Building trustworthy clinical AI: usable checklists and transparency artifacts tested in real-world health systems

The introduction of artificial intelligence (AI) to healthcare has had a rapid rise, but concerns about transparency, bias, and clinician trust affect the sustainable introduction of these technologies. This paper presents an implementation of the HTI-1 Decision Support Intervention (DSI) model, in combination with a NIST AI Risk Management Framework (AI RMF) profile, for the purposes of designing and testing usable checklists and transparency artifacts for clinical AI in two health systems. Artifacts included model fact sheet, calibration tracking, and decision-support explanations and structured for compliance to ONC HTI-1/HTI-2 regulations and interoperability standards such as FHIR and TEFCA. Simulation and pilot testing revealed improvements including clinician understanding of alerts, successful detection of calibration drift and 90% pass rate in bias auditing. These results suggest that a combination of regulatory anchors and practical usability tools can be used to operationalize trustworthy AI at point of care.