We have conducted a review of observational epidemiological studies analysing the relationship between the urban environment and the prevalence/incidence of NCDs – specifically, Type 2 Diabetes Mellitus (T2DM) and Cardiovascular Disease (CVD) – and associated risk behaviours. See the review here: https://horus-urbanhealth.eu/wp-content/uploads/2024/05/HORUS_Knowledge-Hub-Booklet_v1.0.pdf
In this longitudinal studies with a quasi-experimental design are a valid standard for estimating causal effects of a well-defined urban intervention – natural experiment – on a primary outcome, whether a health or a behavioural outcome. Some longitudinal studies do not have a dichotomous definition of the exposure factor, but are appropriate for establishing causal relationships by incorporating one or more time-varying continuous variables as a treatment.
Cross-sectional studies allow the observation of time-static associations between urban environmental characteristics and a primary health or behavioural outcome. However, the attribution of causal nature to the observed association is less straightforward than in the case of longitudinal studies, and must rely to a greater extent on elements exogenous to the study design, such as consistency, plausibility or dose-response criteria, among others (see Bradford-Hill, 1965).
Results obtained in the framework of studies that use clustered data at the territorial unit level as the observational unit should be interpreted more cautiously than in the case of studies with individual-level data, as they can more easily lead to the ecological fallacy. However, this level of data specification is appropriate when individual cohort data (in longitudinal studies), individual survey data (in cross-sectional studies) or other primary observational data are not available, and data must be obtained through alternative sources, such as health or administrative records (as in McGavock et al., 2022).
Methodological and study design features
- We have found several research studies with a quasi-experimental design estimating the effect of a well-defined intervention – a natural experiment – or a feature of the urban environment on a behavioural outcome.
- Most of this research focuses on the construction or refurbishment of parks and urban green spaces, although there is also research focusing on active transport interventions and the renewal of urban areas.
- Many of the studies do not control or weight the estimated causal effect of the intervention by treatment propensity. Some authors, such as Mölenberg et al. (2019), acknowledge that baseline conditions were not inter-homogeneous between control and treatment groups, neither in the primary outcome nor in other relevant independent covariates.
- We have found little research with a quasi-experimental design estimating the effect of a well-defined intervention on a health outcome as the primary outcome. We only found McGavock et al. (2022), which estimates the effect of expanding multi-use trails for physical activity in Winnipeg (Canada) on the incidence of CVD. We interpret this scarcity of studies, compared to the previous case, as being due to the greater difficulty of estimating relevant and significant average causal effects on health outcomes. The causal pathway between treatment and outcome is more complex and there are more potential mediators, confounders and independent pathways.
- There is a study protocol (Pearson et al., 2020) aiming to estimate the effect of ecological restoration of urban parks on physical exercise as a primary outcome, salivary cortisol as a secondary outcome and other biomarkers considered by the literature as surrogates of different health outcomes, such as glycosylated haemoglobin A1C (A1C) and C-reactive protein (CRP). Subsequently, the authors reported difficulties in implementing the study (Pearson et al., 2023), such as the clear delineation of an ‘intervention year’ as an analytical cut-off point, considering that restoration benefits are achieved through increased species diversity and slow development of vegetation cover. This illustrates one of the potential difficulties inherent in quasi-experimental studies: the exogenous character of the natural experiment.
- Some studies with a quasi-experimental design do not include a control group, sometimes unintentionally, as is the case in Schultz et al. (2017). This omission can significantly increase the probability of a false-positive result.
- In some longitudinal studies, exposure is defined as a non-binary time-varying variable, based on indicators that often express composite characteristics of the urban environment. This approach diffuses the boundaries between control and treatment groups, although it allows valid conclusions to be drawn on the marginal effect of the exposure variable.
- Some studies use non-parametric tests, such as the Wilcoxon rank sum test, to assess pre-post intervention changes between groups. However, this approach needs to be adequately justified, since, as a general rule, the null hypothesis refers to stochastic equivalence between two samples. This is less intuitive to interpret compared to a direct comparison of measures of central tendency or counts.
- Several studies focus on comparing observational counts of the use of a public space after the implementation of a change, using tools such as SOPARC. However, if not enough observational periods are included, the analysis may lack statistical power. Some authors reach plausible conclusions about the effect of an intervention based on these observational counts, but the statistical analysis used may not be appropriate. For example, employing a two-way ANOVA for a count dependent variable, with counts equal to or close to 0, may not be appropriate.
- The cross-sectional epidemiological studies providing the most consistent evidence and with the most generalisable results are those similar to Sallis et al. (2016). This study uses a multi-city sample of individuals and includes intra-city variant and between-city comparable environmental exposure variables.
Main Highlights
- Overall, we found evidence of mixed results on the effect of interventions on the urban built environment and NCD-associated behaviours, possibly attributable to heterogeneity in study design and methodological shortcomings. More research is needed on the impact of interventions on health outcomes.
- It is plausible to assume a dose-response relationship between the magnitude of the intervention and the outcome. In their study on the impact of multi-use trails on CVD incidence and risk factors, McGavock et al. (2022) find a greater reduction in the incidence of CVD risk factors in areas close (< 400 m) to the longest trail (IRR = 0.85, 95% CI [0.75, 0.96]) compared to control areas. The IRRs of the rest of the trails are closer to 1, indicating less or no effect compared to the control areas.
- Improving urban facilities may increase their use in the short term, but this effect may stabilise or decline moderately over time. Evidence is provided in Veitch et al. (2018) Cohen et al. (2019) or Hosford et al. (2019), among others.
- It is plausible that the impact of interventions is heterogeneous across population groups, which would partially explain why some studies did not find overall associations. However, there is a risk of false positive for multiple comparisons. Evidence is provided in Quigg et al. (2012), Dill et al. (2014) or Schultz et al. (2017), among others.
References
Bradford-Hill, A. (1965). The Environment and Disease: Association or Causation? Proceedings of the Royal Society of Medicine, 58(5), 295-300. https://doi.org/10.1177/003591576505800503
Cohen, D. A., Han, B., Isacoff, J., Shulaker, B., & Williamson, S. (2019). Renovations of neighbourhood parks: long-term outcomes on physical activity. Journal of Epidemiology & Community Health, 73(3), 214-218. https://doi.org/10.1136/jech-2018-210791
Dill, J., McNeil, N., Broach, J., & Ma, L. (2014). Bicycle boulevards and changes in physical activity and active transportation: Findings from a natural experiment. Preventive Medicine, 69, S74-S78. https://doi.org/10.1016/j.ypmed.2014.10.006
Hosford, K., Winters, M., Gauvin, L., Camden, A., Dubé, A. S., Friedman, S. M., & Fuller, D. (2019). Evaluating the impact of implementing public bicycle share programs on cycling: the International Bikeshare Impacts on Cycling and Collisions Study (IBICCS). International Journal of Behavioral Nutrition and Physical Activity, 16(107), 1-11. https://doi.org/10.1186/s12966-019-0871-9
McGavock, J., Hobin, E., Prior, H. J., Swanson, A., Smith, B. T., Booth, G. L., … & Burchill, C. (2022). Multi-use physical activity trails in an urban setting and cardiovascular disease: a difference-in-differences analysis of a natural experiment in Winnipeg, Manitoba, Canada. International Journal of Behavioral Nutrition and Physical Activity, 19(1), 34. https://doi.org/10.1186/s12966-022-01279-z
Mölenberg, F. J., Noordzij, J. M., Burdorf, A., & van Lenthe, F. J. (2019). New physical activity spaces in deprived neighborhoods: Does it change outdoor play and sedentary behavior? A natural experiment. Health & Place, 58, 102151. https://doi.org/10.1016/j.healthplace.2019.102151
Pearson, A. L., Pfeiffer, K. A., Gardiner, J., Horton, T., Buxton, R. T., Hunter, R. F., … & McDade, T. (2020). Study of active neighborhoods in Detroit (StAND): study protocol for a natural experiment evaluating the health benefits of ecological restoration of parks. BMC Public Health, 20(638), 1-14. https://doi.org/10.1186/s12889-020-08716-3
Pearson, A. L., Pfeiffer, K. A., Buxton, R. T., Horton, T. H., Gardiner, J., & Asana, V. (2023). Four recommendations to tackle the complex reality of transdisciplinary, natural experiment research. Frontiers in Public Health, 11, 1240231. https://doi.org/10.3389/fpubh.2023.1240231
Quigg, R., Reeder, A. I., Gray, A., Holt, A., & Waters, D. (2012). The effectiveness of a community playground intervention. Journal of Urban Health, 89(1), 171-184. https://doi.org/10.1007/s11524-011-9622-1
Sallis, J. F., Cerin, E., Conway, T. L., Adams, M. A., Frank, L. D., Pratt, M., … & Owen, N. (2016). Physical activity in relation to urban environments in 14 cities worldwide: a cross-sectional study. The Lancet, 387(10034), 2207-2217. https://doi.org/10.1016/S0140-6736(15)01284-2
Veitch, J., Salmon, J., Crawford, D., Abbott, G., Giles-Corti, B., Carver, A., & Timperio, A. (2018). The REVAMP natural experiment study: the impact of a play-scape installation on park visitation and park-based physical activity. International Journal of Behavioral Nutrition and Physical Activity, 15(10), 1-14. https://doi.org/10.1186/s12966-017-0625-5