AI in Swiss Healthcare and Translational Research: From Promise to Practice

What is the Swiss-specific potential of AI in healthcare and in clinical and translational research?

Bryn Roberts: Across diagnostics, clinical decision support, drug discovery, precision medicine, and preventive care, AI offers the possibility of earlier detection, more targeted treatment, reduced administrative burden, and efficiency gains in a system where per-capita healthcare expenditure is among the highest globally. In clinical and translational research, AI could accelerate the development of new diagnostics, therapies, and preventive measures, an area where Switzerland's academic and industrial ecosystem is well positioned to contribute internationally.

Realising this potential requires interoperable data infrastructure, sufficient high-quality data for model training and validation, and a regulatory framework adapted to AI-specific challenges. In addition, an adequate level of digital literacy across the healthcare workforce is a prerequisite for effective and safe deployment of AI in healthcare settings, where associated risks, such as data and algorithmic bias, liability gaps, data privacy vulnerabilities, and cybersecurity threats, are addressed.

Could you elaborate on this a little more? What Swiss-specific risks hinder the application of AI in healthcare and in clinical and translational research?

Christiane Pauli-Magnus: Several risks deserve attention, some general to AI in health, others shaped by Switzerland's specific context. By ‘risks’, we mean anything that could go wrong once AI is put into use.

Fairness is a foundational consideration. In a multilingual, federalist country where health data are distributed across cantons and institutions, ensuring representative training data requires deliberate effort. Unrepresentative data, based on age, ethnicity, socioeconomic background, etc. can influence outcomes. The fragmented nature of Swiss health data makes this risk particularly acute.

Accuracy and reliability require ongoing scrutiny. Models that perform well in validation settings may behave differently across diverse clinical environments. Sufficient clinical oversight is needed to identify and correct errors before they affect patient care.

Switzerland currently has no AI-specific liability framework, and a draft legislative package is not expected before the end of 2026. This creates a genuine accountability gap when AI contributes to a suboptimal clinical outcome.

Ethical considerations beyond bias and liability also merit attention. These include transparency and informed consent, whether and how patients are made aware of AI's role in their care, explainability about how AI systems reach their outputs, and the right to human autonomy and oversight of AI-assisted decisions.

Data protection and security merit particular attention given the sensitivity of health data, and cybersecurity vulnerabilities in interconnected health systems require pro-active management. One reason is that it is more challenging to assure cybersecurity in highly fragmented data systems, with heterogeneous governance. However, the consequences of breaches may be more contained within a federated systems.

Finally, public trust warrants attention as a risk dimension. Poorly deployed or inadequately explained AI tools can undermine confidence in digital health more broadly, with consequences for the wider adoption of AI across the healthcare system.

What are the Swiss-specific challenges when implementing AI in healthcare and in clinical and translational research?

Bryn Roberts: Switzerland's context shapes its implementation challenges in distinct ways. By challenges we mean the structural and organisational obstacles that hinder the effective use of AI.

The distribution of healthcare responsibility across 26 cantons, each with its own structures and data systems, makes the data access and interoperability that AI requires more complex than in more homogeneous health systems. With four national languages and associated cultural diversity, developing and validating AI tools that perform equitably across language regions requires additional resources that monolingual systems do not face. Access to high-quality, longitudinal health data for the population, including data from healthy subjects, remains limited by existing data silos. Synthetic data offer a partial solution, though their utility depends on how faithfully they represent real-world clinical complexity. Several Swiss-specific structural features therefore continue to act as barriers to implementation and require a coordinated national responses.

The regulatory pathways presents a further implementation challenge. AI tools with a medical intended purpose must be approved as medical devices under the Medical Devices Ordinance, a national framework that was not designed with AI-specific characteristics, such as continuous learning and non-deterministic outputs, in mind. As AI tools can be deployed as static versions, similar to software as medical device (SaMD), this is only an issue for AI solutions that adapt to evolving context or are designed to learn as new data are acquired. However, navigating this pathway requires substantial time and resources, particularly for smaller institutions and research groups.

Education and training present a cross-cutting challenge. Effective and safe use of AI requires a level of digital literacy and AI knowledge, as well as institutional support, and these remain uneven across health professions and clinical settings in Switzerland.

Public trust is also a practical implementation barrier. Adoption depends on transparency, explainability, demonstrable benefit, and meaningful engagement with public concerns.

Finally, the role of health insurers is also relevant. In Switzerland's healthcare system, insurers are key actors whose willingness to reimburse AI-assisted services is a practical prerequisite for adoption at scale. Without clear reimbursement pathways, even well-validated and highly effective AI tools may struggle to reach routine clinical use. On the positive side, AI tools with the potential to augment population health outcomes, such as risk flagging and early detection algorithms, have positive health economic benefits that payors will appreciate, as well as the Swiss population at large.

What are the Swiss-specific solutions for the identified risks and challenges?

Christiane Pauli-Magnus: On data infrastructure, DigiSanté aims to introduce standards for seamless data exchange across institutions and cantons, and to establish a Swiss Health Data Space for the responsible secondary use of health data, both foundational requirements for AI in healthcare and research. However, parliamentary savings targets have reduced DigiSanté's 2027 budget to roughly half of what was originally planned, with the Swiss Health Data Space refocused on core elements and the research data space postponed. Ensuring these components are adequately resourced remains a structural question for Switzerland's AI ambitions in health.

Overcoming data silos requires concrete incentives. Institutions and researchers contributing data for AI training and validation should be recognised through funding mechanisms and shared governance models. The Swiss Personalized Health Network (SPHN) was one initiative working in this direction, but funding for the SPHN Data Coordination Centre is foreseen to end in 2028. What a coordination framework might look like after 2029, provided sufficient funding is available, is not yet defined, but remains important.

Linking the Swiss Health Data Space with the European Health Data Space is important for training, testing and validating AI applications in healthcare. The benefits: This would expand both data volume and diversity and strengthen cross-border research collaboration. This requires not only technical but also regulatory compatibility between the Swiss and the European Health Data Spaces, as well as reliable, long-term relations between the EU and Switzerland, which are currently being discussed at national level in the context of the Bilateral III Agreement.

On fairness and reliability, independent validation and post-deployment monitoring of AI tools across Switzerland's language regions and patient populations should be a standard requirement. While Swissmedic has adopted international guidance for AI as a medical device, remaining questions around liability and the dynamic characteristics of AI systems still need to be addressed.

Generating evidence that AI improves outcomes and increases efficiency requires systematic evaluation from the outset, combining prospective studies with health economic analyses. Switzerland's university hospitals are strong candidates to contribute to this kind of real-world evaluation.

Building public trust requires transparent communication about how AI is used in clinical settings, meaningful patient involvement, and clear governance frameworks. Patient and public involvement should not be limited to communication after decisions have already been made. Patients should be represented in governance structures for health data use, in the design of consent and transparency models, and in the evaluation of AI tools that may affect clinical decisions. This would help to ensure that AI implementation reflects public expectations, and not only technical or institutional priorities. Engagement with health insurers on reimbursement models for validated AI tools is equally necessary to translate evidence into routine practice.

Also, continued investment in AI education across clinical and research training will be essential.

What are the potential long-term consequences for Switzerland if AI cannot be implemented effectively in the healthcare system and in clinical and translational research?

Bryn Roberts: If the solutions outlined above are not pursued and AI cannot be effectively implemented, the consequences would be felt across several areas — though it should be noted that poorly implemented or prematurely deployed AI carries its own risks. Therefore, the goal is effective implementation rather than rapid implementation.

• On system sustainability, rising healthcare costs, an ageing population, and a projected shortage of healthcare professionals create structural pressures that AI is positioned to help address. Without it, these pressures will be harder to absorb through other means.

• On research and innovation, Switzerland's position as a global biomedical and pharmaceutical hub depends increasingly on the ability to integrate AI into research and development pipelines. Countries that build this capacity earlier will attract talent, investment, and industry partnerships, which could place Switzerland at a competitive disadvantage if implementation continues to lag.

• On costs, AI-driven efficiency gains may become an important contribution to cost containment, addressing the cost pressure already noted. However, these savings should not be assumed. They need to be demonstrated through systematic evaluation, including prospective clinical studies and health economic analyses. Without such evidence, AI risks adding complexity and cost rather than reducing them.

• On care quality, there is growing evidence that AI can improve performance in detection, diagnosis, treatment and monitoring of multiple diseases, including cancer, neurological, cardiovascular, metabolic, renal and ophthalmic disorders. Delayed implementation would limit Switzerland's ability to generate further evidence domestically and to benefit from advances demonstrated elsewhere.

Taken together, the long-term consequences are not primarily technological. They concern the sustainability of the healthcare system, Switzerland's capacity to remain a leading location for clinical and translational research, and the country's ability to ensure sustainable, equitable, high-quality care for its population.

What should Switzerland do now? What are the most important recommendations for action?

Drawing on the solutions discussed above, a few recommendations stand out as the most urgent. Switzerland should:

• Develop a national vision and implementation roadmap for AI in healthcare and clinical research.

• Create a health data infrastructure as critical national infrastructure, ensuring compatibility between the Swiss and European Health Data Spaces.

• Define a national framework for safe, trustworthy and clinically useful AI.

• Invest in AI education as part of clinical and research training.

• Develop reimbursement pathways together with health insurers.

• Pursue international partnerships to ensure access to knowhow, technology and the scale, diversity and interoperability of data needed for research and innovation.

Christiane Pauli-Magnus, Bryn Roberts