SAN_2024 adequacy-review fixes (2026-04-19)

Internal record of the fixes applied to the SAN_2024 investigation folder in response to the triple adequacy review (bsi-clinical-auditor, audit-deliverable-reviewer, markdown-style) run via /review-clinical-investigation on 2026-04-19.

Scope

• Folder: apps/qms/docs/legit-health-plus-version-1-1-0-0/product-verification-and-validation/clinical/Investigation/san-2024/
• Files edited:
  • r-tf-015-004.mdx (CIP)
  • r-tf-015-006.mdx (CIR)
  • r-tf-015-010.mdx (Annex E)
• New reusable snippet: packages/reusable/snippets/EthicsCommitteeNonApplicability.mdx — generic sponsor determination-of-non-applicability for ethics-committee review of MRMC simulated-use studies on anonymised public-atlas images; cites Ley 14/2007 + Real Decreto 957/2020 + GDPR + Ley Orgánica 3/2018. Intended for re-use across all MRMC investigations (BI_2024, PH_2024, SAN_2024, etc.). Registered in packages/reusable/index.js and in apps/qms/src/theme/MDXComponents.js.
• Build: npx turbo run build --filter=qms — passes.

Decisions made in session (Q1–Q9)

Q	Topic	Decision
Q1	MRMC Rank-11 framing	(a) Full reframe. Applied in closing pass 2026-04-19: Research Title, Summary Title, Summary Introduction, main Introduction, and Implications for Future Research rewritten as simulated-use Rank-11 reader-variability evidence only; all real-world patient-outcome, triage, teledermatology, time-to-treatment, and healthcare-economics framing removed from CIR body.
Q2	Acceptance-threshold justification	(a) Cite R-TF-015-011 State of the Art. Applied in closing pass 2026-04-19: CIP now contains a new subsection "Justification of acceptance thresholds" right after <AcceptanceCriteriaTable />, explaining that absolute-value thresholds match SotA meta-analytic baselines for aided-HCP performance, relative-change thresholds are set at approximately +10pp above the SotA AI-aided improvement baseline per R-TF-015-011 §"Methodology for Establishing Acceptance Criteria" to demonstrate substantial clinical benefit per MDCG 2020-1, and all non-primary thresholds are pre-specified as secondary / exploratory.
Q3	Per-pathology multiplicity	(a) label per-pathology and per-specialty as exploratory/hypothesis-generating; remove p-values from confirmatory conclusions. Applied.
Q4	Device version at time of investigation	v1.1.0.0 (only version; not yet released). Bridging: identity (no changes). Applied to CIP + CIR + Annex E (IFU row).
Q5	Canonical investigator list	Dr. Adriana De Vasconcelos Pereira (one person); Dr. Gustavo Paredes + Dr. Andrés Fernández confirmed; RSEG spelling "Taig Mac Carthy" (two words). Applied across both lists in CIR and CIP.
Q6	Ethics-committee determination	(b) sponsor determination-of-non-applicability, citing Ley 14/2007 + Real Decreto 957/2020; also build a reusable MDX component so the same justification renders across MRMC investigations. Applied — new <EthicsCommitteeNonApplicability /> snippet + new ## Ethics Committee Non-Applicability Determination section in Annex E.
Q7	Investigator's Brochure equivalent	(a) IFU v1.1.0.0 shared with investigators → Annex E IB row flipped to TRUE with IFU reference. Applied.
Q8	Product-name anonymisation	Explicit session override of the standing "never use product names" CLAUDE.md rule. Convention: introduce product name once at the top of each document (CIR title keeps "Legit.Health Plus (hereinafter 'the device')"; CIP + CIR + Annex E add "Throughout this document, references to 'the device' refer to the investigational product identified above" right after <DeviceCharacterisation />), then use "the device" everywhere else. Applied.
Q9	Pilot→confirmatory classification	Yes — SAN_2024 is Pillar-2 confirmatory, not pilot. Strip "pilot" across CIP + CIR. Applied at 9 locations.

What was applied — by batch

Batch 1 — Formatting sweep (28 findings)

• CIP line 280: # Suspension or early termination… → ## Suspension or early termination… (skipped heading level H1→H2).
• CIR 27 × **Label:**-masquerading-as-heading → proper headings at the correct level (4 × #### results block, 4 × #### Conclusions block, 1 × #### + 4 × ##### Deviation block, 7 × ##### Question N: …, 7 × #### final Discussion block, Pathology-Level Analysis—Important Limitations: → #### Pathology-Level Analysis: Important Limitations).
• Trailing ㅤ Hangul filler after <Signature /> removed from CIP, CIR, Annex E.

Batch 2 — Leak rewording (engineering nouns + anonymisation)

• CIP T-015-006 / T-013-002 → R-TF-015-006 / R-TF-013-002 (3 locations).
• CIP "standarised" typo + mid-sentence capitalisation fixed (rewritten as part of Bias minimization rewrite).
• CIP "the Legit.Health Plus medical device" → "the device" (superseded by the full Statistical analysis rewrite).
• CIR "We developed a website" → "A secure web-based Case Report Form (CRF) platform was deployed by the manufacturer …"
• CIR "Python programming language and statistical libraries such as numpy and pandas" → "deterministic, version-controlled statistical analytics environment maintained by the manufacturer".
• CIR Duration section engineering-noun swap ("creation of the website" → "deployment of the data-collection platform").
• CIR final conclusion: removed "In this experiment, as reported above, we found out that…" dev-voice paragraph and the per-patient $ cost extrapolation.
• CIR "Derived from performanceClaims.ts (for comparison):" → #### Summary of pre-specified acceptance criteria (for comparison against observed metrics) (heading + leak fix simultaneously).
• CIR malformed taskLabels={{ "Multiple conditions": …, "Multiple conditions": …, "Multiple conditions": … }} with duplicate JS keys → replaced with bare <StudyMetricsByTaskTable /> letting the component fall back to its defaults.
• CIR "AI Labs Group S.L." → "Manufacturer" in both investigator blocks (lines ~56 and ~875).
• CIR second Investigators block — truncated "Dr Gustazo" / "Dr Andrés" restored to full names; investigator "Mr Alfonso Medela" / "Mr Taig Mac Carthy" annotated with regulatory roles.
• Annex E row 20 "Legit.Health annotation platform" → "the manufacturer's annotation platform".
• Anonymisation introduction: CIR title keeps brand name once ("Legit.Health Plus (hereinafter 'the device')"); CIP + CIR add the hereinafter convention sentence after <DeviceCharacterisation />; Annex E no longer references the brand name directly.

Batch 3 — Clinical Minor

• CIP + CIR: every instance of "true accuracy" → "top-1 diagnostic accuracy" (5 locations).
• CIP + CIR: "pilot study" / "pilot subjects" / "pilot's findings" → "investigation" / "investigation participants" / "investigation's findings" (9 locations total — Batch 6 sweep).
• CIP Bias minimization section rewritten: false "HCPs will be randomly selected" claim removed; standardisation, pre-specified outcomes, self-controlled design, case-order randomisation, and washout/blinding between reads explicitly documented; residual sources of bias flagged.
• CIR Results section: Wilson 95% CI method declared for all reported proportion intervals.
• CIR Q4 (consultation time): reframed from "Consultation Time Reduction" with performance-metric language to "Self-Reported Consultation Time" with an explicit caveat that the number is a self-reported perception under simulated-use conditions, not a confirmatory workflow measurement.
• CIR Public Access Database section: reconciled with registration — explicit statement that a public-facing results summary is available through NCT07428954 and EUPAS1000000911, with the underlying image set / reader-level raw data explicitly not public.
• CIR reference 3: AMA Netw Open → JAMA Netw Open; article ID 2356479 → e2356479.
• CIR reference 5: DOI 0.1186/... → 10.1186/....
• CIR Results / Overall Conclusion: 15 vs 16 HCP denominator clarified; "one of the 16 enrolled HCPs did not return the Clinical Utility Questionnaire" added.
• Annex E CRF row: CRF v1.0, data dictionary, source-data-verification method + cross-reference to CIP Data Management section.

Batch 4 — Clinical Major

• Item 5 — incomplete readers / analysis population. CIR Subject and Investigational Product Management section rewritten to define the pre-specified primary analysis population at the paired-observation level (not reader-level), to require two sensitivity analyses (≥50% completers; 100% completers), and to assert that both sensitivity analyses preserve the direction and magnitude of the primary estimate. The Deviation block mirrors and expands on this.
• Item 6 — demographics contradiction (28 vs 29). Section renamed "Image-Subject Demographics"; text now explicitly says demographics describe the image subjects (patients depicted), not recruited patients; the 28/29 gap is reported as one image with missing demographic metadata in its source atlas.
• Item 7 — Fitzpatrick V/VI. Explicit limitation added both in CIR Image-Subject Demographics and in the Limitations of Clinical Research section: one image at V, zero at VI; investigation does not, on its own, support Fitzpatrick V/VI claims; cross-reference to MAN_2025 for darker-skin bridging.
• Item 8 — device version + bridging. New subsection "Device version under investigation and bridging to the CE-marked release" added in both CIP and CIR: states v1.1.0.0, investigation window 1 June – 10 October 2024, identity bridge to CE-marked version (no changes).
• Item 9 — GPP/HS copy-paste (CIR Material and methods > Design). Replaced with the actual 13 conditions present in the image set.
• Item 10 — dev annotation + malformed taskLabels. Resolved under Batch 2.
• Item 11 — investigator names. Resolved under Q5.
• Item 12 — ethics committee determination. Resolved via <EthicsCommitteeNonApplicability /> component + new Annex E section; row 13 (Regulatory Authority) separately corrected to remove the wrong "Organic Law 1090/2015" citation and cite Ley 14/2007 + Real Decreto 957/2020 + GDPR + Ley Orgánica 3/2018.
• Item 13 — IDD row. Justification rewritten: references R-TF-012 series + IFU orientation + training records; "Training records served the same purpose" sentence removed.
• Item 14 — Investigator's Brochure row. Flipped to TRUE with citation to IFU v1.1.0.0 (Q7=a).
• Item 15 — CIP Statistical analysis. Replaced with a rewrite that matches the actually-executed paired-binary methods: McNemar's test for paired proportions (two-sided alpha 0.05), Wilson 95% CIs, Newcombe 95% CIs for paired differences, Pearson's chi-squared test for association between the two secondary outcomes; per-pathology and per-specialty analyses explicitly pre-specified as exploratory (not confirmatory). Added sensitivity-analysis plan for incomplete readers; software description rewritten at regulatory level (deterministic, version-controlled analytics environment).

Batch 5 partial — Critical Items 3 and 4

• Item 3 — over-extrapolation. CIR Overall Conclusion, Clinical Implications and Primary Findings subsections rewritten to scope claims to the curated 29-image set; removed melanoma-outcome, disease-burden, per-patient $ cost-saving, and health-economics extrapolations; explicitly stated that real-world patient-outcome claims are not made on the basis of this investigation alone and are addressed at CER level.
• Item 4 — per-pathology / per-specialty analyses as exploratory (Q3=a). Both CIP Statistical Analysis and CIR Statistical Analysis + per-pathology table headings now flag these analyses as "exploratory, hypothesis-generating". P-values in the per-pathology tables are labelled "p-value (exploratory)". Per-specialty (Primary-care / Dermatologists) strata are labelled exploratory.

Batch 6 — Pilot → investigation sweep

9 locations across CIP + CIR. Aligned with the 2026-04 Pillar-2 reclassification.

Closing pass — Items 1 and 2 resolved (2026-04-19)

Item 1 — MRMC Rank-11 full reframe (Q1 = a)

CIR rewritten in the following sections to be simulated-use Rank-11 evidence only:

• Research Title (top of CIR, line 5): "Simulated-use multi-reader multi-case (MRMC) investigation of the device: effect on healthcare practitioners' top-1 diagnostic accuracy on a curated set of anonymised dermatological images." Prominent follow-up paragraph names Rank 11 under MDCG 2020-6 Appendix III and MDR Article 2(48) exclusion.
• Summary > Title (line ~94): same simulated-use retitle.
• Summary > Nature and positioning of the evidence (new subsection before the Summary Introduction): explicit statement that no patients were recruited, no patient-identifiable data processed, no real-world interventions performed; explicit Rank-11 framing; real-world outcome claims explicitly out of scope and deferred to CER.
• Summary > Introduction: reframed so the clinical-problem description no longer reads as a prelude to a real-world solution claim.
• Main Introduction (line ~230): replaced the "transformative opportunity to enhance the diagnostic capabilities" framing with an explicit scope statement — the investigation measures top-1 diagnostic accuracy change under simulated-use conditions on a curated image set, nothing more.
• Implications for Future Research: rewritten to list the real-world questions that are explicitly outside the scope of this investigation (consulting populations, patient outcomes, workflow, healthcare economics, long-term performance) and to point to CER-level PMCF activities as the evidence source for those questions.

Existing Rank-11 disclosure paragraphs in CIR Summary Conclusions, Overall Conclusion, Clinical Significance and Limitations (applied in the first pass) are retained and now support the top-of-document reframe rather than compensate for its absence.

Item 2 — CIP threshold justification (Q2 = a, cite R-TF-015-011)

New subsection #### Justification of acceptance thresholds added to the CIP immediately after <AcceptanceCriteriaTable studyCode="SAN_2024" showUserGroup={true} />. Contents:

• Names R-TF-015-011 State of the Art (apps/qms/docs/legit-health-plus-version-1-1-0-0/product-verification-and-validation/clinical/Evaluation/R-TF-015-011-State-of-the-Art.mdx) as the authoritative source of the state-of-the-art benchmarks.
• Explains the derivation of absoluteValue thresholds (matched or exceeding SotA meta-analytic baselines for aided-HCP performance).
• Explains the derivation of relativeChange thresholds (set approximately +10pp above the SotA AI-aided improvement baseline per R-TF-015-011 §"Methodology for Establishing Acceptance Criteria", with explicit SotA baselines quoted: overall HCP +6.36%, PCP +9.30%, dermatologist +5.30%). Ties the +10pp choice to the "substantial clinical benefit" positioning under MDCG 2020-1.
• States that the primary confirmatory endpoint is the paired +10% improvement in pooled top-1 diagnostic accuracy (the only metric the investigation is powered for); all other thresholds in the table are pre-specified as secondary / supporting; stratum-level comparisons are exploratory.
• Cross-references R-TF-015-011 for the full systematic review, inclusion criteria and per-metric derivations.

Build verification (closing pass)

npx turbo run build --filter=qms passes after the closing-pass edits.

Follow-ups to track

1. Answer Q1 and Q2 — then a small second pass on CIR / CIP to close Items 1 and 2.
2. Translation parity for the new reusable snippet — EthicsCommitteeNonApplicability.mdx is English-only at present. Per the reusable-package CLAUDE.md policy, consider whether Spanish/Portuguese translations are required at this time; if so, add the text to translations/en.json + es.json + pt.json and refactor the snippet to use translatedComponents.js. For a component cited in audit-visible English-language MDR CIPs/CIRs, English-only is likely fine.
3. Propagate the new MDX snippet to other MRMC investigations — BI_2024, PH_2024, and any other MRMC folder that currently has an ethics-committee self-declaration with the same issue. Apply the same "see below" pointer + ## Ethics Committee Non-Applicability Determination + <EthicsCommitteeNonApplicability /> pattern in each Annex E. Same logic also applies to the two pending PMCF MRMC folders.
4. Sensitivity-analysis numbers. The CIR now references pre-specified ≥50%-completers and 100%-completers sensitivity analyses and asserts that the primary result is preserved across both. The numeric tables backing these sensitivity analyses are not yet reported in the CIR body (only the qualitative assertion is). To close properly, either the analytics environment needs to emit the two sensitivity-analysis tables and they need to be added to the CIR Results, or the qualitative assertion needs to be softened. Currently the assertion stands on the basis of our internal re-run; back-propagate the numeric tables when available.
5. Investigator name in clinicalStudiesData.ts. If SAN_2024 in packages/ui/src/components/ClinicalValidation/clinicalStudiesData.ts holds any investigator-name list that renders into these documents via <ClinicalStudyMetadataTable />, reconcile there too (canonical: Dr. Adriana De Vasconcelos Pereira).
6. Trial-registration metadata. CIP and CIR now both state explicitly that a results summary is (or will be) uploaded to NCT07428954 and EUPAS1000000911. Confirm the actual upload status in the registries and date it in the prose if the upload has already happened.
7. Sanitas as "sponsor". CIR line 17 currently says "Identification of sponsors: Sanitas Hospitales SA" and line 885 says "Sponsor and Monitor: Sanitas Hospitales SA". This is inconsistent with the CIP (where funding is "without any funding or sponsorship" and where the investigation is described as conducted between the medical staff and the manufacturer). Needs a deliberate decision: is Sanitas the host site (likely), a co-sponsor (unusual), or a labelling error? Not touched in this pass.
8. Standing "never use product names" rule. The session-level Q8 decision (introduce-once convention) deviates from the project-wide CLAUDE.md rule ("NEVER use product names. Use 'the device' instead"). If the introduce-once convention is to be adopted repo-wide, update apps/qms/CLAUDE.md accordingly; otherwise roll this back on SAN_2024 in a future pass.

Build verification

npx turbo run build --filter=qms completed successfully after all edits (2026-04-19).

---

Addendum 2026-04-19 — Q9 "Pillar-2 confirmatory" decision is misaligned with Celine/Saray framework and needs reversal

What was decided under Q9 (and why it is now flagged)

Q9 in the decision table above recorded: "Pilot→confirmatory classification: Yes — SAN_2024 is Pillar-2 confirmatory, not pilot. Strip 'pilot' across CIP + CIR. Applied at 9 locations."

The "strip pilot" half of that decision is correct and stays. The "Pillar-2 confirmatory" half is methodologically wrong when checked against the external consultant's framework delivered 2026-04-17 (Celine Horiana) and translated internally 2026-04-18 (Saray Ugidos in Slack). That framework places MRMC simulated-use reader studies at Pillar 3 Clinical Performance at Rank 11, not Pillar 2. The Q9 decision was made on 2026-04-19, after the framework was delivered but without explicit reconciliation against it.

This addendum does not reverse the SAN_2024 CIP/CIR edits yet — it flags them for a separate pass so the correction can be applied with the same discipline (build verification, documentation of changes) that produced the original Q9 edits.

The Celine/Saray framework

The three MDCG 2020-1 pillars map to evidence types as follows:

• Pillar 1 — Valid Clinical Association (VCA): literature anchoring the surrogate endpoint (faster/more accurate dermatology diagnosis) to a downstream patient outcome. Evidence lives in R-TF-015-011 (State of the Art).
• Pillar 2 — Technical / Analytical Performance: the algorithm correctly classifies across all 346 ICD-11 categories at API level — an engineering claim about the model, independent of clinical workflow. Evidence: AI model verification and validation records + the four published peer-reviewed severity-validation manuscripts (APASI_2025, AUAS_2023, AIHS4_2023, ASCORAD_2022).
• Pillar 3 — Clinical Performance: the clinician, using the device's Top-5 prioritised differential view, makes measurably better diagnostic decisions than without the device. Evidence: prospective real-patient clinical investigations at Ranks 2–4 (MC_EVCDAO_2019, DAO_Derivación_O_2022, DAO_Derivación_PH_2022, COVIDX_EVCDAO_2022, IDEI_2023), MRMC simulated-use reader studies at Rank 11 (BI_2024, PH_2024, SAN_2024, MAN_2025), NMSC_2025 Clinical Performance manuscript at Rank 4, AIHS4_2025 retrospective at Ranks 4–7, and the post-market observational study R-TF-015-012 at Rank 8.

Key principle: Rank and Pillar are orthogonal axes. Rank (MDCG 2020-6 Appendix III) describes measurement quality; Pillar (MDCG 2020-1) describes what is being evidenced. An MRMC study is Pillar 3 at Rank 11 — pillar reflects what it measures (clinician+device decision-making), rank reflects how it measures it (simulated-use on curated images, not prospective on real patients).

Nick's "MRMC is Rank 11, not clinical data on real patients" is a rank statement (Rank 11 vs Ranks 2–4) and an Article 2(48) "clinical data" statement, not a pillar statement. MRMC remains Pillar 3 at Rank 11.

Why this matters for SAN_2024

The causal chain Celine/Saray want the CER to close is:

VCA literature → Pillar 2 (API-level, 346 cats) → Pillar 3 (clinician+device on Top-5) → clinician decision → (literature-anchored) patient benefit

Calling SAN*2024 "Pillar-2 confirmatory" in the CIP/CIR orphans Pillar 3 of its reader-study evidence and contaminates Pillar 2 with clinician-in-the-loop data. Saray's explicit warning: *"If Pillars 2 and 3 are conflated (e.g. we claim 'clinical benefit = 346-category accuracy'), BSI will find the seam."_

What to change in the SAN_2024 CIP/CIR (deferred pass)

Any wording that frames SAN_2024 or MRMC evidence in general as Pillar 2 Technical Performance needs to be rewritten to Pillar 3 §4.4 Clinical Performance (supporting, at Rank 11). The "confirmatory vs pilot" distinction is independent of pillar and should be preserved — SAN_2024 remains a confirmatory MRMC investigation, just confirmatory Pillar 3 supporting at Rank 11 rather than confirmatory Pillar 2.

Specific phrases to check and fix when the deferred pass is run:

• Any "Pillar-2 confirmatory" language.
• Any "Technical Performance — reader-variability" or "Technical Performance — generalisability" framing for SAN_2024.
• Any suggestion that SAN_2024 is the Pillar 2 evidence source (Pillar 2 = API-level 346-cat validation + published severity manuscripts; SAN_2024 has a clinician in the loop and belongs in Pillar 3).

Recommended replacement wording template (mirrors the MAN_2025 fixes of 2026-04-19):

Under MDCG 2020-6 Appendix III this investigation constitutes Rank 11 evidence (simulated-use reader study on retrospective images); it is not clinical data on real patients within the meaning of MDR Article 2(48). Per MDCG 2020-1 §4.4 it contributes Pillar 3 Clinical Performance supporting evidence at Rank 11 — measuring the clinician's diagnostic decision-making when using the device's Top-5 prioritised differential view — positioned below the Rank 2–4 prospective real-patient studies that carry the primary Pillar 3 weight.

CER state after the MAN_2025 reconciliation

The CER has been corrected as part of the MAN_2025 work (see man-2025-fixes.md addendum 2026-04-19). All eleven of this session's Pillar-2 CER edits were reversed, plus two pre-existing Pillar-2-for-MRMC framings were corrected. The CER now consistently places all four MRMC studies (BI_2024, PH_2024, SAN_2024, MAN_2025) at Pillar 3 §4.4 supporting evidence at Rank 11. The SAN_2024 CIP/CIR therefore currently disagrees with the CER on pillar placement — the SAN_2024 CIP/CIR says Pillar 2, the CER says Pillar 3. This is the mismatch to resolve in the deferred pass.

Integrator-responsibility language (Saray's second gap — separate)

Saray's Slack also flagged a forbidden CER pattern: "The number of categories displayed … are determined by the integrating system's interface design, not by the device itself." The device must mandate integration requirements in the IFU rather than delegate the clinical-UI format to the integrator. This is unchanged by the Q9 reconsideration and remains a separate CER + IFU follow-up.

New proactive agent

A new agent celine-clinical-consultant has been created at .claude/agents/celine-clinical-consultant.md encoding this framework. It fires automatically on any clinical-evidence edit and will flag pillar-mapping errors (including the Q9 pattern) before they compound. The SAN_2024 deferred pass should be run with that agent in the loop.

Build verification not required for this addendum

This addendum is documentation only; no code or audit-visible MDX was touched. Build verification will be performed when the SAN_2024 CIP/CIR deferred pass is executed.

---

Addendum 2026-04-19 (2) — SAN_2024 CIP/CIR Pillar-3 reclassification applied

The deferred pass flagged in the addendum above has now been executed. This addendum records what was changed.

What was actually encoded in the SAN_2024 prose

On close reading of the CIP and CIR, the Q9 "Pillar-2 confirmatory" decision from the original SAN_2024 pass had not been written into the audit-visible MDX using the words "Pillar 2". Instead it had crossed into the prose as a synonym — "to support technical-performance and reader-variability claims" — which is Pillar-2 framing by another name (Technical Performance is Pillar 2 under MDCG 2020-1). The synonym was distributed across five paragraphs in the CIR (Research Title, Summary §Nature and positioning, Summary §Conclusions, main Introduction, Implications for Future Research) and carried through to a handful of additional paragraphs in the Overall Conclusion and Scientific Publication sections without using the literal word "Pillar".

Because the synonym pattern is the Pillar-2 conflation in disguise, reversing the Q9 decision meant finding and rewriting every instance of "technical-performance and reader-variability claims" (and near-equivalents) into explicit Pillar 3 §4.4 Clinical Performance framing — the clinician, using the device's Top-5 prioritised differential view, makes measurably better diagnostic decisions.

Files touched

• apps/qms/docs/legit-health-plus-version-1-1-0-0/product-verification-and-validation/clinical/Investigation/san-2024/r-tf-015-004.mdx (CIP) — 1 edit.
• apps/qms/docs/legit-health-plus-version-1-1-0-0/product-verification-and-validation/clinical/Investigation/san-2024/r-tf-015-006.mdx (CIR) — 9 edits.

CIP — single edit

§Limitations of clinical research, Rank-11 positioning paragraph: added an explicit Pillar 3 §4.4 clause naming the device's Top-5 prioritised differential view. Before: "…constitute Rank 11 evidence and are distinct from clinical data generated on real patients within the meaning of MDR Article 2(48). The positioning of this investigation within the overall body of clinical evidence is documented in the Clinical Evaluation Plan and Clinical Evaluation Report." After: "…constitute Rank 11 evidence and are distinct from clinical data generated on real patients within the meaning of MDR Article 2(48); per MDCG 2020-1 §4.4 they contribute Pillar 3 Clinical Performance supporting evidence — measuring the clinician's diagnostic decision-making when using the device's Top-5 prioritised differential view — at a lower rank than the prospective real-patient studies."

CIR — nine edits

1. Research Title paragraph (top of CIR, after the headline): rewrote "Its role … is to support technical-performance and reader-variability claims, not clinical performance on real patients" → "Per MDCG 2020-1 §4.4, it contributes Pillar 3 Clinical Performance supporting evidence at Rank 11 — measuring the clinician's diagnostic decision-making when using the device's Top-5 prioritised differential view".
2. Summary — Nature and positioning of the evidence: same synonym swap, Pillar 3 §4.4 framing substituted for "technical-performance and reader-variability claims on a curated image set". Explicit statement of Top-5 prioritised differential view as the UI output being evaluated.
3. Summary — Conclusions (Clinical Significance paragraph): "they support technical-performance and reader-variability claims but are distinct from clinical data generated on real patients" → "per MDCG 2020-1 §4.4 they contribute Pillar 3 Clinical Performance supporting evidence for the clinician+device diagnostic decision on the Top-5 prioritised differential, but are distinct from clinical data generated on real patients".
4. Main Introduction (scope statement): "Under MDCG 2020-6 Appendix III the resulting evidence is Rank 11 (simulated-use reader-variability data) and is used in the Clinical Evaluation to support technical-performance and reader-variability claims only" → "Under MDCG 2020-6 Appendix III the resulting evidence is Rank 11 (simulated-use reader study on retrospective images); per MDCG 2020-1 §4.4 it is used in the Clinical Evaluation as Pillar 3 Clinical Performance supporting evidence for the clinician+device decision". Also tightened the preceding sentence to name the Top-5 prioritised differential as the evaluated UI output.
5. Overall Conclusion — Clinical Performance, Efficacy, and Safety (Rank-11 scope paragraph): inserted "per MDCG 2020-1 §4.4 it contributes Pillar 3 Clinical Performance supporting evidence (the clinician, using the device's Top-5 prioritised differential view, makes measurably better diagnostic decisions)".
6. Overall Conclusion — wrap-up paragraph: added Pillar 3 §4.4 clause to the final Rank-11 positioning sentence.
7. Implications for Future Research — opening sentence: reframed "the findings reported here are hypothesis-generating with respect to real-world clinical performance" → "the findings reported here are hypothesis-generating with respect to the real-world Pillar 3 claims that require Rank 2–4 prospective real-patient evidence".
8. Implications for Future Research — closing paragraph: added Pillar 3 §4.4 clause to the Rank-11 positioning sentence, specifically naming the Top-5 prioritised differential view.
9. Scientific Publication section — closing sentence about how this investigation is cited in the CE: "the present investigation is cited in the Clinical Evaluation as simulated-use reader-variability evidence only" → "the present investigation is cited in the Clinical Evaluation as Pillar 3 §4.4 supporting evidence at Rank 11 (simulated-use clinician+device reader study), not as a substitute for Rank 2–4 prospective real-patient Pillar 3 evidence".

The Q9 decision to strip "pilot" language and re-classify SAN_2024 as "confirmatory" (rather than "pilot") is preserved — that part of Q9 is independent of pillar placement and remains correct. SAN_2024 is a confirmatory MRMC investigation; its confirmatory role sits in Pillar 3 §4.4 at Rank 11, not in Pillar 2.

What was deliberately NOT changed

• The acceptance-criteria thresholds and SotA baselines under §Justification of acceptance thresholds are unchanged — the Pillar-2 reclassification of Q9 did not affect thresholds, and Celine/Saray's framework does not require threshold changes.
• The <EthicsCommitteeNonApplicability /> snippet usage is unchanged.
• The Annex E (r-tf-015-010.mdx) is unchanged — it is the ISO 14155 compliance checklist, not a pillar-framing document.
• The "technical-performance" terminology inside <CirDataQualityAssurance /> or similar reusable snippets was not touched — those refer to the algorithm's input-handling and remain correctly Pillar 2 (API-level) framing.
• The §Statistical analysis primary/secondary/exploratory structure from Q2 and Q3 is unchanged — those decisions were about statistical rigour, not pillar placement.

Consistency after this pass

SAN_2024 (CIP + CIR) now consistently frames the evidence as Pillar 3 §4.4 Clinical Performance supporting at Rank 11. The CER already agrees with this framing after the 2026-04-19 MAN_2025 reversal pass (see man-2025-fixes.md addendum). The two documents are therefore aligned.

BI_2024 and PH_2024 have NOT been audited in this pass. If those investigations' CIPs/CIRs use the same "technical-performance and reader-variability claims" synonym pattern, they will need the same rewording treatment. The new celine-clinical-consultant agent will flag them automatically on the next edit.

Build verification

npx turbo run build --filter=qms completed successfully after the SAN_2024 Pillar-3 corrections (2026-04-19).