Study data
The population-based IDEFICS study, was conducted from 2006 to 2012 to investigate lifestyle-related diseases in European children and infants from eight countries (Belgium, Cyprus, Estonia, Germany, Hungary, Italy, Spain, and Sweden) [15]. The baseline survey (T0) took place between September 2007 and June 2008 including 16,229 2- to 9.9-year-old children [15]. The first follow-up survey was conducted 2 years later from September 2010 to May 2011, where 11,041 children aged 4 to 11.9 years participated in the follow-up examination (T1) and 2555 children were newly recruited [19]. In addition, a second follow-up (T2) was conducted that only assessed the penetration of the intervention messages by mail and that did not comprise the survey protocol of T0 and T1. Participants of the IDEFICS study (T0/T1) were re-invited to participate in the I.Family study for an enhanced third follow-up (T3) in 2013/ 2014 where 7105 children, their siblings and parents provided information based on an extended survey protocol [20] aiming to investigate entire families. In each country, the participating centres obtained ethical approval from the local ethics committees. Parents provided written informed consent for all examinations. Each child was informed orally about the measurements by field workers and asked for his/her consent immediately before the examination. The Pan-European IDEFICS / I.Family children cohort is registered under ISRCTN62310987.
The present analysis is based on data from baseline and follow-up surveys of the IDEFICS study as well as the I.Family study (T3) from seven study regions in three countries. We considered N =6185 observations of n =3287 children and adolescents who participated in the IDEFICS/I.Family cohort and wore an accelerometer device in at least one of the surveys (NT0 =2934, NT1 =1933, NT3 =1318). Environmental variables could not be calculated for N =1968 observations of children who did not directly live within the study area. Further, N =823 had to be excluded due to invalid or unavailable accelerometer measurements, leaving N =3394 observations of n =2488 participants. Most of the participants (n1 =1685) provided one valid observation, while two observations were provided by n2 =700, and three observations by n3 =103 participants. In this sample, only three variables had a small number of item missings, i.e. ISCED: 3.1%, safety concerns: 5.8%, and sports club membership: 6%, for which we included a missing category, each.
Physical activity
Habitual PA was assessed using Actigraph accelerometers (Actigraph,LLC, Pensacola, FL, USA). In IDEFICS (T0 and T1), either ActiTrainer or GT1M monitors were used, while in I.Family either GT1M or GT3x + devices were used. Participants were asked to wear the accelerometers for at least 3 days (including 1 weekend day) at T0 and T1 and for 7 days at T3. Accelerometers were mounted on the right hip during waking hours of each child using an elastic belt adjusted to ensure close contact with the body.
Details on processing of accelerometer data in the IDEFICS study as well as first descriptive results of accelerometer data of the IDEFICS study can be found in Konstabel et al. [21]. Valid measurements were defined as recording more than 360 min of at least one weekday and one weekend day after exclusion of non-wear time according to Choi et al. [22]. Nonwear time was identified using a 60 min. Window for each epoch to detect 30 min consecutive zero counts allowing breaks of 2 min of non-zeros. The threshold for valid measurements of 360 min. at least for at least one weekday and one weekend day was chosen as a trade-off between accuracy and sample size and is discussed in Konstabel et al. [21].
Before assigning intensity ranges, we here used a penalized expectile regression to smoothen the accelerometer counts that has been recently proposed in Wirsik et al. [23]. This method is able to identify underlying activity patterns similar to hidden Markov models (HMM) that were also proposed to improve modelling of accelerometer data. The penalized expectile regression was compared with the commonly used cut-off point methods and HMMs based on labeled data and outperformed the latter [23]. MVPA and LPA in minutes per day were then derived based on Evenson cut-off points for smoothed counts per minute (light: 104–2295, moderate: 2296–4011 cpm, vigorous: > 4011 cpm) [24].
BMI was calculated based on objectively measured height and weight that were assessed to the nearest 0.1 cm and 0.1 kg, respectively [19]. Age- and sex-specific BMI z-scores and categories for overweight and obesity were derived according to the extended IOTF criteria [25].
Covariables
Season of assessment was categorized as spring/summer if the accelerometer device was worn between March and September, and as autumn/winter, if assessment took place between October and February.
Education and qualification of parents were classified according to the International Standard Classification of Education (ISCED) [26]. We collapsed ISCED-levels into three categories, i.e. low (lower secondary education and less), medium (upper and post-secondary education), and high (tertiary education). We further added a category for missing values in ISCED-levels.
Parents were asked to respond to statements regarding safety concerns, i.e. “I restrict my child’s outdoor activities for safety reasons” and “I don’t like to let my child walk/cycle to kindergarten, pre-school or school for safety reasons” on a four-point Likert scale, i.e. disagree, moderately disagree, moderately agree, and agree. While both statements were part of the first two surveys (T0, T1), only the latter was included in the third follow up survey (T3). Agreement or strong agreement to any of the two statements was categorized as having safety concerns, while strong disagreement and disagreement was categorized as no concern.
Sports club membership (yes/no) was reported by parents for baseline and follow up (T0, T1). In T3 this was proxy reported by parents if the child was younger than 12 years, or self-reported, if the child was older than 12 years.
Spatial data
In seven different study regions of three countries, geographical data were collected and processed to objectively assess built environment characteristics by means of a moveability index, i.e. Germany (Delmenhorst and Wilhelmshaven), Italy (Avellino, Atripalda, Mercogliano), and Sweden (Partille and Mölndal), using a geographical information system (GIS) (ESRI 2011. ArcGIS Desktop: Release 10.2 Redlands, CA: Environmental Systems Research Institute). Geographical data were processed to calculate the moveability index based on administrative data as well as open source databases.
The moveability index is an extension of the walkability index [27, 28] and quantifies opportunities for PA, in particular for active travel and leisure time PA, in the home neighbourhood of children. This index showed a positive association with MVPA in children based on cross-sectional IDEFICS data from one German study region [8].
In all study regions administrative data were provided by the land registry office of the local municipality or the federal state. Land use data were provided as adjacent polygons and condensed with regard to six different types including residential, commercial, industrial & agricultural, recreational, and miscellaneous. Residential density was obtained on district and subdistrict level. Geographic line data of the footpath network were obtained from the OpenStreetMap project (OSM) (www.openstreetmap.org – Open Data Commons Open Database License (ODbL)) and validated using administrative data. In all study regions a footpath network was built to calculate service areas and to derive intersections as point data. Bus stops and recreational facilities, i.e. playgrounds and parks, were digitally processed based on available maps and lists provided by the public transport companies and the civil service for green space of the municipalities [29].
Home neighbourhoods
Addresses of participants were geocoded for each survey to derive network-dependent home neighbourhoods. We further accounted for residential relocation which, however, was not observed in participating children who provided two or three observations over time. If children relocated after participating first in either the baseline survey or the first follow-up and the new residential location was outside of the study areas this led to exclusion in the environmental analysis for the following surveys. Especially in the German study regions, it was not permissible to use the exact address coordinates to calculate individual home neighbourhoods due to data protection requirements. Therefore, we carried out spatial blurring based on a Gaussian error that was inversely proportional to the underlying residential density and conducted a simulation study, where spatial blurring shifted original coordinates by approximately 50 to 100 m in densely-populated areas induced only small differences in moveability measures [30]. We conducted the network analyses using the network analyst in ArcGIS 10.2 and calculated the spatial blurring in R 3.4.3 [31] using the rnorm function. Individual-level home neighbourhoods were defined based on network-dependent areas around the place of residence using a distance of 1250 m that was chosen based on previous research [30].
Moveability index
The moveability index consists of the following five main components:
Residential density
Residential density, i.e. number of residents per area, was provided in districts or subdistricts of the considered study regions. For each home neighbourhood residential density was then derived as a weighted mean considering the size of the fraction of districts overlapping the home neighbourhood.
Land use mix
Percentages of land use types, i.e. residential, commercial, industrial & agricultural, recreational, and miscellaneous, in each network-dependent neighbourhood were derived to calculate land use mix based on the entropy formula [27].
Point characteristics
Point characteristics such as intersections, public transit stations and public open spaces were assessed using an anisotropic kernel intensity measure that provides consistent results over varying sizes of the home neighbourhood and tends to reduce bias through scaling and zoning [30]. This way, intersection density, i.e. street connectivity, as well as availability of public transport and public open spaces were calculated as mean intensity per home neighbourhood.
In order to compare opportunities for PA in each region instead of comparing the moveability between countries, z-scores were calculated separately for each region using the corresponding mean and standard deviation (SD) of the moveability index and its components, respectively. We further used the z-scores to dichotomise the moveability index and environmental variables into high (z-score ≥ 0) and low (z-score < 0). Spatial analyses were conducted using the spatstat-package [32] in R 3.4.3 [31].
Statistical analyses
Descriptive statistics, i.e. percentage or mean (SD) and range, of outcome, exposure variables and covariables were calculated based on the first examination of each of the n =2488 participants.
Age-dependent trajectories of MVPA and LPA were estimated using linear mixed models including two levels (repeated measurements nested within individuals) that allow to model different intercepts and age effects, i.e. these models allow study subjects to have their own trajectory over time, where individual trajectories for participants providing only one observation are calculated using supplement information by estimated population level trajectories. These models can easily handle unbalanced data with varying numbers of repeated measurements per subject, as well as subjects measured at different ages. Moreover, these models allow for change in scale and variance of the outcome measurements over time [33].
For each outcome variable, i.e. MVPA, and LPA, six models were estimated to investigate the effect of the moveability index as well as its five components on PA intensities with age. The model included a random intercept and random linear slope for age. Further, repeated measurements were taken into account by means of a random effect on the residual side. For each environmental variable a fixed effect as well as an interaction effect with age was included to model the effect of the built environment on MVPA and LPA over age. All models were adjusted for age (centred at 8 years), maximum ISCED level of both parents, parental safety concerns, sports club membership, valid wear time and season of accelerometer measurements, as well as study region.
In addition, all models were estimated stratified by sex, to investigate the effect of environmental variables to take into account the well-documented differences in PA intensities between girls and boys.
Estimated linear trajectories across age were depicted for high moveable and low moveable home neighbourhoods in boys and girls. Differences in these trajectories were quantified based on least square means (LSM) and 95% CIs that were calculated in each model for chosen age values, i.e. 4, 6, 8, 10, 12 and 14 years covering the age-range of our analysis.
We conducted sensitivity analyses by estimating linear trajectories over age similarly as described above using a reduced study sample of 1709 observations of 803 participants by only including children and adolescents who provided at least two measurements.
Statistical analyses were conducted in SAS 9.3 (SAS Institute Inc., Cary, North Carolina, USA) and mixed models were estimated using the glimmix procedure. All results are presented at a significance level of α = 0.05 without adjusting for multiple testing.