When a doctor at the Gregorio Marañón Hospital in Madrid prescribes a medication to a patient who is already taking five other different drugs prescribed by three different specialists, an artificial intelligence system analyzes the complete combination in real time and alerts the doctor if it detects a potentially dangerous interaction before the prescription is confirmed. Since this system came into operation in 2025, it has prevented more than 15,000 serious drug interactions in Andalusia. It is not science fiction or an experimental pilot. It is part of the daily functioning of the health system.
In 2026, more than 900 medical devices and algorithms with artificial intelligence have approval from the US FDA, and the European Medicines Agency has authorized more than 200. The number is significant, but more significant is the diversity of applications: diagnostic imaging in radiology, dermatology and ophthalmology; prediction of cardiovascular, diabetic and sepsis risk; early detection of cancer; support for prescription and reduction of medication errors.
Where medical AI works best
Experts identify three areas where medical AI is consistently demonstrating clinically relevant results. The first is medical image analysis: systems trained to detect lesions in mammograms, CT scans, MRIs and fundus photographs have demonstrated in multiple studies a diagnostic sensitivity comparable or superior to that of the human specialist in specific tasks. Most importantly, they detect findings that the human eye misses, not because the doctor is incompetent, but because they process images at a speed and volume that no human being can match.
The second area is risk prediction. Systems trained with data from thousands of patients can identify weeks in advance which hospitalized patients are more likely to develop sepsis, which diabetics are at greater risk of kidney or cardiovascular complications, or which patients with heart failure are more likely to be readmitted within 30 days of discharge. This prediction allows for preventive interventions that reduce both patient suffering and system costs.
Medical AI in numbers — 2026
- FDA approved algorithms: more than 900
- EMA authorizations: more than 200
- Spain — electronic prescription with AI (Andalusia): 15,000+ dangerous interactions avoided
- Spaniards who consider the impact of AI on health positive: 44% (Cigna International Health Study)
- Predicted reduction in diagnostic costs with AI: up to 50% (Harvard School of Public Health)
- Spanish investment in AI for health (National Strategy 2024–2027): 120 million euros
The distinction that the patient needs to understand
Doctors at Osakidetza, the Basque health service, have recently published a warning that deserves attention: there is a fundamental difference between clinical medical AI and general-purpose generative chatbots. The systems that hospitals use to detect sepsis or analyze mammograms are algorithms specifically trained with validated clinical data, evaluated in trials, integrated into medically supervised workflows, and subject to strict health regulation. ChatGPT, Claude, Gemini or Copilot are systems designed to produce coherent text from large volumes of information: they may be useful to understand an already established diagnosis or prepare questions for the consultation, but they do not have the reliability or validation necessary to make diagnostic decisions.
The problem is that the 44% of Spaniards who consider the impact of AI on healthcare to be positive do not always distinguish between both categories. The same study indicates that some people interpret the responses of generative chatbots as conclusive diagnoses and act accordingly without consulting a professional, which is generating a worrying increase in risky self-medication.
Share this article
