Clalit Health Services is among the first organizations worldwide to develop a structured organizational mechanism for evaluating artificial intelligence systems before they are implemented in healthcare. The World Health Organization (WHO) presents its OPTICA model as one of four leading approaches in a new international report analyzing the real-world use of AI applications within European health systems.
The organization recently published a comprehensive report examining how health systems across Europe are actually implementing artificial intelligence, rather than simply planning to do so. The report was developed by a dedicated working group as part of the WHO’s strategic initiative on data and digital health. It examined eleven case studies from different countries, three of which came from Israel, including two from Clalit.
Gallery
Visualization of Clalit's C-Pi artificial intelligence system
(Photo: Clalit Health Services)
Most of the cases focus not on whether AI should be implemented in healthcare, but on how it can be implemented without compromising patient safety or the trust of medical professionals.
Among the tools presented in the report alongside the Israeli case studies is OPTICA, a framework developed by Clalit Health Services. The model is described as a structured 77-item checklist divided into 13 sections and four evaluation domains: clinical suitability, data assessment, development and performance evaluation, and deployment and monitoring planning.
Designed to enable the safe implementation of AI tools, OPTICA requires the involvement of five different types of stakeholders, ranging from the clinical expert to the individual responsible for the organization’s AI infrastructure. It also requires an ongoing monitoring mechanism following implementation, rather than relying on a one-time approval process.
Prof. Ran Balicer, Chief Innovation Officer & Head of the Innovation Division at Clalit, explains the rationale behind the model:
“We created an organizational process called OPTICA, in which every AI product intended for implementation at Clalit goes through a checklist of up to 77 items. These items address every aspect of the product, including safety, privacy protection, accuracy, and continuous monitoring over the years, because the product itself may change.”
Prof. Ran Balicer, Chief Innovation Officer and Head of the Innovation Division at Clalit
(Photo: Clalit Health Services)
Two additional Israeli cases featured in the report involve a model developed by researchers at the Clalit Research Institute that analyzes 25 years of patient record data to identify individuals at high risk of undiagnosed hepatitis C infection.
In a mass screening survey involving approximately 50,000 individuals, only dozens of carriers had previously been identified. In contrast, the targeted approach proposed by the model identified 38 carriers among approximately 500 people considered at risk, representing a 100-fold improvement in detection efficiency.
The report’s authors emphasize that the success of the initiative relied primarily on the gradual integration of the model into clinical workflows and on the transparency of the explanations provided by the system to physicians.
The third Israeli case focuses on a central infrastructure platform developed by Israeli startup Aidoc. The platform enables Clalit to simultaneously run multiple algorithms from different vendors for medical imaging, against the backdrop of a shortage of radiologists.
Here too, the report’s central conclusion is organizational rather than technological. It highlights the importance of appointing responsible clinical stakeholders alongside continuous monitoring following every equipment upgrade.
Beyond the Israeli cases, the report describes how the Helsinki University Hospital evaluated 60 AI products, of which only 15, approximately one quarter, progressed to actual implementation.
The report’s authors note that only a minority of the projects evaluated used formal assessment frameworks linking the use of AI to actual clinical outcomes, a gap they identify as one of the key challenges facing the field as a whole.
According to the report, even when a technology demonstrates its effectiveness, its successful implementation depends on factors that extend beyond the technology itself. These include integration into existing clinical workflows, early involvement of clinical professionals, and a governance infrastructure that enables continuous monitoring after implementation, rather than only before it.
The findings reinforce a central principle emerging in the deployment of AI in healthcare: the challenge is no longer simply developing AI that works, but creating the organizational frameworks, clinical processes and oversight mechanisms needed to ensure that AI continues to work safely and effectively in the real world.