Using Multilevel Mixtures to Evaluate Intervention Effects in Group Randomized Trials.

Area of Science:

• Biostatistics
• Psychology
• Public Health

Background:

• Behavioral intervention effectiveness can vary significantly across different individuals.
• Current methods for evaluating these differential outcomes are underdeveloped.
• Understanding subgroup impacts is crucial for optimizing intervention strategies.

Purpose of the Study:

• To propose finite mixture models for evaluating intervention effects in specific participant subgroups.
• To assess whether interventions, particularly in group randomized trials, yield differential impacts based on participant characteristics or problem behavior levels.
• To demonstrate the application of these models using a preventive intervention targeting problem behaviors.

Main Methods:

• Utilized finite mixture models with latent classes derived from clustering individuals based on targeted behaviors.
• Employed a multilevel mixture model incorporating random effects to analyze intervention impacts.
• Illustrated model application through an example evaluating a preventive intervention's effect on illicit substance users versus other problematic substance users.

Main Results:

• Demonstrated the identification and independent validation of latent classes within the population.
• Showcased the specification of random effects in multilevel mixture models.
• Illustrated the estimation of statistical power for detecting intervention effects within identified latent classes.

Conclusions:

• Finite mixture models provide a robust framework for evaluating heterogeneous intervention effects.
• The proposed methodology allows for nuanced understanding of how interventions impact distinct subgroups.
• This approach can be broadly applied to assess the effectiveness of various interventions across diverse populations.