Abstract
OBJECTIVES: This study aimed to explore the potential of synchronously acquiring wrist pressure pulse wave (PPW) and limb lead electrocardiogram (ECG) signals for the development of an identification model for coronary heart disease (CHD) and its associated comorbidities. METHODS: A custom-designed device equipped with pressure and ECG sensors, was utilized to synchronously collect wrist PPW and limb-lead ECG signals from 748 participants (463 for modeling and 285 for external validation). Features were extracted from these two types of physiological signals to form distinct datasets, and RF models were built based on different datasets. The top-performing RF model was then selected and compared against the Feature-Selected (FS-RF), Support Vector Machine (SVM) and Bagged Decision Tree (BDT) models. Ultimately, the optimal model for predicting coronary heart disease (CHD) and its comorbidity was determined based on evaluation metrics. RESULTS: The RF model that integrated both PPW and ECG features demonstrated significantly higher effectiveness compared to the RF model that relied on a single physiological signal. Furthermore, when benchmarked against the feature-selected RF(FS-RF), SVM and DBT models, the FS-RF model demonstrated the best performance, achieving an accuracy of 76.32%, an average precision of 75.82%, an average recall of 76.11%, and an average F1-score of 75.88%, all of which were higher than those of other models. Notably, the selected feature by FS-RF encompassed both PPW and ECG features. CONCLUSION: This study highlights the importance of synchronously acquiring of PPW and ECG signal, along with feature selection, in enhancing the performance of the FS-RF model for identifying CHD and its associated conditions. These findings provide a scientific basis for the application of wearable devices in clinical settings, highlighting their potential to aid in the early detection and management of cardiovascular disease.