Why Implementation Must Become a Learning Cycle

Road to Bilbao | Reflections inspired by the Life Cycle of Health Technologies 2.0

One persistent habit in health innovation is to think of implementation as the end of the journey. We research a technology, develop it, assess it, regulate it, reimburse it and eventually introduce it into practice. Once it reaches routine use, there is a temptation to think the difficult work is largely complete.

A recently published paper by Maximilian Otte, Iñaki Gutiérrez-Ibarluzea and Hans-Peter Dauben suggests something much more interesting. Their Life Cycle of Health Technologies 2.0 asks us to move away from thinking primarily about the lifecycle of a product and towards thinking about the lifecycle of the knowledge surrounding that technology.

It is a subtle shift, but an important one for the Road to Bilbao.

Because if health technologies are constantly generating new knowledge through their interaction with clinicians, patients, organisations and health systems, implementation cannot really be the final stage. It becomes one of the places where learning accelerates.

From products to knowledge

The authors start from the recognition that contemporary health technologies do not live independently from the systems into which they are introduced. Medicines, diagnostics, digital applications, AI systems, advanced therapies and organisational innovations are all shaped by regulatory environments, professional practices, workflows, infrastructure, patient experience and societal expectations.

The paper describes health technologies as sociotechnical phenomena. Their effects arise not simply from what the technology can technically do, but from the interaction between the technology, the people using it and the context in which it is used.

This helps explain something we repeatedly encounter in digital health. A technically successful solution does not necessarily become a successful health-system intervention. A tool can perform well and still fail to fit the workflow. The data can exist while the organisation remains unable to use it. A model can demonstrate accuracy without earning clinical trust. Regulation can allow something to happen without creating the organisational conditions that make it sustainable.

Implementation therefore generates knowledge that could never have been fully produced before implementation.

That insight sits at the heart of LC 2.0.

The framework is represented as an infinity loop linking Needs, Idea, Research, Development, Implementation and Usage. The particularly interesting part is what happens after a technology reaches routine use. The loop does not stop. Experience feeds into reassessment; reassessment reveals gaps; those gaps become renewed needs and new questions for research, development and innovation. The diagram on page 7 makes this continuous movement particularly clear.

As the authors put it, real-world use becomes an "active learning environment."

That phrase deserves attention.

Real-world data is part of the learning mechanism

We often speak about real-world data as an additional source of evidence available after clinical development. LC 2.0 invites a broader interpretation. Data generated during routine use becomes part of the mechanism through which a health system learns whether a technology is actually doing what was expected of it.

Clinical trials answer some questions extraordinarily well. They cannot answer every question that emerges once an intervention enters complex health systems.

Routine use starts revealing different things: variation between populations, long-term outcomes, implementation difficulties, organisational consequences, unexpected effects, patient experience and sometimes gaps between theoretical efficacy and real-world effectiveness. Real-world evidence therefore becomes part of a continuous feedback process rather than simply an evidentiary appendix attached to the end of development.

This connects directly to Real Data, Better AI.

Better AI cannot simply mean applying increasingly powerful models to ever larger volumes of data. The quality of the learning depends on whether the system can understand what happens when technology meets clinical reality, feed that knowledge back into decision-making and adapt accordingly.

In that sense, real data matters not because it is "real" in some abstract way, but because it connects innovation with the lived complexity of healthcare.

Context is not background noise

Another aspect of the paper that resonates strongly with our conversations on the Road to Bilbao is the way LC 2.0 treats context.

The framework deliberately operates across macro, meso and micro levels. Regulation, policy and reimbursement matter. So do organisational readiness, interoperability, workforce capacity and workflow integration. And so do the experiences and perspectives of the clinicians and patients who ultimately interact with a technology.

These are not peripheral variables to be managed after the "real" technological work has been done. They influence whether innovation produces value at all.

This is particularly relevant as Europe moves from the regulatory creation of the European Health Data Space towards implementation.

EHDS can establish rights, obligations, infrastructures and mechanisms for health-data access and secondary use. But its real value will emerge in how these mechanisms interact with actual health systems.

Can organisations use them?

Can professionals work with them?

Can researchers obtain meaningful evidence?

Can patients understand and trust what is happening?

Can experience generated through implementation change what the system does next?

These are learning-system questions as much as they are data-governance questions.

EHDS needs learning systems, not merely compliant systems

This distinction may become increasingly important.

A health system can become compliant with new requirements without necessarily becoming better at learning. It can implement new infrastructure, establish procedures and satisfy regulatory expectations while continuing to reproduce old organisational behaviours.

A learning health system should be capable of something more demanding. It should observe what happens after implementation, recognise unexpected consequences, identify gaps between intended and actual outcomes, listen to users and patients, and translate that accumulated experience into subsequent decisions.

This is where the LC 2.0 infinity loop becomes especially useful as a metaphor for EHDS.

The success of EHDS should not ultimately be measured only by how much data can move, how many datasets become discoverable or how many access requests are processed. Those indicators will matter, but the larger question is what Europe can learn because its health data has become more usable.

Does better access to health data improve research?

Does it shorten the journey between evidence and patient benefit?

Does it reveal inequalities that were previously invisible?

Does it improve health-system decisions?

Does it help identify technologies that should be adapted, scaled — or abandoned?

Those are much harder questions. They are also much closer to the reason for creating health-data infrastructure in the first place.

The Road to Bilbao

This brings us directly to Bilbao.

The Health Data Forum Global Hybrid Summit on 24–25 September is deliberately focused on moving from health-data policy towards real-world implementation. Genomics, imaging, trusted secondary use, cybersecurity, interoperability and infrastructure will all be part of that conversation.

But underneath those subjects lies another question.

What kind of health system do we need in order to make meaningful use of all this new data capability?

The Life Cycle of Health Technologies 2.0 offers one possible answer: a health system that treats implementation and usage as sources of knowledge rather than endpoints.

This is particularly important for AI and high-dimensional data. Genomic information gains value when it can be connected responsibly with phenotype, treatment and outcomes. Imaging data becomes more valuable when routine use feeds learning about pathways, diagnostics and clinical decisions. AI becomes better when the data and experience generated through its use can inform reassessment and adaptation.

Implementation therefore becomes less about installing something and more about creating the conditions for continuous learning.

Perhaps, then, the question we should carry to Bilbao is not simply:

How do we implement EHDS?

It is something more ambitious:

How do we create health systems capable of learning continuously from the data, technologies and experiences they generate?

That is where the paper by Otte, Gutiérrez-Ibarluzea and Dauben feels particularly timely.

The ultimate value of health data does not lie in storing it, moving it or even making it interoperable. Its value lies in what we can learn from it—and whether that learning finds its way back into better decisions, better technologies, and ultimately better health.

Implementation is therefore not where innovation ends.

It is where the next cycle of knowledge begins.

Reference

Otte, M., Gutiérrez-Ibarluzea, I., & Dauben, H. P. (2026). From product to knowledge: the life cycle of health technologies 2.0 as a framework for health system improvement. Frontiers in Pharmacology, 17, 1869110. https://doi.org/10.3389/fphar.2026.1869110