Introduction

As a result of increasing urbanisation, combined with a planning policy of spatial densification, more people face the prospect of living in residential environments with little green space. At the same time, increasing evidence shows that green space has beneficial effects on people’s health. Evidence has been found for a positive relation between green space and self-perceived health,1 2 3 4 longevity,5 number of symptoms and the risk of psychiatric morbidity.1 Access to a garden and shorter distances to green areas from the dwelling were associated with less stress and a lower likelihood of obesity.6 Experimental studies showed that there is a positive relation between green space and restoration from stress and mental fatigue. More specific, exposure to nature has been found to have a positive effect on mood, concentration, self-discipline and physiological stress.7 8 9 10 These studies indicate that there is a relation between green space and self-reported general indicators of physical and mental health. Thus, people living in greener environments report better physical and mental health. The decrease in green space could therefore have health consequences. However, it remains unknown whether living in residential environments with little green space also has negative consequences for objective health. In this explorative study we will go one step further than other studies and investigate whether the prevalence of several physician-assessed morbidity clusters is also related to the amount of green space in people’s living environment. This is the first study to investigate the relation between green space and prevalence of physician-assessed morbidity. This study has an explorative character and takes into account a broad number of diseases highly prevalent in society.

To gain more insight into the relation between green space and physician-assessed morbidity we analysed this relation separately for different age groups and different socioeconomic groups. We hypothesise that the relation is stronger for elderly people and children than adults because, as a result of their lower mobility, they spend more time in the vicinity of their home, resulting in higher exposure to green space in their living environment. The same applies to people with a lower socioeconomic status (SES), whose activities and social contacts are situated close to their homes.11 12 Therefore we also hypothesise that people with a lower SES are more exposed to the green space in their living environment. Finally, the relation was analysed for different levels of urbanicity to investigate whether the relation varies between urban and rural areas.

Methods

For this study data from two different datasets were combined. Morbidity data were collected within the framework of the second Dutch National Survey in General Practice (DNSGP-2), which included a nationwide, representative sample of 104 general practices with 195 GPs and a practice population of approximately 400 000 enlisted people, who were representative for the Dutch population in terms of age, gender and type of health insurance.13 For this study data from 96 practices that recorded morbidity for a full period of 12 months or more were used. This selection had no significant effect on the representativeness of the data, because after the selection the sample was still representative for the Dutch population.13 Only people who had been registered with their current GP for longer than 12 months prior to the study (n = 345 143) were included, because we assumed that people will have to live for at least 12 months in the same living environment before any effect of it would be noticeable.

Environmental data were derived from the National Land Cover Classification database (LGN4) in 2001, which contains the dominant type of land use of each 25×25 m grid cell in The Netherlands.14 The two datasets were matched on the basis of the x and y coordinates of the respondent’s six-character postal code (on average about 15–20 households have the same six-character postal code). The dataset included 50 187 postal codes and on average 6.9 respondents resided in each postal code area.

Morbidity data

Morbidity data were derived from routine primary care electronic medical records. In The Netherlands morbidity presented in general practice is a good indicator of morbidity in the population.15 Basically all non-institutionalised people are registered with a GP. Furthermore, GPs have a gate-keeping role for secondary care and are usually the first point of contact with the healthcare system. The data have been validated for obtaining prevalence estimates.16

During a 12-month period, data on all GP consultations with patients were extracted from the electronic medical records. These data included contact diagnoses and indications (diagnoses) for medication and referral to secondary care. Prevalence rates are based on contacts that were classified by the GP according to the International Classification of Primary Care and subsequently clustered into episodes of disease.17

The most prevalent episodes were combined into 24 disease clusters. These disease clusters have been used in several other studies18 19 and cover the full range of the most prevalent diseases in general practice (prevalence >10 per 1000) (see table 1). The 24 disease clusters have been distributed over seven disease categories, namely cardiovascular diseases, musculoskeletal diseases, mental diseases, respiratory diseases, neurological diseases, digestive diseases and miscellaneous. The principal aim of this study was to explore possible associations between green space and specific diseases. Therefore, chosen disease clusters were not based on a presumed mechanism behind this association beforehand. Instead, we chose a common basis categorisation covering the full range of morbidity most frequently presented in general practice, according to the bodily system involved.

Not all disease clusters were relevant for all age groups; therefore, the epidemiological denominator varied (table 1). A prevalence rate for each cluster was calculated by dividing the number of patients with at least one disease episode in 2001 belonging to the cluster by the population at risk. The population at risk was based on age groups in which the diseases occurred. Some disease clusters, like for instance high blood pressure, were only present in the older age groups. Therefore, for high blood pressure the prevalence rate was calculated by dividing the number of patients with at least one high blood pressure episode in 2001 by the older age groups and not for the population as a whole.

Characteristics of the respondents’ living environment

The LGN4 database discriminates 39 land use classes including crop types, forest types, water, various urban classes and semi-natural classes and has been proven to be valid and accurate.14 20 The total percentage of green space in the respondents’ living environment was measured within a 1 km radius and within a 3 km radius around the postcode centroid of a respondent’s home, to see whether there is a stronger relation for green space close by than green space further away. Only green spaces that dominate the land use in the 25×25 m grid cell (more than 50% of the grid cell is green) have been classified as green space in the dataset. Small-scale green spaces, such as street trees and roadside vegetation were only included as green space if they were dominant in the grid cell.

Urbanicity

Another environmental characteristic is urbanicity. This variable consists of five categories, ranging from very strongly urban (1) to non-urban (5); it was measured at the municipal level and was derived from Statistics Netherlands. The indicator is based on the number of households per square kilometre and is commonly used in The Netherlands.21

Demographic and socioeconomic characteristics

Part of the relation between green space and health may be the result of direct or indirect selection. Direct selection would take place when people’s health is related to their chances of living in a green environment. Indirect selection takes place when people with certain characteristics related to well-being (such as income) tend to live in a green environment.22 As migration flows are related to such sociodemographic characteristics as age, income and education,23 we decided to rule out indirect selection effects as far as possible by controlling statistically for demographic and socioeconomic characteristics.

The demographic characteristics taken into account were gender (female  = 1) and age (which was taken into account as a polynomial until the third order because there was no linear relation between the disease clusters and age), and were derived from the patient lists of the participating practices. To find out whether the relation between green space and morbidity differed between age groups, age was divided into six categories: children (<12 years), adolescents (13–17 years), youths (18–25 years), young adults (26–45 years), older adults (46–65 years) and elderly people (65+ years).

Socioeconomic characteristics were collected by a registration form that was sent by mail to all people listed in the participating practices in the DNSGP-2 (n = 380 000, response 76.5%)13 and included education, work status and healthcare insurance type. For a number of people these socioeconomic characteristics were unknown. To reduce the number of missing entries we included a category unknown in the analyses. Education was measured as the highest level of completed education (unknown, no education completed, primary education, secondary education, higher education). Work situation was categorised as work situation unknown, paid job, attending school/studying, housewife/houseman, retired, disability pension, unemployed. Socioeconomic status was additionally implicitly measured by type of healthcare insurance (unknown, public or private). The type of healthcare insurance can be regarded as an indicator of SES in the Dutch context in 2001, as only people with a higher income had private health insurance, whereas people with a lower income had obligatory public health insurance.

When testing the relation between green space and the annual prevalence of disease clusters for different SES groups, SES was operationalised as the level of education divided into three categories: higher education (university or higher vocational education), secondary education and primary or no education. Characteristics of the study population are displayed in table 2.

Statistical analysis

The relation between percentage of green space in people’s living environment and morbidity was assessed using multilevel logistic regression analyses, controlling for demographic and socioeconomic characteristics and urbanicity. We included two levels, individuals and practices, because of the hierarchical structure of the data within DNSGP-2. The multilevel logistic regression analyses were performed with MLwiN. The independent variables, including the percentage of green space, were centred around their average. The results thus represent morbidity of the average population living in an area with an average amount of green space. We used interaction effects between respective age groups, SES groups and urbanicity and the green space indicator to investigate the relation for different age groups, SES groups and in different levels of urbanicity. Because of the large dataset we adopted a strict type 1 error criterion of α = 0.01.

Results

On average there is 42.4% of green space in a 1 km radius and 60.8% of green space in a 3 km radius around people’s homes. Table 3 presents the ORs for the annual prevalence rate of the 24 disease clusters for people who have 10% more green space than average. In general, a significant relation between the percentage of green space and the annual prevalence rate was only present for green space in a 1 km radius. Only for anxiety disorders, infectious diseases of the digestive system and medically unexplained physical symptoms (MUPS) the annual prevalence rate was lower in environments with more green space in a 3 km radius.

For 15 of the 24 disease clusters the annual prevalence rate was lower in living environments with a higher percentage of green space in a 1 km radius. This relation is apparent for diseases in all seven disease categories. It is strongest for anxiety disorders and depression. The relationship is negative for none of the disease clusters.

Strength of the relation

An indication of the strength of the relation is given in table 4, which shows the annual prevalence per 1000 for people with average characteristics on the control variables with respectively 10% and 90% green space in a 1 km radius around their home. For anxiety disorders, the annual prevalence for people with average characteristics with 10% green space in a 1 km radius was 26 per 1000 people and for those with 90% green space in a 1 km radius 18 per 1000 people. For depression these figures are respectively 32 and 24 per 1000. In general, the found relation between green space and physician-assessed morbidity is comparable with the relation between age and morbidity. An increase in 1 percentage point of green space on physician-assessed morbidity equals the effect of 1-year lower age.

Relation in different age groups

Further analysis showed that the relation was strongest for children younger than 12 and people between 46 and 65 (not in table). For children the relation was not only apparent for the percentage of green space in a 1 km radius, but also for the percentage of green space in a 3 km radius. For a few disease clusters the relation for children was especially strong, for example for vertigo (1 km OR 0.81, 95% CI 0.74 to 0.90; 3 km OR 0.85, 95% CI 0.77 to 0.94) and severe intestinal complaints (1 km OR 0.85, 95% CI 0.80 to 0.90; 3 km OR 0.89, 95% CI 0.84 to 0.94). The strongest relation for children was found for depression (1 km OR 0.79, 95% CI 0.72 to 0.88; 3 km OR 0.84, 95% CI 0.78 to 0.91).

The relations for the other age groups were similar to the overall relations shown in table 3.

Relation for different socioeconomic groups

Especially the lower educated groups had a lower annual prevalence rate when they had more green space in a 1 km radius around their home. For example, the odds for chronic obstructive pulmonary disease (COPD) were smaller for the lower educated (1 km OR 0.97; 95% CI to 0.95 to 0.99) than for higher educated (OR 0.98; 95% CI 0.96 to 1.00).

Relation for different levels of urbanicity

Concerning the level of urbanicity our analyses show that urbanicity influences the relation between green space and the annual prevalence of disease clusters (not in table). There is often no relation between green space and the annual prevalence of disease clusters in the very strongly urban areas. At all other levels of urbanicity people with more green space in a 1 km radius around their home had a lower annual prevalence rate. The relations between green space and annual prevalence rates were strongest in slightly urban areas.

Discussion

Principal findings

This explorative study shows that the previously established relation between green space and a number of self-reported general indicators of physical and mental health can also be found for specific, doctor-assessed disease categories. The annual prevalence rates for 15 of the 24 investigated disease clusters is lower in living environments with more green space in a 1 km radius. Green space close to home appeared to be more important than green space further away. This is in contrast with our previous studies1 2 which found the relation between self-perceived health and the amount of green space in a 1 km and a 3 km radius around people’s homes to be equally strong. It appears that for the prevalence of these more specific diseases green space close to home is more important. This study differs from other studies, which mainly focused on the relation between green space and self-perceived measures of physical and mental health.1 2 3 4 6 This is the first study to assess the relation between green space and specific diseases, derived from electronic medical records of GPs. This dataset helps better establish the relation between green space and health, because it used physician-assessed morbidity as outcome, because there was no single source bias in the data, and because we used a large dataset that was representative for The Netherlands.

In line with our hypothesis, the relation was strongest for people who were expected to spend more time in the vicinity of their homes, namely children and people with lower SES. However, contrary to our expectations the relation appeared to be stronger for people aged between 46 and 65 than for elderly people. Concerning urbanicity, the relation appeared to be strongest in slightly urban areas. In very strongly urban areas there was no relation with the annual prevalence of disease clusters. This may be related to the fact that green spaces in highly urban areas are more often found to evoke feelings of insecurity,24 thereby inhibiting their use. This study only gives some indications for the relation between green space and morbidity for different subgroups. Further research should focus specifically on one of the subgroups to investigate the relation for subgroups more thoroughly.

This study shows that the role of green space in the living environment for health should not be underestimated. Most of the diseases which were found to be related to the percentage of green space in the living environment are highly prevalent in society and in many countries they are subject of large-scale prevention programmes. Furthermore, in many countries, diseases of the circulatory system and mental disorders are among the most expensive diseases with respect to healthcare costs.25 Our study contributes to the evidence that green space can help fight some major public health threats in Western societies. Healthy spatial planning should take the amount of green space in the living environment into account when endeavouring to improve the health situation, particularly of children and lower socioeconomic groups.

Underlying mechanisms

The results of this study give some indications for the possible mechanism behind the relation between green space and health. Several mechanisms could be responsible for the relation between green space and health, of which the following are most commonly mentioned: recovery from stress and attention fatigue, encouragement of physical activity, facilitation of social contact and better air quality.4 7 25 26 What do the results tell us about the mechanism at work?

The strong relation we found, particularly for anxiety disorder and depression, suggests that mental health in particular might be affected by the amount of local green space. Recovery from stress and attention fatigue then seems a likely mechanism behind the relation between green space and health; also, facilitation of social contacts might contribute. However, there is no reason to discard any of the other possible mechanisms. For example, physical activity is also known to have mental health benefits.27 Furthermore, in living environments with more green space the prevalence of most respiratory illnesses was lower, indicating that air quality could also be a possible mechanism behind the relation between green space and health. For diseases related to physical activity (diabetes, coronary heart disease, musculoskeletal diseases) somewhat less strong relations were found. But as associations were present, physical activity could also be a possible mechanism.

Strength and limitations

This is the first large epidemiological study investigating the relation between the amount of green space in the living environment of people and the prevalence of physician-assessed morbidity. Morbidity data were derived from a different database than the data on green space; consequently, there is no single source or method bias. On the other hand, we don’t have information on exposure time.

The morbidity data are accurate because they were extracted from routine electronic medical records of general practices, and the interobserver reliability of grouping contacts into episodes was high.15 The registration covered a 12-month period for each practice in order to eliminate seasonal influences. Considering the representativeness of the participating GPs and their patients—and the high validity of the data—the results of the present study can be assumed to validly represent morbidity in Dutch general practice.

Furthermore, because general practice in The Netherlands is usually the first point of contact with the healthcare system, and because the GP has a gate-keeping role for specialist care, and because there are no large geographic28 or social differences in access to general practice, morbidity presented in general practice can be regarded as a very close approximation of morbidity present in the general population.

The data used for this study also have some shortcomings. First, our data on green space, although assessed on a small scale, does not take small green spaces in the living environment into account. A 25×25 m grid cell was only regarded as green space when green space dominates in the grid cell. Gardens and small-scale green spaces, such as street trees and green verges, which could also influence people’s health, are not regarded as green space in our study. Consequently, the relation might be slightly underestimated.

Second, because of the cross-sectional design of the study, it is not possible to make strong inferences about the causality of the relations that were found. The observed relations between green space and health may partly be caused by selection. We tried to rule out this possibility by taking socioeconomic and demographic characteristics into account, but the effects of selection cannot be ruled out completely. The results from the subgroup analyses by SES groups, however, make it rather unlikely that selection is the mechanism responsible. The relationship observed between green space and morbidity was stronger for the less well-educated group and this is exactly the subgroup that has fewer options in their choice of neighbourhood of residence. Our results may be influenced by selective migration based on people’s health (direct selection). However, longitudinal studies on health-related migration show that direct selection cannot be held responsible for geographical differences that remain if socioeconomic and demographic factors are taken into account.29 30

Third, we tried to control as much as possible for individual SES. However, we did not have any information on the income of the respondents, which is a relevant indicator for SES. Furthermore, we did not control for other confounders at the neighbourhood level, although different studies have shown that, for example, neighbourhood SES could also influence health.31 32 Because this was an explorative study we chose to keep the design somewhat simple. Further research should try to find out whether a relation can also be found when neighbourhood SES is controlled for.

The aim of this study was to explore possible associations between specific diseases and green space. The disease clusters used in this study were therefore based on bodily systems, and covering the full range of morbidity most frequently presented in general practice. However, this does not mean that we consider all relationships equally plausible (in a causal sense). For some disease clusters it seems more difficult to understand why their prevalence would be positively related to the local amount of green space, for example infectious diseases of the intestinal canal. Furthermore, given the many significant relationships the absence of others relationships is also worthwhile noting. For example, high blood pressure could be hypothesised to be linked to chronically high stress levels as well as lack of physical activity, but was not related to the amount of green space. Further research will have to shed more light on the mechanisms behind the relation between green space and health, and to what extent green space indeed plays a causal role in the observed relationships.