Evidence rank × pillar matrix and clinical validation phases

Internal scratch-pad reference

This page is a working reference for drafting BSI Round 1 responses. It is not shipped to BSI and must not be referenced from any document under docs/legit-health-plus-version-*/. Audit-visible rank/pillar tables live in the CEP (R-TF-015-001, §"Planned evidence classification per study" and §"Clinical Evidence") and in the CER (R-TF-015-003); this page consolidates them into a single glanceable view and adds the clinical-validation phase progression that the CEP refactor introduces.

Methodological framing: Rank and Pillar are orthogonal

The MDCG 2020-6 Appendix III rank (1 strongest → 12 weakest) and the MDCG 2020-1 pillar (1 Valid Clinical Association, 2 Technical Performance, 3 Clinical Performance) are independent axes. This was the central framing agreed with external consultant Celine Horiana on 2026-04-17 and translated internally by Saray Ugidos on 2026-04-18.

Concretely:

• A study can be Rank 11 Pillar 3 (MRMC simulated-use reader study contributing supporting Pillar 3 §4.4 evidence) or Rank 2 Pillar 3 (prospective real-patient pivotal investigation contributing primary Pillar 3 evidence). The pillar does not change with the rank.
• A study cannot be "Rank 11 Pillar 2". MRMC studies measure clinicians using the device — that is Pillar 3 Clinical Performance, not Pillar 2 Technical Performance, regardless of how the simulated-use design ranks on the evidence hierarchy.
• Pillar 2 Technical Performance evidence comes from the API-level analytical claim across the 346 ICD-11 categories — i.e. AI model V&V and the published severity-validation literature (APASI / AUAS / AIHS4 / ASCORAD).

The indirect-benefit defence for Class IIb scrutiny is a continuous causal chain: Pillar 1 (VCA literature) → Pillar 2 (analytical performance) → Pillar 3 (clinical performance, primary + supporting) → surrogate endpoints (diagnostic accuracy, referral optimisation, severity scoring) → patient benefit. Every link must be defensible; every gap must be either filled or declared as an acceptable gap per MDCG 2020-6 §6.5(e).

Integration surface. Pillar 3 evidence is measured on the Top-5 prioritised differential view and the malignancy-prioritisation output the device mandates via IFU integration requirements. The device is validated only when integrated per these specifications; Pillar 3 claims do not delegate UI responsibility to integrators.

Matrix

Summary cross-tab (Rank × Pillar)

Counts of evidence-source rank assignments at each (Rank × Pillar) cell. Rank and Pillar are orthogonal axes per the MDCG 2020-1 / MDCG 2020-6 methodology agreed with Celine Horiana (2026-04-17): a Rank 11 study is not a Pillar 2 study — MRMC reader studies are Rank 11 Pillar 3 §4.4 supporting evidence.

MDCG 2020-1 Pillar → MDCG 2020-6 Rank ↓	Pillar 1: Valid Clinical Association	Pillar 2: Technical Performance	Pillar 3: Clinical Performance	Safety confirmation	Row total
Rank 2: High-quality clinical investigations with some gaps	—	—	6	—	6
Rank 4: Studies with methodological limitations, data still quantifiable	—	—	14	—	14
Rank 5: Equivalence data (reliable / quantifiable)	—	4	1	—	5
Rank 6: Evaluation of state of the art	1	—	—	—	1
Rank 7: Complaints and vigilance data; curated QMS data	—	—	—	1	1
Rank 8: Proactive PMS data (surveys / professional opinion)	—	—	1	—	1
Rank 11: Simulated-use testing with healthcare professionals (MRMC)	—	—	5	—	5
Column total	1	4	27	1	33

Rank-ordered evidence table

One row per (source × rank assignment). Sources with split rank/pillar/tier classifications (for example, a mixed prospective/retrospective design at Rank 2 and Rank 4) appear once per assignment.

Rank 2: High-quality clinical investigations with some gaps

Source	Kind	Pillar	Tier	Phase	Benefit	Role	Note
MC_EVCDAO_2019	investigation	Pillar 3	Tier 1: Malignancy (individual)	P2	7GH 3KX	primary	Pillar 3 primary anchor for melanoma (AUC / sensitivity / specificity).
MC_EVCDAO_2019	investigation	Pillar 3	Tier 3: General conditions (pooled)	P2	7GH	primary
COVIDX_EVCDAO_2022	investigation	Pillar 3	Tier 3: General conditions (pooled)	P2	7GH 3KX	primary	CUS 7.14/10 non-attainment transparently declared; Rank 2 preserved.
DAO_Derivation_O_2022	investigation	Pillar 3	Tier 1: Malignancy (individual)	P2	7GH 3KX	primary	Referral-pathway / malignancy AUC ≥ 0.8.
DAO_Derivation_O_2022	investigation	Pillar 3	Tier 3: General conditions (pooled)	P2	3KX	primary
IDEI_2023	investigation	Pillar 3	Tier 1: Malignancy (individual)	P2	7GH 3KX	primary	Prospective arm at Rank 2; malignancy primary endpoint met on the N=8 prospective malignant subset.

Rank 4: Studies with methodological limitations, data still quantifiable

Source	Kind	Pillar	Tier	Phase	Benefit	Role	Note
IDEI_2023	investigation	Pillar 3	Tier 3: General conditions (pooled)	P2	7GH 5RB	primary	Retrospective arm at Rank 4; androgenic alopecia Ludwig agreement failed honestly (Kappa 0.33).
DAO_Derivación_PH_2022	investigation	Pillar 3	Tier 1: Malignancy (individual)	P2	7GH 3KX	primary	Protocol deviation affecting primary endpoint; data still quantifiable.
DAO_Derivación_PH_2022	investigation	Pillar 3	Tier 3: General conditions (pooled)	P2	7GH	primary
AIHS4 2025	investigation	Pillar 3	Severity assessment	P2	5RB	supporting	Pivotal study per CEP §Confirmatory phase; severity assessment with a limited evidence base (N=2 patients, 16 assessments). Further indication-specific confirmation planned via PMCF Activities B.1–B.5.
R-TF-015-012 — cross-sectional observational study of the equivalent legacy device	pms-corpus	Pillar 3	All tiers	P4	7GH 3KX	supporting	Quantitative outcomes at Rank 4; Likert professional-opinion items at Rank 8.
NMSC_2025	publication	Pillar 3	Tier 1: Malignancy (individual)	P6	7GH	supporting	Specialist-clinic malignancy publication; higher BCC/cSCC prevalence than primary-care context — appraised accordingly.
triaje_VH_2025	investigation	Pillar 3	Tier 3: General conditions (pooled)	P5	7GH 3KX	supporting	Planned PMCF — triage confirmatory study.
CVCSD_VC_2402	investigation	Pillar 3	Tier 3: General conditions (pooled)	P5	7GH 3KX	supporting	Planned PMCF — care-pathway confirmation.
clinical_VH_2025	investigation	Pillar 3	Tier 3: General conditions (pooled)	P5	7GH 5RB	supporting	Planned PMCF — general clinical performance confirmation.
AFF_EVCDAO_2021	investigation	Pillar 3	Severity assessment	P5	5RB	supporting	Planned PMCF — severity-assessment external validation.
acne	investigation	Pillar 3	Severity assessment	P5	5RB	supporting	Planned PMCF — acne severity external validation.
aEASI_HVN	investigation	Pillar 3	Severity assessment	P5	5RB	supporting	Planned PMCF — atopic dermatitis severity external validation.
AGM_2026	investigation	Pillar 3	Tier 3: General conditions (pooled)	P5	7GH	supporting	Planned PMCF — general-performance monitoring.
PMCF-ICD-DXP-2026	investigation	Pillar 3	Tier 3: General conditions (pooled)	P5	7GH	supporting	Planned PMCF — ICD-11 coverage / dark phototype external validation.

Rank 5: Equivalence data (reliable / quantifiable)

Source	Kind	Pillar	Tier	Phase	Benefit	Role	Note
Equivalence with legacy device (MDCG 2020-5)	pms-corpus	Pillar 3	All tiers	P4	7GH 5RB 3KX	supporting	Technical, biological and clinical equivalence per Article 61(5)–(6). Load-bearing route bringing legacy pivotal data and the Rank 7 PMS corpus into scope.
APASI_2025	publication	Pillar 2	Severity assessment	P6	5RB	primary	PASI severity — Technical Performance against expert consensus. Rank 5 drawn from scientificPublicationsData.ts `mdcgRank`; CEP §965 confirms Pillar 2 classification.
AUAS_2023	publication	Pillar 2	Severity assessment	P6	5RB	primary	UAS severity — Technical Performance against expert consensus. Rank 5 from scientificPublicationsData.ts `mdcgRank`; CEP §965 confirms Pillar 2.
AIHS4_2023	publication	Pillar 2	Severity assessment	P6	5RB	primary	IHS4 severity — Technical Performance against expert consensus. Rank 5 from scientificPublicationsData.ts `mdcgRank`; CEP §965 confirms Pillar 2.
ASCORAD_2022	publication	Pillar 2	Severity assessment	P6	5RB	primary	SCORAD severity — Technical Performance against expert consensus. Rank 5 from scientificPublicationsData.ts `mdcgRank`; CEP §965 confirms Pillar 2.

Rank 6: Evaluation of state of the art

Source	Kind	Pillar	Tier	Phase	Benefit	Role	Note
R-TF-015-011 — Systematic State of the Art review (Valid Clinical Association)	pms-corpus	Pillar 1	All tiers	P0	—	primary	Consolidates published evidence anchoring the clinical association between AI-assisted dermatological diagnosis / severity scoring and patient outcomes. Primary source for Pillar 1 VCA; referenced throughout the CEP's Evidence quality hierarchy and the CER's Current knowledge — State of the Art section.

Rank 7: Complaints and vigilance data; curated QMS data

Source	Kind	Pillar	Tier	Phase	Benefit	Role	Note
Legacy predecessor passive PMS corpus (2020–present)	pms-corpus	Safety	All tiers	P4	—	supporting	≈ 4 500 reports; 0 MDR Article 87 incidents; 0 FSCAs; 7 non-serious complaints. Consolidated in the legacy device PMS Report (R-TF-007-003).

Rank 8: Proactive PMS data (surveys / professional opinion)

Source	Kind	Pillar	Tier	Phase	Benefit	Role	Note
R-TF-015-012 — cross-sectional observational study of the equivalent legacy device	pms-corpus	Pillar 3	All tiers	P4	7GH 3KX	supporting	Proactive PMS — physician-reported outcomes across 21 client institutions; Holm-Bonferroni corrected.

Rank 11: Simulated-use testing with healthcare professionals (MRMC)

Source	Kind	Pillar	Tier	Phase	Benefit	Role	Note
BI_2024	investigation	Pillar 3	Tier 2: Rare diseases (grouped)	P3	7GH	supporting	GPP + rare-disease MRMC; child-band tinea corporis automation-bias traced to R-DAG + R-HAX.
BI_2024	investigation	Pillar 3	Tier 3: General conditions (pooled)	P3	7GH	supporting
PH_2024	investigation	Pillar 3	Tier 3: General conditions (pooled)	P3	7GH 3KX	supporting	Primary metric reconciled to 63.70 → 81.85 (+18.15 pp) post-adequacy review.
SAN_2024	investigation	Pillar 3	Tier 3: General conditions (pooled)	P3	7GH 3KX	supporting
MAN_2025	investigation	Pillar 3	Tier 3: General conditions (pooled)	P3	7GH	supporting	Fitzpatrick V–VI external-validity anchor. Phase 1 concordance blocker (62.4 % vs ≥ 70 %) disclosed.

Phase legend

• P0: Non-clinical prerequisite
• P1: Exploratory / proof-of-concept
• P2: Pre-market confirmatory (real-patient pivotal)
• P3: Pre-market supporting (MRMC simulated-use)
• P4: Equivalence route (legacy PMS + observational)
• P5: Post-market clinical follow-up (PMCF)
• P6: Published literature supporting the three pillars

Clinical validation phases

The CEP organises all evidence into six phases. The diagram below shows which sources populate which phase and how they feed the three-pillar causal chain.

Reading the diagram

• Arrows encode the causal chain, not time. Phase 0/6 establish the non-clinical and literature anchors that Phase 2 and Phase 3 build on. Phase 4 is the equivalence route that brings the legacy predecessor's pivotal and PMS evidence into the device's scope. Phase 5 is post-market — it does not back-fill pre-market sufficiency; it confirms the pre-market determination per MDCG 2020-6 §6.3 and §6.4.
• MAN_2025 is the Fitzpatrick V–VI anchor. It is ongoing and its completion, combined with the declared-acceptable §6.5(e) gap for dark phototypes, is the primary external-validity claim.
• R-TF-015-012 appears in Phase 4 with two rank assignments. Its quantitative outcomes contribute Rank 4 evidence; its Likert professional-opinion items contribute Rank 8 proactive PMS evidence. The same physical study produces two distinct entries in the matrix above.
• PMCF activities are listed under the activity codes from R-TF-007-002 (A.1–A.3 triage, B.1–B.5 severity, C.1–C.2 core monitoring, D/E/F paediatric-phototype-autoimmune-genodermatoses). Individual planned investigations are listed separately where applicable.

How to use this page when drafting a BSI response

1. Identify the finding's scope — is BSI asking about a rank, a pillar, a specific disease tier, or a specific benefit?
2. Use the summary cross-tab to confirm the count of evidence at that (rank × pillar) cell. If it is zero, the response cannot claim evidence there without either citing a different cell or declaring an acceptable gap.
3. Drill into the rank-ordered table to list the specific sources. The note column carries the one-liner you most likely need to address in the response (e.g. "CUS 7.14/10 non-attainment declared" for COVIDX).
4. Cross-check against the causal chain in the mermaid diagram — if the response leans on a Pillar 3 claim, verify that Pillar 1 and Pillar 2 have continuous evidence upstream. Gaps are the primary BSI pressure point for Class IIb indirect-benefit defence.

Change control

This page reads from clinicalStudiesData.ts and scientificPublicationsData.ts at build time — if a study code or publication ID is removed upstream, the integrity check in clinicalEvidenceRankMap.ts fails the TypeScript build. When the CEP changes a rank or pillar assignment (for example, reclassifying a study from Rank 4 to Rank 2), update the corresponding row in clinicalEvidenceRankMap.ts; the matrix re-renders on the next build.