Across this very large sample of mid-aged adults in the UK, we examined the relationship between three characteristics of the neighbourhood built environment and hospital admissions due to cancer, over almost 10 years of follow up. We examined whether these associations were modified by area deprivation and household income, with the aim of identifying which neighbourhood characteristics might best be intervened on to improve health without widening existing health inequalities.
We observed very little evidence that any of the three neighbourhood exposures were associated with overall hospitalisations due to cancer. However, investigation of effect modification by household income and area deprivation uncovered interesting patterns that may help to illuminate important elements of the links between the neighbourhood built environment and health. The largely null overall associations appeared to mask potentially important variation in the strength and magnitude of some of those associations by sex, individual socio-economic conditions and area deprivation.
For neighbourhood greenspace and cancer-related hospital admissions – and particularly for women admitted for breast cancer – we found evidence of effect modification by area deprivation, suggesting a greater protective influence of greenspace against cancer in more deprived areas than in less deprived areas. This finding is consistent with some other studies that have previously found relationships between greenspace and health to be stronger in more deprived communities1, including in the UK31,32. In contrast, there did not appear to be any association between formal PA facilities and cancer within any income or area deprivation subgroup, and conflicting patterns of effect modification by income for breast and colorectal cancer specifically.
One pathway through which greenspace is hypothesised to influence health is via physical activity. However, the fact we observed no association between PA facilities and cancer, including in deprived areas, combined with other, not yet published findings in which we found no relationship between greenspace and CVD, suggests that greenspace might influence health generally through pathways unrelated to PA. This contrasts, to some extent, with a 2016 study in the US in which PA was estimated to explained a small proportion (2.1%) of an observed association between greenspace and cancer mortality12, but is consistent with a recent study from Spain that reported an association between urban greenspace and breast cancer unlikely to be mediated by physical activity33. While formal PA facilities are unlikely to influence health via pathways other than through physical activity itself, there is emerging evidence that greenspace may influence health via multiple pathways, including mental wellbeing, immune function, and respiratory health13, as well as PA. Several studies have concluded that greenspace-health relationships, if causal, are mediated by pathways other than PA, most notably psychosocial ones12,34,35. One of the principal mechanisms by which greenspace is thought to influence health is the regulation of cortisol secretion36. Cortisol secretion is an indicator of stress and its dysregulation is associated with various health outcomes including cancer37. A recent study in a deprived setting in Scotland found that the presence of more greenspace near the home was associated with lower levels of stress across objective cortisol secretion measures and subjective measures of stress, but this relationship did not appear to be mediated by physical activity36. Access to greenspace near home may also plausibly mitigate other biological pathways through which chronic psychological stress (more prevalent in deprived populations) influences cancer risk, such as oxidative stress-induced DNA damage and telomere shortening38,39. Similarly, greenspace may mitigate some of the effects on cancer risk of air and noise pollution (also often higher in deprived areas), operating though these and related inflammatory and oxidative stress pathways40,41.
For fast-food proximity and cancer, there was no evidence of an interaction with income, and only very weak evidence that area deprivation modifies the effect of fast-food proximity. The measure of fast-food proximity we have used is somewhat problematic, however, and these results may not be reliable. There is likely to be some systematic misclassification, random error, and geographical inconsistency in quality in the proximity measure we have used, due to our reliance on an off-the shelf measure based on local authority data sources collected for non-research purposes. This highlights some of the trade-offs made in the use of big data and administrative data for the purposes of epidemiological research. Further research repeating this UK-wide analysis using improved measures of the fast-food environment may clarify this relationship.
An important a priori rationale for examining effect modification by factors such as income and area deprivation, when a study is sufficiently powered to do so, is that it is plausible that some groups of people will be more sensitive to their neighbourhood environment than others, and that some may be almost completely insensitive for various reasons. Population-wide, average effect estimates smooth out these differences and potentially lead to erroneous conclusions about the importance of neighbourhood environments for some people in some places. Indeed, in this study, we found very little evidence of association between these neighbourhood exposures and cancer across the study population overall, but stronger evidence for associations were observed within more deprived subgroups. We would only expect small effect sizes overall, given the complexity and multitude of causes of cancer, and how distal the outcome is from the exposures, but nonetheless, in some cases these population-average null findings contradict what we might expect based on previous research. In particular, evidence from food environment research in the UK has been mounting of a detrimental effect of excessive exposure to unhealthy food outlets15,42−44. Limitations of the fast-food proximity measure are described above, and are also likely to have led to conservative estimates. Similarly, the greenspace measure may not adequately capture the full extent of relevant greenness of one’s neighbourhood, as it does not include smaller parcels of greenspace such a street trees, or reflect ‘quality’ of greenspace.
There are several other limitations of the current study. First, the hospital admissions data only captures inpatient care, so any early detection of cancer that occurs in primary care settings after baseline and is then effectively treated without admission to hospital will not be counted. Such cases are probably more likely to occur in higher income or less deprived subgroups45, and this may have contributed to lower risk of hospital admission in those groups, potentially distorting the magnitude of effect modification on the additive scale. In the future, when GP records are fully linked to the UK Biobank cohort, it will be possible to examine this potential source of bias. Related to this, if some types of health care have shifted to outpatient settings over the course of the follow-up period, it may result in some dilution of the true association overall and between subgroups. We have not distinguished between elective and emergency admissions, and differences in these may also be socially patterned.
Second, it is unclear what period of follow-up is likely to be necessary to capture the effect of interest, given that people will have been exposed to their baseline neighbourhood conditions for varying lengths of time depending on how long they have lived at that address, and whether relevant changes had occurred in their neighbourhood during that time, and the nature of previous neighbourhood exposures. We adjusted our analyses for years living at baseline address to attempt to deal with this, and are reassured by the long average time people have lived at the address we are using (median = 15 years), but there may be remaining imprecision, and potential bias of estimates in either direction, that we cannot overcome using observational data of this kind. Longer follow up may prove to be more revealing, and that will become possible in future years, but ideally future work would also account for changes in the built environment over that period. UK Biobank would be made richer by the addition of measurement of neighbourhood exposures at one or more post-baseline time points. Our sensitivity analyses using a shorter follow-up period to account for the timing of the exposure ascertainment showed that most point estimates were robust to this specification, but there was a loss of precision presumably driven by the substantial reduction in the number of hospital admissions occurring during the shortened follow-up period (Supplementary Table S9).
Finally, we cannot rule out self-selection into more health promoting neighbourhoods by people more disposed to healthy behaviours. We can, however, by the longitudinal nature of the study and exclusion of people with prevalent disease at baseline, rule out active self-selection prior to baseline into neighbourhoods on the basis of prevalent disease (e.g. following a cancer diagnosis earlier in life, deciding to relocate to a neighbourhood more supportive of a healthy lifestyle). This means that we likely minimise masking of the true effect via this avenue, but may still have some residual positive confounding that could bias the association away from the null, despite our comprehensive adjustment for observed potential confounders. However, UK Biobank is a residentially very stable sample, and most of our strongest findings were within more deprived subgroups, where financial resources enabling relocation for health purposes are presumably the least. In additional robustness checks (not shown) we also confirmed that further model adjustment for baseline hypertension, BMI, and medications for hypertension or cholesterol, made no material difference to our findings (these were not included in the main analysis because of ambiguity regarding temporal precedence i.e. they may be on the causal pathways from neighbourhood environment to cancer if neighbourhood exposure predates them, rather than being confounders).
Overall, despite no estimated protective effect of greenspace on cancer across the mid-aged English population taken as a whole, subgroup effects were observed. Living in a neighbourhood with a greater percentage of greenspace is associated with lower risk of cancer-related hospitalisation among people living in more deprived areas. There is some evidence of the same being true for reduced fast-food proximity and cancer. Greater availability of PA facilities close to home is not associated with lower risk of cancer for any of the analysed groups. Improving deprived neighbourhoods by increasing the amount of public and private greenspace and limiting the proximity of fast-food outlets to residential areas, may improve health outcomes in the population.
Taken together, these results suggest that improving access to greenspace may have a greater public health impact in more deprived areas, but the pathway(s) by which these benefits might arise require further elucidation and should not be assumed to be restricted to the promotion and facilitation of physical activity. We also show that by examining effect modification by multiple socioeconomic indicators in parallel, potentially important insights can be gained that may be missed when we focus only on a single measure of either household or area-level socioeconomic conditions. Understanding the potentially different ways in which different aspects of the socioeconomic conditions of people’s lives influence their relationship with the built environment and its effects on cancer risk may help to avoid intervention-generated inequalities when neighbourhood-based built environment interventions are designed.