P
Performance
Metrics, thresholds, calibration, performance by population, comparators and consistency with the stated use case.
- Appropriate metrics
- Operational threshold
- Calibration
- Error analysis
QSTOM-IT assessment framework
PRISM - A methodological framework for assessing health AI
PRISM organises technical and methodological assessment of an AI system across five complementary dimensions. The framework aims to make questions, evidence and tests explicit, versioned and reproducible.
From claims to evidence.
PRISM is the methodological framework used by QSTOM-IT to structure the technical and methodological assessment of artificial intelligence systems for health. It organises questions, available evidence and verification activities across five complementary dimensions.
PRISM is not a regulatory certification and does not replace a clinical study, conformity assessment, legal advice or medical advice.PRISM
Five complementary dimensions
R
Robustness
Generalisation, drift, temporal variation, degraded data and the conditions under which performance deteriorates.
- Stress tests
- External validation
- Drift
- Data sensitivity
I
Integrity
Data quality and provenance, cohorts, train/test separation, leakage, traceability and pipeline reproducibility.
- Data lineage
- Cohort separation
- Data leakage
- Reproducibility
S
Safety
Critical errors, false negatives, edge cases, silent errors, human oversight and consequences of an incorrect output.
- Critical cases
- Oversight
- Failure modes
- Use limitations
M
Methodology
Evaluation question, hypotheses, experimental protocol, comparators, validity of conclusions and study limitations.
- Explicit hypothesis
- Protocol
- Comparators
- Scope of claims
PRISM Evidence Levels
How far has a claim actually been verified?
PRISM does not produce an aggregate score out of 100. Each assessed element is associated with an evidence level describing the depth of verification achieved within the assessment scope.
The evidence level is not a quality score. It indicates how far QSTOM-IT has been able to verify a claim, result or property of the system.
No conclusion is drawn on this point.
The method or result is described by the assessed team.
QSTOM-IT has run or reviewed a targeted test within the engagement scope.
The result could be reproduced from accessible materials using a documented protocol.
!
Non-compensation principle
A critical weakness on one PRISM dimension is not offset by strong results on other dimensions. For example, major data leakage does not become acceptable because reported performance is high.
Robustness diagnostic
Does your protocol already present methodological weaknesses?
Before a full assessment, our diagnostic helps identify several potential issues related to data, cohorts, validation and metrics in just a few minutes.
5 questionsAround 2 minutesNo patient data requested
PRISM Assessed
Document exactly what has been assessed
A PRISM assessment describes the evaluated system, its version, the scope of the analysis, the elements examined, the evidence levels achieved and any critical risks identified. The PRISM Assessed designation indicates that an assessment was performed using an identified version of the framework. It is not a regulatory certification.
Versioned framework
A traceable and evolving methodology
PRISM is designed as a versioned framework. Its criteria, assessment methods and reporting rules evolve with scientific knowledge, field experience and changes in health AI systems. Each assessment can therefore be linked to a specific version of the framework.
PRISM and regulation
A complementary perspective on the EU AI Act
Some PRISM dimensions overlap with themes in the European framework, including data quality, robustness, documentation and human oversight. This mapping helps structure technical evidence but does not constitute a conformity assessment.
QSTOM-IT
Would you like to apply PRISM to a health AI system?
Tell us about the system, its maturity, the decision context and the results you want to verify.