Wolters Kluwer Releases Validation Framework for Evaluating Clinical AI at Point of Care

The framework goes beyond binary review of answers to assess context, uncertainty, and clinical impact; designed to support governance committees

WALTHAM, Mass.: #ai--Wolters Kluwer Health today released its approach to validating and evaluating clinical AI that is designed to reflect how clinicians use and rely on AI at the point of care. Designed to help support governance committees, this framework moves beyond basic output evaluation to assess whether AI-generated outputs are clinically useful, grounded in trusted knowledge, and reliable to use in complex, real-world point of care scenarios.

How can hospital governance committees evaluate answers provided by clinical AI?

As generative AI becomes more prominent in clinical workflows, traditional evaluation methods such as benchmarks, test questions, or user ratings fall short because they do not capture whether an answer aligns with clinical intent, whether it omits critical information, or whether it behaves appropriately under uncertainty. By contrast, UpToDate’s approach addresses these gaps through a multi-method framework that evaluates the answers clinicians receive and interpret when making care decisions.

The approach, detailed in the new report, A Measured Approach to Evaluating Clinical AI at the Point of Care, evaluates AI performance across three core dimensions: clinical intent, knowledge integrity, and clinical impact. Together, these dimensions provide hospital governance committees with a meaningful assessment of clinical reliability that goes beyond generic, measurement benchmarks.

“Assessing the reliability and clinical validity of AI is complex and inadequately captured by current benchmarks,” said Peter A.L. Bonis, MD, Chief Medical Officer at Wolters Kluwer Health. “What matters most is whether an AI's response is consistently faithful to trusted, evidence-based medical knowledge, tailored to the specific clinical context, and nuanced enough to respect uncertainty and complexity. This framework is essential for healthcare governance teams tasked with safely deploying enterprise AI.”

Clinical AI stress tested by physicians and AI experts; continuously monitored for reliability

The evaluation framework applied to UpToDate Expert AI combines automated testing with structured human review by leading physician editors and clinical AI experts. It includes rubric-based assessment, stress testing, “red teaming” (where professionals work to “break” a system), and ongoing monitoring to detect omissions, unsupported claims, loss of context, and other output anomalies that generic evaluations may miss.

In the most recent evaluation using this model, UpToDate Expert AI was tested on 1,669 clinical queries comprising more than 15,000 criteria. The results showed that UpToDate Expert AI provided clinically aligned information for 99.9% of assessed criteria. UpToDate Expert AI also had a significantly lower rate of omissions when compared with two general-purpose LLM comparators; both comparators had a rate of omission that was 15% higher than UpToDate Expert AI.

What are the key elements of a governance-ready clinical AI validation approach?

• Clinically meaningful evaluation: Solution should be tested against point-of-care criteria written by top physician experts in their field, rather than relying solely on generic benchmarks.
• Performance where it matters: The evaluation approach emphasizes clinical intent by measuring whether the response itself is clinically relevant and includes the information that matters most at the point of care.
• Grounded answers: Responses are evaluated for knowledge, integrity and traceability to trusted databases and clinical content, e.g., UpToDate.
• Risk-aware by design: “Red teaming,” expert review by medical specialists, bias testing, and regression monitoring are used to identify and reduce potential output issues. The most recent evaluation of UpToDate Expert AI had 200 hours of adversarial red-team testing.
• Clinical reasoning embedded: Solution should be designed to help clinicians think through next steps using evidence, assumptions, and expert judgment – and not simply return an AI-generated answer. As concerns increase about de-skilling, the overreliance on AI tools, which reduce clinicians’ ability to exercise independent clinical judgment, preserving access to a transparent view of all steps involved in the reasoning process becomes vital. This system-level approach reflects growing expectations from health systems, regulators, and clinicians for transparency, accountability, and governance in clinical AI.

UpToDate Expert AI is enabled by Wolters Kluwer Expert AI. Solutions built on Wolters Kluwer Expert AI use deep domain expertise and a decade of AI experience, empowering customers to work faster and make smarter decisions based on trusted and verified content.

As of April 30, more than half of U.S. UpToDate Enterprise Edition customers, representing approximately 2,000 hospitals, have signed up for UpToDate Expert AI as part of UpToDate Enterprise, with momentum building toward 70% adoption by mid-year.

For more information about Wolters Kluwer, please visit: www.wolterskluwer.com.

About Wolters Kluwer

Wolters Kluwer (EURONEXT: WKL) is a global leader in information solutions, software and services for professionals in healthcare; tax and accounting; financial and corporate compliance; legal and regulatory; corporate performance and ESG. We help our customers make critical decisions every day by providing expert solutions that combine deep domain knowledge with technology and services.

Wolters Kluwer reported 2025 annual revenues of €6.1 billion. The group serves customers in over 180 countries, maintains operations in over 40 countries, and employs approximately 21,100 people worldwide. The company is headquartered in Alphen aan den Rijn, the Netherlands.

Wolters Kluwer shares are listed on Euronext Amsterdam (WKL) and are included in the AEX, Euro Stoxx 50, and Euronext 100 indices. Wolters Kluwer has a sponsored Level 1 American Depositary Receipt (ADR) program. The ADRs are traded on the over-the-counter market in the U.S. (WTKWY).

For more information, visit www.wolterskluwer.com, and follow us on LinkedIn, Facebook, YouTube, and Instagram.

Fonte: Business Wire