Power Play

Blueprint

Every tool in the lab is documented the same way, from the operational problem to the MLR questions a reviewer will ask.

1. 1

   Operational problem

   Medical Affairs handles statistics every day without a statistician in the room: an investigator-initiated study proposal with an unjustified sample size, a doctor asking what a hazard ratio of 0.75 means for their patients, a diagnostic claim quoted only as sensitivity and specificity.

   The options are expensive statistical software or single-purpose calculators that explain nothing. None shows how a trial behaves before it is run.

2. 2

   Hypothesis

   If one free tool connects design, simulation and interpretation, medical teams will challenge assumptions earlier, explain results in absolute terms more often, and spot fragile or over-optimistic evidence before it reaches a slide.

   Watching a design play out thousands of times builds an intuition for power, uncertainty and the winner's curse that a formula alone does not.

3. 3

   Solution architecture

   • Statistics engine (stats.js, 27 KB): pure functions with no dependencies, shared by the page, the simulation worker and the Node tests.
   • Monte Carlo in a Web Worker, streamed to the page so trials appear as they finish. Seeded xoshiro128** random numbers make every run reproducible.
   • Charts drawn on canvas, with no chart library.
   • Validation: 88 checks against SciPy, statsmodels and lifelines (node tests/validate.js); 21 of them re-run on every visitor's device in the Methods tab.
   • Sharing through the link fragment: inputs travel after the # and never reach a server.
   • Installable PWA, fully offline, strict Content Security Policy.
4. 4

   HCP and patient impact

   HCPs: clearer conversations about trial results in absolute terms (how many patients avoid an event, at what baseline risk), with honest uncertainty.

   Patients: indirect. Better-designed investigator-initiated studies and more accurate communication of benefit and harm.

   No patient-level data is processed at any point.

5. 5

   Regulatory and MLR considerations

   • No data leaves the device; no personal or HCP data is requested.
   • A planning and teaching aid, not validated software for regulatory submissions. Final sample sizes belong to a qualified statistician.
   • Any output used externally (slides, materials) goes through Medical, Legal and Regulatory review.
   • Methods and references are listed in the app, and results are reproducible from the seed.