Proposed Topic (Most preferred): :
Research and Innovations (new projects / technology / innovations / service models)
Authors (including presenting author) :
Dr Emmanuel WONG(1)(2), Ms Jiaxi LIN(3), Dr Celine SL CHUI(3)(4)(5)
Affiliation :
1. Department of Cardiology, Queen Mary Hospital, Hospital Authority, Hong Kong 2. School of Clinical Medicine, Li Ka Shing Faculty of Medicine, The University of Hong Kong 3. School of Nursing, Li Ka Shing Faculty of Medicine, The University of Hong Kong 4. School of Public Health, Li Ka Shing Faculty of Medicine, The University of Hong Kong 5. Laboratory of Data Discovery for Health (D24H), Hong Kong Science and Technology Park, Hong Kong
Introduction :
Cardiovascular disease (CVD) is a global disease burden and a leading cause of mortality, particularly in developing countries. Applying risk prediction models is a cost-effective mathematical method to predict the probability of healthcare outcomes and prognosis disease. However, common risk models are based on multiple ethnicities without concurrent medication effect along with time-varying variable.
Objectives :
With the application of machine-learning (ML) technology, we developed Personalized CARdiovascular DIsease risk Assessment for Chinese (P-CARDIAC) with wide arrays of parameters for estimating incident (primary model) and recurrent CVD risk (secondary model).
Methodology :
We included patients who had used healthcare services provided by the Hong Kong Hospital Authority since 2004. A comprehensive list of those risk variables, including clinical laboratory tests, disease/medication history and other variables were retrieved by a Cox proportional hazards model (CPH) and least absolute shrinkage and selection operator (LASSO) regression for the statistically significant risk variables shortlist. We used multivariate imputation with chained equations (MICE) to replace the missing values and the XGBoost Cox model while dealing with heterogeneous tabular data. We compared the model performance with PCE, PREDICT, China-PAR, Framingham (Asian), TRS-2°P and SMART2 with 1000 bootstrap replicates.
Result & Outcome :
193,556 patients were included in the primary model derivation while 48,799 patients were included in the secondary model derivation. Good performance of internal validation with C-statistic of 0.87 and 0.77 for the full and basic primary model, respectively. Moreover, study showed satisfying performance with a C-statistic of 0.69 and 0.66 in full and basic secondary model. Other existing risk models had lower C-statistic in the comparison of model performance with P-CARDIAC. We believed P-CARDIAC had the potential to provide guidance in the early intervention and prevention for CVD patients, and served as a useful risk stratification tool for healthcare resource allocation in the future.
