.jpg){kind=link}
- Published
- Updated
A device algorithm may receive different inputs while its code remains unchanged. A different imaging system, user group or acquisition method can create conditions unlike the development data. Do not limit performance monitoring to software versions.
Define the assessment
Clarify intended input conditions and known limitations. A distribution change is an investigation signal, not proof of clinical failure or a safety event by itself. Interpretation requires examination of data quality, technical compatibility and clinical context together.
Worked example
For example, lower-resolution images arrive from a new site. Average outputs look similar, but confidence decreases in some records. Rather than automatically accepting results, the team examines acquisition conditions and relevant groups. It distinguishes evidence representing the new inputs from assumptions carried over from previous validation.
Practical checks
- Record changes in input formats, quality ranges and clinical conditions of use.
- Define which signal initiates review and who is responsible for assessing it.
- Examine affected input groups alongside overall performance using a reasoned approach.
Decision and limitations
The decision should explain whether changed conditions remain within intended scope and whether further evidence is needed. Absence of a detected signal does not establish coverage of every new condition. Where necessary, authorised reviewers reconsider use limitations, data collection or validation and retain the rationale.
Frequently asked questions
- Is distribution shift a device failure?
- Not by itself. Investigate its effect on actual performance and risk.
- Does unchanged code eliminate reassessment?
- No. Changed inputs and use context may place the same code in materially different conditions.