Neighbourhood socioeconomic context and self reported health and smoking.

Objective: Many studies have shown that poor health status and harmful health behaviour occur more frequently in deprived neighbourhoods. Most studies show modest associations between area level socioeconomic factors, the neighbourhood context, and health outcomes. However, estimates for the contextual effects vary. It is unclear if this variation is attributable to differences in methodology. This study examines whether contextual neighbourhood differences in health outcomes really vary between cities or that differences in methodology may account for these differences.

Design: Secondary analysis of data from health interview surveys in seven large Dutch cities in the 1990s comprising 23 269 residents of 484 neighbourhoods, using multilevel logistic models.

Setting: General population aged 16 and over.

Main outcome measures: Self reported health, smoking of cigarettes.

Main results: The socioeconomic context of neighbourhoods is associated with health outcomes in all large Dutch cities. The strength of the association varies between cities, but variation is much smaller in the age group 25-64. Furthermore, neighbourhood differences vary in size between native and other residents. Contextual neighbourhood differences are about two times larger for self reported health than for the smoking of cigarettes, but for native Dutch people they are of similar size.

Conclusions: A comparatively large improvement in health may be gained in deprived neighbourhoods, because of the poorer health status to which the context of these neighbourhoods also contributes. Health promoting interventions should be aimed at the residents and at the context of deprived neighbourhoods, taking differences between ethnic groups and age groups into account.

**********

Many studies have shown that poor health and harmful health behaviour occur more frequently in deprived neighbourhoods. (1-7) These differences may be caused by factors at different levels. Firstly, they can simply be the result of individual socioeconomic (SE) health differences, as the socioeconomic status (SES) of residents of these neighbourhoods is lower than the average, and lower SES is associated with more unfavourable health and lifestyles in general. (8,9) This is also called the compositional effect. Secondly, area level factors, the so called neighbourhood context, may contribute to the unfavourable health position of deprived neighbourhoods (the contextual effect). Examples of such area level factors are an adverse physical environment and poorer community services in deprived neighbourhoods.

In a recent review, (1) Picket and Pearl conclude that most available studies show at least some associations between (social-)contextual factors and health outcomes; they are generally of modest size and smaller than the compositional effects. Picket and Pearl only report statistically significant associations, but even then the range of estimates for contextual effects is rather wide. They hypothesise that this wide range may be attributable to heterogeneity of study designs and of analytical methods, or to random variation. However, real differences between cities, regions, and countries may also explain this diversity. (10)

The aim of this study is to examine the impact of the "city level" on the size of the contextual effects regarding neighbourhood health outcomes. It focuses on neighbourhood level SE differences in health outcomes in big cities in one country, the Netherlands, on the basis of a standardised analysis of previously performed health surveys. Central questions are:

* does the size of neighbourhood level SE differences in health outcomes vary between cities (after adjustment for differences in age and gender);

* do these differences persist after adjustment for individual SES--that is, does the size of the contextual effects vary between cities.

METHODS

The study concerned a secondary (multilevel) analysis of data on self reported health and smoking, derived from health surveys that were performed in Dutch big cities in the past decennium, in people aged 16 and over.

Data

Individual data

Individual data came from surveys that were performed in the period 1991-2000 in the larger Dutch cities and met the following criteria:

* they contained data on educational level, self reported health and/or smoking, age and gender, and neighbourhood of residence of the respondent;

* they covered at least the age range 20-64;

* they had a mean of at least 15 respondents per neighbourhood in that age range;

* they used the population register of the city concerned as sampling frame (all Dutch residents have to register in the municipality in which they live);

* they concerned Dutch cities with at least 100 000 residents (for these cities, the Dutch government has declared a separate policy, which implies that national measures can be taken in response to conditions that are specific for a given city). (11)

Data from seven cities met these criteria and were made available by the Municipal Health Services concerned. Details of these surveys as well as some characteristics of these cities are presented in table 1. The quality of each survey was assessed by the following criteria (between parentheses the number of points for quality, leading to a quality score):

* response (60% and over: 1 point; lower: none);

* method of data collection (uniform strategy regarding all respondents: 1 point; otherwise: none);

* selectiveness of non-response (shown to be lacking regarding age and gender: 1 point; otherwise: none).

After restriction to the relevant age range (16 and over) and to records providing all relevant variables (background characteristics and self reported health and/or smoking), the analysis concerned 23 269 people in 484 neighbourhoods in seven cities.

Health outcomes concerned self reported health and smoking of cigarettes; in the analysis both were dichotomised, and poor health and smoking of cigarettes were predicted in the models. Self reported health concerned three different types of questions. The appendix contains a description of the wording of all questions and the method of dichotomisation. Educational level concerned the highest degree earned, in four levels: primary school, lower secondary school, higher secondary school, post-secondary education.

Area data

Area data concerned the socioeconomic position of the neighbourhood in 1995. Measures were: general practitioner (GP) deprivation score, mean income per earner (further: "mean income"), and proportion of residents aged 16-64 who were dependent on social benefits (further: "social benefits").

The GP deprivation score was computed as the sum of the standardised scores of each neighbourhood regarding degree of urbanisation, proportion of ethnic minorities, "mean income" and "social benefits". Regarding the 5% of the population that lives in areas with the highest scores, Dutch GPs receive an additional fee for each patient,(13) comparable to the UK Jarman system. (14) For this study. I computed a deprivation score per neighbourhood although the original score concerns postcode sectors. The reason for choosing neighbourhoods is that postcode sectors have a logistic origin--that is, adequate post delivery--and were designed at a national level. In contrast, neighbourhoods consist of areas with similar housings, often delineated by natural boundaries. Because of this, they are socio-culturally rather homogenous. In the planning of health and other local services, neighbourhoods are usually the lowest level that is considered; these neighbourhoods are often used as equivalents for local communities. I u sed a cut off that identifies those 5% of the Dutch population that lives in the most deprived areas according to the GP score. Regarding the analyses on self reported health, 19% of the respondents lived in such areas; regarding the analyses on smoking this concerned 18%. Similar percentages apply to the …