Background: Power calculations for research designs are critically dependent on estimates of variability, but there are little empirical longitudinal dta on the within-individual variability of various standardized somatometric measures for children and adolescents. Sources of the longitudinal-within- individual variability could be genuine longitudinal differences and/or bias/measurement errors. Large within-individual variability can obscure true treatment effects of therapeutic or preventive interventions for childhood obesity research and reversely, ineffective interventions could spuriously appear to be effective. Studies should have appropriate sample sizes to account for this variability and somatometric measures that are less sensitive to this variability should be selected as study endpoints. Our hypothesis i that a significant percentage of children/adolescents in a pragmatic primary care setting will have large longitudinal-within-individual variability. Design: Retrospective cohort study using electronic health record data (EHR) from relatively healthy children and adolescents in a primary-care- organization over the past 14 years.
Specific aim 1 a): Explore the size of the longitudinal within- individual variability of standardized-somatometric-measures. The variability-metrics we will probe are the slopes of the regression lines of serial-standardized-measures vs. age (slope`0 indicates deviation from original growth-trajectory) and the root-mean-square-error (RMSE) of the standardized-measures (using piecewise-linear-regression models). We will study variability-metrics for weight and BMI-for-age z-scores, percentiles and centiles and explore the correlation of those variability-metrics between those different somatometric-measures. We will estimate the percentage of children with large variability. We will explore the changes in slope of standardized-measures across different age-subgroup-categories. 1b) Generate reference standards of variability-metrics for different standardized-somatometric- measures (according to age/gender /race/ethnicity/socioeconomic status). 1c) Explore how much of the longitudinal within-individual variability could be explained by certain individual-specific or setting- specific factors, using Monte Carlo simulations. Our cohort, with weights for ~231000 individuals and with e2 weights in each of the following three age-subgroups (2-5/6-12/13-20 years) for ~65000 individuals, is well powered to model and explore this within-individual-variability. Potential impact: The generation of reference-standards for variability-metrics would fill an important research gap in the design appropriately powered studies. Identification of somatometric-measures that are less sensitive to within-individual-variability would lead to selection of better study endpoints that reflect the true growth-trajectories and capture the impact of certain interventions/factors in comparative effectiveness research and association-epidemiologic studies using routinely collected EHR data.
We want to explore the size of the longitudinal within-individual variability of standardized weight and BMI measurements of children and adolescents seen in a pragmatic primary care setting. Our findings will have implications in the design of studies in the field of childhood obesity research using routinely collected electronic health record data. Large variability could cause true effects, of effective interventions, to completely disappear and reversely, ineffective interventions could spuriously seem to be effective if this variability is not accounted for in the study design.
