Prof Dato ' Dr Hanafiah Harunarashid is the Master of the Academy of Medicine of Malaysia, and Chief Medical Director of KPJ Healthcare.

" When an institution flattens its clinical records entirely to match an external template, it strips away the very nuances that make its data valuable."

The implicit assumption was simple: aggregate enough data into massive central repositories, expose it to deep learning algorithms, and transformative clinical insights will automatically follow.

Yet, as clinical practice confronts reality, a far more sophisticated truth is emerging. What medicine requires at the bedside is not abstract, uncurated " Big Data," but context-rich, high-fidelity Real-World Data( RWD)— data that faithfully reflects the complex, multi-ethnic, and unique clinical realities of our local populations.

As we navigate this transition, however, we must remain exceptionally vigilant to a subtle, structural trap that is quietly taking root in global digital health partnerships.

THE UNSPOKEN ASYMMETRY

Well-intentioned, highly idealistic global health consortia frequently invite local hospitals and research centres to participate in international AI initiatives. The invitation appears compelling: share local data, gain access to advanced tools, and join the global research vanguard.

Yet, a closer look at the operational dynamics reveals a stark imbalance. First, health records are inherently messy and not immediately research-ready. The work to transform these into research data is labour-intensive, expensive, and continuous, on top of the requirements to maintain data privacy. Inevitably, this is borne almost entirely by the local clinical institution and its staff. Second, the high-value distillation of this labour— the refined algorithmic weights, predictive models, and primary intellectual property— is frequently consolidated elsewhere. Third, in exchange for substantial local investment and patient trust, local clinical teams are often rewarded with little more than a polite mention as junior co-authors on a multi-centre paper, or tempo-

rary licensed access to software built upon their own clinical heritage.

Without deliberate reflection, healthcare systems risk falling into a form of digital vassalage. In this dynamic, local institutions bear the high operational costs of cultivating the raw clinical land, while external entities harvest the valuable mathematical yield.

THE FLATTENING TRAP

To counter this extraction, many have pointed to Federated Analytics as the ideal solution. In a federated framework, raw patient data remains behind the hospital ' s secure firewall; rather than exporting data, the algorithm travels to the hospital, trains locally, and returns only encrypted mathematical updates to a central engine.

Federated analytics is an essential step forward for privacy. Yet, we must recognise a deeper, often overlooked hazard: even with federated analytics, an institution can still lose out if it succumbs to the trap of over-flattening its data.

To enable federated queries across diverse global sites, data is commonly converted into standardised Common Data Models( CDMs). While CDMs create necessary interoperability, the process of fitting complex clinical records into generic schemas inherently flattens the data.

When an institution flattens its clinical records entirely to match an external template, it strips away the very nuances that make its data valuable: qualitative clinical narratives, unique regional disease presentations, and subtle local practice variations.

If we retain only the flattened data model while discarding or neglecting our rich contextual dataset, we give away our unique clinical edge and leave our native intelligence behind.

THE PATH FORWARD

True clinical sovereignty requires that we preserve both interoperability and contextual richness. We must move toward a system that preserves the contextual richness of local data that we can use to generate locally relevant research.

At the same time, we must also create datasets that allow local institutions to run federated queries across global and national networks safely, contributing

to broader scientific research without exposing raw data.

By maintaining this dual structure, local healthcare institutions ensure that their primary clinical assets remain intact. The flattened layer serves the network, but the rich, pure layer remains an enduring, sovereign asset for local innovation, local IP generation, and local clinical stewardship.

CLINICAL WISDOM AND EQUITABLE VALUE

A dual-tier approach transforms our position from passive data donors to active, equal partners in digital health: Protecting Patient Trust: Sensitive patient narratives and contextual health data remain under strict local stewardship in compliance with national privacy frameworks. Sustaining Local Research Capability: Local clinicians retain the full, un-flattened dataset necessary to conduct deep, context-specific medical research tailored to our own population ' s needs. Equitable Value Retention: When external collaborators seek advanced insights, the local institution retains the leverage and capability to generate high-value intelligence, ensuring fair economic and academic return on local investments.

A THOUGHTFUL HORIZON

Global scientific collaboration is vital, and technological exchange must be encouraged. However, genuine partnership must be built on symmetry, mutual respect, and shared sovereignty.

As we shape national policies around clinical AI and health data governance, our focus must extend beyond basic compliance. We must ask: " How do we build a data architecture that allows us to participate globally while preserving our clinical context and native wisdom locally?"

By pairing Federated Analytics with a Dual-Tier Data Strategy, we chart a dignified, forward-thinking course. We ensure that as digital health advances, our institutions do not merely supply raw material for others to refine, but remain the sovereign guardians and primary beneficiaries of their own clinical legacy.- The HEALTH