Interoperability and Trust: The Hidden Backbone of Healthcare AI

When I look at the global landscape — revisiting recent articles in The Lancet Digital Health, MIT Technology Review, public AI papers in The Lancet, and technical studies — I see a clear common thread: we are at an inflection point where the viability of AI in healthcare will be defined more by the maturity of data ecosystems than by the sophistication of the models themselves. It’s like trying to run a marathon wearing elite shoes but stepping in mud: you might be fast, but you won’t move forward.

At the heart of this debate lie three intertwined dimensions: interoperability, trusted governance of models, and business/regulatory frameworks. Looking internationally, the MIT Technology Review report showed that nearly all healthcare leaders claim to be ready for digital solutions — yet an overwhelming majority admit that the Achilles’ heel is interoperability (91%), and many see it as “hard” to overcome. This exposes a central paradox: the will exists, the money exists, the technology exists — but the communication infrastructure between systems remains fragile.

In The Lancet Digital Health, recent papers have highlighted the urgency of developing medical AI that is auditable, explainable, and trustworthy. An article titled “One Shot at Trust” discusses emerging initiatives for responsible AI networks, emphasizing that clinical acceptance will depend on models that are not opaque black boxes.

On another front, studies on AI in public health argue that it can enhance surveillance, resource allocation, and proactive response — but only if it relies on robust, transparent, and representative data. Otherwise, bias becomes inevitable.

Technical experts have long documented that interoperability is not a “nice-to-have” but a prerequisite for a digital future: it fuels AI and big data, enables communication between physicians and systems, and facilitates collaborative research and cross-border cooperation. To confront the dilemma of data centralization and power concentration (whoever controls the datasets controls the AI), new proposals have emerged — such as federated learning, blockchain, and differential privacy — all designed to train models without moving raw data.

Now, bringing this global map to Brazil: we’re standing at the crossroads of a huge opportunity. We have public scale (SUS), meaningful hospital consolidation, innovative startups, and a latent demand for digitalization. Yet much of the AI applied here still hits the same wall of data silos: hospitals with “closed” medical records, regional networks that don’t share information, vendors with proprietary APIs. Many AI projects get stuck in pilot mode — and never scale.

If we want to move forward, we need to unlock three strategic vectors:

1. National interoperability infrastructureSophisticated AI is useless if data input is chaotic. Public policies (and public-private partnerships) must urgently build standard APIs, incentives for adopting shared representations (such as FHIR and openEHR), and compatibility certifications. Brazil could lead in developing an integrated digital network connecting primary care, hospitals, labs, and surveillance — linking our “digital islands.”

2. Governance and trust in AI modelsModels can’t remain black boxes. We need independent audits, transparency, traceability, and explainability mechanisms. Federated learning and differential privacy approaches make it possible to train models across institutions without exposing raw data. That’s essential to earn the trust of physicians, patients, and regulators.

3. Payment models and smart regulationAI as a “standalone tool” doesn’t pay off. To scale, we need value-based and outcome-based payment models — which only work if reliable metrics can be shared across multiple data sources. Regulation must evolve to recognize and certify digital models, define accountability (who’s responsible for an algorithmic error?), anticipate audits, and promote regulatory sandboxes. Regulators worldwide are already starting to demand interoperability and transparency from innovative health models — for example, UK reports criticize how disconnected regulations and risk-averse cultures delay the adoption of digital therapies.

In clinical practice, imagine a medium-sized hospital in the countryside that currently keeps its records locally closed. If that hospital connects its database to a regional network, an AI algorithm could instantly retrieve exam histories from other facilities, combine them with population data, and deliver individualized risk scores. That changes everything: AI shifts from being a “local trick” to becoming a bridge across networks. Or imagine a startup developing AI-assisted image diagnosis — but it only gets paid based on real performance (e.g., reduced rework, earlier diagnosis, fewer hospitalizations). For that to happen, interoperability and reliable metrics are non-negotiable.

From a regulatory standpoint, Brazil could lead by creating a framework for “Responsible AI in Health” — enforcing interoperability requirements, algorithm certification, and clear liability rules. At the same time, it could foster innovation through regulatory sandboxes, allowing startups to experiment in real-world environments with reduced risk.

My vision for the future: by 2030, we’ll see hospitals, labs, and care networks integrated through national data meshes. Physicians will use explainable AI systems that draw insights from interconnected data, with no boundaries between systems. Startups will have plug-and-play models that “connect” to any certified hospital network — and they’ll be compensated based on real, validated outcomes. We need to build this ecosystem with purpose, not hype.

The most relevant question for Inova na Real, therefore, is not “which AI model to use”, but “how to build the groundwork (data + interoperability + regulation) so that AI becomes a tool, not a promise.” At this crossroads, our mission is to push Brazil not to stumble over the same global challenges — but to leap ahead.