The integration of artificial intelligence (AI) and big data to analyse extensive medical records and lifestyle factors is shaping the future of smart healthcare. Leveraging healthcare data from Hong Kong and various regions of mainland China, the School of Nursing and Health Sciences at Hong Kong Metropolitan University (HKMU) has collaborated with researchers from multiple institutions to develop AI-enhanced predictive models. These models aim to assess the risks of cardiovascular and cerebrovascular diseases (CCVDs) in the general population, as well as in patients with specific diseases, such as diabetes mellitus and cancers. This will enable healthcare professionals to identify high-risk patients and provide timely, personalised medical interventions.

Prof. Gary Tse, Associate Dean (Innovations and Research) of the School of Nursing and Health Sciences, has partnered with scholars from universities and hospitals, including The University of Hong Kong, The Chinese University of Hong Kong, The Second Hospital of Tianjin Medical University, The First Affiliated Hospital of Dalian Medical University, Coventry University and The University of Glasgow, as well as local industry partners, such as Apex Medical Group and the Power Health Research Institute, to develop, test, validate and implement these predictive models.

He explained that while existing tools for assessing CCVD risk are based largely on Western population data, the team has long recognised the need for models tailored specifically for the Chinese population.

To address this gap, the team started developing Chinese-specific predictive models in 2019. They utilised anonymous data from over 150,000 patients attending public family medicine clinics in Hong Kong. By comparing 10 different AI algorithms, they developed “PowerAI-CVD” in 2023.

This Chinese-specific AI predictive model incorporates data such as blood pressure measurements, disease status, medications and laboratory results to predict 10-year CCVD risks in patients free of cardiovascular disease and those who had suffered strokes or other cardiovascular conditions. The AI model achieved an accuracy rate of 90% in preliminary analyses, as reported in a pre-print study.

The model also demonstrates strong predictive performance, achieving a c-statistic of 0.87. This reflects its effectiveness in distinguishing between high- and low-risk patients. The related findings were published as a conference proceeding in the European Journal of Preventive Cardiology. Employing a novel AI algorithm, called CatBoost, which is based on gradient boosting over decision trees, the model enables accurate risk stratification of patients for both primary and secondary prevention.

In addition, the team developed risk models that are tailored for specific diseases. For example, in collaboration with researchers using hospital data from Dalian, the team performed phenotyping of heart failure, developing risk tools specific for subtypes of heart failure, including those with reduced, preserved and recovered ejection fraction. They employed various machine learning and AI techniques to consolidate and analyse patients' echocardiographic data on left atrial dimension and morphology, disease status and medications.

Furthermore, in collaboration with the urology team at The Chinese University of Hong Kong, the HKMU team co-led several studies to develop prostate cancer-specific CCVD risk models, incorporating different specific prostate cancer medications that may influence CCVD risk.

Prof. Tse added that various conditions, such as high blood pressure, diabetes, dyslipidaemia and cancers, are all associated with increased risks of CCVDs. These predictive risk models can help identify high-risk individuals, allowing for early treatment, significantly reducing the risk of heart attacks, strokes, heart failure and death due to cardiovascular or cerebrovascular causes. By inputting individual-specific data, including blood pressure measurements, disease status, medications and laboratory tests, healthcare professionals can predict a patient's 10-year risk for CCVDs.

Prof. Tse shared that the predictive models that have been validated locally are already in use at private clinics and health check facilities in Hong Kong. The risk estimates generated by the models help clinicians provide risk assessments for patients, enabling medical staff to provide personalised care based on varying risk levels.

He believes that AI-driven predictive models represent a key focus for the future of smart healthcare, deepening our understanding of disease progression and enabling early interventions. “We will continue to refine our models by incorporating socioeconomic data and functional assessment data from multidisciplinary team members, including nurses, physiotherapists and occupational therapists,” he said.

The team also aims to make an impact beyond the local city. “Using hospital data from other Chinese regions, such as Tianjin and Dalian, as well as clinical data from Singapore, the United Kingdom and the United States, we are developing population-specific and ethnicity-tailored models that can accurately predict the risks of CCVDs, as well as conditions such as dementia,” he added.

The team's comprehensive model was first published on 9 October 2023 as a preprint on MedRxiv, titled “PowerAI-CVD – the first Chinese-specific, validated artificial intelligence-powered in-silico predictive model for cardiovascular disease”. The related findings were subsequently published in various academic journals. The latest version was presented at the ESC Preventive Cardiology Congress in April 2024, hosted by the European Association of Preventive Cardiology of the European Society of Cardiology.

The team further refined the model, incorporating new information, such as long COVID and COVID vaccination status.

Major related research papers and conference proceedings:
1. Reverse Atrial Remodeling in Heart Failure With Recovered Ejection Fraction, Journal of the American Heart Association, 17 January 2023.
2. Healthcare Big Data in Hong Kong: Development and Implementation of Artificial Intelligence-Enhanced Predictive Models for Risk Stratification, Current Problems in Cardiology, January 2024.
3. PowerAI-CVD: Chinese-specific artificial intelligence-powered predictive model for cardiovascular disease, European Journal of Preventive Cardiology, 13 June 2024.
4. PowerAI-Diabetes: Review of glycemic and lipid variability to predict cardiovascular events in Chinese diabetic population, npj metabolic health and disease, 1 July 2024.
5. Associations between glucocorticoid use and major adverse cardiovascular events in patients with prostate cancer receiving antiandrogen: a retrospective cohort study, Prostate Cancer and Prostatic Diseases, 10 September 2024.
6. ChineseCVD: a web-based Chinese-specific cardiovascular risk calculator incorporating long COVID, COVID-19 vaccination, SGLT2i and PCSK9i treatment effects, European Heart Journal, 28 October 2024.
7. Harnessing Artificial Intelligence For Predicting Cardiovascular And Disease Complications: An Overview Of Innovative Support Tools, Journal of the Hong Kong College of Cardiology, 28 October 2024.