Study design and participants
The Physical Activity and Nutrition in Children (PANIC) Study is a controlled lifestyle intervention study investigating the effects of a combined physical activity and dietary intervention on cardiometabolic risk factors in a population sample of children from the city of Kuopio, Finland. The Research Ethics Committee of the Hospital District of Northern Savo approved the study protocol in 2006 (Statement 69/2006). A written informed consent was acquired from the parent or caregiver of each child and every child provided assent to participation. This consent or assent could be revoked by the parent or child at any time.
We invited 736 children 6–9 years of age who started the first grade in 16 primary schools of Kuopio in 2007–2009 (Fig. 1). Altogether 512 children (248 girls, 264 boys), who accounted for 70% of those invited, participated in the baseline examinations in 2007–2009. The participants did not differ in age, sex, or body mass index standard deviation score (BMI-SDS) from all children who started the first grade in the city of Kuopio in 2007–2009 based on data from the standard school health examinations performed for all Finnish children before the first grade. We excluded six children from the study at baseline because of physical disabilities that could hamper participation in the intervention or no time or motivation to attend in the study. We also excluded two children whose parents later withdrew their permission to use the data of their children.
We cluster divided the remaining 504 children into an intervention group (306 children) and a control group (198 children) according to schools to be able to organize after-school exercise clubs carried out at school premises only for the intervention group and to avoid non-intentional intervention in the control group. We also matched the intervention and control group according to the size (large vs. small) and location (urban vs. rural) of the schools to minimize sociodemographic differences in baseline characteristics between the groups. We included more children in the intervention group than in the control group because of a larger number of dropouts expected in the intervention group and to retain a sufficient statistical power for the comparison between the groups.
All the measurements conducted at baseline were repeated at 2-year follow-up. Among 504 children who participated in the baseline examinations, 438 (87%) children also attended the 2-year follow-up examinations, 261 children from the intervention group (85%) and 177 children from the control group (89%). The median (interquartile range) follow-up time was 2.11 (interquartile range 2.07–2.16) years in intervention group and 2.13 (interquartile range 2.05–2.22) years in control group. Data on physical activity were available for 432 children (216 girls, 216 boys) at baseline and for 355 children (181 girls, 174 boys) at 2-year follow-up. Data on diet were available for 423 children (206 girls, 217 boys) at baseline and for 389 children (185 girls, 204 boys) at 2-year follow-up (Fig. 1).
Physical activity and dietary intervention
The goals of this 2-year individualized and family-based physical activity and dietary intervention were based on national recommendations for physical activity and nutrition [10,11,12]. The goals were to (1) increase total physical activity by emphasizing its diversity, (2) decrease total and particularly screen-based sedentary behavior, (3) decrease the consumption of significant sources of saturated fat and particularly high-fat dairy and meat products, (4) increase the consumption of significant sources of unsaturated fat and particularly high-fat vegetable oil-based margarines, vegetable oils, and fish, (5) increase the consumption of vegetables, fruits, and berries, (6) increase the consumption of significant sources of fiber and particularly whole grain products, (7) decrease the consumption of significant sources of sugar and particularly sugar-sweetened beverages, sugar-sweetened dairy products, and candy, (8) decrease the consumption of significant sources of salt and the use of salt in cooking, and 9) avoid excessive energy intake, for example, by recommending regular consumption of main meals and avoiding frequent snacking.
The intervention included 6 physical activity and diet counseling sessions consisting of 30–45 min of physical activity counseling and 30–45 min of dietary counseling for the children and their parents during the 2-year follow-up [13]. The 6 counseling sessions occurred 0.5, 1.5, 3, 6, 12, and 18 months after baseline. In these counseling sessions, the children and their parents received individualized advice from a specialist in exercise medicine and a clinical nutritionist on how to increase physical activity, decrease sedentary behavior, and improve diet among children in everyday conditions. Each counseling session had a specific topic on physical activity, sedentary behavior, and diet according to the goals of the intervention and included practical tasks on these topics for the children. In the counseling sessions, the children and their parents also received fact sheets on physical activity, sedentary behavior, and diet, verbal and written information on opportunities to exercise in the city of Kuopio, and some material support for physical activity, such as exercise equipments and allowance for playing indoor sports. Of all 306 children in the intervention group, 266 (87%) participated in all 6 counseling sessions, 281 (92%) in at least 5 counseling sessions, and 295 (96%) in at least 4 counseling sessions.
We encouraged the children in the intervention group, particularly those who did not attend organized sports or exercise, to participate in after-school exercise clubs supervised by trained exercise instructors. Of all 306 children in the intervention group, 254 (87%) children participated in at least one of the after-school exercise clubs and 124 (41%) children attended the after-school exercise clubs at least once a month.
The children and their parents in the control group received general verbal and written advice on health improving physical activity and diet but no active intervention.
Assessment of body height, body weight, body fat percentage and puberty
Body height and weight were assessed after the children fasted for 12 h. Body height was assessed using a wall-mounted stadiometer and body weight using the InBody® 720 bioelectrical impedance device (Biospace, Seoul, Korea), with the weight assessment integrated into the system. We computed age- and sex-standardized BMI-SDS using Finnish references [14] and defined overweight and obesity using the International Obesity Task Force criteria for children that are based on centile curves passing through adult BMI cutoffs at 25 for overweight and at 30 for obesity [15]. We measured body fat percentage with the children being in the supine position, having emptied the bladder, and being in light clothing by dual-energy X-ray absorptiometry using the Lunar Prodigy Advance® dual-energy X-ray absorptiometry device (GE Medical Systems, Madison, WI). A research physician classified the girls as entered clinical puberty if their breast development had started and the boys if their testicular volume assessed by an orchidometer was ≥ 4 ml [16] according to the criteria described by Tanner [17].
Assessment of lipids and lipoproteins
Fasting plasma concentrations of total cholesterol, LDL cholesterol, HDL cholesterol, and triglycerides were analyzed using a clinical chemistry analyzer (Hitachi High Technology Co, Tokyo, Japan). Before analysis, VLDL was separated by ultracentrifugation (37 000 rpm, 15 h). Concentrations of plasma total cholesterol and triglycerides as well as LDL, HDL, and VLDL cholesterol and triglycerides were measured by colorimetric enzymatic assays.
Assessment of diet
We assessed the consumption of food and drinks using one food record at baseline and one food record at 2-year follow-up [18]. The food record at both timepoints covered 4 predefined and consecutive days, including at least 1 weekend day. In addition at 2-year follow-up, 0.5% of all food records covered 3 days and consisted of 2 weekdays and 1 weekend day and they were also included in the analyses. Clinical nutritionists checked the filled food records together with the family and added any missing information. We calculated food consumption and nutrient intake using the Micro Nutrica® dietary analysis software, Version 2.5. The software is based on detailed information about the nutrient content of foods in Finland and other countries [19]. Moreover, a clinical nutritionist updated the software by adding new food items and products with their precise nutrient content based on new data in the Finnish food composition database [20] or received from the producers.
Assessment of physical activity and sedentary behavior
Physical activity and sedentary behavior were assessed by a combined movement and heart rate sensor (Actiheart®; CamNtech, Cambridge, UK) [21]. The sensor was attached on the chest with 2 standard electrocardiogram electrodes and set to record in 60-s epochs. The participants were requested to wear the sensor continuously for a minimum of 4 consecutive days, including 2 weekdays and 2 weekend days, and instructed to carry on with their usual behavior and to wear the sensor during all daily activities, including sleep, shower, sauna, and swimming. Upon retrieving the sensor, heart rate data were first cleaned [22], then individually calibrated with parameters from a previously performed maximal cycle ergometer exercise test [23] and combined with trunk acceleration using branched equation modelling to produce intensity time-series [24, 25]. Whilst minimizing diurnal bias caused by any potential non-wear episodes [26], physical activity energy expenditure was calculated by time-integration of the intensity time-series, and the time distribution of activity intensity was generated using standard metabolic equivalent of tasks (METs) in 0.5 increments. For these analyses, the equivalent of 3.5 mL O2 /kg/min (71.2 J/kg/min) was used to define 1 MET, and data were summarized as sedentary behavior (≤ 1.5 METs excluding sleep), light intensity physical activity (> 1.5 and ≤ 4.0 METs), and moderate-to-vigorous physical activity (> 4 METs). Physical activity records were included in the analysis if there was a minimum of 48 h of activity recording in weekday and weekend day hours that included at least 12 h from morning (3 am–9 am), noon (9 am–3 pm), afternoon (3 pm–9 pm), and night (9 pm–3 am) to avoid potential bias from over-representing specific times and activities of the days.
Assessment of parental education
We collected data on parental education level by a questionnaire. Parental education level was defined as the highest completed or ongoing degree of the parents (vocational school or less; polytechnic or university).
Power calculations
We determined the number of children required to detect at least a 0.30 standard deviation difference in the primary outcomes between the intervention group (60% of children) and the control group (40% of children) with a power of 80% and a two-sided p-value for the difference between the groups of 0.05, allowing for a 20% loss to follow-up or missing data. According to these power calculations, we would need a sample size of at least 275 children in the intervention group and at least 183 children in the control group at baseline.
Statistical methods
We performed all statistical analyses using the IBM SPSS Statistics® software, Version 25.0 (IBM Corp., Armonk, NY, USA). P-values < 0.05 were used to indicate statistical significance, based on two-sided testing. The distributions of each continuous variable were examined by observing histograms and logarithmic transformation was performed for plasma VLDL cholesterol, plasma triglyceride, and plasma VLDL triglycerides, because of their skewed distribution. We compared baseline characteristics between the groups by the t-test for independent samples or the Chi-Square test. We studied the effects of the intervention on plasma lipids and lipoproteins using the intention-to-treat principle including all 504 children in the statistical analyses. We used the linear mixed-effects model analyses according to a three-level structure, i.e., repeated outcome measures (baseline and follow-up) were clustered within children who were considered as subjects in the mixed model structure and children were clustered within schools. The linear mixed-effect models are especially suitable for analyzing longitudinal datasets containing correlated and unbalanced data. We started with a model adjusted for age at baseline, sex, and pubertal stage at both time points, including main effects for time and time-by-study group interaction: OUTCOMEit = (β0 + ui) + β1age + β2sex + β3 cpubertal stage + (β4 + vi)time + β5study group x time + εit, where OUTCOMEit are observations for subject i3 at baseline and follow-up; β0 is the intercept; β1, β2, β3, β4, and β5, are the regression coefficients for age, sex, pubertal stage, time, and study group x time, respectively; ui are random, subject specific intercepts and vi are corresponding random slopes for follow-up time; and εit is the error for subject i at time t. Time was treated as a continuous variable to allow for a slight variation in follow-up time (1.8–2.5 years) among the children.
We used a Bayesian information criterion (BIC) as a measure of model adequacy. The BIC value penalizes the likelihood of the observed data based on the total number of parameters in a model. A lower BIC value indicates a better model with a better balance between complexity and good fit. We fitted all possible models with allowing or ignoring the possible clustering on subject and/or school level for each dependent variable. However, we a priori chose the model with the lowest BIC value as our final model for a given variable. Thus, we did not force the three-level data structure to our model since it did not improve model fit but resulted in unnecessary complexity.
The data for all lipids and lipoproteins showed the best fit with the model in which a random intercept and a random regression coefficient of time were modeled on the subject level using an independent variance structure, but no random effect for intercept or regression coefficient of time on the school level was included.
One of the typical problems related to the use of the time-by-study group interaction is the phenomenon of regression to the mean due to the differences between the intervention and control group at baseline. We had no differences in plasma lipids and lipoproteins between the study groups at baseline (p = 0.108–0.729). Therefore, we did not include the study group in the model to allow for the regression to the mean phenomenon. Instead, baseline values in the study groups are reflected in the intercept of the model.
In further analyses, we studied the possible confounding factors on the effects of the intervention on plasma lipids and lipoproteins adjusting the analyses for body fat percentage at both time points or for parental education level.
We also studied how physical activity, sedentary behavior, and diet contributed to the observed effects of the intervention using linear mixed-effect models adjusted for age at baseline, sex, and pubertal stage at both time points and entering the physical activity, sedentary behavior, and diet variables, according to the aims of the intervention, one by one as potential explanators for the effects. Changes in the regression coefficients after entering the physical activity, sedentary behavior, and diet variables in the models are presented to show the magnitude of the effect of these adjustments.