Module 01 · IVPT · FDA draft guidance, Oct 2022
Size your FDA IVPT pivotal study from pilot data — defensibly, not by guess.
Import the pilot, confirm the data, and get a recommended donors × sections design with its Joint Passing Rate under the FDA procedure, a sensitivity table across your assumptions, and a Justification Document your statistical reviewer can sign off on.
GuidanceStat is a qualified study-planning application designed for regulated pharmaceutical use, supplied with a Supplier / QA Evidence Pack to support your own risk-based computer system validation.
Built with a CRO running IVPT studies today. Study data never leaves your browser.
The gap
The guidance tells you how the pivotal will be judged. Not how big it has to be.
The FDA IVPT draft guidance publishes the SAS and R code for the pivotal acceptance test. It does not prescribe a sample-size method. The applicant remains responsible for adequate power.
Under the scaled criterion the acceptance limits are themselves random, and two endpoints must both pass. There is no closed form. The honest answer requires simulating the whole procedure — and showing how it moves when the pilot estimates are off.
So the question lands on a biostatistician, who is scarce, or on a formulation scientist with a deadline and a guess. A pivotal that is underpowered, or over-provisioned “to be safe”, is paid for twice: once in donors and sections, once in review cycles.
In FDA’s own review of ANDA submissions presented at GRx+Biosims 2024, only about 40% were approved on the second cycle, with bioequivalence among the disciplines generating the most major deficiencies. FDA and the Center for Research on Complex Generics have run dedicated workshops on recurring issues in IVRT/IVPT studies.
The workflow
Enter through the regulatory job, not the statistical method
You never choose a test, a model or a distribution. GuidanceStat applies the guidance, shows what it fixes and with what force, isolates the few decisions that are yours, and runs the statistics underneath.
Import the pilot dataset
CSV or spreadsheet, one row per skin section: treatment (T/R), donor, section, Maximum Flux, Cumulative Amount. Nothing is uploaded — the file is read inside your browser.
Confirm the data interpretation
Every detected column mapping, donor count, excluded section and unit is shown for explicit confirmation. No calculation runs before you confirm. Ambiguities are surfaced, never guessed.
See what the guidance fixes, decide what it leaves to you
Acceptance limits, the scaled-criterion threshold and the confidence level are preset from the guidance, read-only, each with its citation and provenance class. You decide the assumed Test/Reference ratio scenarios, the variance factor, the target Joint Passing Rate, donor and section ranges, expected integrity loss and optional unit costs.
Run the design grid
Monte Carlo over donors × sections × scenarios, seeded and deterministic. Each candidate design gets a Joint Passing Rate — the fraction of simulated studies where both endpoints pass under the FDA procedure.
Read the decision screen
Recommended design, why it was selected, what is limiting it (precision, variability or the assumed effect), the passing-rate matrix, failure decomposition by regime, a sensitivity table across scenarios, alternatives and the cost frontier.
Export the Justification Document and the Project File
A printable, DOCX-exportable document in English carrying guidance version, engine release, dataset hash, variance components with their origin, every decision, seed and the recommendation. The Project File reproduces the result exactly on any machine.
Real output
On the guidance’s own published example
Pilot: the six-donor balanced example published in Appendix II of the FDA IVPT draft guidance. Scenario: assumed Test/Reference ratio 1.00, variance factor 1.00, target Joint Passing Rate 90%. Engine release 0.3.0, seed 20260904, 2,000 repetitions per candidate design, 228 candidate designs computed in 3.4 seconds.
Recommended design
12 donors × 6 sections
Joint Passing Rate 91.9% against a 90% target. Fewest donors, then fewest sections, since no unit costs were supplied.
Next alternative meeting target: 13 × 6 — 93.3%.
At 20% expected integrity loss: 8 dosed sections per product per donor to provision, plus one non-dosed control — 17 per donor.
These are the application’s actual outputs for the published example under the scenario stated above, not an illustration. Your pilot, your scenarios and your ranges produce your own grid.
What would your pilot recommend? Bring it to a session and find out in 45 minutes.
Deliverables
What you leave the session with
Six artefacts. The document is the deliverable; the rest exists to make it defensible.
Recommended Design
Donors and skin sections per donor that meet your target Joint Passing Rate — lowest cost when unit costs are supplied, otherwise fewest donors, then fewest sections.
Design Limitation
Whether adding donors will actually reduce the risk — or whether the limiting factor is within-donor variability, or the assumed ratio itself. Stated as an action, not as statistics.
Sensitivity Table
Donors required across assumed-ratio and variance-factor scenarios, so the protocol presents a range with its assumptions rather than a single point.
Justification Document
The regulatory deliverable: method, presets with citations, decisions, variance components labelled estimated or reviewer-override, seed, repetitions, results. HTML/print and DOCX.
Project File
Dataset, its SHA-256 hash, decisions, guidance pack version, engine release and seed in one JSON file you own. Reopen it anywhere and every value is recomputed and compared.
Release Verification Report
Engine release, build hash, guidance pack versions, supported environment and qualification corpus result — the evidence your QA lead asks for before allowing use.
Qualification
Built for the way regulated work is reviewed
A number that cannot be reproduced, or whose provenance cannot be shown, is not a recommendation. Everything below is a property of the application, not a claim about your validation status.
Deterministic and reproducible
Same dataset, same decisions, same guidance pack, same engine release, same seed — same result, on any machine in the supported environment. An engine release is immutable; a correction ships as a new release, and no study is silently recalculated.
Qualified against the guidance’s own worked example
The evaluator reproduces every published value of the FDA IVPT draft guidance Appendix II example, balanced and unbalanced, to within 1e-5. Quantile functions are checked against an independent reference to 1e-9. The qualification corpus runs on every release.
Study data never leaves your browser
No upload, no server-side computation, no telemetry. Network traffic is limited to authentication and application assets. Client-confidential pilot data stays where it is.
Decision support, not a system of record
The application recommends. It never approves or signs. Your statistical reviewer can override variance components, select an alternative design with a written rationale, and the document records both. GAMP categorisation and electronic-record applicability remain your quality unit’s determination.
Who it is for
CROs and testing laboratories
Buys capacity
Study volume grows faster than biostatistics headcount. Turn a pilot into a documented pivotal design in one session, and reserve specialist hours for the cases that need them.
Generic and topical pharma
Buys defensibility
A questioned or failed pivotal costs a submission cycle. Regulatory affairs and development get a Justification Document that shows every assumption, its source and its sensitivity.
Biostatisticians
Buys leverage
Inspect variance components, model form and evaluator intermediates in Statistical Detail; override, re-run, endorse or record a written exception. The routine case stops consuming the expert.
Frequently asked questions
Design partners
Bring a pilot. Leave with a pivotal design.
We are working with a small number of laboratories that run IVPT studies to understand how pivotal design is handled today and what would make this worth adopting. A 45-minute facilitated session, your data or the guidance’s example, no charge, no obligation. You keep the Justification Document and the Project File. The roadmap is decided by design-partner evidence, not by a feature list: Dissolution and IVRT are the modules under consideration for 2027.
Defensible statistical design, from protocol to dossier. IVPT module available; Dissolution and IVRT on the 2027 roadmap.