The risk clause that comes after every hazard clears

Clause 6 and Clause 8 are not the same acceptability test

Clause 6 asks whether the risk of a specific hazardous situation, before controls, is already acceptable against the criteria the risk management plan set out; where it isn't, Clause 7's risk control measures reduce it and the same per-hazard question gets asked again of what's left over. Every one of those judgments is local — one hazard, one residual risk, one acceptability call. Clause 8 runs after every one of those local judgments has already closed. It asks whether the overall residual risk of the finished device, all hazards considered together, is still acceptable — a question no amount of individually-acceptable rows in the risk table answers by itself.

Why the individual risks can't just be added up

Clause 8's evaluation isn't arithmetic. Each hazard's residual risk carries a different harm, a different severity, and a probability estimated with a different degree of confidence — some from field data, some from bench testing, some from expert judgment alone. Summing numbers that aren't measuring the same thing produces a number, not an evaluation. What Clause 4.4 requires instead is that the risk management plan define, in advance, the actual method the file will use to reach this holistic judgment and the criteria that judgment will be checked against — and that method doesn't have to match the criteria used for any individual hazard. ISO/TR 24971 sketches several acceptable approaches without prescribing one, which puts the burden on the file to state, and defend, whichever method it picked.

When the overall picture fails, the benefit has to carry it

If that holistic evaluation concludes the overall residual risk isn't acceptable, Clause 8 doesn't stop at a finding. It requires the manufacturer to gather and review data on the expected medical benefit of the device's intended use, and to determine whether that benefit outweighs the overall residual risk the evaluation just found. This is the step where ISO 14971:2019's formal definitions of “benefit” and “medical benefit” — new in the 2019 edition — actually get used, rather than sitting in the definitions clause unreferenced. Where the benefit doesn't outweigh the risk, the device isn't acceptable at this level regardless of how each individual hazard scored on its own.

A file needs a visible answer, not an inferred one

Because Clause 8 is a separate step with its own method and its own criteria, it needs its own visible conclusion in the risk management file — not a note that the risk table is complete and every row passed. An auditor checking the file against the standard is checking for a specific, discrete record: the method used, the criteria applied, the conclusion reached, and, where relevant, the benefit-risk comparison that resolved it. A risk management report that lists every hazard's residual risk as acceptable and never states the overall conclusion separately is missing a record the standard asks for by name, not a general impression the rest of the file is expected to imply.

Where this meets the rest of the file

This is a different rule than the one the EU's own “as far as possible” duty imposes on individual risk reduction, and the two shouldn't be read as versions of each other: MDR and IVDR Annex I still forbid cost from justifying a stop in reducing any single hazard's risk further, full stop, and Clause 8's benefit-weighing doesn't reopen that door for individual hazards — it's a distinct, later judgment about the assembled device, not a route back to cost-based reasoning at the hazard level. And the discipline both clauses share — a stated method, checked against stated criteria, with the record kept separately from the hazard-by-hazard table — is the same structure a risk management file already has to carry hazard by hazard, applied one level up.

A risk management report template that keeps the overall residual risk evaluation as its own visible section, distinct from the per-hazard table and stating its own method and conclusion, is previewed in the launch catalog. If your program runs this evaluation differently, the shelf takes that correction directly.