Distance to parks and non-residential destinations influences physical activity of older people, but crime doesn’t: a cross-sectional study in a southern European city

Setting

Located in the northwest of Continental Portugal, Porto municipality had approximately
240,000 inhabitants in 2008 33], distributed across 41.7 km
²
. Porto is limited by the Atlantic coast, and extends along the Douro River estuary.
It is an industrial and port town situated in the Porto Metropolitan Area, the second
largest metro area of Portugal with roughly 1.3 million inhabitants 34].

Participants

The EPIPorto Cohort encompasses a representative sample of 2485 adult (?18 years old)
inhabitants of Porto. Baseline evaluation was conducted from 1999–2003 35]. Participants were recruited by random digit dialing using households as the sampling
unit. After assessing the number and age of the residents of each household, randomization
was applied to select one eligible person among the permanent adult residents.

The follow-up evaluation took place from 2005–2008. 1943 participants were contacted
but 261 participants refused to participate, resulting in a response rate of 86.6 %.

The Ethics Committee of the Hospital de São João approved the study protocol. The
study was carried out according to the Helsinki Declaration and all participants completed
the informed written consent form.

Google Earthâ„¢ was used to georeference all addresses. For the present study, we included
only adults aged 65 or more at the follow-up evaluation, i.e., 582 out of 1682 participants.
Five participants were excluded because they moved outside of Porto.

Outcome: Leisure-time physical activity

Physical activity was evaluated using the EPIPorto Physical Activity Questionnaire
to measure time and intensity of different types of activities, such as rest, transport
to/from work, occupational, household and leisure 36]. A previous study assessed the validity, reproducibility and seasonal bias associated
with past-year PA reporting, and it showed it is a valid and reproducible instrument
for the brief assessment of different types of PA among adults.

In our study we focused on leisure time physical activities. In the EPIPorto Physical
Activity Questionnaire, these included sedentary (playing cards, watching TV), light
(e.g. brisk walking, golfing, snooker), moderate (e.g. walk at moderate pace, dancing,
stretching) and vigorous (e.g. running, soccer, basketball) leisure activities. Because
older adults benefit from PA even if light 37], we considered LTPA as the sum of the time (minutes/day) spent in non-sedentary leisure
activities.

Two measures of LTPA were defined: time spent (minutes/day) in LTPA and participation
in LTPA – inactive (0 min/day) and active (0 min/day). We followed this approach
because we theorized that the time active individuals spend in LTPA might be more
influenced by neighborhood characteristics, whereas participation in LTPA might be
more related to individual characteristics than to the neighborhood’s 38].

Information about LTPA was available for 533 participants (out of 577), but one outlier
observation had to be excluded, making a final sample of 532 participants.

Covariates: Individual variables

Individual characteristics were obtained through a structured questionnaire. We considered
as confounders the following individual correlates of LTPA: age; marital status (married/non-marital
union, single, widowed and separated/divorced); educational attainment (number of
schooling years); retirement status (not retired/retired); smoking status (smoker,
occasional smoker, non-smoker and ex-smoker); comorbidities (absence/presence of at
least one of the following conditions – cardiovascular, respiratory, osteoarticular
and musculoskeletal disorders, cancer, depression, cirrhosis and hypo/hyperthyroidism);
residence in Porto for 20 years or more (yes/no); and body mass index (classified
according to World Health Organization cut-offs).

Covariates: environmental variables

Neighborhood characteristics included as independent variables in the statistical
analysis were: 1) socioeconomic status (SES) of the census tract of residence (three
classes from the most to the least deprived 39]); 2) population density of the census tract of residence; 3) distance from the residence
to the nearest park (24 parks); 4) distance to the nearest sport space (71 sport spaces);
5) distance to the nearest non-residential destination (includes churches, shops,
libraries, museums and other points of interest) (421 non-residential destinations);
6) distance to the sea/riverside; 7) density of street intersections within 200 m
of the residence (considered as the walkable distance for older individuals); 8) density
of bus/metropolitan stops within 200 m; 9) average land gradient within 200 m. Since
individual data refer to follow-up evaluation (2005–2008), all neighborhood characteristics
were collected for a year within this time-window. The collection of the above mentioned
variables and the georeferencing procedures were previously described 38].

The map of the participants’ residence and neighborhood characteristics is displayed
in Fig. 1.

Fig. 1. Spatial distribution of the participants’ residences and built and socio-environmental
features (Porto, 2005–2008)

Covariates: crime

Data about crime were obtained from the Public Security Police of the Metropolitan
Command of Porto, which provided records of all crimes in Porto during 2008. The dataset
included a description of the crime and the place of occurrence (street, neighborhood,
street segment and, occasionally, exact position).

There were 17,790 records, from which 296 could not be georeferenced due to poor quality
location information and 1776 were excluded because they corresponded to crimes (e.g.
fraud, jobbery, copyright crimes) that were unlikely have an impact on the population’s
fear of crime and, consequently, PA.

Based on previous studies 10],23], we classified the remaining 15,718 crimes into the following categories: 1) incivilities
(drug, vandalism, prostitution); 2) criminal offenses with violence, i.e., with approach
to the victim (robbery, homicide, rape); 3) criminal offenses without violence, i.e.,
with no approach to the victim (theft, verbal offences) and 4) traffic (drunk/dangerous
driving, speeding).

Further details about the georeferencing procedures and categorization of crime records
can be found as additional material (additional file 1 and 2).

We calculated crime rates (/1000 inhabitants), by category, for each census tract;
then a crime rate was attributed to each participant. Fig. 2 shows the spatial distribution of crimes rates across Porto municipality by category.

Fig. 2. Spatial distribution of recorded crime (Porto, 2008). Spatial distribution of the crime rates (crimes/1000 inhabitants) by category

Statistical analysis

Descriptive statistics were computed for all variables, by sex and participation in
LTPA (active vs. inactive). Mann–Whitney U and Chi-square tests were employed to compare
distributions and proportions; the significance level was set at 0.05.

Generalized Additive Models (GAM) were used to estimate the association between LTPA
and covariates. GAM extends generalized linear models to include nonparametric smoothing.
This approach allowed us to model the spatial distribution of LTPA, and therefore
to control for the presence of possible spatial autocorrelation.

For data modeling, LTPA was used as a dependent variable and individual and neighborhood
characteristics as covariates. Firstly, the association between spatial location of
residence and LTPA was evaluated by applying a bivariate smoothing spline function
on the pair of coordinates. Secondly, univariable analysis was conducted and all covariates
with p-values ?0.10 were included in the initial multivariable model. Then, each covariate
was removed step by step until the final adjusted model was attained, eliminating
consecutively those with the highest p-values. The final model included only covariates with p-values ?0.05.

The presence of interactions was evaluated by including interaction terms between:
1) sex/marital status and area variables and 2) crime and other environmental variables.

Two models were fitted to test the hypotheses that 1) neighborhood characteristics
were related to participation in LTPA and 2) neighborhood characteristics affect the
time spent on LTPA among already-active persons. The first model, logistic regression,
(eq 1) included the whole sample and assessed LTPA as a dichotomous variable (active/inactive).
The second, linear regression, (eq 2) contained only active individuals, and assessed LTPA as a continuous variable (minutes/day).
Given its skewed distribution, the variable LTPA (minutes/day) was log-transformed.
The equations are presented below:

(1)

(2)

where y _i
and z _i
are the response variables, ??’?s are the coefficients of the model, x _ik
are the explanatory variables, f(north _i
,?east _i
) is a smooth function of the coordinates and e _i
are the residuals.

Due to the presence of interactions between sex and some neighborhood characteristics,
sex-stratified models were built.