GMV highlights algorithmovigilance as key to safe AI in healthcare at Ennova Health Day 2026

GMV took part in the 4th Ennova Health Day, organized by Diario Médico and Correo Farmacéutico. The event brings together healthcare professionals, public officials, technology companies, and innovation experts to examine the main challenges involved in transforming the healthcare system.

Under the theme “The Reality of AI Use in Spanish Healthcare,” this year’s gathering focused on the practical application of artificial intelligence (AI) and the complexities of integrating it into clinical practice and healthcare management.

Juan Carlos Sánchez Rosado, Health Business Executive at GMV, joined the session “Tangible Innovation: Success Stories Transforming the Healthcare System.” He presented GMV’s approach to algorithmovigilance, emphasizing the need for continuous monitoring of AI models once they have been deployed in healthcare settings.

From AI to AI-driven iatrogenesis

Under the title “Algorithmovigilance vs. AI-driven Iatrogenesis,” the presentation raised a fundamental question: how can we know whether a model that performed correctly when it was validated remains reliable as patients, data, or healthcare settings change?

To illustrate this challenge, GMV introduced the concept of “AI-driven iatrogenesis,” a term used to describe clinical harm caused or exacerbated by an AI system. Such harm may not be apparent when examining individual patients, becoming visible only through longitudinal analysis or the study of specific population subgroups.

Unlike other clinical incidents that can be identified in individual cases, certain algorithmic deviations or biases may remain hidden until the system’s performance is analyzed across a broader population. This underscores the importance of mechanisms for continuously monitoring how models perform once deployed in real-world conditions.

Monitoring to detect warning signs and take action

In response to these challenges, algorithmovigilance provides a continuous process of monitoring, signal detection, analysis, and action.

Key areas to monitor include fairness, to identify differences in model behavior across population subgroups; drift, to detect changes between current conditions and those under which the model was trained; quality and performance; and explainability, to understand which variables are influencing predictions.

GMV illustrated this approach with a practical example involving a clinical decision support system in primary care. The example showed how a change in patient demographics could cause the model to underperform in a particular age group, a problem that might go unnoticed when examining cases individually.

By monitoring results at an aggregate level, teams can detect warning signs and launch a joint investigation involving technical and clinical specialists to determine what is happening and why. Based on the findings, corrective and preventive measures can be taken. These may include retrospectively reviewing potentially affected cases, informing healthcare professionals, adapting or retraining the model, and increasing monitoring for specific groups.

Governance does not end with deployment

GMV’s participation in Ennova Health Day highlighted a challenge that will demand greater attention as AI becomes more widely used in healthcare: AI governance does not end once a model has been validated and deployed.

Ensuring its safety and reliability requires continuous monitoring of its performance in real-world conditions, detecting potential changes, and having the processes, metrics, and teams in place to intervene when necessary.

As patient populations, data, and clinical practices evolve, models can begin to behave differently than expected. The key is having the right tools and mechanisms to detect these deviations in time and turn every warning sign into an opportunity to learn and improve.