Photo: Thomas M. Barwick INC/Getty Images
The National University Health System in Singapore is moving to scale AI and predictive analytics across its cluster, shifting from pilot use cases to enterprise deployment tied to value-based care outcomes, operational performance, and future reimbursement models.
This comes amid a network-wide push to operationalise data for quality and safety management, using predictive models and real-time analytics to guide clinical decisions, manage patient risk, and measure care outcomes more consistently across institutions.
NUHS's progress is documented in a newly released Clinical Quality & Patient Safety e-book, cataloguing innovations, lessons, and best practices in using data and digital tools to strengthen care reliability across the academic health system.
Much of this work, NUHS said, is coordinated through its Institute of Clinical Quality, which provides the governance, training, and data infrastructure required to evaluate, implement, and scale quality and patient safety innovations across the cluster.
In an interview with Healthcare IT News, Dr Ling Zheng Jye, Chief Medical Informatics Officer at National University Hospital (NUH), flagship hospital of the NUHS, discusses the cluster's shift from AI pilot projects to enterprise deployment and the governance and workflow redesign needed to operationalise real-time decision support.
Dr Ling, who is also senior assistant director of the NUHS Regional Health System Office, also shares how NUHS is shaping its workforce for algorithm-assisted care and addressing structural barriers to scaling data-driven safety innovations across a regional health system.
Q. Across the NUHS cluster, how would you characterise the current maturity of AI and predictive analytics deployment within value-based care models, particularly in moving from pilot use cases to system-wide operational and reimbursement impact?
A. Across the NUHS cluster, AI and predictive analytics have progressed from early experimentation into a more structured scale-up phase, particularly in areas that support population health management and chronic disease care. In the past, many AI initiatives focused on discrete pilots, such as patient load prediction, risk stratification, or documentation support, often led by individual institutions or departments. Today, there is a clearer emphasis on institutionalisation, where AI solutions are evaluated for cluster-wide relevance, sustainability, and alignment with care plan and performance management objectives.
While direct reimbursement impact from AI remains an evolving area, NUHS has laid the foundation for this transition. Predictive analytics are increasingly linked to value-based care indicators, such as chronic disease control and avoidable utilisation reduction, rather than being treated as standalone technical tools. This shift is critical because reimbursement and pay-for-performance models ultimately depend on consistent measurement, reliable data pipelines, and clinical adoption at scale.
At the same time, NUHS has become more selective about what scales. There is a strong preference for AI capabilities that can be embedded within core enterprise platforms – particularly the NGEMR (Epic) – or supported as shared cluster services. This reduces fragmentation and ensures that insights can be operationalised consistently across care settings.
Overall, NUHS is moving deliberately from “proof of concept” to proof of value, recognising that system-wide impact requires not just algorithms, but governance, workflow integration, and measurable outcomes that can support future reimbursement models.
Q. Many health systems struggle to convert quality and safety data into real-time clinical decision support. What governance, workflow redesign, or digital infrastructure changes have been necessary within NUHS to ensure data insights are acted on at the bedside?
A. NUHS's experience reinforces a key lesson in digital health: data does not change care unless it is embedded into clinical workflow and supported by governance. Several changes have been essential in moving from retrospective reporting to real-time clinical action.
First, NUHS has established a formal AI governance framework to oversee the development, deployment, and monitoring of AI in healthcare settings. This provides clarity on accountability, risk assessment, and post-deployment review, especially for AI that influences clinical decisions. Governance has shifted the focus from "can we build this?" to "should this be used, and under what safeguards?"
Second, there has been a deliberate workflow redesign anchored in NGEMR (Epic). Rather than relying on separate dashboards or reports, NUHS surfaces insights at the point of care – during pre-consult preparation, within the consultation, or immediately post-consult. This ensures that clinicians receive decision-support recommendations at the exact point of care.
Third, NUHS has strengthened its data platforms and validation processes. Reducing data fragmentation, improving data quality, and shortening the time from data capture to insight are critical for quality and safety use cases. Without trusted and timely data, clinicians will not act on recommendations, regardless of how sophisticated the analytics are.
Together, these changes reflect a shift from passive data consumption to active, workflow-integrated decision support, supported by governance that builds clinician trust.
Q. As AI becomes embedded in clinical risk detection and decision-making, what new workforce capabilities, training frameworks, or accountability structures are required to ensure clinicians can safely interpret and operationalise algorithm-driven insights?
A. As AI becomes more visible in everyday clinical workflows, NUHS has recognised that workforce readiness is as important as technical accuracy. The key requirement is not for clinicians to become data scientists, but to develop practical competencies in interpreting and safely using AI-enabled insights. For example, the Kent Ridge Office of Innovation has trained 10% of NUH staff in AI literacy, and there is an ongoing structured framework to increase this figure progressively.
NUHS has focused on building applied AI literacy, helping clinicians understand an algorithm's purpose, limitations, and how to critically appraise its outputs. Training emphasises real clinical scenarios, such as risk flags, recommendations, or summarised insights encountered during consultations, rather than abstract AI concepts.
A train-the-trainer model has been important. Clinical leaders and informatics champions are equipped first, so they can contextualise tools and AI for their teams and reinforce good practice. This approach supports consistent adoption while respecting differences across specialties and care settings. This year, our allied health professionals are being trained in tools like speech-to-text and Epic personalisation to improve care delivery.
Clear accountability structures remain critical. NUHS upholds the principle that AI supports but does not replace clinical judgment. Clinicians remain responsible for decisions, and AI outputs must be validated within existing clinical governance frameworks. This expectation is reinforced through policy, training, and oversight mechanisms. Ultimately, safe operationalisation of AI depends on a workforce that is confident, critical, and accountable – able to use AI as a decision aid while retaining professional responsibility for patient care.
Q. Lastly, what have been the biggest barriers to scaling data-enabled quality and patient safety innovations consistently across multiple NUHS institutions, and how is the health system addressing variation in digital maturity, interoperability, and change management?
A. Scaling data-enabled quality and patient safety innovations across NUHS has highlighted several persistent challenges. The most significant barriers have been variation in digital maturity, data fragmentation, interoperability constraints, and the human aspects of change management across diverse institutions.
Different teams within NUHS are at different stages of digital readiness, which affects how quickly innovations can be adopted and sustained. To address this, NUHS has prioritised common platforms and shared governance, reducing reliance on bespoke solutions that cannot scale. Enterprise systems and cluster-level standards help narrow variability while allowing local adaptation where necessary.
Interoperability remains a technical and organisational challenge. NUHS has taken a pragmatic approach by focusing first on high-value use cases and strengthening data pipelines that support them, rather than attempting to solve all interoperability issues at once. This incremental strategy enables learning while delivering tangible benefits.
Change management is equally critical. Clinician engagement, leadership sponsorship, and clear articulation of clinical value are essential for adoption. NUHS increasingly applies stepwise roll-outs, pilots with clear success criteria, and structured feedback loops before scaling cluster-wide.
Together, these approaches reflect an understanding that sustainable scaling is not just a technology problem, but a system-level transformation that requires alignment across people, processes, and platforms.
