Short answer

A CAPA for medical software has four parts that are easy to blur. Contain and correct the problem you found. Decide, with a recorded reason, whether a CAPA is needed at all. Find the cause in the process, with evidence. Then prove the cause is gone, using criteria set before you implement anything: a measure, a population, an observation window and a reviewer. “Staff trained” is an action, not evidence of effectiveness. The two worked examples show each step on fictional records, and the request, plan and report forms below give you the structure.

What the requirements say

ISO 13485:2016 covers corrective action in clause 8.5.2 and preventive action in clause 8.5.3. In outline, corrective action must deal with the causes of nonconformities to prevent recurrence, without undue delay, and must be checked so that it does not affect the ability to meet regulatory requirements or the safety and performance of the device. Effectiveness must also be reviewed. Use your licensed copy for the exact wording. The EU MDR requires the manufacturer’s quality management system to address “management of corrective and preventive actions and verification of their effectiveness” (Article 10(9)(l)).

MDCG 2024-12 (October 2024) is written for notified bodies and their assessors, not manufacturers. It covers how a notified body builds and has reviewed a CAPA plan for its own joint-assessment findings. Its structure is a useful model all the same: corrections including containment, root cause analysis, corrective and preventive actions, and verification of effectiveness with SMART criteria.

Correction, corrective action, preventive action

CAPA terms
TermWhat it means in practiceExample SaMD A
ContainmentStop the problem spreading while you investigateRelease review of candidate 1.0 stopped
CorrectionFix the problem you foundMissing test written and executed
Root causeThe process condition that allowed the problem, supported by evidenceWorkflow let a requirement reach “Done” without a test link
Corrective actionRemove the cause so the problem does not recurWorkflow validator; requirement-level release check
Preventive actionRemove the cause of a potential problem that has not happened yetApplying the same validator to change requests before a gap appeared there
Effectiveness checkEvidence, against criteria set in advance, that the cause is goneThree later candidates with zero unexplained missing links

Does this need a CAPA?

Not every software defect needs a separate CAPA. Problem resolution and change control (IEC 62304 clause 9 and 8.2) handle most bugs. A CAPA is for causes that sit in the process and can produce the problem again. Typical routes:

Correction only, or CAPA?
SituationTypical route
A coding defect found in testing before release, with no sign of a process weaknessProblem resolution and change control. No CAPA.
The same type of defect keeps escaping to releasesCAPA: something in review or testing lets it through
An audit finds a requirement released without verificationCAPA if the process allowed it; correction only if the cause was a one-off and the rationale is recorded
A supplier changes an interface without the agreed noticeCAPA if the notice process failed; supplier re-evaluation as well
A complaint shows a safety-relevant misbehaviour in the fieldCAPA, plus vigilance and field-action assessment under the applicable rules

Whatever you decide, write down the decision, who made it and why. A pile of “correction only” decisions with no rationale reads, in an audit, like a CAPA process that never triggers. Look at trends too: five unrelated “one-offs” of the same kind are a pattern.

Root cause for software problems

Start from records, not opinions. Reconstruct the timeline. Compare the records that went wrong with the ones that went right. Walk through the process with the people who use it. Ask “why” until you reach something in the process you can change. Then test the explanation against the evidence, and write down the alternatives you considered and why you rejected them.

  • “Human error” is where the investigation starts, not where it ends. Ask what let the error through or failed to catch it.
  • Look at the tools. In software teams, the cause is often a workflow, permission, automation or pipeline rule that allows the wrong path.
  • Check the extent. Where else could the same cause act: other products, other change types, other suppliers?

Actions that match the cause

Each action should name the cause it addresses, an owner, a due date, and how its implementation will be verified. Prefer actions that make the wrong path impossible or visible, such as a workflow rule or an automated check, over actions that rely on people remembering. Then ask what the action itself could break. A changed workflow in your tracker is a change to QMS software and may need revalidation. A changed product needs verification, risk review and a regulatory assessment of the change.

Effectiveness checks you can measure

Set the effectiveness criteria in the plan, before implementation, so the result cannot be fitted to the data. Each criterion needs:

  • a measure: what you will count or check
  • a population and sample: which releases, changes or records
  • an observation window long enough for the problem to have recurred
  • an acceptance criterion that must be met
  • a reviewer, independent of the work where practical
  • what happens if the criterion is not met
Effectiveness criteria: weak and measurable
WeakWhy it failsMeasurable
“Team trained.”Shows the action happened, not that the problem stopped“Next three release candidates: every in-scope requirement has a linked, passed test on the release baseline. Checked by QA, independent of development.”
“Procedure updated.”Implementation only“Two negative test cases on the production workflow are both blocked.”
“No recurrence.”No population, window or detection method“At least three supplier interface changes over at least four weeks: all notified before implementation and accepted before deployment.”

In the downloadable forms, the planned effectiveness check is recorded in the CAPA plan, and its result in the CAPA report, next to the assessment of any new hazards created by the actions.

Example 1: traceability gap

Fictional example

Example SaMD A is a fictional laboratory reporting application. An internal audit found a requirement in release candidate 1.0 with no linked test. The finding is described in our internal audit guide, and the matrix before and after in the traceability matrix example.

EX-CAPA-001 at a glance
StepRecord (fictional)
SourceInternal audit EX-AUD-01, finding EX-NC-001, 3 August 2026
ProblemRequirement EX-REQ-006 had no linked test or result in candidate 1.0; five of six were linked
Containment and correctionRelease review stopped; extent check of earlier candidates and backlog found no other gap; missing test written and run
Need for CAPAYes: the same path was open for every requirement and release
CauseWorkflow allowed “Done” without a test link; release checklist counted tests instead of checking coverage per requirement. Contributing: late requirements skipped test planning
ActionsWorkflow validator; requirement-level trace export in the release checklist; same rule for change requests; tool revalidation; team briefing (support only)
Effectiveness criteriaZero unexplained missing links in each of the next three candidates; validator blocks two negative cases
Result (simulated)Met in candidates 1.0.1, 1.0.2 and 1.1. The new check also caught a baseline mismatch, fixed before review
Side effect foundValidator blocked requirements verified by inspection; fixed by allowing a “verification record” link
Closure2 October 2026

Note what does not count as evidence: the team briefing. It supported the change, but effectiveness was shown by three later candidates and by the validator blocking the wrong path.

Example 2: supplier change control

Fictional example

The same product depends on a supplier-maintained API. The supplier’s approval and agreement are covered in our supplier qualification guide.

EX-CAPA-002 at a glance
StepRecord (fictional)
SourceAcceptance test failure, 24 August 2026, supplier EX-SUP-001
ProblemSupplier deployed an API version with a renamed field to the test environment without the agreed prior notice
Containment and correctionDelivery rejected and rolled back; supplier deployments paused until accepted; extent check found one earlier unnotified schema change
Need for CAPAYes: the notice process failed twice
CauseSupplier’s notice trigger tied to major version numbers; agreement definition never built into its checklist; no schema check at delivery
ActionsSupplier checklist update (supplier action, verified); automated schema comparison at acceptance; agreement amendment; closer monitoring
Effectiveness criteriaNext three relevant changes over at least four weeks all notified in advance and accepted before deployment; zero unnotified schema changes at delivery
Result (simulated)Three changes notified and accepted; nine deliveries, no unnotified changes
Side effect foundSchema check flagged documentation-only changes; rule tuned
Closure22 October 2026; supplier retained with monitoring

The cause sat on both sides. The supplier’s trigger did not match the agreement, and the manufacturer had no independent check on delivery. Actions addressed both. The manufacturer verified the supplier’s action instead of assuming it.

Fictional examples, simulated results. Organisations, records, dates and results are invented to show how the forms are completed. They are not a client case and not evidence of certification.

Closing a CAPA

  • Every action implemented, with verified evidence.
  • Every effectiveness criterion compared with actual results, and exceptions explained.
  • Side effects of the actions looked for and handled.
  • Risk, change and regulatory impact reviewed.
  • Remaining actions transferred to another controlled process, with a reference.
  • A closure decision by an authorised person, and the conditions that would reopen it.

If a criterion is not met, extend the observation or reopen the investigation. Do not quietly redefine success.

Common mistakes

  • The cause written into the problem statement before any investigation.
  • “Human error” as the root cause, and “retraining” as the action.
  • Correction and corrective action merged into one line.
  • Effectiveness criteria written at closure, after the data are in.
  • Closing on the date the last action was completed.
  • No extent check, so the same cause keeps acting elsewhere.
  • A CAPA for every bug, which buries the few that matter.

Download the forms

Free to download, no sign-up. Open in Word or LibreOffice Writer.

  • CAPA request — formDOCX · 14 KB · v1.0 · Nov 2026 · TTM-TPL-CAPA-01Request date, initiator and CAPA number; corrective or preventive; source (complaint, audit, management review, QMS data, recurring issue and more); description of the issue; approval.
    Download DOCX
  • CAPA plan — formDOCX · 14 KB · v1.0 · Nov 2026 · TTM-TPL-CAPA-02Root-cause investigation prompts (defect, process, references, supplier data, past CAPAs, change history, complaints, risk analysis), action plan with owners and dates, planned effectiveness check, three-step approval.
    Download DOCX
  • CAPA report — formDOCX · 13 KB · v1.0 · Nov 2026 · TTM-TPL-CAPA-03Results with objective evidence, the effectiveness check result, assessment of new hazards against the risk management file, closure and approval.
    Download DOCX

Each template has a document control block and [placeholders] for your company name, product name and document IDs. Adapt it to your own procedures before use. Completing it does not establish conformity or guarantee certification. The forms do not replace your controlled CAPA procedure, risk management or vigilance assessment.

Need help with findings from a Notified Body or certification audit, or with a CAPA backlog? Root cause analysis, CAPA follow-up and responses to audit findings are part of ongoing QA/RA support. If you are setting up the CAPA process itself, see ISO 13485 QMS implementation.

Sources