighbourhood deprivation in two capital cities. London, UK, has a population of over 7 million¹⁵ and covers an area of 1500 km².¹⁶ There are substantial inequalities in unemployment, housing quality, and health across London. For example, 7% of all London dwellings are unfit to live in, ranging from 1.6% in an outer London borough to 18.3% in a borough in the east end.¹⁷ Over the past two decades, approaches to public housing management have tended to widen inequalities. There has been a policy of transferring public housing to the private sector through the “right to buy” (Housing Act 1980, 1984, 1986) and a strategy of offering the least popular properties to the “least deserving” tenants.¹⁸ Initiatives to broaden the diversity of household types within areas have not been widespread in the UK.

The metropolitan area of Helsinki, Finland, has a population of 965 000 and covers 769km².¹⁹ Relatively deprived and affluent neighbourhoods can be distinguished, but local housing and city planning policies along with redistributive state policies have limited income inequalities in the country as a whole and locally. A stated aim of these policies has been to reduce socioeconomic segregation by building publicly funded housing and encouraging integration of different types of home ownership within areas.²⁰ Income inequalities are larger in the UK than in Finland.²¹ Given the information suggesting greater socioeconomic segregation in London and a greater prominence of welfare state provision in Helsinki, we hypothesised that health differences between neighbourhoods and the effect of neighbourhood deprivation would be larger in London than Helsinki.

DATA AND METHODS

The data for this comparative study come from cohorts of public sector employees in London and Helsinki.

The London sample

All civil servants aged 35–55 working in London offices at baseline (1985–8) were invited to participate in the Whitehall II study²² yielding a total of 10 308 participants (response rate 73%). The individual data used here were collected by self completion questionnaire at phase 3 (1991–3) as this coincided most closely with collection of the area level data. Participants living within the M25 (an orbital motorway encircling Greater London) were eligible. Those lost through sample attrition by phase 3 (n = 1492), not living in the focus area (n = 3173), or with missing socioeconomic or health data (n = 342) were excluded giving a total sample of 5301 participants.

The Helsinki sample

Helsinki health study participants provide data for Helsinki (http://www.kttl.helsinki.fi/projektit/helsinki/helsinkitutkimus.htm). Municipal employees aged 40–60 working for the City of Helsinki in 2000–1 were invited to participate. With a response rate of 68%, 6243 returned self completion questionnaires. According to non-response analysis, the data are reasonably representative of the target population.²³ Participants living outside the Helsinki metropolitan area (municipalities of Helsinki, Espoo, Vantaa, and Kauniainen, n = 1914) or not having complete health, socioeconomic or area data (n = 42) were excluded leaving 4287 employees. The Helsinki and Whitehall II studies shared several instruments.

Individual level measurement of health

In both samples self rated health was obtained from the question “In general would you say your health is” with five responses ranging from excellent to poor. Self rated health reflects medical and other health information ranging from knowledge of existing diseases to cognitive and affective characteristics.²⁴ It has good test-retest reliability,²⁵ predicts mortality,²⁶ and has been recommended for comparative purposes.²⁷ The outcome was categorised as 0 = excellent, very good, good, and 1 = fair and poor to examine “less than good health”.

Individual level measurement of socioeconomic position

In both samples information on employment grade, our measure of socioeconomic position, was obtained using the following hierarchy: (1) managerial/administrative, (2) professional and semi-professional, (3) clerical, and (4) manual workers.

Area level measurement

In London, 1991 census data were used and census ward boundaries (average population 8000) defined neighbourhoods. In Helsinki, 2000 census data were used (with the exception of proportion of manual workers from the 1995 census). The classification used by the municipalities of Helsinki metropolitan area defined neighbourhoods (average population 4000). In both studies, neighbourhood areas are such that most residents could walk across in 15–20 minutes. These areas are used for the planning of amenities by local government and health service providers. We contend that these are suitable boundaries to define and compare neighbourhoods, although we note that administrative boundaries may not correspond to residents’ experiences and perceptions.

Unemployment rates, proportion of single households, proportion of single parent households, and proportion of manual workers were used to describe neighbourhoods using the same definitions in London and Helsinki (table 1). These four indicators capture the socioeconomic characteristics of neighbourhoods (income is lower in single households and in single parent households) and were chosen for comparability between the two cities. Data on income and educational attainment were not easily comparable or reliable and were not included. Within each city, neighbourhoods were split into four equally sized groups (quartiles) for each neighbourhood indicator.

Statistical methods

The index of dissimilarity captures residential segregation, measuring the degree of spatial separation of two population subgroups across neighbourhoods within a city²⁸ and taking values between 0 and 100 (maximum segregation). For unemployment it is calculated as:

x_i = number of unemployed residents in neighbourhood i, X = total number of unemployed people in all study neighbourhoods, y_i = number of employed residents in neighbourhood i, Y = total number of employed people in all study neighbourhoods.

The association between neighbourhood deprivation and self rated health was investigated using multilevel logistic regression models (MLwin v1.10, Institute of Education, London) built up sequentially:

π_ij = probability of i^th participant in j^th neighbourhood having less than good health, β₀ = log odds of less than good health in London, β₁ = log odds of less than good health in Helsinki, u_0j = difference in health in neighbourhood j compared with London average, and u_1j = difference in health in neighbourhood j compared with Helsinki average (all adjusted for age and sex). The model was fitted with an overall intercept for London (β₀) and another for Helsinki (β₁) and with random intercepts for each neighbourhood in London (u_0j) and in Helsinki (u_1j).

The total variation in self rated health is partitioned into that between neighbourhoods (σ_u0² in London and σ_u1² in Helsinki) and that within neighbourhoods. σ_u0² and σ_u1² can be interpreted through the intraclass correlation coefficient (ICC), which is the proportion of total variation in health that occurs between neighbourhoods.

Model 2: Base model + neighbourhood indicator as continuous score* + city x neighbourhood indicator interaction (separate model for each of the four indicators)

Model 3: Base model + neighbourhood indicator quartile

Model 4: Model 3 + adjustment for individual socioeconomic position

Model 2 allows us to compare the health effect of a 1% increase in the level of the neighbourhood indicator in the two cities. For presentation, the models were repeated using quartiles of neighbourhood indicators. Effects of age, sex, and socioeconomic position were allowed to vary by city by the addition of interaction terms. Statistical significance was assessed using the Wald test.³⁰

Ethical approval for human research was obtained for each study.

RESULTS

Unemployment was higher in London than in Helsinki and the difference in unemployment rate between the top and bottom quartiles was larger in London (table 1). This is reflected in a higher segregation index for employed compared with unemployed residents in London (23.3) compared with Helsinki (17.1) and similarly for manual compared with non-manual workers. The extent of socioeconomic separation was larger in London than in Helsinki. Single households were separated from multiple person households slightly more in Helsinki. Single parent households were separated from two parent households to a greater extent in London.

Table 2 summarises the study participants. An average of 6.1 participants per neighbourhood in London and 19.3 per neighbourhood in Helsinki had complete data. There were no manual workers in the London sample. In both cohorts, reporting of less than good self rated health increased with age and was slightly higher for women than for men. Participants in lower socioeconomic positions were more likely to report less than good health than those in higher positions. The social gradient was steeper in London. This was allowed for by including interaction terms between socioeconomic position and city (that is, study cohort).

There were no differential effects of neighbourhood unemployment, proportion of residents in manual occupations, or proportion of single households in Helsinki and London. In pooled data, residence in a neighbourhood with highest unemployment was associated with an age and sex adjusted odds ratio of less than good health of 1.78 (95% CI 1.54 to 2.06) (table 3, model 3). Proportion of manual workers and single households were also associated with self rated health. Living in a neighbourhood with the highest proportion of single parent households was associated with a 21% higher odds of reporting less than good health in Helsinki (OR 1.21, 95% CI 0.97 to 1.50) and a 67% higher odds in London (OR 1.67 95% CI 1.09 to 2.55) (test for interaction, p = 0.006).

Adjustment for individual socioeconomic position attenuated the effects of neighbourhood (table 3, model 4). Part of the increased odds of less than good health in neighbourhoods with highest unemployment was attributable to the lower employment grade of residents there. However, independent effects of neighbourhood unemployment remained. Modest effects of proportion of residents in manual occupations and proportion of single households also remained. A slightly larger effect of proportion of single parent households in London remained, but the test for interaction was not statistically significant. These data are suggestive of a small contextual effect of single parent households in London but not in Helsinki.

In the age and sex adjusted model, the variation between neighbourhoods in London was significant (Wald test p<0.01) (table 4). In both cities, a relatively small proportion of the total variation in self rated health was between neighbourhoods (3% in London and 0.4% in Helsinki). Adjustment for individual socioeconomic position reduced between neighbourhood variation to 2% of the total in London and 0.1% of the total in Helsinki. There was greater variation in self rated health between neighbourhoods in London, and this excess was not entirely attributable to greater spatial separation of high and low grade participants in London. Adjustment for neighbourhood indicators further reduced variation between neighbourhoods.

DISCUSSION

The spatial separation of residents in high and low socioeconomic positions was higher in London than in Helsinki. Income inequality is higher in the UK than in Finland and findings from the US show it is highly correlated with socioeconomic segregation^31,32 and so we hypothesised that segregation would be higher in London. We have shown this empirically and we suggest that policies aimed at preventing socioeconomic segregation in the Helsinki metropolitan area may have had some success when combined with national policies to reduce income inequality. Note though that residential segregation is greater in more populous places (although segregation in London is smaller than expected given its size³³). London and Helsinki differ in population size as well as in policy approaches.

Between neighbourhood differences in self rated health were also greater in London. Other studies point to large health differences over small distances in London. Life expectancy in Westminster was five years higher than in Newham, six underground stations away.³⁴ Both individual socioeconomic position and contextual effects (captured by unemployment rate, proportion of manual workers, single households, and single parent households) explained health differences between neighbourhoods. The spatial separation of residents may have emergent effects—at a certain level, the concentration of disadvantaged residents may begin to have effects on other residents irrespective of their own characteristics. This separation may affect health by isolating residents in lower social positions from economic opportunities, from groups with political power, and from potentially health enhancing public and private services that depend on local demand and local income, and/or by weakening social connections and social cohesion.³⁵

This study provides further evidence that features of the residential environment are likely to be associated with self rated health, over and above individual characteristics. These contextual effects were small to moderate in size; residence in high unemployment neighbourhoods was associated with a 50% increase in the likelihood of reporting less than good health. We found no difference in the magnitude of the deprivation health slope for three of the neighbourhood indicators. An association between proportion of single parent households and health was seen in London but was less evident in Helsinki. Although lone motherhood is likely to be a disadvantaged position also in Finland, lone mothers are almost as often employed as married mothers. In Britain, being a lone mother strongly reduces the likelihood of being employed.³⁶ In Finland, the employment of women, including lone mothers, has been supported by social policies, for example by providing widely available and subsidised public childcare. The municipality is required by law to provide childcare for all children aged less than 3 years. A recent international comparison found a variety of ways to support families in Finland, but in Britain such supports are clearly below the OECD average level.³⁷ A comparison of policies that may influence the health of lone parents indicates that Britain also lags behind Sweden.³⁸ The correlation between proportion of single parent households and unemployment rate was 0.83 in London and 0.33 in Helsinki. These findings suggest that inadequate resources and opportunities for single parents may have implications not only for those single parents and their children but also for the wider population.

As a proportion of the total variation, variation between neighbourhoods was small. This may be partly attributable to low statistical power to estimate between neighbourhood variation. However, small differences in self rated health between neighbourhoods combined with the large and consistent effects of individual socioeconomic position on self rated health found here (data not shown) and in other studies^22,39 highlight the importance of structural forces beyond the immediate neighbourhood that contribute towards shaping individual health.

This study was limited to public sector employees and is not representative of the general population. Neighbourhood effects may be smaller for employed participants who are less exposed to their neighbourhood and more exposed to other contexts (especially the workplace). Furthermore, some attrition in the Whitehall II sample means that participants included here were in higher socioeconomic positions than the original sample. Associations between neighbourhood characteristics and health seem to be less important for those in higher socioeconomic positions.⁴⁰ The associations presented here may be conservative, especially for the London sample. However, as our investigation was restricted to public sector employees, participants in the two cities may be more similar on unmeasured variables, which may improve the precision with which neighbourhood effects can be detected and compared. Two other sources of bias may also lead to underestimation of the differences between London and Helsinki neighbourhoods. Neighbourhood data from London were collected in 1991, and from Helsinki in 2001. Geographical variations in health have increased in the UK over the past 10 years so we might expect a comparison of the two cities in 2001 to show larger differences than found here. Additionally, within neighbourhood heterogeneity tends to increase (and between neighbourhood differences to decrease) with neighbourhood population size. If it were possible to use smaller neighbourhoods in London then greater differences between London neighbourhoods may be found.

Residual confounding by individual socioeconomic position may induce spurious associations at the neighbourhood level. Employment grade is shown to be a strong and consistent predictor of morbidity in the Whitehall II and Helsinki health studies.^9,22,41–43 While we cannot ignore the possibility that employment grade does not capture a person’s economic wellbeing completely, evidence suggests it is a good indicator in these working cohorts.

This study used neighbourhood indicators based on administrative data that are proxies for factors such as facilities and services, physical hazards, and employment opportunities that are thought to influence health. The four indicators were studied separately rather than as a summary index in order to identify which indicators were important for health in London and Helsinki. Comparative studies such as this can highlight points for further work—further exploration of the different policies towards single parents and their implications for parents and for wider society is indicated. This needs to be followed up with direct measurement of specific features of the residential context.⁴⁴

In summary, differences in the prevalence of less than good self rated health across neighbourhoods were larger in London than in Helsinki. Greater separation of residents in high and low socioeconomic positions in London may have emergent effects at the neighbourhood level. As with all comparative studies of public policy, there are many historical and current differences that lie outside the immediate remit of government and that cannot be controlled for.⁴⁵ Nevertheless, our findings suggest that local and national social policies may help reduce, or at least restrict, inequality and segregation along socioeconomic lines.