Urban Densification and Physical Activity Change: A 12-Year Longitudinal Study of Australian Adults Free

Abstract

Urbanization, a major force driving changes in neighborhood environments, may affect residents’ health by influencing their daily activity levels. We examined associations of population density changes in urban areas with adults’ physical activity changes over 12 years using data from the Australian Diabetes, Obesity and Lifestyle Study (1999–2012). The analytical sample contained 2,354 participants who remained at the same residential address throughout the study period in metropolitan cities and regional cities (42 study areas). Census-based population density measures were calculated for 1-km–radius buffers around their homes. Population density change was estimated using linear growth models. Two-level linear regression models were used to assess associations between changes in population density and changes in self-reported walking and physical activity durations. The average change in population density was 0.8% per year (range, −4.1 to 7.8) relative to baseline density. After adjustment for confounders, each 1% annual increase in population density was associated with 8.5-minutes/week (95% confidence interval: 0.6, 16.4) and 19.0-minutes/week (95% confidence interval: 3.7, 34.4) increases in walking and physical activity, respectively, over the 12-year study period. Increasing population density through urban planning policies of accommodating population growth within the existing urban boundary, rather than expanding city boundaries, could assist in promoting physical activity at the population level.

Abbreviations

 
• AusDiab

  Australian Diabetes, Obesity and Lifestyle Study

 
• CVD

  cardiovascular disease

Physical activity has a broad range of established health benefits (1–3). Despite the evidence and long-standing public health initiatives, physical inactivity remains highly prevalent worldwide (4). Since educational and motivational approaches appear to have limited public health impacts, the development of activity-friendly neighborhood environments has been advocated as a beneficial and equitable strategy for promoting physical activity on a large scale (5).

Urbanization, an increase in urban population levels due to migration (6), is a major force driving changes in the built environment. It can be managed by population density increases in established neighborhoods (densification), expansion of city boundaries with low-density outer-suburb developments (sprawling), or both (7). Higher population density supports the viability of local destinations such as retail and service outlets, public transportation, and recreational facilities (8). This can facilitate physical activity behaviors due to greater accessibility of destinations (9).

Cross-sectional studies have shown consistent positive associations between neighborhood population density and physical activity (10, 11). However, longitudinal studies examining the associations of population density changes with physical activity changes are necessary to produce robust evidence. There are only a limited number of longitudinal studies examining these associations (12). These studies have focused predominantly on residential relocation—that is, examining associations between changes in population density and changes in active travel behaviors due to moving from one place to another (13–15). However, given that urbanization and increases in population density in cities are a global trend with implications for residents’ daily behaviors and health, it is of interest to examine the impact of densification on residents’ physical activity. Such investigation is distinct from examining the impact of population density changes due to residential relocation, which can also be attributable to factors related to residential self-selection, such as the preference to live closer to work and recreational facilities (16). In contrast, urban densification is a naturally occurring external phenomenon with little or no control by residents. One US study found area-level associations between population density changes and transport walking changes (17). However, this was an ecological, repeated cross-sectional study in which the walking rates of 2 different samples were compared. Prospective cohort studies are needed to clarify the potential causal impacts of densification of areas on changes in physical activity among persons who have stayed in those areas.

Using data from a multisite cohort study of Australian adults who remained at the same residential address for over 12 years, we examined associations between changes in population density and changes in physical activity.

METHODS

Data source and study participants

We used data from the Australian Diabetes, Obesity and Lifestyle Study (AusDiab), a cohort study that examined the national prevalence of, incidence of, and risk factors for diabetes and cardiovascular disease (CVD) among Australian adults. AusDiab investigators commenced baseline data collection in 1999–2000 (AusDiab1), with a 5-year follow-up in 2004–2005 (AusDiab2) and a 12-year follow-up in 2011–2012 (AusDiab3). Ethics approvals were obtained from the Alfred Hospital Ethics Committee (Melbourne, Victoria, Australia). Details on AusDiab have been published previously (18). Briefly, using a 2-stage stratified cluster sampling method, participants were recruited from 42 study areas (6 each from the 6 states and the Northern Territory) across Australia. Study areas were chosen from metropolitan cities and regional cities, and each consisted of 1–4 adjoining Census Collection Districts (the smallest spatial unit used in the Australian Census at that time, averaging approximately 220 dwellings) (19).

Eligible participants were noninstitutionalized adults aged ≥25 years without any physical/intellectual disabilities who had resided in a private dwelling for at least 6 months prior to baseline data collection. Written informed consent was obtained from all participants. In AusDiab1, 11,247 participants provided data (response rate = 55.3%). Of these individuals, 6,400 (follow-up rate = 58.5%) and 4,614 participants (follow-up rate = 44.6%) provided data in AusDiab2 and AusDiab3, respectively. Of those who provided 12-year follow-up data, 2,369 resided at the same addresses throughout the study period (stayers) and 2,164 relocated (movers); addresses were unavailable for 81. As discussed above, we focused on the stayers because we were interested in examining the effect of urban densification on residents’ physical activity. After exclusion of 15 people who reported pregnancy at any data collection point, 2,354 participants remained for analysis.

Outcome variables

The outcome variables were changes in the duration (minutes/week) of walking and total physical activity over the 12-year study period. At each data collection point, participants reported the amount of time spent in multiple physical activities during the previous week, using the validated Active Australia Survey (20). These activities included walking (for recreation and transportation), moderate-intensity physical activity (e.g., golf, gentle swimming), and vigorous-intensity physical activity (e.g., jogging, cycling). Recreational physical activity was included because high-density neighborhoods can facilitate not only transportation-related physical activity but also recreational activity due to easier access to exercise facilities. The survey instrument used is shown in Web Appendix 1 (available at https://doi.org/10.1093/aje/kwab139). Total duration of physical activity was calculated as the sum of the amounts of time spent in walking and moderate activity, plus double the amount of vigorous activity, following survey protocols (21). We truncated the weekly amount of walking at 800 minutes and total physical activity durations at 1,680 minutes, following the guidelines (20).

Exposure variable

The exposure variable was the change in population density during the study period. Two geographical units—circular buffers with a 1-km radius around participants’ residences and the administrative areas where participants resided—were used to calculate population densities at 3 time points concordant with the 3 AusDiab data collection periods. Individual circular buffers were used because they can capture residents’ immediate neighborhoods. We used circular buffers instead of street-network buffers in order to have consistent areas (3.14 km²) during the study period, since it is important to measure density change using the same geographical units over time. A 1-km distance radius was chosen because it is a typical distance within which most home-based walking takes place (22). For administrative areas, we used State Suburbs, which represent Gazetted Localities in Australia (23). This geographical unit, typically containing a retail area and surrounding residential areas, is relevant because local planning initiatives often use this unit for implementation (24). Census data collected in 2001, 2006, and 2011 corresponded to AusDiab1, AusDiab2, and AusDiab3, respectively. The population density of each geographical unit (buffer or suburb) was calculated by dividing the summed population count of all of the smallest Census geographical units (Census Collection Districts in 2001; mesh blocks in 2006 and 2011) within its boundary by the total area (persons/km²). The AusDiab stayers lived in 34 suburbs in metropolitan cities (n = 1,166) and 39 suburbs in regional cities (n = 1,188), according to Remoteness Areas classification (25). Regional suburbs (mean area = 23.2 (standard deviation, 38.7) km²) were much larger than metropolitan suburbs (mean area = 4.4 (standard deviation, 3.2) km²). Since most regional suburbs were considered too large to accurately assess the relationship of population density with residents’ physical activity, we limited our investigation of administrative areas to metropolitan suburbs only.

Statistical analyses

Our analytical strategy involved 2 steps. The models used in each step are described in detail in Web Appendix 2. First, we fitted unconditional linear growth models (26) to estimate the change in population density in individual buffers and suburbs. This was done to quantify explicitly the measures of density change, which is required to make the findings easily interpretable when considering implications for urban planning. For each geographical unit, population density values calculated in 3 Census years during the study period were modeled with corresponding time metrics (⁠|${T}_1$| = 1 for 2001, |${T}_2$| = 6 for 2006, and |${T}_3$| = 11 for 2011). These models included random intercepts and random time slopes that were allowed to vary between participants for individual buffers and suburbs for administrative areas. The estimated random intercepts and random slopes were used as the values of baseline population density (⁠|${T}_0$| = 0 for 2000) and annual change in the population density of the geographical unit, respectively. We examined 2 measures of density change as exposure variables. One was an absolute measure of density change (change in persons/km²), and the other was a relative measure in percentage ((absolute density change/baseline density) × 100). We considered the latter measure the primary measure, since the same amount of absolute change can have different implications depending on the baseline density.

In the second step, 2-level random intercept linear regression models (participants at level 1 and study areas at level 2), estimated by the maximum likelihood method, were used to examine associations between exposures and outcomes by adjusting for potential confounders and accounting for area-level clustering (27). Growth models were not used for repeated measurements of physical activity variables, as these variables exhibited positively skewed distributions with many 0 values. We defined the dependent variables of our regression models as the 12-year change in walking or total physical activity duration, calculated by subtracting the corresponding baseline measure from the 12-year follow-up measure, which did not exhibit skewed distributions. Regression coefficients can be interpreted as the average 12-year change in walking or total physical activity duration corresponding to a 1-unit increase in the population density variable (28). In models examining population density changes in individual buffers (level 1 exposure variable), intercepts were allowed to vary across study areas. In models examining population density changes in suburbs (level 2 exposure variable), intercepts were allowed to vary across suburbs.

Two sets of models were fitted for each exposure-outcome association. A directed acyclic graph (29) was used to identify the potential confounders (Web Figure 1). Model 1 adjusted for the baseline population density of the specific geographical unit, since this was a fundamental characteristic that could influence subsequent density change measures. Model 2 further adjusted for baseline individual sociodemographic characteristics (sex, age, education, marital status, employment status, household income, presence of children in the household) and area-level socioeconomic status. For this area-level measure, we used the Index of Relative Socioeconomic Disadvantage score, which is a Census-based measure reflecting the level of disadvantage in an area through indicators such as income, educational attainment, unemployment, and automobile ownership (30). Higher scores of this measure indicate relatively greater socioeconomic advantage. We calculated this measure for study areas as the population-weighted averages of the 2001 Census-based Index of Relative Socioeconomic Disadvantage scores of adjoining Census Collection Districts. In a sensitivity analysis, models further adjusted for secondary confounders, such as baseline health factors (overweight/obesity, type 2 diabetes, history of CVD) and changes in time-varying sociodemographic characteristics.

To explore the factors associated with study attrition, we fitted a 2-level logistic regression model for 12-year follow-up status (1 = remained in the study vs. 0 = dropped out) with baseline measures of outcome, exposure, and confounding variables as predictors.

All statistical analyses were conducted in STATA, version 15.1 (StataCorp LLC, College Station, Texas). In R 3.6.3 (R Foundation for Statistical Computing, Vienna, Austria), the package “ggmap,” version 2.7.9, was used to geocode study participants’ residential addresses. For all other geographic information systems analyses, we used ArcGIS Pro, version 2.3.3 (Environmental Systems Research Institute, Inc., Redlands, California).

RESULTS

Table 1 shows the key characteristics of the study sample at baseline. The mean duration of follow-up was 11.9 years (range, 11.0–12.4). A comparison of the baseline characteristics of stayers, movers, and persons who dropped out of the study after AusDiab1 is shown in Web Table 1. The movers had slightly higher proportions of women, workers, and singles than stayers. In comparison with stayers, persons who dropped out of the study were more likely to be older, to have lower educational attainment and household income, and to be unemployed, single, and not living with children. In the analysis that examined factors associated with study attrition, we found that attrition was related to baseline neighborhood population density and some confounders but not to the baseline duration of walking or total physical activity.

Table 2 shows the mean duration of physical activity and mean population density at baseline and 12-year follow-up and the change between them. Over the course of the study period, participants increased their walking by 53 minutes/week and their total physical activity by 71 minutes/week, on average. The average annual relative increase in population density was 0.8% (range, −4.1 to 7.8) in 1-km buffers and 1.0% (range, −3.2 to 3.5) in metropolitan suburbs. The mean annual absolute densification was 9 persons/km² (range, −20 to 124) in 1-km buffers and 17 persons/km² (range, −7 to 104) in metropolitan suburbs.

Table 3 shows the results from the regression models. After adjustments (model 2), each 1% annual increase in population density, measured within 1-km–radius buffers, was associated with an 8.5-minutes/week increase in walking and a 19.0-minutes/week increase in total physical activity over the 12-year follow-up period. Similarly, each 1% annual increase in density in metropolitan suburbs was associated with a 15.5-minutes/week increase in walking and a 28.7-minutes/week increase in total physical activity. We observed similar associations for the absolute measure of density increase. Additional adjustments for secondary confounders did not affect the results (Web Table 2).

DISCUSSION

We examined the potential impacts of population density changes in established urban areas of Australia on changes in walking and total physical activity over a period of 12 years. On average, participants in this cohort increased their durations of walking and total physical activity by 53 minutes/week and 71 minutes/week, respectively, over the 12-year period. The study areas were diverse in terms of population-density change. For instance, annual relative density changes, measured within 1-km–radius buffers around participants’ homes, ranged from −4% to 8%. We found that higher levels of densification were associated with greater increases in walking and total physical activity.

Our current findings on the associations between population density changes and physical activity changes are consistent with previous longitudinal studies conducted in the United States and Australia (13–15, 17). However, these studies are not directly comparable to our study because 3 of them examined changes in neighborhood population density due to relocation and subsequent changes in active travel behaviors (13–15) and 1 was an ecological study (17). This appears to be the first study that has examined the impacts of urban population densification occurring over an extended period in established neighborhoods on individual-level changes in walking (including transportation-related and recreational walking) and total physical activity (including walking and moderate- and vigorous-intensity recreational activities). Our study provides policy-relevant evidence to support the beneficial effects of urban densification on residents’ physical activity levels in the Australian context.

Population density underpins links between urban form and physical activity (31). Higher population density supports the viability of mixed types of destinations in local areas (8). The presence of such destinations, which can provide opportunities for active travel and recreation, is conducive to walking and physical activity (32). In a recent cross-sectional study of older adults in Hong Kong, China, Cerin et al. (33) reported that the association between residential density and walking can be explained by the availability of food and retail shops, public transportation, and recreation facilities in the local area. In the context of our study, it is possible that such new destinations, which may have been added in densified study areas, may have contributed to increases in physical activity. However, this cannot be ascertained using the current study data, since it was not possible to source comparative built environmental data for the baseline and 5-year follow-up periods. Investigating the mechanisms through which densification influences physical activity is an important area of future research that could help to advance our understanding of the topic.

We used 4 different measures of densification: relative and absolute changes in 2 geographical units (individual buffers and suburbs). Both relative and absolute change measures are relevant to urban planning: Percent increase is often used to describe the rate of growth of a city, while the number of residents added can be used for planning of community infrastructure. We found consistent results for both relative and absolute densification measures, which suggests that there are robust relationships between densification and increases in physical activity, regardless of the level of baseline population density. Densification measures using individual buffers and metropolitan suburbs also produced similarly significant findings. Suburbs differed in size; most were larger than individual buffers. It can be argued from the findings that suburb-level development plans, which aim to increase population density, may provide additional physical activity benefits to residents, at least in metropolitan cities. Many Australian cities are experiencing population growth: The annual increase in 2018–2019 was 2.3% in Melbourne, 2.1% in Brisbane, and 1.7% in Sydney (34). However, population increases do not necessarily mean density increases. Accommodating population increases by expanding urban boundaries (little impact on the density of existing suburbs) not only misses an opportunity to increase physical activity in existing neighborhoods but also creates low-density, automobile-dependent suburbs where residents are more likely to be physically inactive. Our findings support planning initiatives for creating higher-density, compact local areas (e.g., “20-minute neighborhoods”), which are proposed in many cities around the world, including Australia (35).

The study’s strengths include the longitudinal design with a 12-year follow-up period, a sufficiently large sample, multiple measures of densification, and diverse study areas across Australia. One key limitation is that we used self-reported physical activity. Although the average amount of time spent in each activity (walking, moderate activity, vigorous activity) by the study sample at baseline was similar to the national average at that time (21), self-reported measures can involve measurement error. Another limitation was the unavailability of domain-specific physical activity variables. Neighborhood population density has been shown to be more strongly related to transport-related physical activities than recreational physical activities (36). It would therefore be informative to examine the effects of densification on domain-specific physical activities separately. The attrition rate (55%) was relatively high at 12-year follow-up. However, our analysis found that attrition was not related to the baseline duration of walking or total physical activity. When attrition depends only on the baseline or subsequent values of the exposure or confounding variables but not on the outcome variable, it may be considered that the lost data were missing at random (37). The multilevel modeling used in this study has been shown to be less likely to produce severely biased estimates of effects under the missing-at-random mechanism for up to a 60% attrition rate, if all variables related to attrition are included in the model (27, 37). Further, we used a 1-km–radius circular buffer to calculate population density for 3 time points in a consistent manner. Although circular buffers include areas that are beyond 1 km in distance by road network, they were chosen because street-network buffer sizes can change over time if new roads have been added. Our findings may be generalizable to other cities characterized by lower population density (e.g., North American cities) but may not apply to those with higher density (e.g., Asian cities). Population increases in cities that are already high in density may not contribute to physical activity increases. Future research can explore whether there is a threshold above which additional density increases may not provide benefits in terms of physical activity.

Our previous study using this cohort examined the associations of urban densification with changes in CVD risk markers (38). It found that higher levels of densification were related to smaller increases in body weight and waist circumference but not consistently associated with markers of hypertension, hyperglycemia, and dyslipidemia. In another study conducted using this cohort, we reported that greater 12-year increases in total physical activity were related to smaller increases in CVD risk markers (39). By combining the evidence from the current study and these 2 previous studies, it can be argued that physical activity is a potential mechanism through which urban densification may lower obesity risk. However, lack of consistent associations with other risk markers suggests that physical activity may be one of the multiple mechanisms linking urban density and cardiovascular health. Other potential pathways involved in this relationship can include unhealthy dietary intake due to the presence of fast-food outlets, psychological stress due to overcrowding, low exposure to nature, and high exposure to noise/air pollution (5). Further research is needed to disentangle such complex relationships between urban densification and CVD risk.

Many global cities are undergoing urbanization and experiencing associated changes in the built environment in urban areas. Our study demonstrates that urbanization, if managed properly and regulated toward increasing population density, is a potential opportunity to address physical inactivity, which is highly prevalent worldwide and contributes to the burden of chronic disease. Further research is needed to understand the impacts of urban density on various aspects of human health.

ACKNOWLEDGMENTS

Author affiliations: Centre for Urban Transitions, Swinburne University of Technology, Melbourne, Victoria, Australia (Manoj Chandrabose, Neville Owen, Nyssa Hadgraft, Takemi Sugiyama); Baker Heart and Diabetes Institute, Melbourne, Victoria, Australia (Manoj Chandrabose, Neville Owen, Nyssa Hadgraft, Takemi Sugiyama); and Centre for Urban Research, Royal Melbourne Institute of Technology, Melbourne, Victoria, Australia (Billie Giles-Corti).

N.O. and B.G.-C. were supported by Senior Principal Research Fellowships (grants 1003960 and 1107672) from the National Health and Medical Research Council (NHMRC) of Australia. N.O. was also supported by an NHMRC Centre for Research Excellence Grant (grant 1057608) and the Victorian Government’s Operational Infrastructure Support Program.

The data analyzed in the current study are not publicly available but can be accessed upon request.

We acknowledge the investigators and sponsors of the Australian Diabetes, Obesity and Lifestyle Study, which is coordinated by the Baker Heart and Diabetes Institute (Melbourne, Victoria, Australia).

The findings of this study were presented at the Australian Walking & Cycling Conference (Newcastle, New South Wales, Australia) on October 2, 2020.

Conflict of interest: none declared.

REFERENCES