Optimal Partitioning for Linear Mixed Effects Models: Applications to Identifying Placebo Responders

Area of Science:

• Clinical Research
• Biostatistics
• Psychopharmacology

Background:

• Distinguishing specific drug effects from non-specific placebo effects in clinical research is a persistent challenge.
• Linear mixed-effects models are standard for analyzing longitudinal data in clinical trials.

Purpose of the Study:

• To present an optimal partitioning methodology for linear mixed-effects models to identify placebo responders.
• To compare this new partitioning strategy with existing growth mixture modeling approaches.

Main Methods:

• The optimal partitioning methodology generates prototypical outcome profiles from longitudinal data.
• This approach accommodates both continuous and discrete covariates.
• Applied to a two-phase depression trial involving fluoxetine treatment and a placebo discontinuation phase.

Main Results:

• The optimal partitioning methodology successfully identified distinct prototypical outcome profiles in the first phase of the depression trial.
• Survival analysis on partitioned data from the second phase revealed differences in relapse rates based on continued fluoxetine treatment versus placebo.

Conclusions:

• The proposed optimal partitioning methodology effectively differentiates patient responses attributable to specific drug effects versus placebo effects.
• This method aids in understanding treatment response heterogeneity and can inform clinical trial design and analysis, particularly in distinguishing true drug efficacy.