Transforming Healthcare with AI: The NUHS model and its Global Implications
Jordan Zeiger, Tensility Intern and Undergraduate Student at Cornell University College of Engineering
Professor Ngiam Kee Yuan, Group Chief Technology Officer at National University Health System
Armando Pauker and Wayne Boulais, Managing Directors at Tensility Venture Partners
Lee Chuen Ting, Venture Partner at Tensility Venture Partners
Introduction
The National University Health System (NUHS) in Singapore is leading an AI transformation and rollout to enhance patient care and operational efficiency across the hospital system. The NUHS results are remarkable, not only because of the live AI applications, but because of the arduous process it took to achieve this system overhaul – unifying data systems, building trust across the organization, and establishing oversight committees to keep everything running smoothly. This blog post delves into the challenges NUHS overcame to create and sustain its AI-driven healthcare system, offering lessons that could serve as a playbook for other institutions globally.
NUHS is one of Singapore's three healthcare systems providing care through a network that includes multiple hospitals, specialist centers, and polyclinics. As a public institution, NUHS is an important part of the nation’s healthcare system, not only in terms of patient care, but also as a leader in medical research and education, having pioneered a number of medical breakthroughs since its formation. The NUHS integrated approach, combining cutting-edge clinical services with research and teaching, has allowed it to implement a fully functional AI approach to improve patient care.
Implementation Process and Challenges
Data Privacy and Ownership
In Singapore, patient data is typically owned by the healthcare institution, such as NUHS, under the Personal Data Protection Act (PDPA). This centralized ownership allows hospitals to more easily aggregate and utilize patient data for AI-driven healthcare applications. In contrast, the United States has more stringent data privacy laws, where ownership of patient data lies with the individual, governed by laws like the Health Insurance Portability and Accountability Act (HIPAA). This decentralized ownership complicates data sharing and aggregation, as healthcare institutions may often require explicit patient consent to use patient data. This may hinder the implementation of AI systems, as institutions must ensure compliance with more limiting requirements, potentially limiting the scope of data-driven innovations.
The difference in data ownership and regulatory environments between Singapore and the US has significant implications for AI implementation. In Singapore, the hospital's control over patient data simplifies the process of building and deploying AI systems, enabling quicker adoption and more comprehensive use of data. In the US, however, the regulatory landscape may pose challenges. For US institutions aiming to replicate the success of NUHS, understanding and navigating these legal nuances will be critical to utilizing the full potential of AI in healthcare.
Data Consolidation and Governance
The creation of a comprehensive, secure, and accessible data platform at NUHS was a colossal task, taking over three years to complete. The governance was the most difficult part of this process. Initially, data was scattered across multiple platforms, locations, departments, and formats within and across hospitals. Standardizing the Electronic Health Record (EHR) systems across all departments drove the move towards data standardization. The challenge went beyond just consolidating this data to ensuring it could be effectively used by AI systems without compromising patient privacy or security.
Three important technical processes were instituted to enhance data governance:
Data Security - NUHS implemented a secure data platform that adhered to strict policies and procedures for data management. Robust cybersecurity measures, such as remote desktop monitoring and controlled access with no copying or downloading, were also put in place to safeguard sensitive information.
Data De Identification - This involved establishing universal de-identification mechanisms to protect patient privacy while enabling data linkage. To do this NUHS employed one way hashes to de-identify data, while keeping the hash tables secret, administered by an independent trusted third party.
Data Sharing Protocols - When data is shared across multiple departments for research purposes, shared data is put onto a virtual machine that only the sharing groups can access. Once insights are drawn from the data, users may request for processed information to be extracted through the system administrators. However, raw data cannot be extracted from the virtual machine and is deleted at the end of the project. so only the insights can be used and accessed moving forwards. The data consolidation effort was foundational, setting the stage for the advanced AI applications that followed. Without this unified data platform, the subsequent innovations would have been impossible.
Organizational Processes
The human element was as critical as the technical one. Convincing data holders and stakeholders across NUHS to adopt the new system required more than just demonstrating its technical capabilities—it required building trust. Data security, transparency, and a clear demonstration of the system's benefits were paramount in this trust-building process.
Data holders across NUHS did not have to be coerced into using this new data system. Over time, people naturally built trust with the system and added their data to it so they could experience the benefits of the added security, centralized compute, and easier data sharing.
The implementation of the AI systems was also not easy. These novel processes cannot just be put in place and left to run unsupervised. AI systems require continuous monitoring, updating, and refinement. Committees were set up during the planning stages to help set accurate parameters and continue to meet regularly after implementation to constantly monitor, review and update the AI systems’ operation and evolution. These committees of experts meet regularly to oversee the implementation, ethics, usage and accuracy of all systems.
Innovations in AI at NUHS
National University Health System (NUHS) in Singapore is leading a transformation in healthcare by implementing cutting-edge AI systems across its entire network. This public healthcare group, serving a diverse and extensive patient base, is not content with minor improvements; instead, NUHS is overhauling existing processes and introducing innovative systems to create a more efficient, accurate, and patient-centric healthcare experience. Here's an overview of the AI-driven advancements at NUHS, organized by their impact on various hospital operations. NUHS has built their own streaming AI platform known as ENDEAVOUR AI, on which all of the following AI tools operate. They have also established their own private AI model training cloud platform, DISCOVERY AI, running a cluster of NVIDIA DGX A100s as a supercomputing system. For more information on their hardware see reference [1].
Patient Care & Outcomes
The 30-Day Readmission Prediction Model at NUHS is transforming personalized medicine by predicting the likelihood of a patient being readmitted within 30 days of discharge. This tool enables healthcare providers to customize care plans, proactively preventing readmissions, improving patient outcomes, and reducing hospital costs.
In addition, Disease Progression Modeling allows clinicians to anticipate the progression of diseases, enabling them to intervene earlier, particularly in the management of chronic diseases, such as chronic kidney disease. Timely interventions can significantly alter the course of a disease, thereby improving the patient's quality of life.
The Pharmacogenomics Alerts System further personalizes patient care by providing medication recommendations based on the patient’s genetic profile. This minimizes the risk of adverse drug reactions and tailors treatments to the individual’s genetics, pushing the boundaries of precision medicine.
Emergency Department Efficiency
Managing patient inflows in a fast-paced environment like the emergency department is challenging. NUHS has addressed this with Pathfinder: The Emergency Department Predictive Dashboard, an AI-driven system that predicts patient wait times, manages inflow, and helps ED staff allocate resources more effectively. The result is reduced wait times, enhanced patient satisfaction, and improved care quality.
Hospital Resource Management
Resource management is optimized through the Estimated Length of Stay Model, which predicts how long a patient is likely to stay in the hospital. This allows for more effective planning and allocation of resources, optimizing bed usage, staff deployment, and overall resource management. NUHS ensures that each patient receives timely and appropriate care through this approach.
Population Scale Chronic Disease Management
For chronic disease patients, NUHS has introduced the CHAMP chatbot system, an innovative tool that keeps patients on track with their treatment plans through whatsapp chatbot nudges. Using data from the EMR system, as well as patient’s responses, the system automatically analyzes patient risk levels based on test results and other factors. CHAMP then sends risk appropriate reminders and follow-ups to keep patients engaged and proactive in managing their health. The system has proven successful, with significantly higher rate of enrolment, and lower dropout rates compared to similar programs at other hospitals.
Enhancing Patient Communication
NUHS has also enhanced patient interaction and accessibility through several Chatbot Systems, including RUSSELL-GPT, NUHS Chatbot, and the NCIS Oncology app. These chatbots offer instant responses to queries, schedule appointments, and provide personalized health information. RUSSELL-GPT is based on a proprietary architecture, utilizing a variety of fine-tuned and commercial GPT models, includingRAG routing mechanisms to tailor responses to the users—be it a researcher needing detailed data or a patient requiring simpler explanations—while keeping sensitive information secure.
Integrating AI into Hospital Systems
All these AI tools are seamlessly integrated with the Epic EMR system, providing a single AI dashboard that provides healthcare providers with comprehensive insights. This integration enhances decision-making and patient care, offering a holistic view of hospital operations.
The AI-driven advancements at NUHS are a testament to the institution's robust data infrastructure and organizational commitment. By streamlining processes, improving patient care, and optimizing resource utilization, NUHS is setting a new standard for healthcare institutions worldwide. Their holistic approach ensures that all these tools work together seamlessly, creating a more efficient and patient-focused healthcare system.
Lessons Learned and Global Applicability
NUHS’ journey in implementing AI-driven healthcare systems offers invaluable lessons for any institution aiming to embark on a similar path. The success of such a transformation goes far beyond the technology itself; it involves a deep, multifaceted commitment to foundational work, trust-building, and ongoing maintenance. This section outlines the key lessons learned from NUHS’s experience and provides a playbook for institutions looking to replicate their success, of which a summary can also be found below in Figure 1.
Figure 1: The four key lessons learned from NUHS’s experience for AI implementation
Data Infrastructure is the Foundation
The first and perhaps most crucial lesson is the importance of building a robust data infrastructure. Before any AI systems can be effectively deployed, it is essential to consolidate and secure data across the institution. At NUHS, this process took three years and required a meticulous approach to gathering data from disparate sources, standardizing it, and ensuring that it was both accessible and protected. The creation of a secure data platform, DISCOVERY AI, which adhered to strict privacy regulations and incorporated advanced cybersecurity measures, was foundational. Without this comprehensive data infrastructure, the AI applications at NUHS would have lacked the necessary depth and accuracy to be truly effective. Institutions must be prepared to invest significant time and resources into this foundational work, recognizing that it is the bedrock upon which all subsequent innovations will be built.
Implementation Requires Organizational Trust
Another critical lesson from NUHS’s experience is the necessity of building and maintaining trust across the organization. Implementing a new, AI-driven system is not just a technical challenge; it is a cultural one. Data holders and stakeholders across the institution need to be convinced of the system’s security and benefits.
Committees Provide Constant Human Oversight
NUHS’s system is supported by multiple committees of experts who moderate the content and maintain the medical protocols. These committees play a vital role in ensuring that the AI models remain relevant and effective, regularly reviewing and updating them based on the latest data and clinical insights. This level of oversight is essential for maintaining the system’s integrity and effectiveness over time. Institutions must be prepared to establish similar governance structures, recognizing that the success of an AI-driven healthcare system is not a one-time achievement but a continuous process.
AI is a Long Term Commitment
If it isn’t clear already, any institution seeking to implement a similar system must be ready for a long-term commitment. At NUHS, building the data infrastructures took years alone. On top of that, it required meticulous planning and significant time commitment to build trust throughout the organization, before it was possible to build each individual AI innovation. Further, these innovations are continually monitored and modified by committees. This is a thorough process that is never truly complete, requiring organizational commitment to long-term digital transformation of its systems and processes. Institutions must be prepared to commit significant time and resources that will need to be dedicated to the project.
Conclusion
The transformative potential of AI in healthcare is evident through the pioneering efforts of NUHS. The advancements made by NUHS serve as a model for what is possible, and their global applicability suggests a promising future for AI-driven healthcare innovation. However, the real innovation lies not just in the AI applications but in the comprehensive, integrated system that makes these applications possible. For more insights and details on NUHS’s initiatives, you can explore the work of Prof Kee Yuan Ngiam [2], who has been instrumental in driving these advancements.
References
https://issuu.com/brilliant-online/docs/brilliant_magazine_december_2021_issue/s/14335092
https://nuhsplus.edu.sg/article/meet-the-doctor-whose-healthcare-innovations-are--out-of-this-world
https://nuhsplus.edu.sg/article/shorter-hospital-waiting-times-with-artificial-intelligence
About the National University Health System (NUHS)
The National University Health System (NUHS) aims to transform how illness is prevented and managed by discovering causes of disease, development of more effective treatments through collaborative multidisciplinary research and clinical trials, and creation of better technologies and care delivery systems in partnership with others who share the same values and vision.
Institutions in the NUHS Group include the National University Hospital, Ng Teng Fong General Hospital, Jurong Community Hospital and Alexandra Hospital; three National Specialty Centres – National University Cancer Institute, Singapore (NCIS), National University Heart Centre, Singapore (NUHCS) and National University Centre for Oral Health, Singapore (NUCOHS); the National University Polyclinics (NUP); Jurong Medical Centre; and three NUS health sciences schools – NUS Yong Loo Lin School of Medicine (including the Alice Lee Centre for Nursing Studies), NUS Faculty of Dentistry and NUS Saw Swee Hock School of Public Health.
With member institutions under a common governance structure, NUHS creates synergies for the advancement of health by integrating patient care, health science education and biomedical research.
As a Regional Health System, NUHS works closely with health and social care partners across Singapore to develop and implement programmes that contribute to a healthy and engaged population in the Western part of Singapore.
For more information, please visit www.nuhs.edu.sg.
About Tensility Venture Partners
Tensility Venture Partners is an early stage venture capital fund focused on AI-enabled enterprise software companies. The firm focuses on cyber security, healthcare, vertical applications, enterprise infrastructure, space and robotics software where AI can significantly innovate and transform businesses. Since 2010, the GPs have invested in over 50 startups including realized unicorn exits in Duo Security (sold to CSCO), DocuSign (N:DOCU) and several other companies.
For more information, please visit www.tensilityvc.com.