Implementation Science
New Model Predicts Cardiovascular Disease Risk in a Chinese Population
Dr Celine Chui Sze-ling |Assistant Professor|School of Nursing and School of Public Health
Cardiovascular diseases (CVD), including coronary heart disease and stroke, are the most common fatal non-communicable diseases globally and were responsible for an estimated 18.6 million deaths in 2019. In Hong Kong, CVD is also a burden to the health care system. The Hospital Authority reported that in 2020, 71,300 inpatient discharges and inpatient deaths in all hospitals and 6,561 registered deaths were caused by CVD. A 2021 report by the Centre for Health Protection showed that the rate of heart disease increases steadily from age 35 to affect 9.1% of those aged 65 and over. To tackle the problem, the Hong Kong Government launched the “Towards 2025: Strategy and Action Plan to Prevent and Control NCD (non-communicable diseases) in Hong Kong”, with targets to prevent premature mortality from NCD (particularly CVD) and to prevent heart attacks and strokes.
Clinical guidelines suggest that using a risk prediction model on patients with established CVD could mitigate recurrent CVD. While existing popular risk predictions, such as TIMI and SMART2, were developed on multi-ethnic cohorts and standard statistical modelling based on a points system has been put into clinical use, researchers have found that their estimated results and performance among Asians are relatively poor compared with Caucasians. Standard statistical modelling is also unable to cope with a wide array of variables that could be time-varying as concurrent medications affect risk for recurrent CVD. This has given rise to a need to develop a risk prediction algorithm based on Machine Learning (ML) with dynamic function to address this issue and reduce the healthcare burden among Chinese.
The School has responded to that need in collaboration with the HKU Department of Computer Science, to develop a risk prediction algorithm, P-CARDIAC (Personalized CARdiovascular DIsease risk Assessment for Chinese), to predict the 10-year CVD risk for a Hong Kong Chinese population. The novel algorithm, based on 48,799 participants, was designed by Cox proportional hazards model with XGBoost to achieve better performance and interpretability for medical use. More than 120 risk factors were included that relate to CVD, including demographic factors, disease history, clinical laboratory tests, treatment exposure and other information from local electronic health records. The outcome was defined as the diagnosis of recurrent CVD, i.e., a composite of coronary heart disease, stroke, peripheral artery disease, and revascularisation. To evaluate the performance of the model, validations were conducted on two other independent local cohorts with more than 250,000 individuals who were assessed using TIMI risk score and SMART2.
The novel algorithm had good discrimination performance. Validations showed P-CARDIAC had a better performance than TIMI and SMART2. Our preliminary results demonstrated that P-CARDIAC is an appropriate tool to predict a Chinese population’s CVD risk. P-CARDIAC also allows a more personalised approach for CVD prevention with consideration of the dynamic baseline risk and effects of concurrent medication.
The Government Policy Address 2022 pledged to strengthen the primary healthcare system in Hong Kong and this prediction tool has the potential to play an important role in routine clinical practice. It can raise awareness of the CVD risk for earlier intervention and reduce the healthcare burden brought by CVD. The algorithm is expected to be adaptable to various clinical settings to produce a more accurate estimated result for preventing recurrent CVD.