Presentation (CTI)

OBJECTIVES: Repeated measures are common in clinical trials for assessing treatment effects over time. MMRM, particularly linear mixed-effects models, effectively handle such data, incorporating treatment-time interactions and managing missing data more flexibly than traditional methods. While various open-source packages exist for MMRM analysis, an integrated interface offering a standardized solution was lacking. This study aimed to develop an open-source R-Shiny application for non-programmers to perform Frequentist and Bayesian MMRM analyses in a more streamlined and standardized manner.
METHODS: The R Shiny application, designed in a modular structure using the “shinydashboard” package for frontend and “brms.mmrm” and “nlme” package for backend analysis. The application directly accepts data from ADAM datasets used in clinical trials, ensuring compatibility with standard clinical data formats. For data security, no data is stored on the cloud server, and all uploaded data is erased from the platform at the end of each session.
RESULTS: An R Shiny tool was developed and deployed on Amazon Web Services (AWS) with Auth0 authentication for user security, offering a user-friendly interface for MMRM analyses. Integrating the “nlme” and “brms.mmrm” packages, it supports repeated measures and mixed-effects models, with features for handling correlated data, random effects, and model diagnostics. Users can customize covariance structures, interaction terms, and model criteria, with interactive visualizations and diagnostic plots aiding model assessment. Results are exportable in multiple formats i.e., Excel, Word, and PDF.
CONCLUSIONS: The developed R Shiny application successfully bridges the gap in MMRM analyses by providing a standardized, user-friendly, and secure platform for both Frequentist and Bayesian modeling. Future updates will include additional covariance structures, support for non-linear models, multi-arm trial analyses, enhanced visualizations, integration with CDISC standards and machine learning for model selection.