Related Experiment Videos
Assessing response profiles from incomplete longitudinal clinical trial data under regulatory considerations
Craig H Mallinckrodt1, S W Scott Clark, Raymond J Carroll
1Eli Lilly & Co., Lilly Corporate Center, Indianapolis, Indiana 46285, USA. mallinckrodt_Craig@Lilly.com
Journal of Biopharmaceutical Statistics
|May 6, 2003
Summary
Likelihood-based mixed-effects model repeated measures (MMRM) offer better statistical control for clinical trials than traditional last observation carried forward (LOCF) methods. MMRM improves analysis of longitudinal data with missing values, enhancing treatment effect evaluation.
Area of Science:
- Biostatistics
- Clinical Trial Methodology
- Longitudinal Data Analysis
Background:
- Longitudinal data analysis is crucial for evaluating treatment effects over time.
- Missing data in clinical trials pose significant analytical challenges.
- Last Observation Carried Forward (LOCF) has been a traditional but limited approach for handling missing data.
Purpose of the Study:
- To review statistical methodologies for handling missing data in longitudinal studies.
- To compare different approaches for modeling time effects and correlations in repeated measures.
- To propose and justify the use of mixed-effects models for analyzing acute phase clinical trial data.
Main Methods:
- Comparison of statistical approaches for missing data imputation.
- Evaluation of methods for modeling time trends and within-subject correlations.
- Application of likelihood-based mixed-effects model repeated measures (MMRM) under a missing at random assumption.
Main Results:
- MMRM approaches demonstrate superior control of Type I and Type II errors compared to LOCF.
- LOCF relies on a more restrictive missing completely at random assumption.
- MMRM provides a more robust analysis for longitudinal data in regulatory settings.
Conclusions:
- Likelihood-based MMRM is recommended for analyzing longitudinal data in acute phase clinical trials.
- MMRM offers improved statistical validity over traditional LOCF methods.
- Unstructured modeling of time trends and correlations may be advantageous in specific trial designs.