EffTrial

Web platform that ingests FDA histories and your interim data to generate adaptive, Bayesian, regulator-ready trials -- fewer patients, lower cost, and exportable, auditable approval strategy.

Comment

Turn on subtitles for the best demo experience

Discover the full proposal with all the details on Google Drive

Search Your Data, Chain LLM Calls, Build a Multi-Step Flow

Inspiration

Clinical development is slow, risky, and expensive. Under-powered trials fail; over-powered trials waste patients and money. Regulators now accept adaptive and Bayesian approaches, but most teams lack a simple, auditable way to use them. EffTrial aims to make rigorous, regulator-ready design accessible in the browser.

What it does

  • Ingests evidence: Upload FDA reports (PDF/CSV). The system pulls out study arms, endpoints, and adverse events into tidy tables.
  • Adds your data: Optionally upload interim outcomes (e.g., a small CSV). These combine with public evidence to set data-driven priors.
  • Simulates designs: Estimate power, error rates, expected sample size, time to decision, approval odds, total cost, and expected profit/NPV.
  • Recommends a plan: Outputs a clear two-stage design with early-stop rules, plus one-click exports (PDF/CSV/JSON) for review.
  • Shows trade-offs: A clean dashboard puts science and business metrics on one screen so teams decide faster with fewer patients.

How we built it

  • Web platform + cloud engine: 100% browser-based front end with a cloud design engine that runs simulations.
  • LLM-assisted extraction: Large-language models help parse trial descriptions and FDA PDFs/CSVs into structured fields.
  • TiDB backbone: Extracted fields, priors, interim aggregates, and simulation traces are stored in TiDB Cloud for fast queries, versioning, and audits.
  • Practical delivery: One-link sharing, one-click exports, and smooth hand-offs to Python workflows.

Challenges we ran into

  • Turning messy FDA documents into clean, reusable tables without manual re-typing.
  • Keeping priors, assumptions, and early-stop rules transparent and easy to audit.
  • Presenting statistical and business outcomes together in a way teams can read at a glance.
  • Making the whole flow work in a browser while keeping data private and secure.

Accomplishments that we're proud of

  • A pure web experience that goes from uploaded evidence to an auditable two-stage design quickly.
  • Transparent reasoning: Priors and assumptions are shown next to results with printable boundaries.
  • One-screen clarity: Approval odds, patients needed, timelines, cost, and expected profit/NPV in one place.
  • Reusable data: TiDB-backed storage creates a clean dataset for reproducible audits and future ML.

What we learned

  • Accessibility and auditability are as important as algorithms—clear, shared views speed up internal and regulatory reviews.
  • Storing extracted fields and simulation traces unlocks reuse across programs and future machine-learning improvements.
  • Simple inputs and outputs help teams adopt adaptive/Bayesian methods with confidence.

What's next for EffTrial

  • Pilot rollouts: Start with sponsors and CROs in Singapore/SEA, then expand to the U.S./EU.
  • APIs & libraries: Expose ingestion, priors, and simulation endpoints; add indication-specific templates.
  • Enterprise options: Offer compliance-focused, private-deployment packages alongside SaaS.
  • Learning loop: Keep enriching the TiDB dataset to improve priors, cost/enrollment forecasts, and design ranking over time.

TiDB Cloud account Email: hujiajun19961024@gmail.com

Built With

Share this project:

Updates