Patient safety · evidence hierarchy
A patient-safety story should identify whether its evidence is a verified incident record, a surveillance estimate, a quality measure, a voluntary report, claims analysis or research study. These sources answer different questions.
Match the source to the claim
AHRQ describes itself as the lead federal agency for patient-safety research and links to research, quality measures, implementation tools and the Network of Patient Safety Databases. Its public index shows why “patient safety data” is not one uniform dataset.
An incident claim about a named facility requires a record that supports that incident. A national surveillance estimate can describe a measured pattern but not create a local event. A toolkit can explain a prevention practice but does not prove that an organization used it or achieved an outcome.
A five-level source card
- Event evidence: what happened, where, when and according to which verified record.
- Measurement evidence: definition, denominator, reporting system and coverage.
- Research evidence: design, population, comparison and limitations.
- Guidance evidence: issuing body, intended setting and recommendation status.
- Implementation evidence: whether the practice was adopted, followed and evaluated.
Voluntary and administrative systems can contain incomplete, delayed or differently coded information; their documentation defines what can be concluded.
What remains uncertain
Underreporting, duplicate reports, changing definitions and incomplete clinical context can affect counts and trends. An association between a practice and an outcome does not alone establish that the practice caused the result.
What readers can verify next
Editors can attach a source label to every material claim and state what evidence would change the conclusion. Organizations can use AHRQ tools for quality improvement without presenting adoption as proof of success. Patients with immediate safety concerns should contact their care team or the appropriate reporting channel rather than rely on a general data article.