Study design and participants
The current work was part of the International Study of Childhood Obesity, Lifestyle and the Environment (ISCOLE) survey, designed to examine the associations between lifestyle behaviors and obesity in children [20]. The participants were 9–11-year-old children who were recruited from schools located in urban and semi-urban areas of 12 countries, including Finland. The detailed protocol of recruitment and data collection has been published elsewhere [20]. All ISCOLE participants in Finland were also invited to the national ancillary study on allergic diseases. The current work used the data collected from Finnish ISCOLE participants alone. In brief, a cross-sectional survey was conducted between September 2012 and April 2013 in primary schools in the metropolitan area of Finland. The sampling frame comprised a complete list of primary schools of the cities of Helsinki, Espoo, and Vantaa, stratified by indicators of socio-economic status (primarily parental educational level, if available, otherwise income level). Children aged 9 to 11 years in the 4th grade were selected from each stratum of schools. One class in a school was hence the smallest unit of recruitment, and all pupils from the class were invited. Consequently, a total of 39 schools were approached, and 25 (64%) consented to participate. From these consenting schools, 789 children were invited, of whom 542 (69%) from 25 primary schools participated in the ISCOLE study in Finland (Fig. 1).
Flow chart presenting the sample sizes in the ISCOLE sub-study on lifestyles and allergic diseases
a had complete data on lifestyle factors
b 30 children missing asthma and eczema symptoms, 31 children missing symptoms of allergic rhinitis
c primary analytic sample
d secondary analytic sample
Abbreviations: FFQ, food frequency questionnaire; IgE; immunoglobulin E; MVPA, moderate-to-vigorous physical activity
The study was conducted according to the Declaration of Helsinki and approved by the Helsinki and Uusimaa Hospital District Ethics Committees. Informed parental consent as well as child assent were obtained for each participant.
Lifestyle factors
In the current work, we focused on the following four dimensions of lifestyle: dietary patterns, physical activity, nightly sleep duration, and screen time.
Dietary assessment
The children’s usual food consumption frequency was recorded using a 23-item non-quantitative food frequency questionnaire (FFQ; i.e., portion sizes were not asked), which showed acceptable validity against a pre-coded food diary [21] and reliability in assessing dietary patterns [9]. The FFQ inquired how many times children usually ate foods and beverages using the following seven frequency alternatives: never; less than once a week; once a week; 2–4 days a week; 5–6 days a week; once a day; or more than once a day.
For further analysis, consumption frequencies were recoded into numeric weekly consumption frequencies as follows: ‘never’ into 0, ‘less than once a week’ into 0.5, ‘once a week’ into 1, ‘on 2–4 days a week’ into 3, ‘on 5–6 days a week’ into 5.5, ‘once a day’ into 7, and ‘more than once a day’ into 10. Based on these numerical weekly consumption frequencies, principal component analysis (PCA) was applied to identify dietary patterns among the children. The PCA combined the consumption frequencies of foods and beverages that correlated with each other, and had been previously performed for the total international ISCOLE dataset and for the data of each country separately. In the current work, we used the country-specific dietary patterns of Finnish ISCOLE participants. Detailed description of the PCA procedures can be found elsewhere [9]. In brief, after the first run of PCA, two components were selected based on eigenvalues and scree plot inspection. Fruit juice consumption was excluded from the PCA due to its low validity [21]. A rerun with the two components was performed with an orthogonal Varimax rotation, and standardized principal component scores were computed for each child. The two components together explained 33% of the total variance in FFQ data and were labelled ‘unhealthy dietary pattern’ and ‘healthy dietary pattern’ based on the food loadings with an absolute value of > 0.30 (Supplementary Table 1, Additional File 1). The unhealthy dietary pattern was characterized by high loadings of fast foods (e.g., hamburgers, pizza), fried food (nuggets, fish sticks), French fries, potato chips, ice-cream, and sugar-sweetened sodas, whereas the healthy dietary pattern comprised high loadings for dark-green vegetables, orange vegetables, vegetables in general, fruits and berries, wholegrains, and fish.
Sleep duration and physical activity
Objective measurements of children’s nightly sleep duration and time spent in moderate-to-vigorous physical activity (MVPA) were obtained by 24-hour accelerometry. The children wore an Acti-Graph GT3 × 1 accelerometer (Pensacola, FL, USA) on a waistband and were encouraged to wear it 24 h per day for at least seven days, including two weekend days, while maintaining their normal daily routines.
Nightly sleep duration was estimated using a refined, validated algorithm that excluded extended episodes of nocturnal non-wear and wakefulness and avoided misclassified daytime sleep episodes [22]. Briefly, sleep time was defined as time between algorithm-determined sleep onset and sleep offset. A night was considered valid if overnight sleep time was ≥ 160 min. At least three valid nights of sleep, including at least one weekend night (Friday or Saturday), was a minimum criterion for including sleep data in the analysis. Each child’s mean nightly duration of sleep from all valid sleep nights was considered in the current analyses.
For recording physical activity, a valid day was defined as ≥ 10 h of wear time excluding sleep time and awake non-wear time (any sequence of ≥ 20 consecutive minutes of zero activity counts). At least four valid days, including at least one weekend day, were required to include physical activity data in the analysis. In the current work, we focused on MVPA since recommendations for school-aged children are based on this level of intensity [23]. Using commonly employed cut-points by Evenson et al. [24], MVPA was defined as ≥ 574 counts per 15-second period. Each child’s mean daily duration of MVPA from all valid days was considered in the current analyses.
Screen time
As objective measures cannot distinguish the type of sedentary behavior, we relied on the children’s own reports on their screen time. They were asked to fill in a questionnaire inquiring (1) how many hours they watched TV; and (2) how many hours they played video or computer games or used a computer for something that was not schoolwork in the last week. Both questions were asked for a typical school day and weekend day separately. Answer options were 0, < 1, 1, 2, 3, 4, and ≥ 5 h per day. To obtain mean daily screen time scores, the answer options ‘<1’ and ‘≥5’ were recoded to ‘0.5’ and ‘5’, respectively. Then, the answers on time spent with the screen devices were weighted and added up as follows: hours of TV on school days x 5/7 + hours of TV on weekend days x 2/7 + hours of video games or computer on school days x 5/7 + hours of video games or computer on weekend days x 2/7. This produced a possible range of from 0 to 10 h per day.
Outcomes
Allergic disease symptoms
The prevalence of allergic symptoms, i.e., those of asthma, allergic rhinitis, and eczema, was recorded using a modified International Study of Asthma and Allergies in Childhood (ISAAC) allergy questionnaire [25] completed by the parents. The prevalence of asthma symptoms was defined as a positive answer to one or both of the following questions: ‘Has the child had prolonged cough (for more than 6 weeks)?’ and ‘Has the child ever had wheezing sounds while breathing or appeared to have difficulty in breathing?’. The prevalence of rhinitis symptoms was defined as a positive answer to one or both of the following questions: ‘Do pollen or animals cause sneezing or runny or blocked nose?’ and ‘Do pollen or animals cause red, itchy, or swollen eyes?”. The prevalence of eczema symptoms was defined as a positive answer to the question ’Does the child have dry, red, and itchy skin that requires regular care?’.
IgE measurements
In order to assess allergen-specific IgE responses, blood samples were collected between January 29th and June 3rd 2013 from 173 children (52% of the invited; 32% of the Finnish survey sample) (Fig. 1). A blood specimen of 9 ml was taken from the antecubital vein. Specific IgE responses to common inhaled and food allergens were analyzed using the Pharmacia CAP-fluoroenzyme immunoassay. Measurements were performed in the laboratory of Helsinki University Hospital. Allergic sensitization was defined as an allergen-specific IgE concentration of ≥ 0.35 kU/L, based on which the following two outcomes were derived: sensitization to ≥ 1 inhaled allergen (birch, timothy grass, mugwort, cat, dog, horse, house dust mite [Dermatophagoides Pteronyssinus], mold [Cladosporium Herbarum]); and sensitization to ≥ 1 food allergen (cow’s milk, egg white, codfish, wheat, soy, peanut).
Confounding variables
In the current work, we considered ten potential confounding variables based on theory or earlier findings. For body mass index (BMI), trained research staff measured the children’s height and weight in the schools such that the children were without shoes and heavy clothing. Height was measured using a Seca 213 portable stadiometer (Hamburg, Germany). Weight was measured using a Tanita Body Composition Analyzer SC-240 scale (Arlington Heights, Illinois). BMI was computed as kg/m2. Using age- and sex-specific reference data from the World Health Organization [26], BMI Z-scores were then calculated and used as a continuous variable in the current analyses. Parents reported the child’s sex and eight other background variables through questionnaires. We treated the age when solid foods were introduced and the age when completely stopped being breastfed as continuous variables. As a surrogate for the child’s birth order, we considered the number of older biological siblings the child had. This was categorized in the descriptive analyses (0, 1, ≥ 2 older siblings) and used as a count variable in the multivariable modelling. As dichotomous data, we used the information on furry pets at home or daycare facility (yes, no), maternal smoking during pregnancy (yes, no), and current parental smoking (one or both of the parents, none). Parental allergy history was defined as having a history of at least one of the following: asthma, pollen or animal allergy, food allergy, or atopic eczema. Then, parental allergy history was divided into four categories as follows: both parents, mother alone, father alone, or neither of the parents had a history of allergic disease. We divided parents’ educational level into three categories (high school or less, college, bachelor’s degree or postgraduate degree), of which the highest achieved by either parent was considered in the current work.
Statistical methods
All analyses were carried out with two-tailed tests using the R statistical programming language version 4.3.2 [27]. We considered P-values below 0.05 to be statistically significant.
Cluster analysis
To identify groups of children with similar lifestyle behaviors in the data, a cluster analysis was employed. As input variables, we used unhealthy and healthy dietary pattern scores, MVPA, nightly sleep duration, and screen time, which were standardized into Z-scores due to different measurement units. Children who provided complete data on all five lifestyle variables were included in the analysis. Since all lifestyle variables were continuous, we considered a widely used K-means algorithm [28] to be suitable for the current work. It performs an iterative process of assigning observations to groups based on their distances from the pre-selected number of cluster centers [28]. We used the function kmeans, available in the R package stats. We set a random initial seed to initialize cluster centers and used 25 different random starting assignments to optimize the allocation to clusters of similar features. To determine the number of clusters identified, a four-cluster solution was used as a starting point, as suggested by the Elbow method, and clustering was then repeated with three- and two-cluster solutions. Distinguishing features of the cluster solutions were identified by comparing the values of the resulting cluster centers, i.e., average Z-scores. The final cluster solutions were based on interpretability and degree of distinction, inspected visually, and a cluster membership was then recorded for each child. To examine the robustness of the final cluster features, we omitted univariate outliers, i.e., those with a Z-score of > 3 or < − 3 in any of the standardized lifestyle variables, and repeated clustering. We also randomly shuffled the observation rows and repeated clustering several times to confirm whether the cluster features remained the same. Lastly, we verified the stability of the cluster solutions by randomly dividing the data into halves, i.e., two equal subsamples on which clustering was repeated. Then, Kappa degrees of agreement (Ƙ) were computed between the cluster memberships of the subsamples and those of the total sample.
Descriptive analyses and modelling
To compare background characteristics between the included and excluded children and across binary allergy outcomes, we used the independent samples t-test for continuous variables, Mann-Whitney U-test for skewed continuous variables, and chi-squared or Fisher’s exact test for categorical variables. A Venn diagram was drawn to illustrate the co-occurrence of allergic symptoms. Logistic regression models were used to explore the associations of lifestyle cluster memberships with the prevalence of allergic symptoms and sensitization, with results presented as odds ratios (OR) and their 95% confidence intervals (CIs). The models were fit separately for each allergy outcome. The unadjusted models included cluster membership as a predictor variable. Variables included in the adjusted models were sex, number of older siblings, parental history of allergic disease, exposure to furry pets, age when solid foods were introduced, age when breastfeeding was stopped, current parental smoking, maternal smoking during pregnancy, BMI Z-scores, and the family’s highest educational level, based on their theoretical importance. We also conducted data-driven analyses using the following exploratory outcomes in modelling: manifestation combinations of allergic symptoms (≥ 2 symptoms, all three symptoms), higher-threshold sensitization (IgE ≥ 0.70 kU/L), polysensitization (specific IgE ≥ 0.35 kU/L to two or more allergens), and symptomatic sensitization (specific IgE ≥ 0.35 kU/L to any allergen, accompanied by symptoms of any allergic disease).
