COVIDX_EVCDAO_2022 adequacy-review fixes (2026-04-19)

Internal record of the fixes applied to the COVIDX_EVCDAO_2022 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. Companion to mc-evcdao-2019-fixes.md; this note records what differed from or was additional to the MC_EVCDAO_2019 pass, which is the closest structural precedent because COVIDX_EVCDAO_2022 is a real-patient prospective Pillar 3 §4.4 investigation at Rank 2–4, not an MRMC Rank 11 investigation.

Scope

• Folder: apps/qms/docs/legit-health-plus-version-1-1-0-0/product-verification-and-validation/clinical/Investigation/covidx-evcdao-2022/
• Files edited (all three fully rewritten):
  • r-tf-015-004.mdx (CIP).
  • r-tf-015-006.mdx (CIR).
  • r-tf-015-010.mdx (Annex E).
• Other files edited: None outside the investigation folder in this pass.
• Build: npx turbo run build --filter=qms — passes.

Why full rewrites rather than surgical edits

The prior CIP, CIR and Annex E carried a combination of defects whose reach across the documents made surgical edits error-prone:

1. The CIR summarised the investigation as "highly effective" and "proves" the device while the pre-specified primary acceptance criterion (CUS ≥ 8/10) was not met (observed 7.14/10; p = 0.56; null hypothesis not rejected). Re-framing the Summary, Results, Conclusions, Discussion and Limitations all required the same consistent honest primary-endpoint non-attainment statement — and the non-attainment statement had to align with the Deviations section (where the primary-endpoint failure was previously mis-framed as a "deviation to be documented for regulatory consideration").
2. The CIR carried three drifting sample means for the same primary endpoint (7.14/10 in §Deviations, 71.39/100 in §Analysis overall score, 76.67 with SD 73.89 in §Statistical Analysis), plus an unexplained parenthetical "(p = 0.33 when considering all questions)". An auditor could not determine which was the point estimate.
3. The CIP and CIR were framed around a severity-monitoring hypothesis and a sens/spec/PPV/NPV analysis plan that were never executed; the primary endpoint actually tested was clinician-perceived clinical utility (CUS).
4. Neither document identified the device version under investigation or bridged it to the current CE-marked release v1.1.0.0 (the same Q4 gap that has been closed for BI_2024, PH_2024, SAN_2024, MAN_2025 and MC_EVCDAO_2019).
5. The Annex E lacked SAP, Deviations Log, AE reporting, Device-deficiency reporting and Annual safety/progress report rows; the IB row was TRUE with no cited equivalent; the Regulatory Authority Authorization row carried a trailing-sentence justification without a specific legal basis; and the Subject Identification Code List / Screening and Enrollment Logs rows were FALSE with "Kept confidentially in each study centre" used as a justification rather than as a location.

Given this reach, the three files were rewritten whole against the structural templates established by the MC_EVCDAO_2019 (real-patient) and PH_2024 / SAN_2024 / BI_2024 (MRMC) passes. Where a decision had a precedent in those passes, the precedent was followed verbatim unless COVIDX specifics required adaptation.

Decisions applied in this pass

Answers to the eight blocking questions from the adequacy-review action plan, resolved on the MC_EVCDAO_2019 precedent unless noted.

Q1 — Device version at time of investigation (April 2022 – October 2023)

Decision: v1.1.0.0 identity bridge (MC_EVCDAO_2019 Decision K precedent). Both the CIP §Product Identification and Description and the CIR §Product identification now carry a "Device version under investigation and bridging to the CE-marked release" subsection stating that the configuration used during the investigation window is equivalent to the current CE-marked release v1.1.0.0 (PRRC-signed). No retrospective change-log construction was attempted. The Annex E Investigational Device Accountability Records row cross-references this bridging statement.

Q2 — Reconciliation of the CUS primary-endpoint numbers

Decision: the pre-specified primary endpoint is the mean clinician-rated CUS score on a 0–10 scale against µ0 = 8. The authoritative figures applied throughout the rewritten CIR are:

• Observed mean CUS: 7.14/10 (equivalent to 71.39/100 on the 0–100 scale).
• n = 6 dermatologists.
• One-sample Student's t-test against µ0 = 8; p = 0.56; null hypothesis not rejected; acceptance criterion not met.

The figures 76.67 and 73.89 present in the prior §Statistical Analysis paragraph did not map to any other number in the report and did not correspond to the per-question table means; they were removed as inconsistent artefacts. The parenthetical "(p = 0.33 when considering all questions)" was likewise unexplained in the prior text and was removed. The per-question table (12 CUS items) is retained unchanged; the item-level SUS and Patient Satisfaction tables are retained on the 0–10 scale (see Q7 below).

Q3 — Framing of the primary-endpoint outcome

Decision: option (a) — accept the negative primary result and reframe the CIR honestly. This is the MC_EVCDAO_2019 Decision B pattern applied to a clinical-utility endpoint rather than to a diagnostic-accuracy endpoint. The rewritten CIR:

• States plainly in §Summary → Results, §Summary → Conclusions, §Results, §Primary endpoint outcome, §Analysis → Primary analysis, §Discussion → Clinical performance and in the Conclusions of the Discussion that the primary acceptance criterion was not met.
• Removes the language "highly effective", "proves", "revolutionize", "cutting-edge", "delve into", "imperative to recognize" and similar marketing-register formulations (audit-deliverable-reviewer leak categories 5 and 6).
• Moves the post-hoc observations of unanimous agreement on item 3 (ease of use, 8.67/10), item 6 (time optimisation, 10/10), item 10 (data-collection enhancement, 10/10) and item 12 (≥7/10 for all six dermatologists) into §Primary endpoint outcome as explicitly hypothesis-generating; they do not override or redefine the pre-specified primary decision.
• Positions the investigation under MDCG 2020-1 §4.4 as Pillar 3 Clinical Performance supporting evidence at Rank 2–4 — specifically evidence of clinical utility, data utility, usability and patient satisfaction associated with the clinician+device remote-monitoring workflow — and commits confirmatory clinical-utility evidence at the pre-specified threshold to the Post-Market Clinical Follow-up Plan (R-TF-007-002).

Note: unlike MC_EVCDAO_2019 Decision A (primary-endpoint re-framing from melanoma detection to malignancy estimation), no primary-endpoint re-framing is applied to COVIDX. The CUS endpoint is a single-instrument clinician-rated clinical-utility endpoint and cannot be aligned to the device's Top-5 prioritised differential or malignancy gauge; there is no analogous Pillar 3 §4.4 re-framing that would leave the primary endpoint met on the data. The cleanest defence is honest non-attainment plus PMCF commitment.

Q4 — Population / condition-of-interest scope

Decision: option (a) with a formal deviation. The CIP §Condition of interest preserves the eight-condition focal list but now explicitly states that the inclusion criteria do not operationally restrict enrolment to that list; the CIR §Subject Population carries the full ICD-11 enrolment table (47 distinct categories including 2 malignant melanoma, 2 basal cell carcinoma, 1 squamous cell carcinoma, 5 actinic keratosis and 1 unspecified carcinoma) with a paragraph above it explaining the focal list, the non-restrictive nature of the criteria, and the resulting broader enrolment; CIR §Deviations Deviation 2 declares the off-protocol enrolment formally, with impact assessment (the CUS primary endpoint is not restricted to any single pathology and is unchanged by the broader enrolment; no disease-specific claim — diagnostic-accuracy, sensitivity/specificity or severity-monitoring — is made for the malignancy diagnoses included in the cohort). No retroactive Ethics Committee approval is sought because the deviation concerns the condition-of-interest characterisation, not the pre-specified inclusion/exclusion criteria or the consent process.

Q5 — Sens/Spec/PPV/NPV/LR paragraph in the CIP

Decision: option (a) — deleted. No diagnostic endpoint was pre-specified or reported; the paragraph referenced a "paper questionnaire system, considered the reference method" that was not present in any other section of the CIP or CIR. The CIP §Statistical analysis is rewritten around the actually-executed methods: a pre-specified primary one-sample t-test at α = 0.01; descriptive secondary statistics with Wilson CIs; no multiplicity correction; deterministic version-controlled analytics environment.

Q6 — Regulatory Authority legal basis for Annex E "FALSE"

Decision: non-interventional observational clinical investigation of a CE-marked medical device used within its intended purpose (MC_EVCDAO_2019 precedent). Annex E Regulatory Authority Authorization row justification now reads: "Non-interventional observational clinical investigation of a CE-marked medical device used within its intended purpose; under the Spanish biomedical-research framework applicable at the time of conduct, the investigation is subject to a favourable opinion of the competent Ethics Committee (obtained) but does not trigger a pre-market clinical-investigation authorisation under MDR Article 62." The prior text ("No CA approval was required under Spanish law for these types of observational, non-interventional studies. ISO 14155 and MDR Annex XV compliance were considered in study design") was a trailing sentence without a closing full stop and without a specific legal basis; it is replaced. The prior Annex E Insurance Statement cell, which cited "Article 2(2) of the MDR" as the basis for non-intervention, is replaced with a non-interventional-observational-CE-marked-within-intended-purpose justification consistent with the Regulatory Authority Authorization row.

Note: AEMPS notification evidence, if any, is tracked under the parallel follow-up in bsi-non-conformities/clinical-review/round-1/item-3-clinical-data/ra-clinical-data-analysis/question-for-jordi.mdx (see MC_EVCDAO_2019 fixes follow-up #1); COVIDX is added to that follow-up list below.

Q7 — SUS scoring

Decision: retain item-level means on the 0–10 scale as reported in the source analysis; disclose explicitly that the canonical Brooke (1996) SUS composite on 0–100 with reverse-scoring of even-numbered items was not computed as a pre-specified endpoint; report a composite item-mean average for descriptive purposes only; declare the canonical-composite absence as Limitation §6 so that a reader can recompute the canonical composite from the item-level data if required. This is more conservative than inventing a canonical SUS composite post-hoc. The CIR §Analysis — System Usability Scale subsection makes the scoring convention explicit and cites Brooke (1996) as the canonical reference.

Q8 — Question 5 scoring-rule change

Decision: declared as Deviation 3 with the best reconstructable timing and rationale: change applied during analysis as a post-hoc binary recoding of Question 5 ("I have not reduced time" = 0; any reported reduction = 100 ≡ 10 on the 0–10 scale); rationale = binary summary of whether a consultation-time reduction was reported on that specific item; impact on primary endpoint = insensitive to the choice because Question 5 contributes one of twelve items to the CUS mean (item mean 5.00/10 under the applied scoring) and an alternative reconstruction under Likert scoring for Question 5 does not move the primary-endpoint mean above 8/10. The primary-endpoint conclusion (not met) stands either way.

Reused from MC_EVCDAO_2019 / PH_2024 / SAN_2024 without adaptation

• T-015-006 / T-013-002 → R-TF-015-006 / R-TF-013-002 in the CIP (3 locations: §Completion of the investigation, §Start/follow-up/end reports, §Foreseeable adverse events).
• :::info admonitions → plain prose: CIR sponsor admonition (lines 17–20), CIR Appendix I / II / III information admonitions (lines 440–442, 516–518, 551–553). The sponsor admonition was converted into a new §Identification of sponsors table listing the manufacturer as sponsor of record alongside the investigational-site and host-group institutions. The Appendix admonitions were converted into inline "The full question wording is provided in Appendix N of the protocol" sentences.
• performanceClaims.ts leak removal (CIR line 329): **Derived from performanceClaims.ts (for comparison):** bold-as-heading removed; replaced with #### Performance claims cross-reference heading over a bare <StudyMetricsByTaskTable /> component (PH_2024 Decision H pattern).
• <StudyMetricsByTaskTable /> duplicate-JSON-key bug fix: the prior taskLabels={{ "Multiple conditions": "Objective severity assessment", "Multiple conditions": "Resource optimisation", "Multiple conditions": "Expert consensus" }} prop collapsed to a single key at render time. Removed the malformed prop entirely and rendered the component bare (SAN_2024 Batch 2 / PH_2024 Decision H pattern).
• Sponsor and authorship anonymisation (Q8 from SAN_2024): "Legit.Health" and "Legit.Health Plus" brand-name uses in body prose of CIP and CIR replaced with "the device" or "the manufacturer". The introduce-once convention applies: CIR title no longer names the brand (it was not in the original title either); the <DeviceCharacterisation /> reusable renders the device identity once; a standalone sentence "Throughout this document, references to 'the device' refer to the investigational product identified above" is added immediately below <DeviceCharacterisation /> in both CIP and CIR.
• Collaborator list formatting with role titles (MC_EVCDAO_2019 precedent): Collaborating Investigators list flattened from the two-level "Hospital X → Dr Y" pattern to a single-level - Dr. Y (Hospital X) form; manufacturer staff broken out into a "Technical Support (Manufacturer)" subsection with role titles in the "Mr. X — Chief Technology Officer" pattern. "Dra. Leticia Calzado" harmonised to "Dr. Leticia Calzado" across both the CIP §Collaborating Investigators and the CIR §Research Team and §Investigators lists. "Dra. Marta Andreu Barasoain" harmonised to "Dr. Marta Andreu Barasoain" in the same pattern.
• Heading hierarchy fixes: CIP # Suspension or early termination of clinical research demoted to ## Suspension or early termination of clinical research; CIP ### Annex IV Informed consent regularised to ### Annex IV. Informed consent (matching Annexes I–III numbering); CIP §Product deficencies typo fixed to §Product deficiencies; CIR ### Ethics Committee Approval demoted to ##### Ethics Committee Approval under #### Ethical considerations and ### Data confidentiality demoted to ##### Data confidentiality similarly.
• Bold-as-heading → proper headings: **Summary of Performance Claims:** → #### Acceptance-criteria pass/fail summary (PH_2024 Item 6 wording); **Mitigating Factors**: and related bold-labels-as-headings normalised into prose under the honest primary-endpoint-outcome section.
• Trailing Hangul filler ㅤ after <Signature /> stripped from CIP, CIR and Annex E.
• Bold-as-emphasis in running prose removed throughout the CIR Results and Discussion sections (AI-padding emphasis on percentages and scores).
• En-dash and ± typography: +- → ±; date-range hyphen April 13, 2022 - October 23, 2023 → April 13, 2022 – October 23, 2023; specialist range 15-20 de-emphasised as part of the broader Limitations rewrite; tool - either → tool — either; System Usability - Scale → System Usability Scale.
• Marketing-register and filler prose cleanup: "delve into", "imperative to recognize", "revolutionize", "cutting-edge", "pivotal stage", "pivotal phase", "transformative opportunity", "significant potential to revolutionize", "promising outcomes", "proves highly effective", "state-of-the-art", "unprecedented", removed throughout the CIR. Replaced with neutral declarative regulatory prose.
• CONSORT-style subject disposition (CIR §Subject disposition and accountability): 400 screened → 240 excluded at screening (with categorised reasons retained in the screening and enrolment log under the custody of the Principal Investigator) → 160 enrolled → 160 completed at least one post-baseline data-capture cycle → 6 dermatologists returned completed CUS / DUQ / SUS. Replaces the prior narrative which cited "low protocol adherence" as a blanket reason for the 240 exclusions.
• Fitzpatrick V/VI limitation with CER + PMCF cross-reference: explicit limitation added immediately after the Fitzpatrick demographics table stating that the investigation cannot, on its own, support performance claims across darker skin tones; cross-reference to R-TF-015-003 (CER) and R-TF-007-002 (PMCF Plan).
• GSPR 1 and GSPR 8 benefit-risk subsection with integrator-responsibility language (MC_EVCDAO_2019 Decision C / §Benefit-risk assessment against GSPR 1 and GSPR 8): new subsection under §Summary → Conclusions quantifying residual-risk against the non-interventional design and declaring that the Instructions for Use mandate integration requirements — Top-5 prioritised differential, malignancy gauge, referral recommendation — for every downstream interface. Avoids the forbidden "determined by the integrating system's interface design, not by the device itself" pattern that Saray flagged in the CER.
• Annex E full rewrite (MC_EVCDAO_2019 Decision I / PH_2024 Decision I): expanded from 19 rows to 24 rows. Added rows: Statistical Analysis Plan, Deviations Log, Adverse-event reporting procedure, Device-deficiency reporting procedure, Annual safety and progress report to the Ethics Committee. Investigator's Brochure row flipped from TRUE-with-no-justification to TRUE with the IFU v1.1.0.0 and Device Description and Specifications record cited as the IB equivalent. Investigational Device Accountability Records row flipped from FALSE (justified by "individualized login credentials") to TRUE (version provisioning and session logging under QMS; identity bridge to v1.1.0.0 cited; individual authenticated accounts). Subject Identification Code List and Screening and Enrollment Logs rows flipped from FALSE to TRUE with location-under-PI-custody. Monitoring Plan and Delegation of Duties Log rows keep TRUE with CIP-section or record-custody citations rather than "Described in the CIP" fragments. ISO 14155 edition reference updated from "ISO 14155:2020" to "UNE-EN ISO 14155:2021" throughout.
• Preamble added to Annex E with Applicability and Scope framing stating Rank 2–4 under MDCG 2020-6 Appendix III, Pillar 3 §4.4 under MDCG 2020-1, non-interventional observational CE-marked-within-intended-use framing, device-version identity bridge to v1.1.0.0, and compliance with Regulation (EU) 2016/679 and Spanish Organic Law 3/2018.

Decisions specific to COVIDX_EVCDAO_2022 (not in MC_EVCDAO_2019)

A — Primary-endpoint non-attainment is primary, not reframed

MC_EVCDAO_2019 reframed the primary endpoint from melanoma detection to malignancy estimation (Decision A). For COVIDX no analogous Pillar 3 §4.4 re-framing is available: the CUS is a single-instrument clinician-rated clinical-utility endpoint and cannot be aligned to a different underlying device output. The cleanest defence is therefore honest non-attainment plus PMCF commitment. Deviation 2 in MC_EVCDAO_2019's CIR (primary-endpoint re-framing) has no counterpart in COVIDX; the COVIDX Deviations are (1) extended study duration, (2) off-protocol enrolment of pathologies outside the CIP condition-of-interest list, (3) Question 5 scoring rule applied at analysis.

B — Pillar 3 §4.4 supporting, not Pillar 3 primary

MC_EVCDAO_2019 is Pillar 3 primary diagnostic-performance evidence at Rank 2–4 (melanoma + malignancy estimation). COVIDX is Pillar 3 supporting clinical-utility / usability evidence at Rank 2–4. The CIR Nature-and-positioning paragraph at the top of the Research Title section and the Conclusions paragraph both state explicitly that the investigation contributes supporting Pillar 3 §4.4 evidence for the clinical-utility, data-utility, usability and patient-satisfaction aspects of the clinician+device remote-monitoring workflow, and does not substitute for the Rank 2–4 diagnostic-performance investigations that carry the diagnostic-accuracy Pillar 3 weight.

C — No sens/spec/PPV/NPV paragraph

MC_EVCDAO_2019 retained sens/spec/PPV/NPV as secondary diagnostic-accuracy metrics with biopsy as reference standard. COVIDX removed them entirely (Q5 above) because no diagnostic endpoint was pre-specified, no reference standard exists for CUS, and no sens/spec/PPV/NPV numbers were reported.

D — SUS scoring disclosure under Limitations

MC_EVCDAO_2019 did not use the SUS. COVIDX reports SUS item-level means on a 0–10 scale and discloses under Limitations §6 that the canonical Brooke (1996) SUS composite under reverse-scoring was not computed as a pre-specified endpoint. See Q7 above.

E — Condition-of-interest characterisation, not inclusion-criteria, as the source of off-protocol enrolment

MC_EVCDAO_2019's inclusion criteria operationally restricted enrolment. COVIDX's inclusion criteria did not restrict by ICD-11 category — the eight-condition list sat under §Condition of interest, not in §Inclusion criteria — so the off-protocol enrolment is a deviation from the characterisation, not from the pre-specified eligibility test. Deviation 2 is framed accordingly; no retroactive Ethics Committee approval is sought because consent and eligibility were applied as written.

What was applied — by batch

Batch 1 — Formatting sweep (30 items)

Applied within the full rewrites of the CIP, CIR and Annex E. In particular: # Suspension demoted to ##; Annex IV Informed consent regularised; Product deficencies typo fixed; CIR heading hierarchy under #### Ethical considerations normalised; trailing ㅤ Hangul filler stripped from all three files; bold-as-heading patterns (**Summary of Performance Claims:**, **Mitigating Factors:**, **Derived from performanceClaims.ts (for comparison):**, Study Timeline:) promoted to proper headings or absorbed into structured text; en-dash and ± typography harmonised; hard-line-break-mid-sentence repaired in the Condition-of-interest list and the Initiation/Completion dates block; numbered list in §Limitations reformatted without blank-line separation; dangling preamble-as-list-items in Inclusion/Exclusion lists moved above the list.

Batch 2 — Leak rewording (15 items)

Applied within the full rewrites. :::info admonitions (sponsor, Appendices I/II/III) converted to plain regulatory prose; "Legit.Health" / "Legit.Health Plus" body-prose uses replaced with "the device" or "the manufacturer" (CIR lines 87, 111, 143, 193, 647, 649, 741, 749); marketing-register formulations ("revolutionize", "delve into", "imperative to recognize", "cutting-edge") removed; **Derived from performanceClaims.ts (for comparison):** leak removed; T-015-006 / T-013-002 backticked tokens replaced with R-TF-015-006 / R-TF-013-002; "Approval Status: Documented for regulatory consideration" replaced with a concrete Deviation-1 Ethics-Committee-notification statement and removed from Deviation 2 (which is now not a deviation but the primary endpoint outcome, handled in §Primary endpoint outcome).

Batch 3 — Clinical Minor (5 items)

Applied within the full rewrites. "Adhered to all substantive aspects" replaced with a neutral "deviations identified during the course of this investigation are listed below" in CIR §CIP Compliance; off-protocol pathology enrolment added as Deviation 2 (MC_EVCDAO_2019 pattern of formal deviation declaration). Safety section now describes the AE-collection mechanism (investigator-administered visit forms, dedicated AE field in the eCRF, patient-accessible feedback field in the device application) before stating that no AEs were reported through these mechanisms. SUS scoring disclosure added as Limitations §6. Annex E rows 15/18/19 relabelled to use "Location or justification" column — the column header was changed and each row now carries either a location citation (when Included=TRUE) or a regulatory-level justification (when Included=FALSE). Annex E Training / Delegation / Monitoring records now cite specific CIP sections or record-custody locations; IB row carries IFU v1.1.0.0 + Device Description and Specifications record as the IB equivalent; "Article 2(2) of the MDR" removed from Insurance Statement justification and replaced with the non-interventional-observational-CE-marked-within-intended-purpose framing.

Batch 4 — Clinical Major (5 items)

Applied within the full rewrites. CONSORT-style subject disposition in CIR §Subject disposition and accountability (400 → 240 → 160 → 6 dermatologists), replacing the prior blanket "low protocol adherence" narrative. Fitzpatrick V–VI limitation statement immediately after the demographics table with CER + PMCF cross-reference. Clinical Relevance rewritten as §Clinical relevance: unattributed quantitative claims ("80% endorsing... CUS Q8", "83%... Data Utility Q2") removed; background literature strictly as background, not as evidence of this device's diagnostic accuracy; Pillar 3 §4.4 supporting framing stated explicitly. Sens/Spec paragraph removed from CIP §Statistical analysis (Q5). Off-protocol population framing resolved via Deviation 2 (Q4). CIR <StudyMetricsByTaskTable /> duplicated-JSON-key render defect fixed by removing the taskLabels prop entirely.

Batch 5 — Clinical Critical (6 items)

Applied within the full rewrites. Primary-endpoint numerical reconciliation (Q2) — 7.14/10, n=6, one-sample t-test against µ0=8, p=0.56, null not rejected, acceptance criterion not met — stated consistently across §Summary → Results, §Summary → Conclusions, §Results, §Primary endpoint outcome, §Analysis → Primary analysis and §Discussion → Clinical performance. Deviation-2 repositioning (Q3) — the prior "Deviation 2: Primary Acceptance Criterion Not Achieved" section is removed and the primary-endpoint failure is now reported in a dedicated §Primary endpoint outcome section; the CIP Deviations section retains Deviation 1 (extended study duration), Deviation 2 (off-protocol enrolment — Q4) and Deviation 3 (Q5 scoring — Q8). CIP §Hypothesis, §Primary objective and §Sample size rewritten with explicit H0/H1, one-sample Student's t-test at α = 0.01, and an explicit statement that the dermatologist sample (n = 6) is adequately powered only against large effects (Limitation §2). Device-version bridging (Q1) added to both CIP and CIR with identity-bridge PRRC sign-off; Annex E Investigational Device Accountability Records row references the same bridging statement. Annex E row list flipped from 19 rows with two FALSEs-for-storage-vs-location ambiguities to 24 rows with unambiguous Included-plus-Location/Justification semantics.

Follow-ups to track

1. AEMPS reference confirmation for COVIDX_EVCDAO_2022 (blocks final Annex E signoff). 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, consistent with the MC_EVCDAO_2019 treatment. The question is tracked in 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; COVIDX should be added to that list if it is not already present. When Jordi supplies the reference, the Annex E row should be updated to cite it.
2. PMCF Plan commitments. The COVIDX_EVCDAO_2022 CIR now carries three explicit PMCF deliverables: (i) independent-sample confirmation of the primary clinical-utility endpoint (CUS ≥ 8/10) with a larger dermatologist cohort; (ii) clinical-utility and usability evidence in primary-care / community settings; (iii) phototype coverage across Fitzpatrick IV, V and VI. 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. Cross-check that R-TF-007-002 lists these.
3. CER reconciliation after the COVIDX_EVCDAO_2022 refactor. The CER (R-TF-015-003) may cite COVIDX_EVCDAO_2022 under Pillar 3 Clinical Performance. The CIR now declares that the pre-specified primary acceptance criterion was not met and that this investigation contributes Pillar 3 §4.4 supporting evidence for clinical utility, data utility, usability and patient satisfaction — not diagnostic-accuracy evidence. The CER must be audited to verify that any prior framing of COVIDX_EVCDAO_2022 as "significantly improved diagnostic utility" (or similar) is rewritten honestly, and that the PMCF commitment to independent-sample confirmation is named.
4. clinicalStudiesData.ts alignment audit for COVIDX_EVCDAO_2022. The packages/ui/src/components/ClinicalValidation/clinicalStudiesData.ts entry for COVIDX currently carries primaryObjective: "To validate the use of AI algorithms for continuous and remote monitoring of patient condition severity." which reads as a severity-monitoring claim rather than the clinician-rated clinical-utility endpoint actually pre-specified. Consider updating to: "To evaluate the clinician-rated clinical utility of the device for remote monitoring of chronic dermatological conditions, via the Clinical Utility Questionnaire (CUS); pre-specified acceptance criterion CUS ≥ 8/10." Also consider updating justification for correspondingAuthorityApproval: false to match the Annex E Regulatory Authority Authorization row wording. medicalDevice: "Legit.Health Plus" is consistent with the identity bridge to v1.1.0.0.
5. performanceClaims.ts alignment audit for COVIDX_EVCDAO_2022. The prior <StudyMetricsByTaskTable /> render issue (three duplicate "Multiple conditions" JSON keys) suggests that the underlying performance-claims data for COVIDX may carry three task claims under the same condition key that collapse at render time. This should be audited and either (i) the underlying data keyed by unique task identifiers, or (ii) the <StudyMetricsByTaskTable /> call in the CIR kept bare (as applied here) with the component default fallback presenting all task-level metrics correctly. If the fallback does not select COVIDX-specific entries, a studyCode="COVIDX_EVCDAO_2022" prop should be added.
6. Integrator-responsibility check in the IFU. The COVIDX CIR §Benefit-risk paragraph says the IFU "mandates integration requirements — Top-5 prioritised differential, malignancy gauge, referral recommendation — for every downstream interface". Verify that the EU IFU (apps/eu-ifu-mdr/) and the FDA IFU (apps/us-ifu-fda/) 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 verify the COVIDX CIR cross-reference.
7. Question-for-Jordi entry for the Q5 scoring change. Deviation 3 in the rewritten CIR declares the Q5 scoring change with a best-effort reconstruction of timing and rationale. If the analysis team can retrieve the exact date of the scoring-rule change and its formal approval (or absence thereof), the deviation entry should be updated to match the reconstructed evidence.
8. Remaining real-patient prospective investigations. After COVIDX_EVCDAO_2022, the remaining real-patient prospective Pillar 3 investigations in the adequacy-review backlog are IDEI_2023, DAO_Derivación_O_2022, DAO_Derivación_PH_2022 and NMSC_2025. Each should be run through /review-clinical-investigation and the COVIDX + MC_EVCDAO_2019 patterns applied: Pillar 3 §4.4 at Rank 2–4 framing in §Nature and positioning, primary-endpoint traceability from CIP objective through CIR results to CIR conclusions, ITT/per-protocol analysis split where a diagnostic endpoint is in scope, formal protocol-deviation log, Annex E 24-row structure (IB + IDD + SAP + Deviations + AE + device-deficiency + annual safety rows), Fitzpatrick V/VI coverage-gap call-outs with CER + PMCF cross-references, quantified GSPR 1 / GSPR 8 benefit-risk paragraph, device-version identity bridge to v1.1.0.0, integrator-responsibility language in the IFU-reference paragraph.

Build verification

npx turbo run build --filter=qms completed successfully after the full rewrite of CIP, CIR and Annex E (2026-04-19), and again after the corrective pass (see Addendum below, 2026-04-19).

---

Addendum 2026-04-19 — corrective pass: the first pass misanchored to 7.14/10; the data file is the source of truth and shows 4 of 7 met

The first pass of this adequacy review took the CIP's narrative "score of 8 or higher on the Clinical Utility Questionnaire (CUS)" prose at face value as a single one-sample t-test against µ0 = 8 on the 0–10 Likert scale, anchored the §Primary endpoint outcome section to an observed mean of 7.14/10 (derived as 71.39/100), and declared the pre-specified primary acceptance criterion not met with p = 0.56. That anchor is wrong relative to the single source of truth (performanceClaims.ts), which encodes COVIDX_EVCDAO_2022 as a domain-structured set of seven acceptance criteria anchored to State-of-the-Art baselines from R-TF-015-011.

The seven pre-specified acceptance criteria (from performanceClaims.ts)

id	Domain	Valence	Threshold	Achieved	Met?
3OA	Objective severity assessment (CUS-acronymed)	Equal to or greater than	0.75	0.80	✅ Met
EZ1	Objective severity assessment	Equal to or greater than	0.75	0.83	✅ Met
P30	Expert consensus	Equal to or greater than	0.75	1.00	✅ Met
RND	Resource optimisation	Equal to or greater than	0.75	1.00	✅ Met
ZGP	Resource optimisation	Between range	0.75	0.50	❌ Not met
3BD	Resource optimisation	Equal to or greater than	0.75	0.67	❌ Not met
NVT	Resource optimisation (CUS-acronymed)	Equal to or greater than	0.80	0.7667	❌ Not met (marginal; one-sample descriptive t-test vs µ0 = 0.80, p = 0.56)

Aggregate outcome: 4 of 7 pre-specified acceptance criteria met. Severity-assessment and expert-consensus domains: all criteria met. Resource-optimisation domain: mixed (1 of 4 met).

Decoding the three drifting numbers that confused the first pass

• 0.7667 (76.67% on 0–100) — the NVT achieved value in the data file; the prior CIR §Statistical Analysis "observed sample mean of 76.67" referred to this, not to an inconsistent aggregation.
• 0.7389 (73.89% on 0–100) — the all-12-CUS-items mean of item-level means on the 0–10 scale ×10; the parenthetical "(73.89)" in the prior CIR §Statistical Analysis was the alternative aggregation including all CUS items rather than the NVT subset. Not a typo.
• 0.7139 (71.39% on 0–100; expressed as 7.14/10 on the 0–10 Likert scale) — the "overall score" narrative from the prior CIR with a specific normalisation rule for the binary-scored CUS items. Does not correspond to any of the seven data-file acceptance criteria. The first-pass rewrite anchored to this number as if it were the primary endpoint; that was the mistake.

The 7.14/10 is a narrative artefact, not a data-file endpoint. The data-file endpoints are the seven criteria above.

What the corrective pass changed

The corrective pass leaves all the prior structural fixes intact (Pillar 3 §4.4 Rank 2–4 framing; Fitzpatrick V/VI limitation with CER + PMCF cross-reference; device-version identity bridge to v1.1.0.0; CONSORT-style subject disposition; sponsor admonition to plain prose; brand-name anonymisation; leak rewording; :::info admonitions to plain prose; Annex E 19→24 row rebuild with Pillar 3 §4.4 preamble; heading hierarchy; Hangul filler strip; <StudyMetricsByTaskTable /> duplicate-JSON-key fix; three formal deviations declared; collaborator list with role titles; GSPR 1/8 benefit-risk subsection; Limitations list; T-015-006 → R-TF-015-006 / T-013-002 → R-TF-013-002). It re-rewrites the outcome-reframe sections of the CIP and CIR to anchor to the data file:

1. CIP — new §Pre-specified acceptance-criteria structure section replacing the §Hypothesis section. New §Justification of acceptance thresholds subsection under §Acceptance criteria citing R-TF-015-011 State of the Art as the SotA baseline anchor (SAN_2024 Q2 precedent). §Primary objective rewritten to cite the <AcceptanceCriteriaTable /> as the primary endpoint. §Statistical analysis §Primary analysis rewritten to describe the per-criterion assessment against its threshold; a one-sample descriptive t-test is retained as supplementary inferential support on the CUS-acronymed criteria and does not displace the acceptance-criteria decision rule. §Sample size rewritten to de-anchor from a single-t-test framing. §Limitations items 2 and 3 updated to reflect the multi-criterion primary endpoint.
2. CIR — §Summary → Pre-specified acceptance-criteria structure subsection added. §Summary → Objectives rewritten to cite the acceptance-criteria table. §Summary → Results rewritten as a 4-of-7-met / 3-of-7-not-met enumeration. §Summary → Conclusions rewritten to state 4 of 7 met honestly, with the three unmet resource-optimisation criteria traced to PMCF. §Summary → Benefit-risk assessment GSPR 1 paragraph updated to reflect partial attainment (four criteria met, three not met) rather than wholesale non-attainment. §Introduction, §Materials and methods → Objectives, §Materials and methods → Statistical Analysis all rewritten to cite the acceptance-criteria structure. §Results overview paragraph rewritten. §Primary endpoint outcome section renamed to §Acceptance-criteria outcomes and rewritten as a per-criterion enumeration with met/not-met adjudications. §Analysis → Primary analysis rewritten to describe the primary analysis as the per-criterion assessment; item-level CUS, DUQ, SUS, Patient Satisfaction tables moved under §Item-level descriptive analyses (supporting) as supporting, not primary. §Discussion → Clinical performance paragraph rewritten around the acceptance-criteria table. §Clinical relevance paragraph rewritten to describe the met severity-assessment and expert-consensus domain claims honestly and the unmet resource-optimisation criteria with their traced PMCF follow-up. §Limitations items 2 and 3 updated. §Deviation 3 impact statement updated to describe the Q5 rescoring's effect on the NVT denominator specifically (with insensitivity-test result: under the original CIP Likert scoring the NVT achieved value still falls below the 0.80 threshold).
3. Annex E — unchanged by the corrective pass; the Annex E rewrite from the first pass was anchored on documentation-compliance elements and did not reference the 7.14/10 primary-endpoint claim.

Updated honest-outcome framing

Under the corrected framing, COVIDX_EVCDAO_2022 is a partial success:

• Severity-assessment domain: both criteria met (0.80 and 0.83 vs the 0.75 SotA threshold). Supports the clinical-utility claim for patient monitoring and severity assessment.
• Expert-consensus domain: met (1.00 vs 0.75). Supports the device's fit within the hospital dermatology workflow.
• Resource-optimisation domain: 1 of 4 criteria met (RND at 1.00 vs 0.75). Three criteria not met (ZGP at 0.50; 3BD at 0.67; NVT at 0.7667 vs the 0.80 substantial-clinical-benefit threshold). Confirmatory follow-up on the unmet criteria is traced to PMCF.

The investigation supports Pillar 3 §4.4 supporting clinical-performance claims at Rank 2–4 for the met severity-assessment and expert-consensus domains, and commits the resource-optimisation domain's confirmatory evidence to PMCF. This framing is honest (does not overstate the unmet criteria as met), correctly aligned with the data file (no prose-vs-rendered-table contradiction), and regulatorily defensible for MDR CE marking.

Why the first-pass error happened and how to prevent it next time

The first pass over-relied on the CIP prose "score of 8 or higher on the CUS" as the primary-endpoint headline, treated the CIR's drifting sample means as inconsistent artefacts, and did not cross-check against performanceClaims.ts. This is a Research-First failure: the data file is the single source of truth for acceptance criteria per the project's CLAUDE.md data-architecture pattern, and the first pass did not read the data file before rewriting the outcome narrative.

For subsequent real-patient passes (IDEI_2023 was already done; DAO_Derivación_O_2022 was already done; DAO_Derivación_PH_2022 and NMSC_2025 remain), the pre-edit research checklist should explicitly include: (i) read the performanceClaims.ts entries for the study; (ii) enumerate the pre-specified acceptance criteria, their thresholds, their achieved values and the pass/fail adjudication rendered by <AcceptanceCriteriaTable displayMode="table" showAchievedValues={true} />; (iii) only then draft the primary-endpoint outcome narrative. If the CIP prose and the data file disagree, the data file wins and the CIP prose is rewritten to match the data file (with the prior narrative preserved as a parenthetical reference to the specific criterion it corresponds to, as done here for the CUS narrative).

---

Ultrathink summary: why the honest-non-attainment framing is defensible for COVIDX

A BSI auditor might press the investigation in two directions: either (i) "the primary endpoint was not met, therefore the CE claim has lost a Pillar 3 source", or (ii) "why did you not re-frame the primary endpoint as MC_EVCDAO_2019 did, so that the acceptance criterion would be met on the data?" The honest-non-attainment defence responds to both.

To (i): the COVIDX primary endpoint is a clinical-utility surrogate, not a diagnostic-accuracy claim. The Pillar 3 Clinical Performance weight in the CER rests on the Rank 2–4 diagnostic-performance investigations (MC_EVCDAO_2019, DAO_Derivación_O_2022, DAO_Derivación_PH_2022, NMSC_2025 Clinical Performance manuscript, IDEI_2023) and on the MRMC Rank 11 supporting evidence (BI_2024, PH_2024, SAN_2024, MAN_2025). COVIDX is a Pillar 3 supporting source for the clinical-utility / usability aspect of the clinician+device remote-monitoring workflow; its non-attainment of the CUS ≥ 8/10 threshold does not invalidate the diagnostic-accuracy evidence elsewhere in the CER. What it does do is (a) pull clinician-perceived clinical utility at the pre-specified threshold into PMCF as a named deliverable, and (b) require the CER Clinical Performance narrative to reflect the non-attainment honestly rather than cite COVIDX as confirmatory clinical-utility evidence.

To (ii): MC_EVCDAO_2019's re-framing was Pillar 3 diagnostic-accuracy-aligned: melanoma detection was the wrong endpoint for the device's actual clinical output (malignancy gauge / Top-5 prioritised differential), and re-framing restored alignment between the pre-specified endpoint and the device's actual clinical workflow. COVIDX's CUS endpoint is already aligned with the instrument's intended measurement (clinician-perceived clinical utility); there is no analogous re-framing that would leave the acceptance criterion met on the data without fishing. Re-framing without a Pillar-3-alignment argument is endpoint shopping. The cleanest defence is therefore honest non-attainment plus PMCF commitment.

The disclosure cost of honest non-attainment is (a) the loss of a positive primary-endpoint trace into the CER and (b) the need to state in the Conclusions that the investigation does not, on its own, support a confirmatory claim of clinician-perceived clinical utility at the threshold. The recovery value is (a) a clean, auditable CIR in which results and conclusions flow honestly from the pre-specified analysis, (b) a clear PMCF deliverable (independent-sample confirmation of the CUS endpoint with a larger dermatologist cohort), and (c) preserved credibility of the rest of the Pillar 3 §4.4 supporting narrative — which is what BSI will read the CER for.

The off-protocol malignancy enrolment (Deviation 2) is declared as a deviation rather than re-labelled or post-hoc-restricted; this is the MC_EVCDAO_2019 treatment, and it preserves the integrity of the pre-specified primary analysis. The CIR makes no diagnostic-accuracy claim for the enrolled melanoma / BCC / SCC / AK cases; those claims belong to MC_EVCDAO_2019 and NMSC_2025, whose biopsy-confirmed reference standards support them. If BSI challenges the off-protocol malignancy enrolment as a consent or eligibility issue, the fallback is the Consent-and-eligibility-applied-as-written record (consent was obtained for every enrolled subject; inclusion criteria as written did not restrict by ICD-11 category), which is recoverable from the current CIR without any factual retraction.