The CSA context-of-use statement for the DOE Study Designer pre-protocol design aid — what it does, what it explicitly does not do, the assurance its risk tier warrants, and the human review that stands between its output and use.
DRAFT · for qualified reviewA browser-based design aid that assembles a proposed DOE structure and a pre-protocol draft document for blending and encapsulation development studies, from parameters a qualified user enters. Technology class: deterministic JavaScript. The labelled “AI pre-screen” is advisory prioritisation of candidate factors and ranges only. GxP process touched: pharmaceutical development (ICH Q8), upstream of any GMP record.
Generate, hold, or transmit a GxP record; constitute a controlled document; perform statistical analysis or compute a design space, PAR, or control-strategy conclusion; substitute for validated statistical software; qualify equipment or materials; apply an electronic signature or maintain a Part 11 audit trail; determine batch disposition. Every output is a draft requiring independent review, completion, and approval in the client’s quality system, which remains the system of record. All computation is client-side; no study data leaves the browser.
Patient safety: none direct — no output reaches product or patient without a qualified human authoring, reviewing and approving a protocol in the eQMS. Product quality: indirect but non-trivial — a defective design could under-power a study whose conclusions inform PAR and Module 3 P.2 content. Data integrity: the material axis — the tool emits a document that will be filed.
Tier: Moderate Higher than a training or rehearsal aid because the artifact is destined for a controlled repository and underpins development claims; below full CSV because the tool is deterministic, makes no GxP decision, holds no record, and every error is detectable at the mandatory human review and approval step that stands between output and use.
Decision-maker of record: the protocol author [real human, named at deployment], with independent statistical review and QA approval per the client’s SOP. Checkpoints: (a) author reviews every generated section and completes all fillable fields; (b) statistical reviewer confirms the design and power basis against validated software; (c) QA approves and issues under document control, at which point the client’s controlled version — not the tool’s output — becomes the record.
Design matrices reproduce reference construction for the stated family and factor count; randomisation reproduces exactly from the recorded seeds; feasibility calculations reproduce from stated formulae and inputs; the input-set digest is stable across regeneration with identical inputs. Quantitative thresholds and the reference test set: to be defined by the deploying organisation.
The tool is not the record. It supports ALCOA+ in the receiving system by stamping every output with tool version, UTC generation time, and an input-set SHA-256, and by printing the randomisation seeds and algorithm so run order is reconstructible. Attributable/Contemporaneous are supplied by the eQMS on filing, not by the tool. Local browser storage is working state, not a record, and is not a retention mechanism. Limitation to state plainly: the tool keeps no audit trail of edits made before export.
Version is pinned and printed on every output. Re-qualification triggers: any change to design construction, randomisation, power/MDE, or feasibility logic; any change to protocol template content with regulatory meaning; any change to the AI pre-screen. Deterministic code, so no model drift; if the pre-screen becomes a generative component, this rationale requires reassessment before release.
Unscripted/exploratory testing with documented evidence, plus scripted verification of the computational core (design construction, randomisation reproducibility, feasibility formulae, digest stability). Proportional to Moderate tier: the failure mode of concern is a document defect, and the mandatory human authoring-review-approval sequence provides the detectability that justifies assurance below full CSV.
FDA CSA for Production and Quality System Software; 21 CFR Part 11; 21 CFR 211; ICH Q8(R2)/Q9(R1)/Q10; GAMP 5 2nd ed. and the ISPE GAMP AI Guide; ALCOA+; EU GMP Annex 11 and 15.
EU GMP Annex 22 (AI in GMP) is in DRAFT, is not in force, and is expected to be finalised in 2026; it is expected to permit static, deterministic models and to exclude generative AI. It is tracked as a watch item, not an applicable requirement. Re-verify CSA and Annex 22 status against the primary source before external issue.
CSA Context-of-Use Statement · DOE Study Designer v1.8 (pre-protocol design aid) · draft v1.0 · 2026-07-19 · aligned to FDA CSA / GAMP 5 (2nd ed.) risk-based principles
Draft for qualified review. This statement accompanies the DOE Study Designer available in the PharmTech.AI account area. It is illustrative until reviewed and accepted by the deploying organisation’s quality unit; content is pending SME and customer sign-off. See also the Mock-FDA-Inspection validation rationale for that separate product’s statement.
PharmTech.AI is an independent service of Aligned Executive Solutions, Inc. and is not affiliated with, endorsed by, or sponsored by the U.S. Food and Drug Administration. © 2026 Aligned Executive Solutions, Inc. · PharmTech.AI · info@pharmtech.ai