Symposium 24 - Statistics and Data Science in Healthcare: Unlocking the Power of Data

In this presentation, I will showcase four digital health solutions recently developed at the Laboratory of Data Discovery for Health (D24H), an InnoHK research center led by HKUMed.
The first solution is a user-centric pharmacy chatbot, developed in collaboration with the HA Chief Pharmacist Office. This chatbot enables users to freely access trustworthy information about medications listed in the HA formulary. It accepts multimodal inquiries and delivers concise answers on usage, side effects, dosage, etc.
The second solution is an LLM-powered chatbot focused on HPV vaccines co-developed with Fudan University. A randomized controlled trial (RCT) involving over 2,500 parents in mainland China showed that this chatbot increased HPV vaccine uptake among their school-aged daughters, particularly in rural locations with limited health literacy and less robust healthcare infrastructure (Nature Medicine 2025).
The third solution is an AI-powered smartphone application designed to detect heart murmurs without the need for additional hardware. In a clinical validation study involving more than 500 patients at Queen Mary Hospital, this app demonstrated an accuracy level comparable to that of commercial electronic stethoscopes (Gold Medal with Congratulations of the Jury, 49th International Exhibition of Inventions of Geneva, 2024).
The fourth solution is an offline LLM application designed to support autonomous cancer staging and risk stratification for thyroid cancer cases using their clinical notes. Trained on 50 pathology reports from The Cancer Genome Atlas - Thyroid Cancer (TCGA-THCA) database, this application achieved an accuracy of over 90% when tested on 289 TCGA-THCA notes and 35 pseudo-clinical cases composed by clinicians at Queen Mary Hospital (npj Digital Medicine 2025).

The Hong Kong Government has been supportive in nurturing the development of a retirement planning market that provides monetization of wealth accumulated during working lifetime for improving retirement living. Hong Kong is one of the few countries that has all three such products available, namely, immediate life annuity, reverse mortgage and policy reverse mortgage, collectively referred to the "decumulation products". The products are being offered by the Hong Kong Mortgage Corporation Limited (HKMC), a public sector entity, with a mission to help develop the retirement planning market. In order to ensure the products are affordable and sustainable, HKMC strives to better manage the risks and promote wider private participation in making more products available. In 2021, the Hospital Authority provided medical data to the HKMC for the development of a life expectancy estimation model that helps to supplement models that are developed based on overseas data. This development will hopefully improve individual's life expectancy estimate and enable development of new products tailored to the specific needs of individuals. Powered by AI technology in many medical related fields, human life expectancy estimates made based on historical data may no longer be adequate for insurance product pricing purposes. More research on new methodology that captures the impact of technological advancement on human life expectancy will be essential for ensuring long-term viability of these retirement products.

This presentation by the Statistics & Data Science Department explores the transformative potential of big data analytics in healthcare, leveraging the vast treasure of information accumulated in the Hospital Authority's IT systems. By applying advanced analytics techniques and tools, the department demonstrates how diverse datasets can be harnessed to generate actionable insights, facilitating strategic decision-making and addressing the growing demand for healthcare services.


The objectives of this work are threefold: (1) to optimize healthcare decision-making through data-driven methodologies; (2) to develop predictive models including the Diabetes Mellitus (DM) Risk Engine, the Uncommon Disorders Database (UDD), and the Chemotherapy Toxicity Risk Model - to enable personalized care and early interventions; and (3) to promote sustainable and efficient healthcare systems using innovative technologies.


The projects employ cutting-edge techniques spanning machine and deep learning methods, such as the DeepHit model for survival analysis, alongside traditional statistical approaches like the Cox proportional-hazards model. Natural language processing (NLP) and LightGBM models were also utilised for text analysis and patient identification within clinical notes, supporting the development of comprehensive databases and risk prediction tools.


Key advancements include: 
 The DM Risk Engine, which enables risk stratification for diabetic patients, fostering timely interventions and fostering personlised care.
 The UDD, a centralized resource for better management of rare diseases, enhancing diagnosis and care delivery for these conditions.
 The Chemotherapy Toxicity Risk Model, which supports clinicians in identifying high-risk lung cancer patients with potential adverse events, ensuring safer treatments and optimizing decisions to reduce complications.


These innovative applications highlight the pivotal role of big data analytics in driving personalized healthcare, improving patient outcomes, and addressing the sustainability challenges faced by modern healthcare systems. The work underscores the importance of integrating advanced analytics into healthcare strategies to deliver impactful, data-driven solutions.