To mark Affected person Security Consciousness Week, which runs March 8-14, ECRI and the Institute for Protected Treatment Practices (ISMP) printed a report that identifies the ten most important affected person security challenges anticipated to affect the healthcare business in 2026, with “Navigating the AI Diagnostic Dilemma” topping the listing.
The nonprofit ECRI, one of many largest U.S.-based affected person security organizations, mentioned that the High 10 Affected person Security Issues listing is knowledgeable by insights from senior executives from throughout the healthcare panorama—together with built-in well being programs, youngsters’s hospitals, rural group well being facilities, and nationwide associations.
In selecting AI used for diagnostics as its prime security concern, ECRI notes that AI continues to be an evolving know-how that raises points associated to reliability, transparency, privateness, legal responsibility, and ethics, and “customers mustn’t deal with it as a substitute for medical experience. Putting an excessive amount of belief in an AI mannequin to diagnose sufferers with out factoring in clinician experience can result in misdiagnosis— the very drawback AI was meant to unravel.”
Regardless of its potential to enhance diagnostic accuracy by automating information retrieval, lowering cognitive load, decreasing cognitive biases, and offering clinicians with info to assist information their choices, ECRI notes that in some circumstances AI can contribute to diagnostic errors. The report gave some examples:
• AI fashions can perpetuate biases current within the information used to coach them. Such biases can lead to incorrect diagnoses and will exacerbate healthcare disparities.
• An absence of transparency associated to the information used to coach the AI mannequin and the event and testing of underlying algorithms can lead to diagnoses primarily based on outdated, inadequate, or incorrect info.
• Points with an AI system’s operation and efficiency can lead to hallucinations (i.e., incorrect, nonsensical, or nonexistent outputs) or system brittleness (i.e., AI’s lack of ability to think about conditions that fall exterior of its coaching information), each of which may contribute to misdiagnosis. That is particularly harmful as a result of AI programs are sometimes skilled to
give solutions to each query—and customers could not understand the solutions are flawed.
• Analysis has proven that over time, over-reliance on AI can erode individuals’s essential considering abilities. This has raised issues that clinicians who recurrently depend on AI to assist diagnose sufferers will lose useful diagnostic abilities, and that clinicians in coaching could fail to develop these abilities solely.
The report recommends that well being programs set up AI utilization insurance policies, pointers, and procedures for workers that define clear roles and tasks for the governance, implementation, oversight, documentation, and monitoring of AI applied sciences.
ECRI additionally means that well being programs be sure that employees are skilled on the correct use of AI programs, significantly those who help in analysis, and inform clinicians of the programs’ capabilities and limitations. They need to require employees to doc situations wherein AI was used for diagnostic functions and the way it affected the medical diagnostic course of.
The report recommends disclosing using AI to sufferers and acquire knowledgeable consent earlier than utilizing generative AI in affected person analysis or importing affected person info to an AI system. Well being programs ought to embrace opt-out clauses in consent agreements, ECRI recommends.
One other advice is to foster a simply tradition and encourage employees to talk up if points with AI-based applied sciences happen. Well being programs ought to take issues associated to the operation and use of AI programs critically and take steps to analyze and tackle them, the report says.
Right here is the complete ECRI High 10 Affected person Security Issues matter listing for 2026:
1. Navigating the AI Diagnostic Dilemma
2. Diminished Entry to Rural Healthcare Will increase Well being Dangers and Disparities
3. Rising Charges of Preventable Acute Illnesses in Communities and Healthcare Settings
4. Results of Federal Funding Cuts on Healthcare Operations and Affected person Security
5. Lack of Recognition and Reporting of Hurt Occasions
6. Structural and Systemic Boundaries Inhibit Equitable Ache Administration for Girls
7. Persistent Workforce Shortages Proceed to Burden Employees and Limit Entry to Care
8. The Influence on System Enchancment When a Tradition of Blame Hinders Studying
9. Emergency Division Boarding Contributes to Worse Affected person Outcomes
10. Persistent Gaps in Producer Packaging and Labeling Design Proceed to Undermine Treatment Security Efforts
