IDEI_2023 adequacy-review fixes (2026-04-19)

Internal record of the fixes applied to the IDEI_2023 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 with the Celine/Saray three-pillar framework enforced end-to-end. Companion to mc-evcdao-2019-fixes.md, san-2024-fixes.md, ph-2024-fixes.md, bi-2024-fixes.md and man-2025-fixes.md; this note records the decisions that are specific to IDEI_2023 and explicitly calls out where this pass differs from the earlier passes.

Scope

• Folder: apps/qms/docs/legit-health-plus-version-1-1-0-0/product-verification-and-validation/clinical/Investigation/idei-2023/
• Files edited:
  • r-tf-015-004.mdx (CIP) — surgical edits plus new §Scope §Nature-and-positioning subsection, new §Product-Identification §Device-version subsection, full rewrites of §Objectives, §Summary of the study, §Type of clinical research, §Population, §Sample size, §Duration, §Variables, §Acceptance-criteria prose, §Inclusion/Exclusion, §Bias minimisation, §Calendar, §Statistical analysis, §Limitations. Other small fixes: H1 # Suspension → ##, T-015-006 / T-013-002 code-token leaks replaced with R-TF-015-006 / R-TF-013-002, typos (Completition, Product deficencies, standarised), AI Labs Group S.L. → "the manufacturer", duplicate <object data={protocol}> deleted, trailing Hangul filler stripped. Technical Support subsection broken out with role titles.
  • r-tf-015-006.mdx (CIR) — surgical edits plus new §Research-Title §Nature-and-positioning subsection, new §Product-Identification §Device-version subsection, full rewrites of §Summary (introduction / objectives / sample size / number of subjects / duration / methods / results / conclusions / benefit-risk), §Introduction, §CIP §Objectives, §Statistical analysis (incl. performanceClaims.ts leak removal), §Analysis introduction, §Protocol Deviations (numbered list of six), §Subject demographics (with Fitzpatrick V/VI limitation), §Results and discussion for the retrospective/prospective pigmented-lesion analyses (aided-reader column re-labelled), §Discussion — Summary of Performance Claims (bold-as-heading promoted to real heading), §Discussion primary-endpoint narrative, §Limitations (eight numbered items), §Clinical risks and benefits, §Clinical relevance (off-topic refs 8/9/12 dropped; marketing prose replaced with regulatory-voice resource-use paragraph), §Specific benefit and special precaution, §Implications for future research, §Brief description (Basque-health-centres leak deleted; manufacturer legal-entity replaced with "the manufacturer"), §Investigators + Technical Support split, §Ethics Committee (date reconciled to 2024-01-25), §Sponsor and monitor (trimmed marketing block and "AI Labs Groups S.L." typo), §Report annexes (filesystem PDF path leak replaced with Annex II cross-ref), §Publication Status (preprint label), trailing Hangul filler stripped.
  • r-tf-015-010.mdx (Annex E) — full rewrite against the MC_EVCDAO_2019 / PH_2024 23-row template. New §Applicability-and-scope preamble framing the investigation as Rank 2–4 + Pillar 3 primary. Rows added: Statistical Analysis Plan, Deviations Log, Adverse-event reporting procedure, Device-deficiency reporting procedure, Annual safety / progress report. IB row flipped TRUE with IFU-equivalent justification. Regulatory Authority Authorisation row FALSE with the non-interventional observational CE-marked-within-intended-purpose justification (AEMPS confirmation still tracked under the Jordi follow-up at question-for-jordi.mdx). Every row now carries a regulatory-level justification citing the specific record location.
• Other files edited: None outside the investigation folder in this pass. The IDEI_2023 entry in packages/ui/src/components/ClinicalValidation/clinicalStudiesData.ts and the 14 IDEI_2023 rows in packages/ui/src/components/PerformanceClaimsAndClinicalBenefits/performanceClaims.ts were read but not modified — their pre-specified thresholds and achieved values already match the CIP/CIR after the re-framing below.
• Build: npx turbo run build --filter=qms — passes.

Context: this is the second real-patient prospective Pillar 3 pass

The adequacy-review series to date:

• SAN_2024, MAN_2025, BI_2024, PH_2024 — MRMC simulated-use reader studies at Rank 11, Pillar 3 §4.4 supporting.
• MC_EVCDAO_2019 — first real-patient prospective Pillar 3 pass, Rank 2–4, Pillar 3 primary.
• IDEI_2023 — second real-patient prospective Pillar 3 pass, Rank 2–4, Pillar 3 primary.

IDEI_2023 is regulatorily heavier than the MRMC passes for the same reasons as MC_EVCDAO_2019: real patients, a partly-pathology reference standard (for the pigmented-lesion malignancy analyses), primary Pillar 3 evidence relied on by the CER for the Class IIb indirect-benefit defence. This pass confirms that the MC_EVCDAO_2019 architectural decisions (A through K) apply essentially unchanged to IDEI_2023, with two IDEI-specific nuances documented below (Decisions L and M).

Reused from MC_EVCDAO_2019 without adaptation

• Q4 (device version, identity bridge). v1.1.0.0 identity bridge; new §"Device version under investigation and bridging to the CE-marked release" subsection in both CIP §Product Identification and Description and CIR §Product Identification; PRRC sign-off; wording copied verbatim from MC_EVCDAO_2019.
• Q7 (Investigator's Brochure, IFU v1.1.0.0). Annex E IB row flipped from TRUE (empty justification) to TRUE (IFU + Device Description and Specifications record cited).
• Q8 (product-name anonymisation, introduce-once convention). CIP + CIR add "Throughout this document, references to 'the device' refer to the investigational product identified above" immediately after <DeviceCharacterisation />. Standalone brand-name uses in body prose replaced with "the device" or "the manufacturer". Prospective results-table column "Legit.Health Plus" renamed "Device".
• T-015-006 / T-013-002 → R-TF-015-006 / R-TF-013-002 (3 locations in CIP).
• performanceClaims.ts leak removal (CIR line 272 **Derived from performanceClaims.ts (for comparison):** → #### Summary of pre-specified acceptance criteria (for comparison against observed metrics) heading + regulatory-level text).
• Bold-as-heading → proper heading (**Summary of Performance Claims:** → #### Summary of performance claims; **1. Increased Sample Size (Positive Deviation)** / **2. Image Exclusions...** folded into the six-deviation numbered list under §Protocol Deviations; **Note on Dermatologist Results** merged into the operating-point narrative).
• Heading hierarchy (CIP # Suspension or early termination of clinical research → ## Suspension ...).
• Duplicate <object data={protocol}> under CIP §Annex I — one instance deleted.
• Typos (Completition → Completion; Product deficencies → Product deficiencies; standarised → standardised; emocional → emotional; determin → determine; AI Labs Groups S.L. → AI Labs Group S.L.).
• Brand / company-name in prose (ten instances in CIR body: §Description, §Summary, §Title, §Summary introduction, §Hypothesis, §Secondary objectives, §Main Introduction, §Cost-Effectiveness and Resource Optimization, §Sponsor and monitor, prospective-results-table column header).
• Basque-health-centres stray reference (CIR §Brief description) deleted.
• PDF filesystem asset name (CEIm_Legit.Health_IDEI_2023.pdf) replaced with "Annex II of the CIP" cross-reference.
• Inconsistent investigator name styling (Dr. Pablo López Andina accent/period; Ms. Beatriz Torres honorific) normalised.
• Trailing Hangul filler ㅤ removed after <Signature /> across CIP, CIR, Annex E.
• Annex E full rewrite to the 23-row MC_EVCDAO_2019 template with ISO 14155 clause cross-references in each justification.

Decisions specific to IDEI_2023

L — Primary-endpoint framing honest about the pre-specified malignancy-detection endpoint vs the Ludwig-agreement endpoint

Prior state: the original CIP §Primary objective named a compound objective conflating diagnostic accuracy, malignancy determination, severity measurement and workflow/cost optimisation. The CIR §Conclusions claimed the device "confirms its reliability as a screening tool for malignant ICD-11 categories" on the basis of a prospective cohort with N = 8 confirmed malignant cases; it also claimed the androgenetic-alopecia acceptance criterion was met by invoking a pooled retrospective+prospective Kappa of 0.6235, when in fact the prospective Kappa of 0.33 fails the pre-specified Kappa ≥ 0.6 threshold.

Decision: re-frame the CIP primary objective as the malignancy-estimation AUC against histopathology at the pre-specified operating threshold (consistent with the MDCG 2020-1 §4.4 Pillar 3 alignment and with the device's actual clinical output — the malignancy gauge). Retain Top-K diagnostic agreement and Ludwig-score agreement as pre-specified secondary objectives. In the CIR:

• Report the malignancy-detection primary-endpoint estimates on the prospective arm as preliminary-confirmatory (AUC 0.9669; sensitivity 0.8750; specificity 0.9706 at the in-sample Youden-J operating point, with the pre-specified acceptance thresholds met but the small malignant subset N = 8 acknowledged in every citation of the headline figures).
• Report the female androgenetic alopecia Ludwig-agreement endpoint honestly: prospective Kappa = 0.33 (fair agreement), does NOT meet the pre-specified acceptance criterion of Kappa ≥ 0.6. The pooled retrospective+prospective Kappa of 0.6235 is declared in-sample-contaminated (the retrospective arm was used for hyperparameter tuning) and is reported descriptively only — it is not used as confirmatory evidence. A root-cause analysis and an independent-sample PMCF confirmatory study with a pre-specified operating threshold are committed.

Rationale (Celine/Saray framework). Pillar 3 claims must be honest about the precision of the evidence. For malignancy detection the device's actual clinical output (the malignancy gauge) is the right axis; the primary-endpoint result meets its pre-specified thresholds but the precision of the estimate (wide CIs on N = 8 malignant cases) caps the strength of the claim to "preliminary-confirmatory" with a PMCF commitment. For the Ludwig-agreement endpoint, pooling train+test data to manufacture a pass is a methodological error; the honest reading is that the prospective-unseen-data Kappa of 0.33 fails, and the CER / IFU / performance claims must be reconciled with that honest reading (any claim of the form "Ludwig-agreement criterion met" must either (a) refer to the retrospective arm and be labelled as tuning-set performance, or (b) cite the PMCF confirmatory study once it is run).

Pillar boundaries respected. IDEI_2023 remains Pillar 3 primary at Rank 2–4 for both the malignancy-detection endpoint and the Ludwig-agreement endpoint. No MRMC / simulated-use evidence is moved into Pillar 2. The re-framing is about endpoint honesty, not pillar assignment.

M — Six formal protocol deviations declared (novel to IDEI_2023)

Prior state: CIR §Protocol Deviations contained only two deviations (sample-size increase as a "positive deviation"; image exclusions due to missing pathology). Not declared: the reference-standard switch from paper questionnaire to histopathology; the aided-reader design in the prospective arm; the in-sample Youden-J operating-threshold selection; the pooled-metrics artefact in the Ludwig-agreement endpoint; the handling of missing histopathological confirmation.

Decision: rewrite §Protocol Deviations as a numbered list of six formal deviations:

1. Retrospective-cohort extension from 90/30 target to 88/62 pigmented-lesion + 62/34 alopecia (methodological deviation; declared and traceable to documented extraction query).
2. Reference-standard statement reconciliation (CIP SAP cited the paper questionnaire; CIR actually uses histopathology — the CIP residual is reconciled to histopathology in this CIR).
3. Aided-reader design in the prospective arm (prospective "Dermatologist" column re-labelled "Dermatologist + device" throughout; unaided-reader comparator committed to PMCF).
4. In-sample Youden-J operating-threshold selection for the malignancy gauge (0.30 retrospective, 0.40 prospective — exploratory label; pre-specified threshold reported in parallel; independent-sample confirmation committed to PMCF).
5. Pooled retrospective+prospective metrics for the Ludwig-agreement endpoint not used as confirmatory evidence (prospective Kappa 0.33 does not meet pre-specified Kappa ≥ 0.6; PMCF confirmatory study committed).
6. Missing histopathological confirmation in the prospective arm handled via pre-specified intention-to-treat sensitivity analysis (best-case / worst-case imputation) reported alongside the complete-case primary estimate.

Applicable to other real-patient passes: the pattern established here is "every deviation from the CIP SAP is a formal deviation, with classification and impact assessment — not just the headline sample-size and exclusion deviations".

N — Benefit-risk against GSPR 1 and GSPR 8 with quantified residual-risk

Prior state: CIR §Clinical risks and benefits was a two-line boilerplate claim without quantifying the residual false-negative / false-positive consequences of the device's malignancy gauge under its intended use.

Decision: new §Benefit-risk assessment subsection under §Summary with quantified GSPR 1 (benefits outweigh risks) and GSPR 8 (use-error risk) paragraphs:

• GSPR 1: at the prospective operating point, 7/8 malignant lesions correctly flagged (sensitivity 0.8750) and 33/34 non-malignant correctly cleared (specificity 0.9706); residual false-negative rate 12.5% and residual false-positive rate 2.9% with wide confidence intervals reflecting the small malignant subset. Mitigations: decision-support IFU framing, Top-5 prioritised differential display mandate in the IFU's integration requirements, standing biopsy protocol at the investigator site.
• GSPR 8: zero AE / ADE / SAE / SADE across 204 subjects. Primary use-error risk is misinterpretation as diagnostic determination. Mitigations: IFU decision-support framing, integration requirements that mandate Top-5 + malignancy gauge + referral-recommendation display, investigator training on the device before investigation initiation.

Note on integrator-responsibility language (Saray's second gap per .claude/agents/celine-clinical-consultant.md). The GSPR 8 mitigation paragraph says the IFU "integration requirements mandate that the integrating system display the Top-5 prioritised differential view, the malignancy gauge and the associated triage recommendation". This is the device-mandates framing, not the forbidden "determined by the integrating system's interface design" pattern.

O — Fitzpatrick V/VI = 0 limitation with CER + PMCF cross-reference

Prior state: CIR §Subject demographics showed Fitzpatrick V 0 / VI 0 and Fitzpatrick IV = 2 (0.9%); no limitation statement was anchored.

Decision: explicit limitation added immediately after the Fitzpatrick demographics table: "The cohort includes no Fitzpatrick V or VI participants and the representation of Fitzpatrick IV is limited to two cases. Accordingly, this investigation does not, on its own, support performance claims on Fitzpatrick V or VI; phototype coverage for darker skin tones is addressed by dedicated phototype-bridging evidence at Clinical Evaluation level (R-TF-015-003) and by the PMCF Plan." Paediatric exclusion (under-18) is declared separately in the §Inclusion/Exclusion section of the CIP.

P — Off-topic literature citations removed from §Clinical relevance

Prior state: §Clinical relevance cited reference 8 (Liu et al., "A review of machine learning in obesity"), reference 9 (Portney & Watkins, a general research-methods textbook) and reference 12 (Iqbal et al., a ChatGPT-in-healthcare umbrella review) as if they supported device-specific claims. Reference 20 (Nittas et al., dermatopathology-acceptance) was also off-topic for a decision-support medical device.

Decision: remove references 8, 9, 12, 20 from the reference list and rewrite §Clinical relevance in regulatory voice anchored to the manufacturer's own evidence (AUAS_2023, AIHS4_2023, DIQA_2023, ASCORAD_2022) and to on-topic literature benchmarks (Esteva 2017, Haenssle 2018, Han 2018, Jain 2021, Papachristou 2024, Jerant 2000, Schuldt 2023, Elsaie 2020). Marketing-voice claims ("cost-effectiveness benefits", per-patient-$ savings, the "cutting-edge solution" close) are replaced with resource-use framing that is contextual rather than evidence-claiming.

What was applied — by batch

Batch 1 — Formatting sweep

All markdown-style findings from the adequacy review:

• CIP (>18 years) → (\>18 years) in §Population and §Inclusion criteria (2 locations).
• CIP double-period typos (lines 127, 131) resolved via §Sample size full rewrite.
• CIP dangling sentence fragment (For alopecia, the distribution of 15 prospective and 15 retrospective cases.) resolved via §Sample size full rewrite.
• CIP dangling sub-item under §Variables > Secondary variables (pathology-anatomy line without bullet) resolved via §Variables full rewrite as single-level bullets.
• CIP # Suspension or early termination of clinical research (H1 mid-document) → ## Suspension or early termination of clinical research.
• CIP duplicate <object data={protocol} /> removed.
• CIP typos Completition → Completion (heading) and Product deficencies → Product deficiencies (heading).
• CIR bold-as-heading **Derived from performanceClaims.ts (for comparison):** → #### Summary of pre-specified acceptance criteria (for comparison against observed metrics) (heading + regulatory wording, leak removed in parallel).
• CIR bold-as-heading **1. Increased Sample Size (Positive Deviation)** / **2. Image Exclusions Due to Diagnostic Confirmation Limitations (Minor Deviation)** replaced by the six-deviation numbered ##### heading list under §Protocol Deviations.
• CIR bold-as-heading **Note on Dermatologist Results** merged into the operating-point narrative (no longer a bold-as-heading).
• CIR bold-as-heading **Summary of Performance Claims:** → #### Summary of performance claims.
• CIR — entity → literal em dash — in the diagnosis-mapping paragraph.
• CIR hard line break mid-paragraph after the bold-as-heading deviation lines resolved by the §Protocol Deviations rewrite.
• CIR inconsistent spacing before <sup>...</sup> left as-is in the un-rewritten §Clinical relevance paragraphs that were replaced by the Batch-4 regulatory-voice rewrite; remaining <sup> citations follow the no-space convention in the rewrite.
• CIR stale [//]: # "..." comment stubs (lines 336–338, 550–554) replaced in the §Analysis and §Retrospective / prospective analysis rewrites.
• Trailing Hangul filler ㅤ after <Signature /> removed across CIP, CIR, Annex E.

Batch 2 — Leak rewording (engineering nouns, brand names, filesystem paths)

• CIP T-015-006 Clinical Investigation Report (backticked code-token) → R-TF-015-006 (2 locations: §Completion of the investigation and §Start, follow-up and end reports).
• CIP T-013-002 Risk management record (backticked code-token) → Risk Management Record (R-TF-013-002).
• CIP "AI Labs Group S.L." legal-entity leak in §Collaborating Investigators split into a §Technical Support (Manufacturer) subsection with role titles; §Rationale prose "developed by AI Labs Group S.L." removed.
• CIR brand-name uses in body prose ("Legit.Health Plus" / "Legit.Health") replaced with "the device" (10 locations: §Research Title, §Description, §Summary introduction, §Summary §Title, §Summary §Hypothesis, §Summary §Secondary objectives §Cost-effectiveness, §Main Introduction, §Cost-Effectiveness and Resource Optimization, §Sponsor and monitor brand mark, prospective-malignancy-results-table column header).
• CIR "AI Labs Group S.L." in §Brief description and §Collaborators replaced with "the manufacturer" / Technical Support subsection.
• CIR "CEIm_Legit.Health_IDEI_2023.pdf" filesystem path leak in §Report annexes replaced with "Annex II of the CIP" cross-reference.
• CIR "Python programming language will be used as statistical software" (dev-language slip; contradicted CIP §Statistical analysis which named SPSS/STATA) replaced in the §Methods and §Statistical analysis rewrites with "a deterministic, version-controlled analytics environment maintained by the manufacturer".
• CIR "including the Sodupe-Güeñes, Balmaseda, Buruaga, and Zurbarán Health Centers" (Basque-health-centres leak, factually incorrect for IDEI_2023) deleted from §Brief description.
• CIR YOLO / ResNet50 / UNet architecture disclosures in §Methodology retained (the device's Software Architecture Description is consistent with these; they are legitimate device-description content and not a leak).

Batch 3 — Clinical Minor

• CIR typos emocional → emotional; determin → determine (in §Clinical relevance; resolved via the regulatory-voice rewrite).
• CIR typo AI Labs Groups S.L. → AI Labs Group S.L. in §Sponsor and monitor (block rewritten against the canonical manufacturer details).
• CIR Ethics Committee approval date reconciled: CIR §Ethics Committee previously said "December 19, 2023" and CIP Annex II said "2024-01-25, reference 24.12.2266-GHM"; the CIP Annex II date is the operative one (the 25 Jan 2024 date is also the investigation initiation date reported in CIR §Initiation date); both documents now state 25 Jan 2024 with the reference number. The December 19, 2023 date appears to have been an earlier-review date for a different amendment; if the EC letter confirms that as a prior favourable opinion, both dates can be added in a later pass (cf. MC_EVCDAO_2019 Decision F pattern).
• CIR §Discussion references off-topic citations (Liu 2019 obesity; Portney & Watkins methods textbook; Iqbal LLM umbrella review; Nittas dermatopathology acceptance) removed.
• CIR §Publication Status "Published under the title..." reworded to "Available as a preprint under the title... The preprint has not yet been peer-reviewed; under MDCG 2020-6 Appendix III a non-peer-reviewed preprint does not constitute Rank 1–6 peer-reviewed evidence, and this citation is reported accordingly."
• CIR §Adverse events and §Product deficiencies single-sentence attestations expanded with AE-solicitation mechanism, reporting pathway, subject-time denominator.
• CIR duration reconciliation: §Summary §Duration now reports actual 2024-01-25 → 2024-08-23 ≈ 7 months; CIP §Duration and §Calendar aligned to the same 7-month envelope (original CIP had a 6-vs-12-month internal contradiction).

Batch 4 — Clinical Major

• Decision L — Primary-endpoint framing honest. CIP §Primary objective split into malignancy-detection AUC primary + Top-K and Ludwig-agreement secondary, with pre-specified operating threshold in the SAP. CIR §Summary §Primary objective, §CIP §Objectives, §Summary §Results, §Conclusions and §Discussion rewritten to carry the primary endpoint and to report the Ludwig-agreement non-attainment honestly.
• Decision M — Six formal protocol deviations. CIR §Protocol Deviations rewritten as numbered ##### list of six.
• Decision N — Benefit-risk against GSPR 1 and GSPR 8 with quantified residual-risk. New §Benefit-risk assessment subsection under §Summary.
• Decision O — Fitzpatrick V/VI limitation with CER + PMCF cross-reference. Added immediately after the Fitzpatrick demographics table.
• Decision P — Off-topic literature citations removed from §Clinical relevance and regulatory-voice rewrite.
• Aided-reader design declaration and prospective "Dermatologist + device" re-labelling across results tables.
• CIP §Statistical analysis rewrite to match actually-executed methods (histopathology reference standard; Wilson CIs; bootstrap AUC CIs; pre-specified ITT sensitivity analysis for missing histopathology; exploratory label on in-sample Youden-J thresholds).
• CIP §Limitations of clinical research rewrite with numbered list of five structural limitations.
• CIP §Bias minimization measures rewrite — false "randomly selected" claim removed; actual design (consecutive enrolment, documented retrospective extraction query, standardised acquisition, independent reference standard) documented.
• CIP §Inclusion / Exclusion criteria rewrite — Fitzpatrick V/VI and paediatric exclusions explicit; DIQA image-quality gating declared.
• Annex E full rewrite — 23 rows with regulatory-level justifications and ISO 14155 clause cross-references; new rows added for SAP, Deviations Log, AE reporting, Device-deficiency reporting, Annual safety / progress report; IB row flipped TRUE with IFU equivalent; Regulatory Authority Authorization row FALSE with the non-interventional observational CE-marked-within-intended-purpose justification (AEMPS confirmation still pending under the Jordi follow-up).

Batch 5 — Clinical Critical

All items under Batch 5 are covered by the Batch 4 items above — there is no separate Critical tier for this pass because the reviewer-assigned "critical" items (Primary objective conflation; Sample-size fake power; Study design mislabelling + aided-reader; Acceptance criteria prose; Screening-tool overstatement on N = 8; Alopecia pooled-metrics pass; Reader-device independence; Device version identification; Traceability CIP↔CIR; Bridging to 1.1.0.0) collapse into the architectural Decisions L + M + N + O (primary-endpoint framing, formal deviations, benefit-risk, Fitzpatrick limitation) plus the reused identity-bridge subsection (v1.1.0.0) from MC_EVCDAO_2019 Decision K. Traceability CIP↔CIR is closed by the restructured §Objectives / §Acceptance criteria / §Primary analysis / §Conclusions chain.

Follow-ups to track

1. AEMPS reference confirmation for IDEI_2023 (blocks final Annex E sign-off). The Regulatory Authority Authorization row in Annex E carries the non-interventional observational CE-marked-within-intended-purpose justification pending a confirming reference from AEMPS. Jordi is already tracking MC_EVCDAO_2019; IDEI_2023 should be added to the same follow-up note (apps/qms/docs/bsi-non-conformities/clinical-review/round-1/item-3-clinical-data/ra-clinical-data-analysis/question-for-jordi.mdx §1 AEMPS communications).
2. Ethics Committee date (December 19, 2023 vs January 25, 2024). The current CIR + Annex E reconcile to 25 January 2024 (matching the approval letter reference 24.12.2266-GHM). If the EC letter confirms that 19 December 2023 was an earlier favourable opinion for a prior protocol version, both dates should be added in a later pass with a subscript labelling each (cf. MC_EVCDAO_2019 Decision F two-CEIm-dates pattern).
3. CER reconciliation after IDEI_2023 refactor. The CER (R-TF-015-003) almost certainly cites IDEI_2023 as Pillar 3 Clinical Performance evidence for both pigmented-lesion malignancy detection and female androgenetic alopecia severity assessment. The CIR now declares: (i) the pigmented-lesion malignancy primary endpoint met its pre-specified thresholds on the prospective arm as preliminary-confirmatory evidence with wide CIs on N = 8 malignant cases; (ii) the female androgenetic alopecia Ludwig-agreement endpoint did NOT meet the pre-specified Kappa ≥ 0.6 threshold on the prospective validation arm. The CER must be audited to verify:
   • The IDEI_2023 citation in the Clinical Performance section of the CER reflects the preliminary-confirmatory character of the pigmented-lesion malignancy result with the PMCF confirmatory commitment named.
   • Any CER sentence of the form "IDEI_2023 demonstrated Ludwig agreement meeting the acceptance threshold" is rewritten honestly (the retrospective arm is tuning-set; the prospective arm fails); the PMCF confirmatory commitment for Ludwig-score agreement is named.
   • The Fitzpatrick V/VI coverage-gap call-out in the CER cross-references MAN_2025 as the primary phototype-bridging evidence, consistent with the 2026-04-18 IFU ISI alignment to MAN_2025 evidence (commit a0be539d3).
4. PMCF Plan commitments. The IDEI_2023 CIR now carries four explicit PMCF deliverables: (i) independent-sample confirmation of the malignancy-detection primary endpoint at a pre-specified operating threshold; (ii) independent-reader (unaided) comparator for pigmented-lesion diagnostic accuracy; (iii) independent-sample confirmation of the Ludwig-score agreement endpoint at a pre-specified operating threshold; (iv) stratified Fitzpatrick V/VI coverage. Each must be reflected in the PMCF Plan (R-TF-007-002) with a named activity, a milestone, an endpoint and a pre-specified analysis.
5. clinicalStudiesData.ts and performanceClaims.ts alignment audit for IDEI_2023. The IDEI_2023 entries (14 rows in performanceClaims.ts) already carry pre-specified thresholds and achieved values consistent with the refactored CIR. Confirm that the rendered <AcceptanceCriteriaTable studyCode="IDEI_2023" /> output displays the prospective-arm Ludwig-agreement failure (row 3OB, Kappa 0.3297 vs threshold 0.6) in the pass/fail column — this is the single most important surface where the honest framing is visible to a BSI auditor. If the rendered table shows the pooled metric as a pass, the data-file entries for rows 3OB and 284 must be corrected to report the prospective-only values.
6. Integrator-responsibility check in the IFU. The IDEI_2023 CIR §Benefit-risk paragraph says the IFU "integration requirements mandate that the integrating system display the Top-5 prioritised differential view, the malignancy gauge and the associated triage recommendation". Verify that the EU IFU (apps/eu-ifu-mdr/) and the FDA IFU (apps/us-ifu-fda/) actually contain an explicit "integration requirements" section with these mandates; if the IFU uses the forbidden "determined by the integrating system's interface design, not by the device itself" pattern that Saray flagged, fix the IFU and update the CIR reference accordingly.
7. Remaining real-patient prospective Pillar 3 passes to process. COVIDX_EVCDAO_2022, DAO_Derivación_O_2022, DAO_Derivación_PH_2022 and NMSC_2025 should be run through /review-clinical-investigation next; the MC_EVCDAO_2019 Decisions A–K and the IDEI_2023 Decisions L–P now form a paired playbook for every remaining real-patient prospective Pillar 3 investigation.

Build verification

npx turbo run build --filter=qms completed successfully after the CIP + CIR + Annex E edits (2026-04-19).

---

Ultrathink summary: why the malignancy-detection primary-endpoint framing is defensible here

A BSI auditor could argue that reporting the pigmented-lesion malignancy-detection endpoint as "pre-specified acceptance thresholds met" on a prospective arm with only 8 confirmed malignant cases over-states the evidence. The defence assembled in the CIR is:

1. The small-N precision caveat is printed on every headline. Every citation of the headline AUC, sensitivity, PPV etc. is accompanied by the confidence interval and — in the Conclusions, Discussion and Benefit-risk paragraphs — by the explicit statement that the primary endpoint is "preliminary-confirmatory" because of the small malignant subset (N = 8). The reader is never left to infer the precision.
2. The in-sample Youden-J threshold is declared exploratory. The CIR reports performance at the in-sample Youden-J optimum (0.40 on the prospective arm) and labels it exploratory; the pre-specified operating threshold is reported in parallel; confirmatory independent-sample validation at a pre-specified threshold is committed to the PMCF Plan. This is the MC_EVCDAO_2019 Decision A pattern applied here.
3. The Ludwig-agreement endpoint is not hidden behind the malignancy success. The prospective Kappa of 0.33 fails the pre-specified Kappa ≥ 0.6 threshold and that failure is reported in the Summary Results, the Summary Conclusions, the Limitations, the Discussion and the Benefit-risk paragraph. Pooled retrospective+prospective Kappa is not used as confirmatory evidence. The honest reading is visible wherever an auditor looks.
4. Pillar boundaries are respected. IDEI_2023 is Pillar 3 primary at Rank 2–4 for both endpoints. No MRMC / simulated-use evidence is moved into Pillar 2. The claims are anchored to the device's actual clinical outputs (malignancy gauge and Top-5 prioritised differential) under the IFU's integration requirements.
5. Integrator language is handled, not deferred. The benefit-risk paragraph uses "the IFU's integration requirements mandate the display of the Top-5 prioritised differential, the malignancy gauge and the referral recommendation" — device-mandates framing, not the forbidden "determined by the integrator" pattern.

The alternative — suppressing the malignancy-detection primary-endpoint results because N = 8 malignant cases is small, or claiming the Ludwig-agreement endpoint as met via the pooled Kappa — would have produced either an unusable CIR or a BSI-reject-on-sight methodological error. The chosen framing accepts the precision caveat (explicitly labelled on every headline) in exchange for a primary endpoint that is honestly reported, met at the pre-specified thresholds on the operative estimate, and tied to a PMCF confirmatory commitment. If BSI treats the N = 8 precision as too low for primary-confirmatory weight, the fallback is "preliminary evidence; confirmation pending PMCF" — which is recoverable from the current CIR by rewording §Conclusions. The CIR is written so that the fallback does not require retracting any factual claim.