Methods for analyzing longitudinal data from randomized pretest-posttest-follow-up trials in behavioral research: a practical guide to latent change models.

Abstract

Randomized pretest, posttest, follow-up (RPPF) designs are widely used in longitudinal behavioral intervention research to evaluate the efficacy of treatments over time. These designs typically involve random assignment of participants to treatment and control conditions, with assessments conducted at baseline, immediately post-intervention, and during the follow-up period. Researchers primarily focus on determining whether the intervention is more effective than the control condition at post-treatment and whether these effects are sustained or change over time. This paper presents Latent Change Models (LCMs) as a practical approach for analyzing randomized pretest-posttest-follow-up (RPPF) trials, directly estimating discrete changes between timepoints and intervention-control group differences. The utility of LCMs is demonstrated through an application to the STAR (Supporting Treatment Adherence Regimens) trial, a pediatric randomized behavioral clinical trial aimed at improving adherence to anti-seizure medications (ASMs) among children with new-onset epilepsy. The results of the trial analyzed via an LCM are contrasted with the results as analyzed by an ANCOVA, a longitudinal linear mixed-effects model, and a latent growth curve model. The tutorial and application to the STAR trial demonstrate that LCMs offer notable strengths, including the ability to estimate discrete changes over time, control for baseline variability in the outcome, and incorporate all longitudinal data within a single, parsimonious model. These models provide an accurate and nuanced understanding of intervention effects in RPPF designs, with implications for clinical intervention research.