Regression testing as a standard software quality practice — re-running verified tests after changes to detect regressions is well-established.
Class B applicability added by A1:2015. Regression testing explicitly triggered after each integration step, not only at release. Scope must be 'appropriate' and documented. [Class B, C]
Regression records tied to each integration step (not just final testing), with documented scope rationale demonstrating appropriateness for the specific integration change.
Maps to
IEC 62304: §5.6.6 Conduct regression tests
ISO 13485: §7.3.6 Design and development verification
Pre-QMSR Part 820 (legacy QSR): §820.30(f) Design verification.
Requirement text
When software items are integrated, the manufacturer shall conduct regression testing appropriate to demonstrate that defects have not been introduced into previously integrated software. Applies to Class B and Class C software.
Why this clause exists
Incremental integration — adding software units to an existing integrated software item — introduces the risk that the new unit disrupts previously verified behavior. Regression testing after each integration step is the mechanism for detecting these disruptions before the cumulative integration produces a system where the source of any failure is obscured by multiple simultaneous changes. In safety-critical software, regression failure is particularly important to detect: a previously verified risk control in an earlier-integrated unit may be silently broken by a new unit that modifies shared state, interrupt handling, or communication timing. Clause 5.6.6 requires regression testing that is 'appropriate' — not a fixed scope — recognizing that the regression scope should be proportionate to the potential interaction of the newly integrated unit with previously integrated components.
What changed
Clause 5.6.6 was present in IEC 62304:2006 and maintained in A1:2015. Amendment 1 added Class B applicability (previously applicable to Class C only). The requirement for 'appropriate' regression testing is unchanged — the standard intentionally leaves scope determination to the manufacturer's risk judgment rather than prescribing a fixed regression approach.
Common gaps (what we see in audits)
- Regression testing conducted only at final integration, not after each integration step — Manufacturers perform regression testing once on the fully integrated software item rather than after each incremental integration step. Clause 5.6.6 triggers regression testing 'when software items are integrated' — each integration step — not only at final assembly. Defects introduced in earlier integration steps may be masked by later failures when regression is deferred to the end.
- Regression test scope not documented or justified — Regression testing is performed but the scope selected is not documented or justified as appropriate to the integration change. Auditors expect to see that the regression scope was deliberately determined based on interaction potential of the newly integrated unit.