Services › Trial Design & Evaluation
Trial Design Review
A three-week stress test of a planned cluster-randomized, stepped-wedge, or ring-vaccination trial for interference, contamination, and identifiability. Delivered before data collection, so the design choices that matter most are still on the table.
You have a trial protocol in hand and standard methods don't quite fit.
Cluster-randomized, stepped-wedge, and ring-vaccination trials in HIV, TB, malaria, and emerging-pathogen contexts almost always violate the standard no-interference assumption. The Trial Design Review is for sponsors and PIs who would rather know now than litigate it in peer review.
What you get
- A structured methods memo (~15–25 pages). Threats to identifiability, interference and spillover concerns, design alternatives, and a recommended analysis strategy — with citations.
- A 90-minute working session with the PI and trial statistician to translate the memo into concrete decisions before the protocol locks.
- Optional analysis-plan co-authorship for engagements that continue past the review.
How the engagement unfolds
Protocol read & scoping
Read the protocol, statistical analysis plan, and any pilot data. One scoping call with the PI and trial statistician. Issue a one-page memo identifying the three to five issues that will drive the rest of the review.
Methods stress test
Formal review of design and analysis under interference. Where appropriate, simulation under plausible network and spillover structures. Identify which estimands are recoverable, which are not, and which design tweaks restore identifiability.
Memo & working session
Deliver the written memo. Run the 90-minute working session with the trial team. Document the decisions the team commits to. Optionally, a follow-on SOW for the full analysis plan.
What this engagement will not do
This is not a validation exercise. The deliverable is whatever the analysis honestly produces. If the recommendation is "do not run the trial as designed," that is what the memo will say. See Operating Principles.
Causal inference under interference is not a side interest.
- Epidemiologic Methods — foundational paper on causal inference under interference (2016)
- NIH R01 PI · $2M, network-informed HIV prevention trial design
- CRAN: treatSens — sensitivity analysis for unmeasured confounding
- Statistics in Medicine — contact network estimation in infectious disease with Goyal et al. (2023)
- Three-time Atlantic Causal Inference Conference data-analysis competition winner
- FDA-contracted Targeted Learning webinar on causal inference with BART