Every day, thousands of people go to health centres throughout Spain. There, they undergo various tests, are offered a diagnosis and treatment. But like many other aspects of life, artificial intelligence (AI) is transforming the healthcare system, offering new tools to improve diagnosis, treatment and epidemiological surveillance. Since the emergence of advanced language models such as ChatGPT and Med-PaLM, AI has demonstrated its ability to outperform medical tests and enhance personalised medicine. Key applications include disease diagnosis, disease prognosis, risk assessment and treatment optimisation, especially in chronic diseases and complex cases.
In the healthcare system as a whole, AI can contribute in areas such as epidemiological surveillance, helping to detect outbreak patterns and modelling the evolution of epidemics. In Spain, where the healthcare system is recognised as one of the best in the world, AI integration is being strengthened with initiatives such as the Digital Health Strategy 2021-2026, which seeks to optimise the use of medical data and foster interoperability between European countries. This allows AI models to be trained on diverse data, increasing their predictive capacity and robustness.
In conclusion, AI promises to be an essential ally in the healthcare system, not to replace professionals, but to work in synergy with them, optimising processes and improving medical outcomes and patient care. Progress in its integration depends on the ethical and responsible use of data, as well as collaboration between institutions and countries to maximise its benefits.
INSIDE. An artificial brain to boost healthcare
Artificial Intelligence (AI) is a technology that has been creeping into different aspects of life. From 2022 onwards, following the take-off of large language models (LLMs) by ChatGPT, we have seen more and more companies and institutions jumping on the AI bandwagon to stay at the forefront of the modern era. However, as with any technology, in order to apply AI in a way that really benefits society, it is first necessary to study very carefully the places where it can fill current gaps. In other words, in order not to waste AI's capabilities, it is important to understand both its advantages and its limitations.
These capabilities are particularly interesting in medical care. On a visit to a hospital, doctors collect a large amount of data from patients, compare it with the knowledge they have acquired or other similar cases, and after reasoning offer the most likely diagnosis. This workflow is also one of the strengths of AI: collecting data, comparing it with their information base, and providing a coherent answer. The potential of these systems to improve medical care and health outcomes can therefore be glimpsed at a glance.
Even more interesting is that, already in June 2022, ChatGPT 3.0 successfully passed (with more than 60 %) some of the US medical licensing exams without any specialised reinforcement.[i]. This was considered a milestone, as the test consists of clinical cases that can be complex and require an understanding of the linkage of different factors. Just 5 months later, in July of the same year, Google presented Med-PaLM[ii], an LLM system that consistently passed the exams with around 65%. And in 2023 it introduced Med-PaLM2, which not only passes the exam, but also scores around 86.5%.[iii], This is a result that only the best students or experts can achieve.
But replacing all doctors with AI is certainly not a sensible option. To reap the full benefits of this technology, the integration process needs to be optimised in a number of ways to create the best synergy between healthcare professionals and AI. In particular, some experts[iv] have identified 8 areas of interest where AI could directly help healthcare professionals in their dealings with patients.
- Diagnosis: AI could help predict the presence or absence of disease as a doctor would. This can be based on symptoms, tests, patient history or other relevant data. These diagnostic predictions make it possible to find a treatment in time to find a cure for the disease.
- Prognosis: In addition to diagnosis, thanks to their huge database, models can help predict both the evolution and the likely outcome of a disease. Such predictions provide insight into how the disease will progress with or without treatment, as well as whether there are potential complications to consider.
- Risk assessment: AIs can collect patient data such as genetics, lifestyle, exposure to environmental factors and other health problems. In this way, they may be able to predict a patient's likelihood of developing a disease in the future and act on it before they do.
- Response to treatment: One of the most interesting ideas where AI models can be used is to enhance personalised medicine. With individual patient data, especially genetic data, much more precise drug doses can be established that take into account genetic or metabolic factors of each patient. In this way, by tailoring treatment to the individual patient, it is possible to maximise benefit while minimising risk.
- Disease evolution: In chronic diseases, such as diabetes, or neurological disorders, it is crucial to predict how the disease will evolve over time in order to anticipate care needs. AI could play the role of a strategy planner for different scenarios and, again, personalise treatment for each individual.
- Risk of readmission: With all the above data, AIs could also help to identify those patients who have a higher likelihood of readmission. Thus, both healthcare professionals and hospitals as a whole can be prepared to act effectively in the event of relapse.
- Risks of complications: In any medical procedure there is a risk of complications. Identifying them, detecting them, and trying to find a remedy so that they occur less frequently could be an interesting field for AI.
- Predicting mortality: Finally, AI could also predict patients' mortality risk in the face of disease. In this way, it could help to make treatment decisions, as well as to plan the palliative care offered to the patient at the end of life.
But these uses only take into account the potential benefits of AI in the medical practice. In other words, it is a kind of companion that can help the doctor or indicate that he or she has missed something.
Finally, one of the key areas of the healthcare system is drug development. In this area, pharmaceutical companies are increasingly using AI tools in the R&D phases. For this reason, the EMA has established in May 2025 a roadmap for the management, analysis and sharing of medical data through a document called «.....«Seizing opportunities in a changing medicines landscape».» Weighing opportunities in a changing medical landscape[v]. The idea is also to establish a coordination framework for dealing with new legislative initiatives in the European Union.
The health system involves many other actors without whose work, most likely, none of the above would work. A clear example is epidemiologists, who are constantly reviewing cases of certain diseases to contain possible outbreaks and alert facilities and citizens to the risks. In Spain, this work falls to the National Epidemiological Surveillance Network (RENAVE).[vi], coordinated by the National Epidemiology Centre of the Carlos III Health Institute (ISCIII). To facilitate its work, an AI specialised in detecting these patterns could allow early detection and establish guidelines at the crucial moments of future epidemics and pandemics: their beginnings. In addition, it would allow predictive models of the evolution of contagions, as well as help in the development of vaccines and drugs.
As the Ministry of Health points out, the COVID-19 pandemic revealed the existence of certain weaknesses and deficiencies in public health surveillance. Especially in the face of exceptional cases such as those that occurred during the pandemic. As a result, a series of transformations are being carried out to respond to present and future risks, based on lessons learned from the mistakes of the past. Figure 1 reviews the different ways in which data can be used to drive these initiatives.
In the decalogue established by the meeting of public health surveillance professionals to advance in the development of the Public Health Surveillance Strategy in 2023 [vii], In the case of electronic records, one can read that the central focus should be the electronic record. Many of these records are standardised, i.e. they follow the same patterns. However, there are still some centres where it has not been possible (due to lack of resources or for other reasons) to follow the patterns of other hospitals. However, standardisation is not only good for the centres. Having data ordered in a consistent way can serve as a stepping stone for a proper input of AI-based virtual assistants that can help with epidemiological data analysis.
In the specific case of our country, this reinforced surveillance is critical, as new diseases are emerging that threaten people's health. As we have seen in recent years, due to climate change and the establishment of tropical species, diseases that were once eradicated, such as malaria, are now becoming more common.[viii], rickettsiosis[ix] or African trypanosomiasis[x] could become more frequent. In this scenario, it is crucial to keep them under surveillance in order to know the level of risk, the hot spots, and to be able to issue early warnings in the event of any case. In addition, it is also necessary to continue with the development of legislation to regulate the surveillance network at state and autonomous community level.
By sharing medical data between countries in an ethical and responsible manner, AIs could be trained on more diverse populations and conditions, thus increasing their robustness and ability to be accurate on any medical issue. They will also be able to warn of potential epidemiological outbreaks that can be quickly transmitted to neighbouring countries and alert health centres so that they can strategise on any mishaps.
To implement the tools, the first step is to create them, and to do so, hundreds of researchers are developing and training AI models that allow them to do in seconds what would take a person decades to do. This is the case of the Biomedical Genomics Lab, led by ICREA researcher Dr Núria López-Bigas at the Barcelona Science Park. In the lab, they have developed a computational tool that identifies the mutations that cause cancer in each type of tumour. In this way, they can personalise the therapy for each type of patient and choose the treatment with the highest probability of success for the specific mutations in each tumour. [xi]
The tool, called BoostDM, is also able to simulate every possible mutation within each gene and indicate how it might evolve, allowing oncologists to stay one (or several) steps ahead of the cancer and make the right decisions.
AI is also currently being used to facilitate the interpretation of X-rays. Specifically, since April 2025, most hospitals in the Valencia Region have been participating in a pilot study. The image, after being analysed by AI, is always interpreted and displayed by a physician, who makes the decision. To guarantee the validity of this algorithm, validation was carried out in two phases. The first consisted of a retrospective study in which images that had already been analysed and diagnosed by radiologists were selected to be read by the artificial intelligence and, in this way, to be able to make a comparison of their effectiveness. Subsequently, in the prospective study, they corroborated these results in healthcare practice.
The results speak for themselves: the AI achieved more than 90% reliability and, more importantly, a high negative predictive value. This means that, if the AI does not detect any pathology, there is a minuscule chance that it is wrong.[xii].
Figure 2 reviews the various applications of AI to improve the healthcare system. It is also being used to reduce the bureaucracy that limits the activity of doctors. In the case of the Madrid region, a tool is being considered for implementation in 2026 that would reduce the time spent by healthcare professionals filling in reports, forms and documents by up to 70%. The idea is simple: in each consultation, an AI will listen to the conversation between doctor and patient and transcribe the relevant data into a document. To ensure privacy, the audios will not be stored, only the transcribed document with the relevant words. The idea behind this idea is to optimise the work of clinical staff and aims on the one hand to reduce the time spent manually entering information into the Electronic Health Record and thus improve the professional's care.[xiii]
So, as in other parts of our lives, it will not be uncommon to see AI making its way into our healthcare system. AI in healthcare offers a whole world of possibilities to explore, a world in which, first, real intelligence is required to find where this companion is really useful. If applied correctly, we can safely say that we are at the beginning of a revolution. A revolution that promises to reshape and make more accessible the resources at our disposal in order to reduce tedious and superfluous work and add value to each and every one of the hours that healthcare professionals dedicate to their work. In other words, AI in healthcare promises to be that companion that you turn to when there are problems, to ask for advice and that offers solutions based on the enormous amount of data it has in its silicon head, commonly known as “servers”.
IN ACTION. The new era of medical-machine collaboration
With 4.5 billion people without access to essential healthcare services, artificial intelligence (AI) could help close the gap, but healthcare is below average in AI adoption compared to other sectors. In the UK, where around 350,000 people are transported by ambulance to hospital each month, a study has found that, in 80% of cases, AI could correctly predict which patients should be transported.[xiv]. Generative AI in the healthcare market is expected to reach $2.7 billion by 2025 and close to $17 billion by 2034.[xv]. Health services will undergo a fundamental transformation from information-centric systems to solution platforms for automation, a revolution that will inevitably shape future operational approaches across multiple sectors.
In the next five years, AI has the potential to transform pandemic preparedness and predict the impact of disease outbreaks on individual patients. Recent advances in AI methodologies work increasingly well even with limited data, a major barrier to adoption to date, and open up new possibilities in both high- and low-income countries.[xvi]. However, the difficulties that still hinder its expansion when a large amount of context is required will have to be overcome.[xvii] to create more complete and accurate epidemic forecasts[xviii]. Organic and social networks, from human immune systems to population mobility, are by definition adaptive, and any perturbations affect the accuracy of predictions. New types of AI and new approaches will be needed to overcome compartmentalised views.
A notable development, at the point of convergence between the possibilities opened up by AI technologies and the dynamic reality to which they must be applied, is the emergence of Regulatory Automation Management Systems (RAMS), which are on the verge of replacing traditional Regulatory Information Management Systems (RIMS) in the United States. Thanks to them, companies such as Weave Bio and Collate are managing to reduce US Food and Drug Administration (FDA) document processing times from months to just days.[xix].
For its part, the European Health Data Space (EEDS)[xx] is presented as a key pillar of the future European Health Union.[xxi] and opens up a wide range of opportunities for the implementation of new AI solutions to help improve system management. It is the first to emerge within the European Data Strategy.[xxii], The new Directive, which entered into force in 2025, promotes the exchange of information for healthcare across the EU, while fostering a true single market for electronic health record (EHR) systems and creating a reliable model for re-using health data for research, innovation, policy-making and regulation.
To facilitate AI deployment, it will be complemented by specialised data infrastructures (such as 1+Million Genomes, Cancer Image Europe or EOSC-Life), together with the high-performance computing network that drives Europe. Up to 39 organisations from the EU health community welcomed the publication of the EEDS in a joint statement.[xxiii]. They said it is essential to avoid unnecessary bureaucracy and workload for healthcare providers and professionals, as well as to overcome fragmented interpretations of legislation. It is complemented by the Recommendation on a European Electronic Health Record Exchange Format (EEHRxF), which sets the framework for secure, interoperable and cross-border access to information.[xxiv].
The European Commission forecasts that improved access and exchange of health services will yield a still modest €5.5 billion in savings over ten years. Several challenges remain to ensure the effective and efficient implementation of AI tools and to expand their equitable and fair adoption in clinical practice. The AICare@EU initiative is designed precisely to remove the main barriers and the EU-funded SHAIPED Project pilots the validation and implementation of AI models and tools using the EEDS infrastructure. In addition, the European Commission supports Member States in health workforce planning, with the aim of retaining professionals and upgrading and retraining workforces, in the framework of the Skills Pact.
AI strategies in healthcare must also be articulated on the assumption that hospitals and healthcare systems are facing a growing wave of cyber-attacks driven by the high economic value of patient data, including EHRs. The healthcare sector has indeed become the most targeted sector in the EU in the last four years, including during the COVID-19 pandemic.[xxv]. 71% of the cyber assaults with effects on patient care, such as delays in treatment and diagnosis, and difficulties in accessing emergency services, are ransomware.[xxvi]. In this respect, it is important to be aware that, as digitisation expands (an average of 79% of EU citizens have access to the Internet), the EU is becoming more and more digitalised. online to their electronic medical records in primary care[xxvii]) increases the attack surface. Add clinical information systems, hospital workflow systems, medical imaging systems and devices used for diagnostic purposes or for patient monitoring, and the range of potential targets for a cybersecurity attack is immense.
The American Medical Association often refers to «augmented intelligence» to emphasise that the purpose of AI should be to assist and not replace healthcare professionals, who are still needed to provide clinical context to algorithms and to translate their findings into decisions for the benefit of patients. However, some uses of AI in healthcare remain outside the scope of the FDA in that country, because certain software is excluded from the definition of a medical device.[xxviii]. Achieving a successful AI governance ecosystem is ultimately one of the greatest challenges facing national health systems. It requires a detailed understanding of the competencies and capacities of the different actors, the establishment of frameworks for action, and the assignment of responsibilities to entities with the necessary operational capacity.[xxix]. Perhaps this is why, as Figure 3 shows, investors are hesitating after the surge in interest following COVID-19.
AI governance is in fact a relatively new concept for health systems.[xxx]. It involves reviewing and evaluating individual AI tools to ensure that they can be used safely and effectively, and that they comply with applicable legislation. A study of just six US health systems found that some had integrated AI governance alongside that of other software tools; others designed a completely separate governance process; and others adopted a hybrid approach. When California Attorney General Rob Bonta asked the CEOs of hospitals in his state by letter in September 2022 to send him a list of all the software tools they use, he saw a clear need for a centralised inventory and a standardised evaluation system.
AI systems inevitably absorb and perpetuate under-representation of certain groups or gender biases in the treatment of the data they are fed. One AI used in several US health systems showed bias in prioritising healthier white patients over sicker black patients for additional care management. It had been trained on cost data, not care needs data.[xxxi] Without appropriate precautions, algorithms can predict lower health risks in populations that have historically had less access to health services, not because they are healthier, but because there is less documented use of health care. This is compounded by climate change, which is exacerbating existing social and health inequalities by increasing vulnerability to the emergence and resurgence of infectious diseases such as malaria, dengue and Zika.[xxxii]. Finally, there is a critical need for data scientists in health systems, who could be asked to take an «Oath of the Data Scientist», similar to the Hippocratic Oath, that enshrines their specific commitment to addressing algorithmic bias.[xxxiii]. Specific blockchain solutions for AI-based healthcare systems offer a solution to the privacy and security issues holding back the widespread adoption of AI applications in healthcare.[xxxiv].
Health systems face three challenges in addressing algorithmic bias: the lack of clear definitions and a standard of fairness; insufficient contextual specificity; and the «black box» nature of algorithms.[xxxv]. Data scientists, clinicians and patients want, need and have the right to know how a particular outcome or prediction was produced by an algorithm. At the same time, teams developing AI applications must be aware of the specificities of the health system context, as well as the different expectations of different groups, to avoid incurring biases while minimising the speed of implementation.
The fact is that, despite the boom in medical AI (the US FDA had cleared 1,016 AI devices for clinical use by September 2024), there is still no evidence that AI is being used in clinical trials.[xxxvi]), doubts persist about their routine use that go beyond technical limitations and focus on trust, not only in AI tools, but also in their creators. Explainable AI« therefore remains a political priority.[xxxvii], The digital tools can perpetuate systemic inequalities within societies if they are not designed and validated with equity.
A Pew Research Center survey found that 60% of US adults, across all demographic groups, express discomfort with the idea of their health care provider relying on AI.[xxxviii]. In Europe, patients do not want their care to be outsourced to a «cookie-cutter» algorithm instead of being guided by human experience.[xxxix]. It calls for support for European technological innovation and greater integration of AI in healthcare settings, ensuring reliable, safe and effective care and diagnosis.[xl].
AI can automate tasks to free up time and allow doctors to focus more on their patients, «humanising» care in new ways.[xli]. Burnout syndrome in healthcare professionals has increased in the United States, especially in primary care, and the use of EHRs is a key factor in this drift. Documentation and administrative burdens, complex usability, electronic messaging and inbox, cognitive load and time demands have inherently changed with respect to paper records.[xlii]. Many administrative tasks have also become additional responsibilities for physicians. The consequences of this burnout for medical care can materialise in the form of medically relevant errors, poor quality of care, safety incidents, reduced patient satisfaction and turnover of primary care staff.
Nearly 80% of hospitals[xliii] and 86% of outpatient clinics in the United States implemented an EHR model in 2015 and 2017, respectively. The prevalence of physician burnout syndrome has increased across all specialties in recent years, reaching significantly higher levels of nearly 50% in primary care in the United States.[xliv]. About 75% of people with burnout symptoms identify EHR as a source of stress.[xlv]. In a survey of 282 physicians at three institutions in California, Colorado and New Mexico, 68% of whom worked in primary care, the most prominent concern (86.9% of respondents) was excessive data entry. In fact, physicians may need up to two additional hours for every hour of direct patient contact for administrative tasks. Physicians who do not have enough time for documentation are 2.8 times more likely to show symptoms of burnout. In some cases clinic schedules are deliberately shortened and slots are closed to prevent this from happening.
The quality of the user experience with a technology can be measured with the System Usability Scale (SUS) which ranges from 0 to 100. A Google search receives 93 points and a usability rating of A, while Microsoft Excel's score is 57, and its usability rating is F, which places it in the lower 22% band.[xlvi]. The healthcare industry average is 68 points and the EHR (Electronic Health Records) system receives a SUS score of 45.9, making it the worst performing 9%, and an F grade, considered «not acceptable» compared to other products. 90% of primary care physicians say that EHRs need to be more intuitive and responsive, and 72% believe that improving user interfaces would be the best way to address their challenges in the immediate future. It is therefore not just a matter of introducing more AI into patient records. For it to be effective, it needs to be complemented by other technological applications. When doctors at Yale School of Medicine switched to an ID card login system to eliminate repetitive typing of username and password, which they had to do up to 140 times a day, they saved 20 minutes a day, as well as an obvious hassle.
In April 2025, at a Chicago convention centre, tens of thousands of attendees watched as GPT-4 was able to turn a patient interaction into a clinical note in seconds.[xlvii]. The healthcare professional recorded the visit in the AI platform's mobile app, where the patient's information was displayed in real time; at the end of the visit, the AI generated the notes and, after review, sent them to the individual's EHR. It is estimated that primary care teams can use AI-based voice solutions to automatically document conversations with their patients for at least 60% of visits. This could allow doctors to see up to nine additional patients per month.[xlviii].
70% of healthcare workers' tasks could, in fact, be reinvented through technological improvement. In nursing alone, automation can free up 20% of repetitive and less complex tasks, generating a potential annual value of nearly $50 billion in the US alone. Approximately 40% of healthcare labour hours are spent on language tasks that can be transformed by generative AI: 17% can be fully automated, while 23% can be scaled up, to optimise human work efficiency. Integrating AI into diagnostics can significantly reduce costs compared to conventional methods.[xlix], The cost savings could free up between 3.3 and 15.2 hours per day and generate cost savings of between USD 1,667 and USD 17,881 per day in a hospital.[l].
From a more systemic point of view, the concept of the digital twin for health (DT4H) promises to revolutionise the entire healthcare model, including the management and delivery of services, the treatment and prevention of disease, and the maintenance of wellbeing. However, while there are a number of initiatives underway, DT4H is still in its early stages.[li]. In trauma management, promising steps are being taken: the digital twin can start acting before the patient arrives at the health centre, because the collection and reception of information already takes place directly at the scene of the accident, allowing for faster response times, which are crucial when time is of the essence.
Siemens Healthineers and General Electric Healthcare are among the companies partnering with healthcare facilities to simulate changes in workflows and medical equipment. During the COVID-19 pandemic, which could be considered a digital stress test, predictive models were able to anticipate the risk of shortages in critical care resources, such as ventilators as well as extracorporeal life support (ECLS), before they occurred. Future advances in high-performance computing could provide the processing power needed for more complex modelling and simulations. They will also benefit from advances in mobile broadband technology in 5G, which will offer faster data transfer speeds and lower latency. In the field, AR/VR (augmented reality/virtual reality) technologies will enable interaction with digital health twins in a more immersive and intuitive way, and blockchain and DLT (Distributed Logging Technology) will ensure decentralised, secure and transparent data storage and transfer.
Among start-ups, US-based Cohere Health uses AI to proactively create data-driven care pathways, enabling pre-approval of services and making it easier for hospitals to manage bed availability. Using AI-based predictive analytics, they can forecast patient admissions and discharges more accurately.[lii]. Danish Corti's platform, meanwhile, automatically summarises emergency calls, streamlines documentation and monitors employee performance. It helps teams identify inefficiencies and provides personalised feedback to employees.
Ensuring a healthy work-life balance for groups such as nurses is no longer just an ideal, but a necessity.[liii]. It is an inherently demanding profession, with long shifts, high-pressure environments and a commitment to patient-centred care, often at the expense of one's own health. The growing shortage of nurses, exacerbated by global health crises, further accentuates the demand for solutions to ensure care for an ageing population and rising operational costs.[liv]. The United States will face a shortage of between 54,000 and 139,000 physicians by 2033, while the global nursing shortage could reach 13 million.
AI applications are emerging to address this key issue. Compared to manually prepared timetables, computer-generated timetables have proven to be better suited to the needs of the workforce.[lv]. But the challenge of making even rigorously tested systems work well in clinical practice remains. The vicissitudes that the Epic Sepsis Model, one of the most widely used clinical alert platforms, has had to overcome eloquently exemplify this. The Epic Sepsis Prediction Model (ESPM) is an algorithm developed by Epic Systems to predict the risk of sepsis in hospitalised patients using electronic health data. It was built on EHR data from 405,000 patients from three health systems. In a large external validation study, the model failed to identify 67% of sepsis cases. Its failure was attributed, in part, to the implementation of EHR models.[lvi]. Advanced machine learning techniques have consistently outperformed traditional statistical methods in identifying ailments, with accuracy rates of between 55.6% and 95%. However, these models have often been considered «black boxes», making their predictions difficult to interpret. The balance between accuracy and interpretability is one of the great opportunities for technological innovation.[lvii].
During triage, healthcare professionals make discharge and admission decisions based on their assessment of patients' needs and available medical resources.[lviii]. Decisions are often subjective and may lead to admitting «low-risk» patients. The ability of AI to process, connect and draw conclusions from large amounts of data can be used to risk-stratify patients according to their personal factors and needs. Pre-diagnostic AI solutions have been shown to accurately differentiate low-risk patients from high-risk patients.[lix], including from self-reported assessments. NHS Improvement (NHSI) has published a number of tools and reports in the UK to help manage patient flow. SAFER reduces delays in adult inpatient units, and is commonly used in conjunction with Red2Green Bed Days, a visual management system to identify lost time and LOS (length of stay).
The regulatory landscape is having to evolve fast, to keep pace with technological advances.[lx] and address new and emerging risks associated with data sharing. Patients may perceive the latter as a violation of their privacy, especially if findings are made public to third parties. But large amounts of data from various sources are required to train AI algorithms in healthcare, and fragmentation across different platforms and systems makes access to them very difficult. The self-learning function of AI software also poses challenges for regulators as algorithms continually evolve with use. In a study by Panel AmeriSpeak[lxi], In the case of women, the majority of respondents showed little confidence in their healthcare system to use AI responsibly (65.8%) and also that their healthcare system would ensure that an AI tool would not harm them (57.7%). Women stood out for their greater scepticism.
It is likely that these population scruples, coupled with the hyper-cautious approach of the health sector, explain why adoption of generative AI is still generally limited to fragmented pilots. Few models have progressed beyond development or retrospective validation, creating what is commonly referred to as the «AI chasm”. Without being applied in the field and without available data, translating AI algorithms from field-based in silico to real-world clinical settings remains a formidable challenge.[lxii]. Algorithms cannot operate in isolation, they must have infrastructure, resources and skilled personnel trained to act on their predictions. This needs to be addressed urgently because the absence of a holistic AI strategy could put healthcare organisations at risk of falling behind. Boston Consulting Group predicts that this will begin to correct itself next year, when the use of AI across entire workflows, rather than just specific tasks, will expand.[lxiii]. For example, intelligent agents will automate a patient's entire care, from admission to treatment plan.
It is striking that, in the face of suspicions about AI's ability to take on health system tasks, consumers are increasingly using chatbots to make queries. A KFF survey from August 2024 reveals that one in six adults chat with them at least once a month on aspects related to their health and well-being, a figure that rises to 25% for adults under 30. Clearly, if this trend is consolidated, it will lead to a reconversion of care platform models. online and will enhance the transition towards proactive approaches linked to Medicine 3.0, which drives prevention, personalisation and participation with a focus on longevity.
If we look at the glass half full, ultimately, AI will also be able to help spread the word. online and massively disseminate information on disease prevention, reaching large numbers of people quickly and even analysing social media texts to predict outbreaks.[lxiv]. For example, one study found that internet searches for COVID-19 related terms correlated with actual cases of infection.[lxv]. In this case, AI could have been used to predict where an outbreak would occur and then help authorities know how best to communicate and make decisions to help stop the spread.
IN SPAIN. AI in healthcare from primary care
Artificial Intelligence (AI) is becoming increasingly integrated into the healthcare landscape in Spain, with the aim of improving patient care. Asebio detects[lxvi] a growing demand, as currently only 11% of healthcare professionals in Spain use AI, but 42% have explicit intentions to do so in the future. The government has launched initiatives to promote the use of AI in healthcare, such as the AI in Health Strategy programme and the AI in Health Observatory. In the field of university research and technology centres, AI-based healthcare solutions are appearing all the time, focusing on areas such as diagnostic imaging and prosthetics. Microsoft's decision to locate its new AI R&D centre in Barcelona will reinforce all these activities. 92 % of the population in Spain has access to information and the information systems based on Primary Care records (SIAP) have anonymised data on around 12 million people. In addition, a hospital network (CMBD) integrates administrative and clinical information on patients in public and private hospitals. Although there is still a long way to go to break down information silos and facilitate interoperability and exchange, significant progress has been made.
A survey of companies in the health services sector reveals that the personalisation of treatments thanks to the use of AI is one of the points for improvement in our country, as it received 37% of affirmative responses from Spanish companies, compared to 45% for the average of the other countries analysed. There is also room for improvement in the area of medical diagnosis and the updating of medical records. Spain outperforms the rest, however, in the use of AI for treatment planning (49% compared to 46%). The report also identifies significant challenges around patient data security and device management, largely due to problems of technology integration and compliance with legacy systems. In any case, the use of AI in healthcare organisations has grown from 61% in 2024 to 81% a year later, no longer as a purely administrative tool, and is being used to improve advanced healthcare services.
In July 2025, the government announced an ambitious aid plan worth 180 million euros for the application of artificial intelligence (AI), including 50 million euros for the RedIA Salud call, promoted by Red.es. It is financed by the European Regional Development Fund (ERDF) and will address the different stages of the healthcare process, from the prediction and diagnosis of illnesses to the treatment and monitoring of patients. Projects focused on clinical, biomedical and pharmacological research, as well as health system management and emergency response, may also be submitted.
The impact of AI is already producing success stories such as that of the Hospital de Sant Pau in Barcelona, which has gone from having six MRI machines to four because the algorithm reduces timescales and makes it possible to act more quickly. It works like a copilot that helps specialists find lesions, both those they are looking for and others that might be hidden. In the emergency department of the same hospital, the AI makes it possible to start diagnoses in the early hours of the morning, even if the specialist has not arrived. During the months of June and July 2025, the Ministry of Health launched a pilot to incorporate AI in primary care consultations in seven autonomous communities that volunteered: Cantabria, Euskadi, Catalunya, Murcia, Comunitat Valenciana, Castilla-La Mancha and Extremadura.[lxvii]. The tool was used to transcribe the conversation between patient and medical staff, thus eliminating the computer barrier. At the end of the consultation, the artificial system provided a summary of the symptoms, history and family situation, a document that was deleted once the professional generated and validated the corresponding report. The practitioner could focus exclusively on the interview without the need for simultaneous typing. The Ministry planned to implement new functionalities in the future, such as appointment management and the elaboration of pre-consultation summaries based on available patient information.
[i] Sallam, M. (2023) ‘CHATGPT utility in healthcare education, research, and practice: Systematic review on the promising perspectives and valid concerns’.’, Healthcare, 11(6), p. 887. doi:10.3390/healthcare11060887. (Accessed 18/06/2025)
[ii] Singhal, K. et al. (2023) ‘Large language models encode clinical knowledge’, Nature, 620(7972), pp. 172-180. doi:10.1038/s41586-023-06291-2. (Accessed 18/06/2025)
[iii] Singhal, K. et al. (2025) ‘Toward expert-level medical question answering with large language models’, Nature Medicine, 31(3), pp. 943-950. doi:10.1038/s41591-024-03423-7. (Accessed 18/06/2025)
[iv] Khalifa, M. and Albadawy, M. (2024) ‘Artificial Intelligence for clinical prediction: Exploring key domains and essential functions’.’, Computer Methods and Programs in Biomedicine Update, 5, p. 100148. doi:10.1016/j.cmpbup.2024.100148.
[v] European Medicines Agency, Seizing opportunities in a changing medicines landscape - The European medicines agencies network strategy 2028, Publications Office of the European Union, 2025 (Accessed 18/06/2025).
[vi] ISCIII - Public Health Surveillance - Renave ISCIII Web Portal. Available at: https://www.isciii.es/servicios/vigilancia-salud-publica-renave (Accessed on 18/06/2025).
[vii] ‘Conclusions of the Meeting of public health surveillance professionals to advance the development of the Public Health Surveillance Strategy’.’ (2023). Ministry of Health (Accessed 18/06/2025).
[viii] Taheri, S. et al. (2024) ‘Modelling the spatial risk of malaria through probability distribution of anopheles maculipennis S.L. and imported cases’.’, Emerging Microbes & Infections, 13(1). doi:10.1080/22221751.2024.2343911. (Accessed 18/06/2025).
[ix] Oteo, J.A., Santibáñez, S. and Portillo, A. (2025) ‘Hard Tick-Borne Diseases in Spain’.’, Clinical Medicine, 164(12), p. 106989. doi:10.1016/j.medcli.2025.106989. (Accessed 18/06/2025)
[x] Menéndez-Capote, R.L. and Bandera-Tirado, J.F. (2022) ‘Human African Trypanosomiasis’.’, Spanish Journal of Clinical Cases in Internal Medicine, 7(3), pp. 9-12. doi:10.32818/reccmi.a7n3a4. (Accessed 18/06/2025)
[xi] Muiños, F., Martínez-Jiménez, F., Pich, O. et al. In silico saturation mutagenesis of cancer genes. Nature 596, 428-432 (2021). doi:10.1038/s41586-021-03771-1 (Accessed 18/06/2025).
[xii] Health introduces La ia in Chest and bone X-rays to aid in the diagnosis of pathologies in emergency and primary care. - Alicante. (Accessed 18/06/2025).
[xiii] Community of Madrid (2025) The Community of Madrid will use artificial intelligence to transcribe clinical information in primary care consultations., Community of Madrid. (Accessed 18/06/2025).
[xiv] “Artificial intelligence: 10 promising interventions for healthcare”, NIHR, 28 July 2023, accessed 16/05/2025.
[xv] Madeleine North, “6 ways AI is transforming healthcare”, World Economic Forum, 14 March 2025, accessed 16/05/2025
[xvi] Moritz U. G. Kraemer et al. Artificial intelligence for modelling infectious disease epidemics, Nature, 19 February 2025, doi.org/10.1038/s41586-024-08564-w
[xvii] Noah Lloyd, “Northeastern network scientists are developing AI tools to predict - and prevent - the next epidemic”, Northeastern Global News, 17 March 2025, accessed 15/05/2025.
[xviii] Amy Roeder, “Harnessing AI to model infectious disease epidemics”, Harvard School of Public Health, 13 March 2025, accessed 15/05/2025
[xix] Kazi Helal, Aaron DeGagne, “Takeaways From the J.P. Morgan Healthcare Conference”, Pitchbook, 21 January 2025.
[xx] European Health Data Space Regulation (EEDS) - European Commission
[xxi] “The European Health Union: acting together for people's health”, European Commission, 22 May 2024.
[xxii] https://digital-strategy.ec.europa.eu/es/policies/strategy-data
[xxiii] “A Call for Effective Stakeholder Engagement and Capacity Building during the Implementation of the European Health Data Space”, VVAA, 5 March 2025.
[xxiv] “GUIDELINE on the electronic exchange of health data under Cross-Border Directive”, European Commission 2011/24/EU, November 2024.
[xxv] “European action plan on the cybersecurity of hospitals and healthcare providers”, European Commission, 15 January 2025.
[xxvi] Reina, V. and Griesinger, C., “Cyber security in the health and medicine sector - A study on available evidence of patient health consequences resulting from cyber incidents in healthcare settings”, Publications Office of the EU, 2024.
[xxvii] “2024 State of the Digital Decade report”, European Commission, 2024.
[xxviii] “Changes to existing medical software policies resulting from section 3060 of the 21st Century Cures Act”, FDA, 2020.
[xxix] Jee Young Kim et al. “Organizational Governance of Emerging Technologies: AI Adoption in Healthcare”, FAccT ’23, 10 March 2023, doi.org/10.1145/3593013.3594089, accessed 16/05/2025.
[xxx] “AI Governance in Health Systems: Aligning Innovation, Accountability, and Trust”, Margolis Institute for Health Policy, 28 October 2024.
[xxxi] Ted A. James, “Confronting the Mirror: Reflecting on Our Biases Through AI in Health Care”, Harvard Medical School, 24 September 2024, accessed 16/05/2025.
[xxxii] Artificial Intelligence applications to support epidemic and pandemic prevention, preparedness and response, IDRC, n. d.
[xxxiii] “Envisioning the Data Science Discipline: the undergraduate perspective: Interim report”, National Academies of Sciences, Engineering, Medicine, 2018.
[xxxiv] Rucha Shinde et al. Securing AI-based healthcare systems using blockchain technology: A state-of-the-art systematic literature review and future research directions, Transactions on Emerging Telecommunications Technologies, 27 October 2023, doi.org/10.1002/ett.4884
[xxxv] Trishan Panch, Heather Mattie, Rifat Atun, Artificial intelligence and algorithmic bias: implications for health systems, J Glob Health, December 2019, doi: 10.7189/jogh.09.020318
[xxxvi] “Health C for D and R. Artificial Intelligence and Machine Learning (AI/ML)-Enabled Medical Devices”, FDA, 20 December 2024, accessed on 16/05/2025
[xxxvii] Madeline Sagona et al. Trust in AI-assisted health systems and AI's trust in humans, npj Health Syst., 28 March 2025, doi.org/10.1038/s44401-025-00016-5.
[xxxviii] Tyson, A., Pasquini, G., Spencer, Funk C. “60% of Americans Would Be Uncomfortable With Provider Relying on AI in Their Own Health Care”, Pew Research Center, February 2023, accessed on 16/05/2025.
[xxxix] Chiara Longoni , Andrea Bonezzi , Carey K Morewedge, Resistance to Medical Artificial Intelligence, J. Consum. Res. 3 May 2019, doi.org/10.1093/jcr/ucz013
[xl] Medical technology industry perspective on the final AI Act, MedTech, 13 March 2024
[xli] Lisa D. Ellis, “The Benefits of the Latest AI Technologies for Patients and Clinicians”, Harvard Medical School, 30 August 2024, accessed on 14/05/2025.
[xlii] Jeffrey Budd, Burnout Related to Electronic Health Record Use in Primary Care, Journal of Primary Care & Community Health,19 April 2023, doi: 10.1177/21501319231166921
[xliii] Adler-Milstein J et al. Electronic health record adoption in US hospitals: the emergence of a digital “advanced use” divide. J Am Med Inform Assoc., 2017, doi.org/10.1093/jamia/ocx080
[xliv] Shanafelt TD et al. Changes in burnout and satisfaction with work-life integration in physicians and the general US working population between 2011 and 2017, Mayo Clin Proc., September 2019, doi.org/10.1016/j.mayocp.2018.10.023
[xlv] Tajirian T et al. The influence of electronic health record use on physician burnout: cross-sectional survey. J Med Internet Res. July 2020, doi: 10.2196/19274
[xlvi] Melnick ER, Dyrbye LN, Sinsky CA, et al. The association between perceived electronic health record usability and professional burnout among US Physicians. Mayo Clin Proc.March 2020, 10.1016/j.mayocp.2019.09.024.
[xlvii] Shashank Bhasker et al. “Tackling healthcare's biggest burdens with generative AI”, McKinsey, 10 July 2023, accessed 16/05/2025.
[xlviii] How talent and technology can help solve the nursing shortage, Accenture, 2023
[xlix] David B. Olawade et al. Artificial intelligence in healthcare delivery: Prospects and pitfalls, Journal of Medicine, Surgery, and Public Health, August 2024, doi.org/10.1016/j.glmedi.2024.100108
[l] N.N. Khanna, M.A. Maindarkar, V. Viswanathan, J.F.E. Fernandes, S. Paul, M.M. Bhagawati, et al.
Economics of artificial intelligence in healthcare: diagnosis vs. treatment, Healthcare (Basel), 9 December 2022, doi.org/10.3390/healthcare10122493
[li] Evangelia Katsoulakis et al. Digital twins for health: a scoping review, npj Digit. Med., 22 March 2024, https://doi.org/10.1038/s41746-024-01073-0
[lii] M. Javaid et al. Significance of machine learning in healthcare: Features, pillars and applications,
Int. J. Intell. Netw., 5 June 2022, doi.org/10.1016/j.ijin.2022.05.002
[liii] Moustaq Karim Khan Rony et al. The role of artificial intelligence in enhancing nurses’ work-life balance, Journal of Medicine, Surgery, and Public Health, August 2024, https://doi.org/10.1016/j.glmedi.2024.100135
[liv] Tejash Shah, Kaveh Safavi, Daniel Owczarski, “Gen AI amplified Scaling productivity for healthcare providers”, Accenture, March 2025.
[lv] Franklin Leung et al. Artificial intelligence and end user tools to develop a nurse duty duty roster scheduling system, International Journal of Nursing Sciences, July 2022, doi.org/10.1016/j.ijnss.2022.06.013
[lvi] Rachel Knevel et al. From real-world electronic health record data to real-world results using artificial Intelligence, Ann Rheum Dis, March 2023, doi:10.1136/annrheumdis-2022-222626
[lvii] Christo El Morr et al. AI-based epidemic and pandemic early warning systems: A systematic scoping review, Health Informatics Journal, 2024, under revision, doi: 10.1177/14604582241275844, accessed 15/05/2025
[lviii] Fatema Mustansir Dawoodbhoy et al. AI in patient flow: applications of artificial intelligence to improve patient flow in NHS acute mental health inpatient units, Heliyon, May 2021, doi.org/10.1016/j.heliyon.2021.e06993
[lix] He, J. et al. The practical implementation of artificial intelligence technologies in medicine.
Nat. Med., January 2019, doi: 10.1038/s41591-018-0307-0
[lx] Margaret Chustecki, Benefits and Risks of AI in Health Care: Narrative Review, Interact J Med Res, 18 November 2024, doi: 10.2196/53616
[lxi] Paige Nong; Jodyn Platt, Patients’ Trust in Health Systems to Use Artificial Intelligence, JAMA Netw Open., 14 February 2025, doi:10.1001/jamanetworkopen.2024.60628
[lxii] Jethro C. C. Kwong, Integrating artificial intelligence into healthcare systems: more than just the algorithm, npj Digital Medicine, 1 March 2024, doi.org/10.1038/s41746-024-01066-z
[lxiii] Ashkan Afkhami et al. “How Digital and AI Will Reshape Health Care in 2025”, Boston Consulting Group, 14 January 2025, accessed 16/05/2025.
[lxiv] “AI in healthcare: The future of patient care and health management”, Mayo Clinic, 27 March 2024.
[lxv] Atina Husnayain, Anis Fuad, Emily Chia-Yu Su, Applications of Google Search Trends for risk communication in infectious disease management: A case study of the COVID-19 outbreak in Taiwan, Int J Infect Dis., 12 March 2020 doi: 10.1016/j.ijid.2020.03.021
[lxvi] “El panorama de la inteligencia artificial en el sector salud en España”, Asebio, 4 April 2024.
[lxvii] Sofía Pérez, “Sanidad comienza un proyecto piloto para usar la AI en las consultas de Atención Primaria de siete comunidades”, eldiario.es, 16 June 2025