SMART trial analysis pipeline
Estimation, inference, and interpretation for the BEST chronic low back pain trial.
A targets-orchestrated R pipeline for the Biomarkers for Evaluating Spine Treatments (BEST) trial (registration ID NCT05396014), covering the full path from raw trial data to an interpretable report.
What it does
- Estimates dynamic treatment regimes via Q-learning and super-learner ensembles.
- Performs CV-TMLE inference for the resulting regimes.
- Produces variable-importance summaries and interpretable models alongside the primary estimates.
- Renders a fully reproducible report, re-run end to end on any change to the upstream data or code.
Stack — R, targets, Quarto, tidymodels.
Developed with the UNC Gillings Center for Artificial Intelligence and Public Health (CAIPH) and the UNC Collaborative Studies Coordinating Center (CSCC).