医疗相关 ISO 标准准备与证据整理

正文

references/iso-14971.md

Research basis: 2026-10-01. This reference summarizes a preparation process and evidence architecture for medical device risk management. It does not reproduce requirements and is not a substitute for the standard.

Copyright and authorized access

ISO publications are copyrighted. Obtain ISO 14971 and ISO/TR 24971 from ISO, an ISO national member, or another authorized source; see ISO copyright. Do not ask an agent to retrieve, transcribe, summarize clause-by-clause, or store proprietary text.

Current edition, the guidance companion, and the European amendment

ISO 14971:2019 is Edition 3, published December 2019, reviewed and confirmed in 2025.
ISO/TR 24971:2020 is Edition 2 and is the guidance companion, following the same clause structure. It is a Technical Report: informative, not normative. Conformity is claimed to ISO 14971, never to ISO/TR 24971, and a TR cannot relax or reinterpret a normative requirement. Record which guidance the organization adopted and why. ISO marks the TR for revision; ISO/AWI TS 24971-1 is stage 20.00, under development. Keep the published 2020 TR as the guidance baseline until an authorized change review; a work item is not a published replacement.
EN ISO 14971:2019/A11:2021 is a European amendment to the EN adoption, cited for MDR and, through the corresponding decision, IVDR support. It does not change the international normative part. Track it only where the European source basis applies.

This profile supports risk evidence inside a named assurance lane

ISO 14971 is a device risk-management process standard. This skill does not define a standalone management-system certification scheme for it. Risk evidence supports other assessments, for example:

an ISO 13485 certification audit examines the risk-management process as part of the QMS;
an FDA inspection examines risk-based decisions against applicable FDA requirements;
an MDSAP audit treats risk management as a foundation across its process areas; and
EU conformity assessment examines the risk-management file as part of technical documentation against the applicable general safety and performance requirements.

If a third party supplies a certificate or assessment referencing ISO 14971, record its actual issuer, scheme, scope, and limitations. Do not treat its label as ISO 13485 certification, product approval, or a regulator decision. Report risk evidence as supporting the declared lane, without issuing an assurance result.

Hard boundary

This skill and its files cannot:

certify, approve, or validate a risk-management file or process;
determine device classification, intended use, reportability, conformity route, market access, or product authorization;
decide whether a risk is acceptable, whether benefit outweighs residual risk, or whether a risk-control measure is adequate or safe;
set or approve risk-acceptability criteria, severity or probability scales, or a risk matrix;
replace top management, the risk-management owner, qualified clinical and engineering reviewers, RA/QA, a notified body, or a regulator; or
infer that a hazard set is complete, or that residual risk is acceptable, from a template, a matrix, a document count, or a script result.

Use outputs as a list of evidence questions for accountable human review. Completeness of hazard identification cannot be established by any check in this skill — a zero-finding structural result says the declared links are present, not that the analysis found everything.

Process and evidence domains

The iso-14971 profile carries these domain labels for manifests and gap reports. They are workflow topics, not clause references:

risk-management-plan, risk-management-file, competence-and-authority, intended-use-and-characteristics, hazard-identification, risk-estimation-and-evaluation, risk-control-option-analysis, risk-control-implementation-and-verification, risk-control-side-effects-review, residual-risk-evaluation, benefit-risk-analysis, overall-residual-risk-evaluation, risk-management-review-and-report, production-and-postproduction-information, change-control.

Each domain needs an owner, status, evidence IDs, source/version reference, recorded approval, and links to open gaps.

The chain that has to hold together

Risk-management evidence is judged as a connected chain, not as a set of documents. For each analysed item, the following must be traceable in both directions:

intended use and reasonably foreseeable misuse → characteristics related to safety → hazard → foreseeable sequence of events → hazardous situation → harm with severity → probability basis → risk evaluation against declared criteria → risk-control option analysis and decision → implemented control → verification that the control was implemented → verification of its effectiveness → review for new or increased risks introduced by the control → residual risk evaluation → benefit-risk analysis where residual risk is not acceptable → contribution to overall residual risk → disclosure of residual risk → production and post-production information feeding back.

Break any one link and the file stops being evidence. The two links most often missing are verification of control effectiveness (distinct from verifying the control exists) and review of risks introduced by the control itself.

Build the traceability register with:

PYTHONDONTWRITEBYTECODE=1 python3 scripts/check_traceability.py \
  /path/to/traceability-matrix.json

That check links intended use, hazard or signal, risk evaluation, risk control, design input and output, verification, validation, production control, and post-market source. It verifies declared linkage, not analytical adequacy.

Where risk management meets the QMS

ISO 14971 work is not separable from the device QMS. Keep these joins explicit and evidenced:

Design and development — risk controls that are design outputs must appear as design inputs with verification, not only as risk-file rows.
Production — controls realized as process parameters, in-process checks, or acceptance criteria need production evidence, and process validation where the control cannot be fully verified on the finished device.
Purchasing — controls that depend on a supplier's process or component need supplier controls and change notification. See scripts/check_supplier_controls.py.
Post-market — complaints, service records, vigilance signals, and production nonconformities must actually re-enter risk estimation, with evidence that the file was reassessed and that changed estimates were acted on.
CAPA — where a risk estimate changes, the resulting action needs effectiveness evidence before closure. See scripts/check_capa.py.
Change control — any change to intended use, design, process, supplier, or materials triggers a documented risk reassessment.

The production and post-production feedback loop is the single most common structural gap: files that were complete at design transfer and never updated by field experience.

Risk-acceptability criteria and the matrix problem

Acceptability criteria belong in the risk-management plan, are set by authorized humans with a documented rationale, and must be applied consistently. Recurring failure modes to look for, none of which this skill can adjudicate:

criteria written after the analysis so that every risk lands acceptable;
probability values asserted with no stated basis and no data;
a matrix whose cells were adjusted per project to move results across the boundary;
risk reduction claimed from labelling or an instruction alone, where a design or protective measure was available;
residual risk described as acceptable with no benefit-risk analysis where the declared criteria were not met;
overall residual risk never evaluated as a whole — only item by item; and
no disclosure of residual risk to users where the analysis assumed it.

Flag these as blockers for authorized review. Do not resolve them.

Shared checks that apply here

scripts/validate_scope_intake.py --standard iso-14971 — declared scope, product families, intended use references, and applicability owners.
scripts/check_traceability.py — the risk/design/production/post-market chain.
scripts/check_capa.py — corrective action with effectiveness evidence.
scripts/check_supplier_controls.py — controls that depend on external providers.
scripts/validate_evidence_manifest.py and scripts/gap_analyzer.py with --standard iso-14971.

Sources

ISO 14971:2019 — catalogue entry
ISO/TR 24971:2020 — guidance companion
Decision (EU) 2022/757 — EN ISO 14971:2019/A11:2021 citation basis
references/iso-13485.md — the QMS lane this evidence usually supports
references/source-ledger.md — dated baseline and provenance limitations
references/assurance-lanes.md — the declared assurance purpose and limits