St. James Ballroom
Analysis of Longitudinal Data Using B-Splines in R and SAS (303911)
Heidi Fischer, Kaiser Permanente Southern California
*Margo A. Sidell, Kaiser Permanente Southern California
Keywords: longitudinal data, marginal models, mixed models, b-splines, time varying data, non-linear data
Time varying data are common in medical research and pose analytical challenges as outcomes may change in a non-linear way. We sought to develop analytical code to handle non-linear trends for use in statistical practice and to visualize results.
Using both R and SAS, we fit covariate adjusted longitudinal models with B-spine functions allowing outcomes to vary non-linearly over time. Knots were placed at data driven boundaries for flexible modeling. We utilized data from two studies. The first consisted of children from 0 to 5 years old, examining the relationship between a 2 or 3 level exposure and BMI. The second followed obese patients with chronic disease, comparing glomerular filtration rate for patients with bariatric surgery versus non-surgical controls.
Codes were written in SAS and R to estimate mean differences in outcomes over time between exposures with 95% confidence intervals (CIs), accounting for correlation within patients. Data visualization methods were developed to aid in communicating complicated model results to non-statistical audiences.