2.1. Study Population
The PREDIMED-Plus trial is an ongoing six-year multi-center, randomized, parallel-group trial, designed to evaluate a lifestyle strategy for the primary prevention of cardiovascular mortality in 6874 senior adults. The PREDIMED-Plus study protocol details have been described elsewhere [16
] (the protocol is available at http://predimedplus.com/
and the trial was registered at the International Standard Randomized Controlled Trial http://www.isrctn.com/ISRCTN89898870
Participants were community-dwelling adults (aged 55–75 in men; 60–75 in women) with BMI ≥ 27 and <40 kg/m2
, enrolled from primary care facilities in Spain between the 5th of September in 2013 and the 31st of October in 2016 who had at least 3 or more metabolic individual components of the syndrome [17
]. The lifestyle intervention consists of an energy-restricted traditional Mediterranean diet (MedDiet), PA promotion, and behavioral support, in comparison to a usual care intervention with only an energy-unrestricted MedDiet (control group).
This is a cross-sectional analysis of baseline data (dataset extracted 1/15/18) of 335 participants enrolled from primary care facilities dependent on University Hospital Son Espases, one of the 23 recruitment centers participating in the PREDIMED-Plus trial. The PREDIMED-Plus study protocol and ethical approval was granted by the Committee of Research Ethics of the Balearic Islands (CEI-IB) review boards and all participants provided written informed consent. The ethics approval number given to this study was IB-2242/14-PI. For this geographic sub-study within the PREDIMED-Plus project, following the regulations of the CEI-IB, an amendment was requested and favorably evaluated.
Participants who reported living outside the city limits of Palma de Mallorca (n = 16), those with no accelerometer data (wore no accelerometer at all; n = 100), and those with fewer than four days of valid accelerometer data (n = 1), were excluded from the analyses, leaving a final sample size of 218 participants.
2.2. Neighborhood Exposure to Walk-Friendly Routes and Public Open Spaces (POS)
POS were evaluated using the geographic area around each participant’s home. The location of each participant’s residence, reported at baseline, was geocoded using the web processing service provided by Carto Ciudad, a project run by the National Geographic Institute of Spain, which is freely available for download (www.cartociudad.es
). Using the software ArcGIS 10.5.1 (ESRI, Redlands, CA, USA), for each participant’s residence location, a sausage network buffer methodology was applied. The full methodology has been described elsewhere [18
], briefly: using only the walking street network, ignoring routes restricted to pedestrians such as freeways, the area within 1 km and 0.5 km walkable street distance of each participant’s residence location was obtained, then, 30 m of each road length were buffered; as a result, it was possible to obtain the space where each participant could move along the street walkable network and a radius at an amount of 30 m either side of the street network (Figure 1
). Next, using the tabulate intersection tool from ArcGIS 10.5.1 software, different POS were quantified within or partially within each participant’s residence sausage network walkable buffer. Details on the steps to build this buffer are included in a protocol available online [19
]. The capacity of these buffers to predict the results of physical activity have been described similar that other buffers (circular buffer or detailed buffer), although sausage buffers have a substantial advantage over others buffers by quantifying the built environment. Sausage buffers only count the features that are accessible from the road network regardless of street network connectivity [18
In order to determine the distance to the closest POS, walk-friendly routes in the absence of POS and walk-friendly route entry points, the boundaries of each polygon or line were transformed into points spaced 20 m from each other. Since the shape and the size of the resources are heterogenous, by converting the polygons and lines into predefined spaced points, we were able to reduce the measurement error [21
] (Figure 1
). The next step was to identify the closest distance to each of the POS boundaries and the nearest walk-friendly routes for each participants residence. This was achieved by calculating the walking street network distance using the Origin-Destination Cost Matrix function from ArcGIS 10.5.1.
Public Open Spaces (POS)—For the POS in this study, we calculated the proximity, count and total area, all of which are commonly used in PA and health research. Our POS were generated for three different data sources for sports facilities, beaches and parks. Sports facilities data was generated from the Municipal Sports Institute of the City of Palma. Only public sports facilities owned and maintained by the City of Palma were considered. A full list is detailed in Colom et al., 2018 [15
]. Since our target area was for public areas only, private sports facilities, such as golf courses, private gyms, and private courts, were not included in this study. In addition, the City of Palma is in close proximity to the coastline and has a considerable number of beaches with lifeguard services during the summer season; all beaches with this service were included. Finally, the dataset containing all of the parks was acquired from the Department of Infrastructure and Accessibility from the Palm City Council.
The number of sports facilities and parks, and the area of parks within or partially within the 1000 and 500 m sausage network walkable buffer were computed separately, as well as the sum of the number and area of all types of POS. The number of beaches and the area of sports facilities and beaches were excluded due to the high number of zeros.
Distance to the nearest POS from each participant’s residence were considered separately (sports facilities, beaches and parks). Distance to any POS was not considered exposure because for 85.3% of the sample, the nearest POS was a park. Proximity of each participant to the coast as a potential resource for doing PA was included, since the city of Palma is a coastal city [22
In addition, distance inside the buffer and proximity to walk-friendly routes for healthy walking and coast line were also considered, as they might directly or indirectly influence active travel [23
]. Walk-friendly routes were part of a project called “Rutas Saludables” launched by local government health policies. For the design of these routes, the coordinators had the collaboration of neighborhood associations and citizens in the neighborhoods involved, as well as municipal managers, and they were promoted at the primary health care centers in the city.
2.3. Outcome Measure: Physical Activity
For this analysis both objectively-measured PA and self-reported PA, assessed at baseline, were evaluated.
Self-reported leisure-time brisk walking—Participants completed the Girona Heart Registry (REGICOR) Short Physical Activity Questionnaire [24
], a validated short version of the validated Spanish Minnesota leisure-time PA questionnaire (MLTPAQ) [25
]. The REGICOR questionnaire collects information on walking, prior analysis of the main mode of physically active transportation [24
], and includes a specific question to collect information about leisure-time brisk walking, frequency (number of days), and duration (min/day) performed during a representative month. Time per day spent on leisure-time brisk walking (accumulated minutes/day) was computed as the sum of frequency multiplied by duration of the activity, divided by 30. Dichotomous outcome measures were computed to represent the recommendations defined by the World Health Organization any ≥150 min per week of moderate-to-vigorous physical activity such as leisure-time brisk walking [27
Objectively-measured moderate-to-vigorous physical activity in bouts of at least 10 minutes (OM-MVPA)—Participants at the baseline visit were asked to wear a wrist-worn triaxial accelerometer (GENEActiv, ActivInsights Ltd, Kimbolton, UK), on their non-dominant wrist nonstop for seven consecutive 24-h days. The accelerometer was sampled at 40 Hz with a ± 8 g
dynamic range, and data were stored in gravity (g
) units (1 g
= 9.81 m/s2
). Accelerometer data were processed and scored using R version 3.3.3 (R Core Team, Vienna, Austria) and R-package GGIR (version 1.2-5), available on CRAN (https://cran.r-project.org
) and managed on servers at the University of Malaga, one of the 23 recruitment centers participating in the PREDIMED-Plus trial. For scoring PA, a ‘valid day’ was defined as ≤2 h non-wear per day. Only participants with four or more valid days were included in analyses [28
]. These methods were consistent with other authors’ recommendations [29
]. The maximum time was nine consecutive 24-h days (both daytime and nighttime). Objectively-measured moderate-to-vigorous physical activity was calculated as accumulated minutes/day in intensities greater than 100 milligravity [30
], and bouts of at least 10 min of moderate-to-vigorous physical activity. Dichotomous outcome measures were computed to represent World Health Organization recommendations of ≥150 min of moderate-to-vigorous physical activity a week [27
2.6. Statistical Analysis
A descriptive analysis for the outcome variables (min/day) of self-reported leisure-time brisk walking and OM-MVPA was performed by comparing participant demographic characteristics and the precipitation variable. A one-way analysis of variance (ANOVA) was used to evaluate differences in baseline self-reported leisure-time brisk walking and OM-MVPA based on demographics.
Associations of objectively-measured POS variables with self-reported leisure-time brisk walking and OM-MVPA were estimated using generalized additive mixed models (GAMMs) [32
]. GAMMs can assume data with various distributional assumptions, and also cover mixed-effects models, accounting for dependency in error terms due to clustering (in our case, participants enrolled from selected primary care facilities), and estimate complex dose-response by handling non-linear, linear, and non-monotonic relationships [33
]. Preliminary analyses indicated that GAMMs with Gaussian variance and identity link functions would be most appropriate for the continuous outcomes: minutes/day of self-reported leisure-time brisk walking, and minutes/day of OM-MVPA 10-min bouts. Regression coefficient estimates of these GAMMs represent the change in outcomes per increment 1 km, 1 km2
, or count in exposure to POS. GAMMs with binomial variance and logit link functions were the most appropriate for the dichotomized outcomes: any ≥150 min per week of self-reported leisure-time brisk walking and any ≥150 min per week of OM-MVPA 10-min bouts. Antilogarithms of the regression coefficient estimates of these GAMMs represent odds ratios.
The curvilinearity relationships were assessed with thin-plate spline smooth terms [32
] adjusted for sex, age (≤65 years vs. >65 years), education level (primary or less vs. more than primary education), and self-rated health (excellent/very good/good vs. fair/poor). For all GAMMs, random intercepts were specified to account for clustering effects at the administrative unit level, as participants were enrolled from selected primary care facilities. Analyses of a curvilinear relationship were conducted for each model observing the Akaike Information Criterion (AIC), differences ≥10 caused the smooth terms of the model to be removed and replaced by a simpler linear term [34
Further, interaction of weather conditions specifically rainy conditions, with the association of objectively-measured POS and walk-friendly route variables with OM-MVPA was also examined. Significance of interactions was assessed by adding cross-product terms between our exposure of interest and rainy conditions (classified in two categories: rainy or non-rainy periods). Then for the walk-friendly route measures that presented significant interactions, stratified analyses were performed, by examining the association of these POS measures and OM-MVPA in rainy periods and non-rainy periods, separately.
All analyses were conducted in R version 3.3.3 (R Development Core Team, Vienna, Austria) using ‘stats’ [35
], ‘mgcv’ [33
], and ‘gamm4’ [36