CAPA effectiveness: evidence that the problem stayed corrected
A revised procedure, a completed training record and a released software fix show that an action was implemented. They do not, by themselves, show that the failure mechanism has been controlled. A useful CAPA effectiveness review connects the original problem to evidence gathered after the change, under conditions in which the problem could reasonably recur.
For a cardiovascular device manufacturer, the consequential question is what supports the decision to close the action. The answer should explain which cause was addressed, what was observed and why those observations are sufficient for the risk involved.
Use the current regulatory framework
FDA’s Quality Management System Regulation became effective on February 2, 2026. [1] Under 21 CFR 820.10(a), manufacturers subject to the provision must document a quality management system meeting applicable ISO 13485 requirements and the other applicable Part 820 requirements. [2]
FDA’s current inspection program maps corrective and preventive action to ISO 13485 clauses 8.5.2 and 8.5.3. In follow-up of previous inspection or compliance issues, it directs attention to verification or validation of corrective-action effectiveness, alongside implementation and investigation. [3, pages 37–38 and 74] The practical method below is Zaliance’s recommended approach to developing that evidence, not a prescribed FDA sampling plan.
Define the failure that the action must prevent
Write the effectiveness question at the same level of specificity as the investigated cause. “Have complaints decreased?” is too broad if the action addressed a particular assembly setup. “Does the revised setup control prevent the identified joint defect across the affected production conditions?” gives the team a testable question.
Keep immediate correction and sustained control distinct in the record. Replacing a defective unit addresses that unit. Changing the process that produced it addresses future production. If the investigation also identified an inspection weakness, evaluate the production control and the detection control separately; an improvement in one should not conceal failure of the other.
The title’s focus is corrective action after an observed problem. For a preventive action, the corresponding question is whether the control addresses a credible potential failure before it occurs. In either case, define the mechanism being controlled before choosing a closure metric.
Separate implementation, performance and sustained effectiveness
Build the review around three different questions:
| Question | Useful evidence | Limit of evidence |
|---|---|---|
| Was the action implemented? | Approved change, deployed version, installed fixture or completed training. | Establishes that the intended change occurred. |
| Does the changed control work? | A justified challenge of the failure mechanism and relevant boundary conditions. | Establishes performance under the evaluated conditions. |
| Has the control remained effective? | Post-change observations covering relevant operating conditions and opportunities for recurrence. | Supports a sustained-control conclusion within the observation scope. |
A single report can contain all three, but it should not treat them as interchangeable. Training attendance, for example, can establish completion of training. If the suspected cause involves an ambiguous setup instruction, observing operators perform the setup is a more direct test of whether the ambiguity has been resolved.
Manufacturing example: challenge the cause across production conditions
Consider a hypothetical catheter assembly problem. Investigation links intermittent bond failures to inconsistent insertion depth during a manual assembly step. The manufacturer introduces a locating fixture, clarifies the work instruction and revises setup verification.
The effectiveness plan should examine whether the new fixture controls insertion depth across the relevant component tolerances and production setups. It should also assess whether the change introduces another problem, such as damage from excessive clamping. The specific tests and acceptance limits must come from the device and process requirements; this example does not establish a catheter test specification.
For follow-up production, identify which lots, operators, shifts and component sources matter to the investigated mechanism. Several lots made by one operator using the same component batch may provide less coverage than the lot count suggests. Record which conditions were actually encountered and which remain unobserved.
If temporary inspection catches defects before shipment, a quiet complaint record could coexist with an unresolved production problem. Examine first-pass results, rejected units and rework alongside released-product outcomes. Do not remove temporary containment merely to create an effectiveness experiment; assess its withdrawal through the established risk and change process.
Software example: confirm both the fix and its use
In a separate hypothetical connected blood-pressure device example, a mobile application occasionally displays an older measurement as the latest result after an interrupted synchronization. The corrective action changes the ordering logic and the handling of incomplete transfers.
Reproducing the original sequence and showing that the new version handles it correctly is useful verification. The effectiveness question is broader: does the failure remain controlled across the supported combinations that can trigger it? A justified challenge set might include reconnects, delayed messages, duplicate transfers and changes in device time.
After release, confirm which installations actually received the corrected version. Where privacy-appropriate operational records are available, define a measure tied to relevant synchronization events rather than total app downloads. Distinguish corrected-version observations from older-version complaints. Otherwise, the team can neither interpret continuing reports nor establish meaningful exposure to the change.
Also look for displaced failure: the app might stop displaying stale results by silently omitting valid measurements. The verification and follow-up plan should therefore examine the intended outcome and plausible adverse consequences of the correction.
Choose enough exposure to support the conclusion
A calendar interval is an administrative boundary, not a measure of opportunity for recurrence. For the assembly example, count relevant production and setups. For the software example, count relevant operations on the corrected version. Explain why the observation period covers the conditions and delay mechanisms that matter.
Zero observed failures needs a denominator and assumptions. As a purely illustrative calculation, if 100 independent trials share one fixed failure probability, zero failures gives an exact one-sided 95% upper confidence bound of approximately 2.95%, using 1 − 0.051/100. It does not establish a zero failure rate. This is a statistical illustration, not a recommended CAPA sample size or acceptance limit.
Production and field observations may violate those assumptions. Repeated operations on one installation or units from one shared setup can have common causes. A large count should not substitute for coverage of the conditions that could defeat the control. Justify the sample, the observation window and any statistical model against the actual question.
Decide the closure rule before reviewing the results
Specify the measure, its denominator, the acceptance criterion, the observation scope and the response to an adverse finding before follow-up begins. Explain the basis for each. A severe failure may need escalation after one recurrence even when an aggregate rate remains below a numerical threshold.
Use three possible dispositions: effectiveness demonstrated, effectiveness not demonstrated, or evidence insufficient. The last category matters when production volume was too low, a relevant configuration never ran or deployment could not be confirmed. Extending observation can resolve insufficient evidence; it does not resolve a recurrence that challenges the cause analysis.
The closure statement should connect the investigated cause, implemented change, evaluation results and residual uncertainty. Name any continuing monitoring owner and trigger. A later recurrence should lead back to that reasoning, so the team can distinguish a failed correction from a different failure mechanism instead of reopening the investigation without a clear starting point.
References
- FDA. Quality Management System Regulation (QMSR). Updated February 2, 2026.
- 21 CFR Part 820. Sections 820.7 and 820.10(a), current eCFR text.
- FDA. Inspection of Medical Device Manufacturers, Compliance Program 7382.850. Implementation date February 2, 2026; pages 37–38 and 74.