Latent class growth modelling for the evaluation of intervention outcomes: example from a physical activity intervention

Abstract Intervention studies often assume that changes in an outcome are homogenous across the population, however this assumption might not always hold. This article describes how latent class growth modelling (LCGM) can be performed in intervention studies, using an empirical example, and discusses the challenges and potential implications of this method. The analysis included 110 young adults with mobility disability that had participated in a parallel randomized controlled trial and received either a mobile app program (n = 55) or a supervised health program (n = 55) for 12 weeks. The primary outcome was accelerometer measured moderate to vigorous physical activity (MVPA) levels in min/day assessed at baseline, 6 weeks, 12 weeks, and 1-year post intervention. The mean change of MVPA from baseline to 1-year was estimated using paired t-test. LCGM was performed to determine the trajectories of MVPA. Logistic regression models were used to identify potential predictors of trajectories. There was no significant difference between baseline and 1-year MVPA levels (4.8 min/day, 95% CI: −1.4, 10.9). Four MVPA trajectories, ‘Normal/Decrease’, ‘Normal/Increase’, ‘Normal/Rapid increase’, and ‘High/Increase’, were identified through LCGM. Individuals with younger age and higher baseline MVPA were more likely to have increasing trajectories of MVPA. LCGM uncovered hidden trajectories of physical activity that were not represented by the average pattern. This approach could provide significant insights when included in intervention studies. For higher accuracy it is recommended to include larger sample sizes.