Characteristics of the Residential Environment and Their Association With Depression in Hong Kong

Key Points Question Are residential environments, specifically livable space and residential density at a building-block and neighborhood level, associated with depression? Findings In this cohort study involving 16 968 participants followed up over 2 waves, each IQR increment in livable area was associated with 16% lower odds of probable major depression and 7% lower odds of depressive symptoms. Each IQR increment in building-block density was associated with 11% higher odds of depressive symptoms, only in single-housing environment models. Meaning These findings suggest that policies to provide more residential livable space and lower residential density at the building-block scale may be associated with less depression.


Housing density exposure -Hong Kong Housing Environment Database (HKHED)
We developed the Hong Kong Housing Environment Database (HKHED), a geospatial database of objectively measured residential density at different scales, namely within apartment at household level, building block, and neighbourhood level). Spatial data for building block footprints, housing estate boundaries, street centerline and city-wide housing address dataset sourced from the Hong Kong Lands Department (spatial database IG1000), as well as Property Agency Limited (CentaMap). We created 3 multi-scalar metrics of crowding and density: • residential liveable space, expressed as liveable floor area of the apartment of the cohort participant (at the micro-level of household), • building block density, measured as the number of residential units in building (apartment) block of the cohort participant (at a meso-level), and • neighbourhood housing density at a macro scale, expressed as the number of residential (apartment) units within pre-specified street catchments of 402 m (0.25-mile) and 805 m (0.5-mile). These radii correspond to distance covered in a 5-10 minute walk from a residence in high density Hong Kong, and have been employed elsewhere in studies of mental health 1,2 . We used Hong Kong LANDS Department's street centerline data to create street network catchments around geocoded cohort participants' home address and residential density (in units per square kilometer) was measured as the number of residential units divided by the catchment area.
The full dataset employed to measure participants' neighborhood residential density comprised 2.5 and 2.6 million housing units in the years of the end of baseline and wave 2 respectively. Our models measured the associations between one interquartile range increment in the density exposures and depressive sequelae.

Other environment
Among the other environment variables, building age was measured from the first occupation date of each building block and expressed as 1945-80 (old), 1981-95, post-1995 (newly built).
Floor level was extracted from the participants' address data. The floor level data had to be further corrected for participants residing in floors that start with "4" (i.e. 40-49), end with "4" (i.e. 4, 14, 24, 34...) and floor level 58, mostly in private housing blocks constructed after the mid-1990s.These building blocks skipped these floor levels due to auspicious reasons in Hong Kong context and adjustments were accordingly made after thorough checking. Floor level was coded as a four-factor variable (0-5, 6-10, 11-20 and >20). The density of public transport was measured as the number of public transport (bus, green minibus, mass transit railway (MTR), light rail, trams and ferries) stops and stations per square kilometer within 402 m and 805 m residential catchments. Terrain is a measure of physical environment that among other things, increases the friction of travel. It was measured from a 5-metre resolution digital terrain model as mean slope within 402 m and 805 m residential catchments. Building age, floor level, density of public and terrain were assessed both at baseline and wave 2. The proportion of working population in the neighbourhood, a proxy of neighbourhood-level economic activity, was measured at the level of large street block groups (LSBG) from the 2011 population census.
LSBGs are the lower level census aggregates containing a mean population of approximately 4,350 people in Hong Kong. The data was sourced from the Hong Kong Census and Statistics Department on enquiry and subsequently matched with the participants' geocoded locations.

Covariates and confounders.
In our study, all covariates and confounders (except sex) were measured at two time points, derived from the baseline and wave 2 survey questionnaires. These included sociodemographic covariates (age, sex, marital status, educational qualification, employment status and personal income); lifestyle (number of family members, smoking status, alcohol intake frequency, current religion and neighbourhood cohesion score) and comorbidities (doctordiagnosed coronary heart disease and high cholesterol) and housing environment attributes (building age, floor level, and public transport density and terrain variability). The cohort participants' age (in years) was calculated from the dates of birth and examination dates at baseline and wave 2. Sex was assessed at baseline. Marital status was derived from a questionnaire: 'What is your current marital status?' and coded as never married; married; and widowed/divorced/separated. Educational level was coded as a 3-factor coded as primary; secondary; and tertiary or higher degrees. Employment status was coded as employed; homemaker/student/others; and retiree/unemployed. Income was derived from the question: 'Including all your income sources and MPF contribution, what is your average monthly personal income? (in Hong Kong dollars; 1HK$=0.13$US)' and coded as a 4-factor variable as <HK$ 5,000; 5,000-9,999; 10,000-14,999; and ≥15,000. Among the lifestyle level variables, the number of family members was coded as a 4-factor variable as one, two, three, ≥four. Smoking status was expressed as non-smoker/past smoker versus current smoker, while alcohol intake frequency was a 3-factor variable coded as non-drinker/former drinker; occasional to 1-3 per month; and 1-3 per week up to daily). Religion was based on the question: 'What, if any, is your religion currently?' and coded as None; Christianity; and Buddhism/others. Neighbourhood cohesion score based on a five-item questionnaire on neighbour's willingness to help, being close-knit, trustworthy, able to get along and sharing similar values was coded as low, medium and high. Self-reported doctor-diagnosed comorbidities (coronary heart disease and high cholesterol) were coded as binary outcomes (Yes versus No). were entered into the model separately in the single housing environment models. **The housing environment variables were simultaneously entered into the model in the composite housing environment model. a Model 1 adjusted for age and sex. b Model 2 represents fully-adjusted models accounting for socio-demographics (age, sex, marital status, employment status, educational level, income), lifestyle (smoking status, alcohol intake frequency, number of family members and current religion), comorbidities (cardiac heart disease, high cholesterol level) and residential environment (residential building age, floor level, density of public transport, terrain, neighbourhood cohesion and proportion of working population in the neighbourhood). a Fully-adjusted models accounting for socio-demographics (age, sex, marital status, employment status, educational level, income), lifestyle (smoking status, alcohol intake frequency, number of family members and current religion), comorbidities (cardiac heart disease, high cholesterol level) and residential environment (residential building age, floor level, density of public transport, terrain, neighbourhood cohesion and proportion of working population in the neighbourhood). The residential environment (neighbourhood residential density, density of public transport and terrain) were measured within 402m of street catchment of geocoded participants' residence. a Fully-adjusted models accounting for socio-demographics (age, sex, marital status, employment status, educational level, income), lifestyle (smoking status, alcohol intake frequency, number of family members and current religion), comorbidities (cardiac heart disease, high cholesterol level) and residential environment (residential building age, floor level, density of public transport, terrain, neighbourhood cohesion and proportion of working population in the neighbourhood). The residential environment (neighbourhood residential density, density of public transport and terrain) were measured within 402 m of street catchment of geocoded participants' residence. The housing environment variables were simultaneously entered into the model in the composite housing environment model.