Modelling placebo response in depression trials using a longitudinal model with informative dropout

Area of Science:

• Clinical Psychology
• Psychiatric Research
• Biostatistics

Background:

• Longitudinal studies in depression frequently encounter participant dropouts.
• Failure to account for missing data can introduce bias and inconsistency in study outcomes.
• Existing models for Hamilton Depression Rating Scale (HAMD-17) scores do not fully address dropout mechanisms.

Purpose of the Study:

• To characterize placebo response in depression trials considering dropouts.
• To identify the most appropriate dropout mechanism for time-varying dropout probabilities.
• To establish a framework for clinical trial simulations in depression research.

Main Methods:

• A meta-analytic approach was applied to placebo data from 6 clinical trials (n=695) involving Major Depressive Disorders patients.
• A non-linear model was extended to jointly estimate HAMD-17 time-course and dropout mechanisms.
• Hazard models were used to evaluate 'Missing Completely At Random', 'Missing At Random', and 'Missing Not At Random' hypotheses.

Main Results:

• The 'Missing Not At Random' model demonstrated superior statistical performance (p<0.01) compared to 'Missing At Random'.
• 'Missing At Random' also outperformed 'Missing Completely At Random' (p<0.01).
• These findings suggest that dropout mechanisms significantly influence the analysis of longitudinal depression trial data.

Conclusions:

• Dropout mechanisms in depression clinical trials are not random and significantly impact results.
• The 'Missing Not At Random' model provides a more accurate approach for analyzing longitudinal depression data.
• This research offers critical insights for improving the validity of analyses supporting new antidepressant drug registration.