Healthcare Innovation Isn’t the Problem—Implementation Is
Medicine has never been better at generating innovation. From mRNA vaccines and CAR-T therapies to AI-assisted diagnostics and precision medicine, the pace of scientific progress continues to accelerate. Yet one pattern persists across each new wave of innovation.
Image: Unsplash
Scientific success does not automatically translate into widespread clinical impact. Many different technologies, for diagnosis, for treatment, for patient management, for facilities optimisation. Same implementation challenge. There are rare exceptions, of course – like the speed with which mRNA (and other vaccines) were taken up by health systems in the face of the urgency of a pandemic. Then most rules were set aside and the world of healthcare astonished itself at how fast innovation could be implemented. But, once ‘normal service was resumed’ and normal processes followed, implementation returned to its sluggish pace.
So, while healthcare often describes itself as facing an innovation problem, I believe that this is a misdiagnosis. I have been reflecting on my many years in the healthcare innovation world – in senior roles in life sciences and the NHS - and am convinced it is.
Innovation Is Not the Bottleneck
There is little evidence that healthcare is running out of scientific ideas.
Breakthroughs continue to emerge across almost every area of medicine. CAR-T therapies continue to redefine what is clinically possible for certain cancers. AI systems have repeatedly demonstrated specialist-level performance in areas such as medical imaging. Precision medicine continues to reshape oncology and rare disease. These advances differ profoundly in their science. Yet they encounter remarkably similar challenges once they leave the laboratory.
Why do scientific breakthroughs continue to take so long to be absorbed into routine clinical practice?
Implementation is often treated as the final stage of innovation. In reality, it is an entirely different discipline, facing different challenges, needing different managerial tools and calling for different leadership styles.
How Implementation Is Different
Scientific discovery and product development depend on creativity, research, experimentation and the accumulation of evidence. Implementation is fundamentally an organisational challenge, demanding adequate resources, rigorous processes, value demonstration and – throughout – time and attention, from both users and management. Let us take each of these in turn.
Resources – the budget has to be there, which requires a horizon-scanning system that anticipates new technologies with major economic cost. We saw this, or the lack of it, when the pharmaceutical industry produced a therapy for Hepatitis C. Thanks to websites like www.clinicaltrials.gov, anyone who cared to look could see these therapies coming through, yet many systems failed to make provision for them.
Rigorous processes – implementation is an engineering problem rather than a scientific one, calling for Gantt charts, identification of dependencies, assembly of trained staff, organisation of their training, etc etc. The originator of the innovation has only some of the tools at their disposal – most of the heavy lifting must come from the health system itself. We saw in the failure of the Inclisiran deal between the UK government, the NHS and Novartis that a high level agreement that something valuable should happen does not magically translate into action, particularly from an unprepared workforce.
Value demonstration is key - reimbursement may have been agreed in principle, but the system typically needs to see the clinical and economic benefits flowing from early adoption before the more skeptical clinicians can be persuaded to change their practices. The use of screening technologies like Oncotype DX to segment cancer populations into responders and non-responders is a case in point.
Which brings me to time and attention. For a long time I thought that money was the main barrier to the uptake of innovation in the UK NHS. Of course, it’s a significant one – but more significant is the dedication of time from those whose clinical practice needs to change and those managing the system in the midst of multiple, and ever-changing, targets and priorities. This seems to have been the main barrier to the ready uptake of insulin pumps for type 1 diabetics.
We now have the challenge of artificial intelligence. Despite thousands of published AI models, relatively few have yet become part of routine clinical care. One recent analysis concluded that fewer than 2% of reported AI models progress beyond prototype development into real-world clinical implementation¹. It is necessary, but far from sufficient, to hold impressive demos, identify clinical champions and make bold predictions of impact. All of the above dimensions of implementation must be put in place or this potentially transformative wave of technology will disappoint us.
The Vital Role of Leadership
In life sciences, competitive advantage has often depended on generating better science, coupled with marketing clout. But – as the volume of potential breakthrough diagnostics, therapies and digital tools increases – it will be the leaders who focus on implementation science and engineering that will win out. And these place different burdens on their organisations, and require new skills, particularly when it comes to AI. In many cases, it will require combinations of innovations to have impact – perhaps a precision diagnostic, a therapy that it targets and AI analysis to home in on the right population. New leadership demands for a new era.
In summary, the organisations that will shape the next decade may not necessarily be those producing the greatest number of scientific breakthroughs. They will be those able to integrate proven innovations into routine practice faster, more consistently and at greater scale, in partnership with innovation-hungry health systems.
Reference
¹ K. Y. Foo et al. “Clinical translation of AI requires implementation science.” npj Digital Medicine (2024). The authors report that only a very small proportion of published AI models progress to routine clinical implementation, highlighting that organisational adoption, rather than technical performance, is increasingly the limiting factor.
