Baseline outcomes are often underused in design
Cross-sectional SW-CRTs may collect baseline and endline outcomes for each participant. Ignoring baseline outcomes can miss opportunities to improve power and reduce required sample size.
Background
Stepped-wedge cluster randomized trials are widely used in pragmatic and implementation research, but standard design and analysis tools often do not fully use baseline information or clearly distinguish the causal estimand being targeted.
Standard analyses can target ambiguous effects
With unequal cluster-period sizes or heterogeneous treatment effects, participant-average, cluster-average, and period-specific effects can differ. Analyses should state and target the intended estimand.
Subgroup analyses need flexible, transparent tools
Patient-centered research often asks what works best for whom. SW-CRT subgroup analyses require methods that accommodate clustering, time, covariates, and treatment-effect heterogeneity.
Aim 4 translates the methods developed in Aims 1-3 into public-facing software
materials, rendered tutorials, and documented source code for broader dissemination.
Summary of Project Outputs
The GitHub project page centralizes four main products. Together they document what was produced for Aim 4 and how each product supports design or analysis of stepped-wedge cluster randomized trials.
Aim 1 Shiny app tutorial
Documents an interactive design calculator for cross-sectional SW-CRTs with baseline outcome measurements. The tutorial explains the design inputs, ICC parameters, baseline-adjusted models, and power-curve interpretation.
Aim 2 `MRStdLCRT` package tutorial
Documents the R package workflow for model-robust standardization in longitudinal and stepped-wedge CRTs. The tutorial summarizes the estimands, adjusted and unadjusted estimators, and analyses using the package's public example datasets.
Aim 2 `swcrtcov` source code
Provides the `swcrtcov` source code and tutorial for model-assisted ANCOVA estimators in stepped-wedge cluster randomized experiments. The tutorial covers weighted average treatment effect estimands, unadjusted and ANCOVA-adjusted estimators, design-based plug-in variance, and cluster-robust standard errors.
Aim 3 annotated R code
Provides annotated R functions for calibrated Bayesian analysis with mixed-effects BART. The workflow supports overall and subgroup treatment-effect estimation, g-computation, model-robust standardization, and cluster-bootstrap calibration.
Rendered Materials
The rendered tutorials and source materials are collected below for the Aim 4 final report.