What makes the evaluation separate
Separation may involve a different setting, time period or population. Fitting and evaluating a method in the same data is not an independent validation. Describe how the evaluation data were obtained and kept separate.
Evaluate more than discrimination
For prediction models, examine calibration as well as discrimination, missing data and relevant groups. For diagnostic tests, consider the reference standard and target population. A single performance measure is rarely sufficient.
Consider transfer to another setting
Compare definitions, prevalence, technology and workflows. Good performance in one validation establishes evidence for that context. Updates to a method should be documented and evaluated again.
Interpretation limits
External validation is not an impact study and does not itself demonstrate benefit from using the method in decisions.
Original references
Continue exploring
Research questions
Select tools by method, population and purpose. A thematic connection does not validate use in every case.
Educational content for framing research. Consult the original document, study population and methodology before interpreting a conclusion. Methodology