Task 3b-5: Autoimmune and Genodermatoses Triangulated-Evidence Package

STATUS (2026-04-19): In progress. Triangulation strategy defined; execution today across five parallel work-streams. Replaces the current CEP/CER §6.5(e) "acceptable gap" + "passive surveillance" framing for autoimmune diseases (Gap 4) and genodermatoses (Gap 5) with a sufficient-clinical-evidence argument under MDCG 2020-6 §6.3, supported by four evidence sources (Pillar 1 literature, Pillar 2 algorithm V&V, Rank 7 legacy PMS, Rank 11 focused MRMC) and confirmed by pre-specified PMCF.

Purpose

Resolve the pre-submission finding A.2.C5 from pre-submission-review-2026-04-19-cep-cer.md: the current §6.5(e) declaration for autoimmune and genodermatoses is backed only by "passive surveillance," which violates MDCG 2020-6 §6.4 (PMCF cannot fill pre-certification evidence gaps) and contributes to a five-gap over-use pattern that weakens every §6.5(e) claim by association.

Decision: keep autoimmune and genodermatoses in the device's intended use (do not narrow the indication). Re-frame the regulatory argument from "Pillar 3 gap → MRMC fills it" (weak, Rank 11 alone) to "sufficient clinical evidence per MDCG 2020-6 §6.3 via multi-source triangulation, with PMCF post-market confirmation" (defensible).

Traceability: A.2.C5 in the pre-submission review · Item 3b in the action plan · CEP lines 900–908 (R-TF-015-001) · CER §Representativeness (R-TF-015-003) · Kanban card: i3b_autoinmunes_genodermatosis.

---

Background

The autoimmune + genodermatoses gap

The current CEP declares two §6.5(e) acceptable gaps for these sub-indication categories:

• Gap 4 (benefit 7GH, sub-criterion a — indication coverage): Autoimmune diseases evidence coverage; approximately 3% of legacy PMS presentations.
• Gap 5 (benefit 7GH, sub-criterion a — indication coverage): Genodermatoses evidence coverage; approximately 1% of legacy PMS presentations.

Both gaps are currently "addressed through passive surveillance through PMS/PMCF data collection." Under MDCG 2020-6 §6.4, pre-certification evidence must be sufficient prior to CE marking — passive surveillance is not a pre-cert evidence source. This is the direct violation flagged by BSI clinical reviewers Erin Preiss and Nick.

The §6.5(e) over-use pattern

The pre-submission review identified five §6.5(e) declarations in the CEP:

1. Autoimmune diseases (this task)
2. Genodermatoses (this task)
3. Fitzpatrick V–VI (separate: task-3b4-mrmc-dark-phototypes)
4. Pillar 3 Clinical Performance for severity assessment (separate review gap)
5. Paediatric population (separate review gap)

Five is too many. Individual declarations weaken by association; the pattern reads to reviewers as indication over-ambition. Resolving Gaps 4 and 5 via triangulation reduces the count and concentrates the remaining declarations where they are genuinely narrow and bounded.

Why MRMC alone does not close §6.4

The "fast MRMC" intuition is the obvious first option, but it fails regulatorily on three layered grounds:

1. Rank hierarchy. MRMC is Rank 11 per MDCG 2020-6 Appendix III. The CEP's own text (line 801) acknowledges MRMC "is not clinical data under the strict MDR Article 2(48) definition." Closing a Clinical Performance gap with evidence the document says is not clinical data is self-defeating.
2. Erin/Nick Round 1 precedent. BSI reviewers pushed back on MRMC-as-primary for dark phototypes (captured in task-3b4-mrmc-dark-phototypes/CLAUDE.md and in the pre-submission review A.2.C6). The same logic applies uniformly — Rank 11 cannot substitute for Pillar 3 real-patient evidence on any named sub-indication.
3. §6.5(e) over-use. An MRMC-backed sixth §6.5(e) declaration reinforces the over-use pattern rather than resolving it.

MRMC can and should be part of the package — but as supporting evidence, not as the gap-closing element. That is the regulatorily correct framing.

Why the structural advantages of this sub-population favour triangulation

Autoimmune dermatoses and genodermatoses have structural features that make a triangulated argument genuinely defensible even without large prospective studies:

• Low prevalence. Large prospective studies are infeasible within reasonable timelines; MDCG 2020-6 §6.3 permits weighting toward alternative evidence sources when design constraints are justified.
• Image-based clinical recognition is an accepted standard for these categories in the dermatology literature (Pillar 1 VCA is easy).
• The device reads images. The test conditions of MRMC (HCPs reading images, assisted vs. unassisted) closely match real clinical workflow for these image-diagnosable conditions.
• Legacy real-world data exists. The 250,000+ legacy report corpus includes presentations from these categories — Rank 7 real-world equivalent-device evidence via Article 61(5)–(6).

---

Strategy: multi-source triangulation, not MRMC alone

Re-frame the regulatory argument from gap-filling to sufficient-evidence-via-triangulation. The package has five ingredients; each alone is insufficient, the combination is defensible.

Ingredient 1 — Pillar 1 Valid Clinical Association (literature)

Targeted SotA literature review (≥10 peer-reviewed references per category group) establishing that image-based clinical recognition of autoimmune dermatoses (lupus, dermatomyositis, pemphigus, bullous pemphigoid, lichen planus, morphea, vasculitis) and genodermatoses (ichthyoses, epidermolysis bullosa, neurofibromatosis cutaneous manifestations, tuberous sclerosis, Gorlin syndrome) is an accepted clinical standard with published diagnostic performance benchmarks.

Appended to R-TF-015-011 State of the Art as a new section. Produces the Pillar 1 VCA summary.

Ingredient 2 — Pillar 2 Technical Performance (algorithm V&V)

Per-category algorithm performance metrics extracted from the existing curated image-labelling dataset, filtered for autoimmune and genodermatoses examples. Report sensitivity, specificity, AUC, Top-1 and Top-5 accuracy for each category group.

Surfaced in the CER as a named Technical Performance sub-analysis. Traceable to the existing V&V documentation (R-TF-028-*).

Ingredient 3 — Rank 7 legacy passive PMS data (Article 61(5)–(6) equivalence)

Filter the legacy 250,000+ report corpus for autoimmune / genodermatoses presentations. Summarise:

• Case volumes per category group
• Complaint and incident rates (with rule-of-three upper bounds where zero events)
• HCP-reported feedback where captured
• Time distribution across the 2020–present window

Reported as Rank 7 real-world equivalent-device evidence flowing into Plus's CER via the MDCG 2020-5 equivalence route (see pre-submission-review-2026-04-19-cep-cer.md A.2.C3, A.2.C4 for the Route A equivalence framing).

Ingredient 4 — Rank 11 fast focused MRMC

A dedicated MRMC study on autoimmune and genodermatoses images with a pre-specified protocol. Purpose: provide supporting Pillar 3 §4.4 evidence that intended users (HCPs) achieve clinically relevant outputs on images representative of these sub-populations.

Methodological floor (below these, the MRMC adds no value and will be picked apart):

• ≥30 images per category group (target ≥50–100 combined per group)
• ≥5 reader HCPs, mix of primary-care and dermatology
• Pre-specified protocol with acceptance criteria before data collection begins
• Comparator: unassisted vs. device-assisted diagnostic performance
• Endpoints: diagnostic accuracy (sensitivity / specificity / Top-1 / Top-5), user-concordance, clinical utility
• Ground truth: atlas-label or expert panel consensus

Additive, not load-bearing. The MRMC strengthens the package; it does not carry it alone.

Ingredient 5 — Pre-specified PMCF activity

A PMCF activity committing to post-market prospective collection with pre-specified thresholds:

• Enrolment target per category group (e.g., N autoimmune cases, M genodermatoses cases within T months post-CE)
• Pre-specified diagnostic-accuracy thresholds
• Pre-specified user-concordance thresholds
• Trigger conditions for unscheduled CER update if thresholds breached

Framing discipline: PMCF "confirms" / "strengthens" an adequately-evidenced base (permitted under MDCG 2020-6 §6.3). PMCF never "fills" / "closes" a pre-cert gap (forbidden under §6.4). The wording of the PMCF commitment in the CEP / PMCF Plan must use "confirms" and "strengthens," never "fills" or "closes."

Narrowed CLAIM language (not narrowed INDICATION)

Keep autoimmune and genodermatoses in the intended use. Frame the device's output for these specific sub-categories at a level the triangulated evidence supports:

"For autoimmune dermatoses and genodermatoses, the device provides probability rankings within its broader ICD-11 output distribution. These outputs should be interpreted as supporting information within the healthcare professional's differential diagnosis workup. Final diagnosis for these categories is based on clinical evaluation and histopathological examination as per the current standard of care, not on the device's probability ranking alone."

This language goes in the IFU (via the Intended Purpose reusable in packages/reusable/snippets/) and in the CEP / CER benefit descriptions for the 7GH sub-criteria affected.

---

§6.5(e) four-test reframing (replaces lines 900–908 of CEP)

Replace the current passive-surveillance framing with an explicit four-test analysis:

1. Is the gap narrow and bounded? Yes — autoimmune 3% + genodermatoses 1% = ~4% of presentations combined.
2. Does the core benefit-risk conclusion depend on this gap evidence? No — the three declared benefits (7GH Diagnostic Support, 5RB Severity Assessment, 3KX Clinical Workflow) are independently evidenced for the remaining 96% of presentations by the six pivotal clinical investigations and the legacy PMS / RWE corpus.
3. Is there adequate residual evidence for the claim at these sub-categories? Yes:
   • Pillar 1: literature review (this task, Ingredient 1)
   • Pillar 2: algorithm V&V per category (this task, Ingredient 2)
   • Rank 7: legacy PMS (this task, Ingredient 3)
   • Rank 11: focused MRMC (this task, Ingredient 4)
4. Is PMCF planned to address remaining uncertainty? Yes — pre-specified PMCF activity with enrolment targets and diagnostic-accuracy / user-concordance thresholds (this task, Ingredient 5).

The visible four-test structure distinguishes a defensible §6.5(e) declaration from a hand-wave. Reviewers can check each test; each test is independently evidenced.

---

Work-streams (parallel execution)

All five work-streams run in parallel today.

Work-stream 1 — Pillar 2 extraction

1. Query the curated labelling dataset for autoimmune and genodermatoses examples.
2. Compute per-category sensitivity, specificity, AUC, Top-1, Top-5 accuracy.
3. Produce evidence-package/pillar-2-algorithm-vv.md with the table and narrative.
4. Cross-reference R-TF-028-* V&V records.

Work-stream 2 — Legacy PMS filter

1. Query the legacy 250,000+ report corpus for autoimmune / genodermatoses keywords and ICD-11 codes.
2. Deduplicate, categorise, and summarise.
3. Apply rule-of-three upper bounds where zero events observed.
4. Produce evidence-package/rank-7-legacy-pms-filter.md with the Rank 7 summary.
5. Cross-reference R-TF-007-003 legacy PMS Report.

Work-stream 3 — Literature review

1. Targeted SotA search: AI / image-based diagnostic recognition of autoimmune dermatoses and genodermatoses.
2. Screen titles/abstracts against CRIT1-7 appraisal.
3. Extract ≥10 references per category group with diagnostic performance benchmarks.
4. Produce evidence-package/pillar-1-literature.md.
5. Appendix to R-TF-015-011 State of the Art.

Work-stream 4 — Fast focused MRMC

1. Pre-specify the protocol (CIP-style) — design, population, endpoints, acceptance criteria, statistical analysis plan.
2. Curate the image set (≥50–100 images per category group; atlas-sourced, anonymised).
3. Recruit reader HCPs (≥5–8; primary-care + dermatology mix).
4. Execute the reader study (unassisted → device-assisted on the same image set).
5. Analyse and report Rank 11 evidence.
6. Produce evidence-package/mrmc-protocol.md and evidence-package/mrmc-results.md.

Note: this MRMC is a separate study from MAN_2025 (task-3b4-mrmc-dark-phototypes). They address different sub-indications (autoimmune/genodermatoses vs. Fitzpatrick V–VI) and cannot share images. Co-location in the Investigation folder under a new study code is expected once the protocol is drafted.

Work-stream 5 — CEP/CER edits + PMCF spec + claim language

1. Rewrite CEP lines 900–908 as the four-test §6.5(e) analysis backed by the four evidence sources.
2. Pre-specify the PMCF activity in R-TF-007-002 PMCF Plan with enrolment targets and thresholds.
3. Update the Intended Purpose reusable (packages/reusable/snippets/intendedPurpose.*) and IFU with the narrowed-claim language for these sub-categories.
4. Cross-reference the Risk Management File (R-TF-013-002) if any risks are affected by the narrowed-claim framing.
5. Produce four-test-rewrite.md, pmcf-activity-spec.md, narrowed-claim-language.md.

---

Deliverables

Produced in this folder:

• evidence-package/pillar-1-literature.md
• evidence-package/pillar-2-algorithm-vv.md
• evidence-package/rank-7-legacy-pms-filter.md
• evidence-package/mrmc-protocol.md
• evidence-package/mrmc-results.md
• four-test-rewrite.md — prose for CEP §900–908 and CER §Representativeness replacement
• pmcf-activity-spec.md — PMCF activity specification with thresholds
• narrowed-claim-language.md — IFU and reusables wording for these sub-categories

Propagated into audit-visible documents:

• CEP (R-TF-015-001) — lines 900–908 replaced with four-test analysis
• CER (R-TF-015-003) — §Representativeness and §Sufficiency determination updated with triangulated-evidence narrative
• SotA (R-TF-015-011) — literature review appendix added
• PMCF Plan (R-TF-007-002) — activity with pre-specified thresholds added
• Intended Purpose reusable (packages/reusable/snippets/intendedPurpose.*) — narrowed-claim language for these sub-categories
• IFU (apps/eu-ifu-mdr/) — consequential update via the reusable
• CIP for the focused MRMC (R-TF-015-004) — new study code
• CIR for the focused MRMC (R-TF-015-006) — new study code

---

What NOT to do

• Do not run a fast MRMC and declare the gap closed by itself — BSI will reject, citing Rank 11 insufficiency (Erin/Nick Round 1 precedent).
• Do not add a sixth §6.5(e) declaration without the four-test structure — the over-use pattern becomes the finding.
• Do not frame PMCF as "filling" or "closing" the gap — PMCF "confirms" / "strengthens" an adequately-evidenced base. Wording discipline matters.
• Do not under-design the MRMC because it is "just supporting" — methodological flaws in supporting evidence are worse than no supporting evidence. Apply the methodological floor (≥30 images per category, ≥5 readers, pre-specified protocol).
• Do not narrow the indication. The strategy keeps autoimmune and genodermatoses in scope. What is narrowed is the CLAIM language for these sub-categories, not the indication.
• Do not disclose any internal implementation detail (dataset extraction scripts, API integration, folder paths, engineering tooling) in the audit-visible documents. Internal mechanics stay in this task folder.

---

Cross-references

• Pre-submission review: ../../pre-submission-review-2026-04-19-cep-cer.md §A.2.C5 (primary driver of this task)
• Related review findings: A.2.C3 (equivalence), A.2.C4 (Route A regulatory pathway), A.2.C6 (MRMC-as-primary for other sub-populations)
• Related tasks:
  • ../task-3b4-mrmc-dark-phototypes/CLAUDE.md — Fitzpatrick V–VI MRMC (MAN_2025); methodological precedent for supporting-evidence MRMC
  • ../task-3b2-3b3-legacy-rwe-study/CLAUDE.md — legacy RWE study (R-TF-015-012); methodological precedent for Rank-appropriate legacy evidence
• QMS documents: R-TF-015-001 CEP · R-TF-015-003 CER · R-TF-015-011 SotA · R-TF-007-002 PMCF Plan · R-TF-007-003 legacy PMS Report · R-TF-013-002 RMF
• Regulatory: MDR Article 61, Article 61(5)–(6) · MDCG 2020-1 §4.4 · MDCG 2020-5 §A2.1 · MDCG 2020-6 §6.3, §6.4, §6.5(e), Appendix III