Noelia Hernández
For weeks, early cancer screening has been in the media spotlight in Spain, with much discussion about the problems faced by the public health system in carrying out these tests and its lack of management.
All this takes me back to several professional projects over the last decade. During this time, I have participated in the development and validation in hospitals of four AI solutions for the early detection of cancer. Both the developers and the hospitals were from different countries, not just the European Union. As a result, my experience and knowledge of healthcare and innovation systems has been deep and varied.
Although they focused on different types of cancer (breast, skin and colon), all four were based on similar premises. The problems to be solved were clear:
- Decrease in the number of healthcare professionals and difficulty in recruiting them in public systems
- Increase in the number of patients undergoing screening due to the ageing population and the increase in incidence among the young population
- Advancement of the age of commencement of screening by decision of the EU
- Need to increase the efficiency of the budget allocated to healthcare expenditure
The increased pressure on healthcare professionals and the inability to increase staffing levels and physical resources (machines, hospitals, etc.) had a direct impact on the quality of medical services. This was common to many countries. Radiology services were a clear focus of patient backlogs.
Therefore, the option of digitising part of the early detection and cancer prevention process was extremely positive and had a direct impact on how the service was perceived by professionals and patients. Artificial Intelligence was the perfect technology for the automatic detection of possible lesions or malignant tumours. For professionals (especially radiologists), it would reduce their workload and the time spent on each patient, as AI would do the initial screening, ruling out healthy patients and detecting/analysing those with possible malignant lesions. For patients, it would reduce the time it took to receive diagnoses and uncertainty, and avoid biopsies and invasive treatments to analyse doubtful cases.
This was the theory and basis on which the development of the solutions was built and the motivation for presenting it to public and private hospitals, managers of these healthcare systems and professionals. All this at a time when there was no ChatGPT and many of today’s supposed AI specialists did not even know what a data repository was.
I developed these projects with companies from different European countries. And they were validated in hospitals in Europe, North Africa, Canada, the Middle East and Latin America. As you can imagine, the healthcare systems were very different, as was their level of technological adoption. But this provided an unparalleled learning curve: working with all kinds of professionals, in environments with very different levels of digitalisation and with different patient profiles. This was essential in order to have the most competitive solution possible, for two reasons: there are differences in cancer incidence depending on race, habits, socio-economic profiles and environmental settings; and in increasingly multiracial societies, it is essential to have tools that analyse behaviour taking these variables into account. This is a mistake I see very often in Spain and Europe: innovation is developed and validated only in the local environment (at most nationally), working only with fellow companies and organisations. This adds a huge bias to the results and limits the target market.
And after all this, we ask ourselves: if these AI solutions would improve the efficiency of the healthcare system and reduce waiting times for screening and testing, why are they not being deployed in many countries?
Well, that’s the question. After developing and validating these early cancer detection solutions in pilot projects, the conclusions are that there is much room for improvement to facilitate technological adoption. Examples include:
- The availability of big data to develop AI algorithms. In systems such as the Spanish one, where even an image test of a patient carried out in another public hospital in the same city is not shared, this is a huge barrier to obtaining anonymised data that would allow for a relevant data repository. The European Health Data Space (EHDS) will arrive a little late, as, for example, medical images will not be available until 2031.
- Interoperability is a nightmare to implement and integrate this type of solution and make the possibilities for scaling up a reality. There are technical difficulties with existing systems, which are often legacy products, fragmented (each local system has its own architecture) and met with reluctance on the part of consultants or supplier companies.
- The periods for certification of solutions. This is a problem that not only causes the demise of start-ups or innovative companies because they cannot move from a pilot to actual commercialisation and deployment, but also delays healthcare systems that want to improve their effectiveness and efficiency. This is particularly serious in the EU. It has been easier and quicker for us to obtain FDA authorisation (which is not exactly known for being easy for non-US companies) than to operate in European countries.
- Reluctance on the part of healthcare personnel. This is often due to their own ignorance about the functionalities of an AI tool, as they cannot operate without the supervision of a medical professional. It is simply a way to gain productivity and avoid automatable work.
- Barriers to launching pilot projects in some public systems. There are many differences from one country to another. While in some it was easy to establish ad hoc agreements, in others we had to assist and guide the public hospital to launch a public procurement of innovation. And, again, this tool can be useful for development and piloting, but today it falls short for more ambitious challenges.
- Agility in adopting new technologies. Here we come up against administrative slowness and the sclerotisation of public systems. It has been much easier for private hospitals to adopt these technologies. Meanwhile, in public systems we are now talking about launching pilot projects with funding from the state or the regions. And this is not just a Spanish issue. You can look up how many examples there are of AI being used in imaging and diagnosis in the EU. I can tell you, having experienced it myself: few and far between. And I’m not just talking about the AI Act (someone will come out and say it), because some of these experiences predate even the idea of publishing it.
Therefore, there are problems in the management of health systems that could be solved and improved with the use of new technologies and innovation.
We can talk and philosophise about the creation of innovation ecosystems, the facilitation of technological entrepreneurship, the commitment to Deep Tech in Europe, putting the patient at the centre of innovation, the importance of the public sector as a lever for innovation. However, many of these barriers to adoption and, therefore, to improving service quality and user satisfaction are administrative, legal and bureaucratic in nature. Those related to education and training can be managed directly by the developing companies, but this is not the case for exogenous agents.
When discussing Europe’s technological leadership and competitiveness gains, it may be necessary to examine and focus on all these issues rather than writing endlessly about the sex of angels and the abstraction of innovation. Because, unfortunately, many of those potential European unicorns in healthcare AI have already fallen by the wayside.