Response

We acknowledge that BSI correctly identified a traceability gap in R-TF-013-002 for risk R-DAG ("The medical device outputs a wrong result"), per ISO 14971:2019 clause 7.2 (verification of implementation of risk control measures). The mitigationRequirements field contained infrastructure SRS codes identical to causeRequirements, and the verificationOfImplementation test cases verified only the transport layer (API port, HTTP status codes, JSON format), not the clinical output mitigations BSI flagged.

The implementations exist and have been verified — the gap was in the traceability documentation, not in the actual risk controls. We have corrected R-TF-013-002 to establish complete traceability from mitigation → requirement → test case → result for each implemented mitigation.

Mitigation-by-mitigation traceability (R-DAG)

Mitigation 1: "Information about device outputs are detailed in the IFU"

Element	Reference
Mitigation requirement	LR-4XK (Read IFU before use), LR-9WR (Device outputs interpretation guidance), LR-4RZ (Warnings and precautions), LR-8YN (Device supervision requirement)
IFU location	apps/eu-ifu-mdr/versioned_docs/version-1.1.0.0/installation-manual/user-interface.mdx — Full JSON output structure including probability distributions, entropy scores, explainability media, clinical indicators, severity scores
IFU location	apps/eu-ifu-mdr/versioned_docs/version-1.1.0.0/troubleshooting/clinical.mdx — Interpretation of distributions, entropy as uncertainty measure, top-5 accuracy approach
Verification	Labeling requirements verification documented in R-TF-012-037; complete evidence provided in M2 Q2 response

Mitigation 2: "The medical device returns metadata about the output that helps supervising it, such as explainability media and other metrics"

Element	Reference
Mitigation requirement	SRS-0AB (Generate per-image ICD analysis with explainability heat map), SRS-K7M (Orchestrate diagnosis support workflow with pixel-level attention indicators)
Test case	C256 (T123): Verify response includes per-image ICD probabilities and heat maps for top five categories — validates explanation.attentionMap objects, colour model data, Base64-encoded image data
Test case	C265 (T132): Verify diagnosis workflow returns ranked ICD-11 codes, binary indicators, and explainability maps — validates entropy, pixel-level attention indicators (heat maps/saliency masks)
Test execution	R-TF-012-033 Software Tests Plan — all tests passed
Model-level verification	AI Models Integration Tests (T307-T379, C466-C539) verify each AI model produces correct explainability outputs

Mitigation 3: "The device returns an interpretative distribution representation of possible ICD categories, not just one single condition"

Element	Reference
Mitigation requirement	SRS-Q3Q (Generate aggregated ICD probability distribution from a set of images), SRS-K7M (Compute normalized probability vector across all supported ICD-11 categories)
Test case	C255 (T122): Verify API returns aggregated ICD probability distribution with structured code details — validates hypotheses array with numeric probability fields, valid ICD-11 code structures
Test case	C265 (T132): Verify diagnosis workflow returns ranked ICD-11 codes — validates top-5 ranked ICD-11 categories, probability sum = 100% across full distribution, entropy, five binary indicators
Test execution	R-TF-012-033 Software Tests Plan — all tests passed
Model-level verification	AI Models Integration Tests (T307-T379, C466-C539) verify each AI model produces correct probability_distribution and icd_distribution data

Mitigation 4: "AI models are subject to retraining under expanded datasets"

This is a prospective lifecycle control, not a runtime software feature. It has no software-level test cases because it is verified through QMS process adherence, not through runtime tests.

Element	Reference
Process definition	GP-028 AI Development, § AI Updates → Retraining: "Retraining is performed when an algorithm's core logic or data foundation is modified"
Change governance	GP-023 Change Control: classifies retraining as minor or major AI model version change
Verification mechanism	R-TF-028-010 AI V&V Checks: mandatory verification before any retrained model is released
Documentation	R-TF-028-007 AI Retraining Report: mandatory output of any retraining activity

Note: No retraining has been performed for v1.1.0.0 (no completed R-TF-028-007 record exists). Retraining is a prospective control triggered by PCCP criteria (e.g., post-market data indicating performance drift, new training data available). The mitigation statement in R-TF-013-002 has been reworded to accurately reflect this prospective nature, per ISO 14971:2019 clause 7.2 note on risk control measures that may include "inherent safety by design, protective measures, or information for safety."

Systematic audit of all risks

BSI noted: "It is unclear if other risks are similarly impacted." A systematic audit of all 62 risks in R-TF-013-002 was performed. 29 risks were identified with traceability gaps analogous to R-DAG. They fall into three categories:

Category A: Infrastructure/API risks with process-level mitigations (21 risks)

These risks had mitigationRequirements identical to causeRequirements with infrastructure-only verification. The mitigations are process-level controls (security best practices, SOUP analysis, QMS procedures) that require process-level verification references, not just software test cases.

Risk ID	Risk name	Process-level verification added
R-T8Q	Data transmission failure from HCP system	Security and availability techniques per R-TF-012-006; error handling in API documentation
R-3N5	Data input failure	Security and availability techniques per R-TF-012-006; error handling in API documentation
R-YF4	Data accessibility failure	Security and availability techniques per R-TF-012-006; error handling in API documentation
R-LRP	Data transmission failure	FHIR interoperability per IFU; added LR-5TG, LR-7XP to mitigationRequirements
R-MWD	Interruption of service	Elastic scaling and backup infrastructure per R-TF-012-006; REST protocol error handling
R-OM1	Data overwrite	REST protocol architecture per R-TF-012-006; request immutability is inherent design feature
R-B63	Inconsistent or unreliable output	Algorithm V&V per GP-012; representative dataset validation per R-TF-028-010
R-VL1	Device failure or performance degradation	Elastic scaling infrastructure per R-TF-012-006; error handling in API documentation
R-72D	SOUP anomaly/incompatibility	SOUP analysis per R-TF-012-023; compatibility evaluation per GP-012; design review records in DHF
R-MQ1	SOUP not maintained nor patched	SOUP monitoring and patching process per R-TF-012-006; SOUP records in R-TF-012-023
R-QLF	Non-compliance with GSPR	GSPR compliance per R-TF-001-006 GSPR Checklist; design per harmonized standards per R-TF-012-006
R-ES8	Absence of risk management process	ISO 14971 implementation per GP-013; risk management records in R-TF-013-001 and R-TF-013-002
R-C6Q	Absence of PMS & PMCF process	PMS plan per R-TF-018-001; PMCF plan per R-TF-015-002
R-27M	Inadequate maintenance	Maintenance activities per GP-012; SOUP maintenance per R-TF-012-006
R-9SS	SOUP cybersecurity vulnerabilities	SOUP analysis per R-TF-012-023; cybersecurity evaluation per SP-012-002; design review records in DHF
R-33B	Electronic IFU tampered	Git workflow with GPG-signed commits per GP-012; RBAC and branch protections in repository configuration
R-GY6	Inaccurate training data	Image selection and HCP labeling process per GP-028; dataset quality records in R-TF-028-001
R-7US	Biased or incomplete training data	Image selection and HCP labeling process per GP-028; dataset diversity records in R-TF-028-001

Category B: Security risk with missing SRS codes (1 risk)

Risk ID	Risk name	SRS codes added	Test cases added
R-HH0	Electronic data and content tampered	SRS-1KW, SRS-SDZ, SRS-WER, SRS-WGF	C332, C333, C343, C344, C345, C351, C352, C353, C354

Category C: Retraining mitigation with no traceability (5 risks)

These risks include AI model retraining as an implemented mitigation but had no process-level verification reference:

Risk ID	Risk name	Corrective action
R-DAG	Wrong result (ICD distribution)	The original BSI finding — added SRS-Q3Q, SRS-0AB, SRS-K7M; C255, C256, C265; retraining process refs (GP-028, GP-023, R-TF-028-010)
R-75H	Incorrect clinical information	Added SRS-0AB, SRS-K7M; C256, C265; retraining process refs; reworded mitigation to prospective form
R-SKK	Incorrect results shown to patient	Fixed typo "retarining" → "retraining"; added SRS-Q3Q, SRS-0AB, SRS-K7M; C255, C256, C265; retraining process refs
R-75L	Stagnation of model performance	Reworded mitigation to prospective form; added retraining process refs (GP-028, GP-023, R-TF-028-010)
R-PWK	Degradation of model performance	Reworded mitigation to prospective form; added manual retraining process refs (GP-028, GP-023, R-TF-028-010)

All affected risks have been corrected in the updated R-TF-013-002.

Summary of changes to R-TF-013-002

R-DAG (Risk: "The medical device outputs a wrong result")

Field	Change
implementedMitigations[3]	Reworded from "AI models undergo retraining using expanded dataset of images" to "AI models are subject to retraining under expanded datasets as governed by GP-028 (§ AI Updates → Retraining) and GP-023 (Change Control), with verification through R-TF-028-010 (AI V&V Checks) before any retrained model is released."
mitigationRequirements	Added SRS-Q3Q, SRS-0AB, SRS-K7M (kept existing infrastructure codes alongside)
verificationOfImplementation	Added test cases C255 (T122), C256 (T123), C265 (T132); added reference to AI Models Integration Tests (T307-T379, C466-C539); added process-level references for retraining (GP-028, GP-023, R-TF-028-010)

R-75H (Risk: "Incorrect clinical information")

Field	Change
implementedMitigations[2]	Reworded retraining statement to prospective form
mitigationRequirements	Added SRS-0AB, SRS-K7M
verificationOfImplementation	Added test cases C256 (T123), C265 (T132); added AI Models Integration Tests reference; added process-level retraining references

R-SKK (Risk: "Incorrect results shown to patient")

Field	Change
implementedMitigations[3]	Corrected typo "retarining" → "retraining"; reworded to prospective form
mitigationRequirements	Added SRS-Q3Q, SRS-0AB, SRS-K7M
verificationOfImplementation	Added test cases C255 (T122), C256 (T123), C265 (T132); added AI Models Integration Tests reference; added process-level retraining references

R-HH0 (Risk: "Electronic data and content are tampered")

Field	Change
mitigationRequirements	Added SRS-1KW (TLS), SRS-SDZ (hashed passwords), SRS-WER (OAuth), SRS-WGF (AES-256 encryption)
verificationOfImplementation	Added test cases C332, C333 (TLS), C343, C344, C345 (auth), C351, C352, C353 (OAuth), C354 (encryption)

R-75L (Risk: "Stagnation of model performance")

Field	Change
implementedMitigations	Reworded to prospective form with explicit governance references (GP-028, GP-023, R-TF-028-010)
verificationOfImplementation	Added process-level retraining references

R-PWK (Risk: "Degradation of model performance")

Field	Change
implementedMitigations	Reworded to clarify manual-only retraining with explicit governance references (GP-028, GP-023, R-TF-028-010)
verificationOfImplementation	Added process-level retraining references

Category A risks (21 risks with process-level mitigations)

For all Category A risks (R-T8Q, R-3N5, R-YF4, R-LRP, R-MWD, R-OM1, R-B63, R-VL1, R-72D, R-MQ1, R-QLF, R-ES8, R-C6Q, R-27M, R-9SS, R-33B, R-GY6, R-7US):

Field	Change
verificationOfImplementation	Added process-level verification references to QMS procedures and records where mitigations are documented
mitigationRequirements	For R-LRP: Added LR-5TG, LR-7XP (FHIR IFU documentation)

Regulatory compliance

The corrective actions address:

• ISO 14971:2019 clause 7.2: Verification of implementation of risk control measures — complete traceability chain now documented
• ISO 14971:2019 clause 7.6: Completeness of risk control — systematic audit confirmed all analogous risks have been corrected
• ISO 14971:2019 clause 7.4: Benefit-risk analysis — conclusions unchanged by traceability corrections
• GSPR 1: Intended performance — mitigations (explainability, distributions, IFU) ensure device output supports HCP decision-making as intended
• GSPR 4: Risk management system per Annex I §3 — traceability chain is now complete
• GSPR 17.2: Diagnostic accuracy — ICD probability distribution and explainability media are the mechanisms by which accuracy/precision are communicated to HCP
• Annex II 5(b): Description and justification of residual risks — R-TF-013-002 demonstrates residual risks acceptable after verified controls
• Annex II 6.1(a)/(b): Evidence of GSPR compliance — verification test cases now clearly map to mitigations
• Annex II 6.2(f): Risk analysis including risk control measures — complete traceability chain fulfils this requirement

Note on traceability matrix (R-TF-012-001)

The Software Requirements Specification (SRS) codes and test cases added to R-TF-013-002 already existed in the traceability matrix R-TF-012-001 (apps/qms/docs/legit-health-plus-version-1-1-0-0/design-and-development/R-TF-012-001.json). Specifically:

• SRS-Q3Q, SRS-0AB, SRS-K7M (clinical output mitigations) → already mapped to test cases C255, C256, C265
• SRS-1KW, SRS-WER, SRS-SDZ, SRS-WGF (security mitigations) → already mapped to test cases C332, C333, C343, C344, C345, C351, C352, C353, C354

The gap identified by BSI was a referencing error in the Risk Management Record (R-TF-013-002), not a missing implementation or verification gap. The requirements were implemented, the tests were executed, and the traceability was documented in R-TF-012-001 — the Risk Record simply failed to reference the correct codes.

R-TF-012-001 has not been modified as part of this corrective action because no changes to the requirement-to-test traceability were necessary.

Documents modified

Document	Path	Changes
R-TF-013-002 Risk Management Record	apps/qms/docs/legit-health-plus-version-1-1-0-0/risk-management/R-TF-013-002.json	Corrected traceability for 29 risks: (1) R-DAG, R-75H, R-SKK — added clinical mitigation SRS codes and test cases; (2) R-HH0 — added security SRS codes and test cases; (3) R-75L, R-PWK — reworded retraining mitigations to prospective form; (4) 21 Category A risks — added process-level verification references; (5) R-LRP — added LR codes for FHIR documentation; (6) R-GY6, R-7US — fixed incomplete mitigation text; (7) Fixed R-SKK typo "retarining" → "retraining"

Documents not impacted

Document	Path	Reason
R-TF-012-001 Traceability Matrix	apps/qms/docs/legit-health-plus-version-1-1-0-0/design-and-development/R-TF-012-001.json	SRS codes and test cases already existed; corrective action was referencing, not implementation