Teun Remmers

Relationship between PA and the development of BMI in children | 27 Strengths and weaknesses One of the strengths of this study is the longitudinal design, containing two periods of repeated measurements, representing a period in childhood that included an important developmental stage in childhood adiposity (27). In total, 105 children contributed to both time periods, providing accelerometry and anthropometry at three subsequent time points. Accelerometers are considered to be a valid measure of children’s daily PA behavior (3). By the parental reports of cycling and swimming we controlled for uncertainty regarding these activity types when using accelerometer measurements. Furthermore, as we measured PA for a minimum of two weekdays and one weekend day, the between-day intraclass reliability coefficient of PA in 4-5 year-old children (irrespective of intensity) was 0.62 when applied for the first three days of measurement. Similarly, the intraclass reliability coefficient for 6-7 and 8-9 year-old children was 0.63 and 0.54, respectively. In addition, we showed that the present sample did not differ significantly from the total KOALA Birth Cohort ( N = 2834) in terms of gender and recruitment group. In total, 13.3% of the children in our sample had an alternative lifestyle. As these children may be exposed to alternative dietary patterns, the relationship between PA and BMI development may be different in these children. However, we found no evidence for a statistically significant interaction in this relationship. Finally, we controlled for the potential confounding effect of the recruitment group in all analyses. By including these participants, our results are generalizable to a population with diverse views of health and life. The present study was also prone to some weaknesses. The type of measurement (e.g. parent-reported versus measured height and weight to calculate BMI) may have biased our results. One study comparing measured and parent-reported height and weight on a national scale in the United States, reported that parents of 2-11 year-old children overestimated their child’s overweight by underestimating their child’s height (1). Another study showed that parents underestimated their child’s overweight by underestimating weight and overestimating height (31). In addition, a study that used 64.4% of the total KOALA cohort reported that parents underestimated their 6-7 year-old child’s overweight (36). When examining longitudinal associations, this potential bias may be differential, depending on the combination of origins present. For example, a child’s height and weight may be measured at 6-7 years and parent-reported at 8-9 years or vice versa. Such differential bias was not possible in the first period, as all participants’ height and weight were measured at 4-5 years old. In the second period, we found that the combination of parent-reported and measured anthropometry over time (e.g. parent- reported at baseline versus measured at follow-up) differed between boys and girls and we therefore included this interaction in the adjusted models. Furthermore, the selection of participants to measure (instead of report) was based on the availability of accelerometry at 6-7 years old, and not based on any PA or BMI trait. This was supported by an intraclass correlation coefficient of -0.08 (95% CI = -0.30 to 0.11) between origins from 6-7 and 8-9 years old. Therefore, we can cautiously suggest that it is unlikely that this bias has significantly inflated our results. In addition, as we have adjusted for the origin of the measurements in all analyses, any potential systematic bias was controlled for.