TabulaTime: Novel multimodal deep learning for Acute Coronary Syndrome prediction through environmental and clinical...

Area of Science:

• Cardiovascular Medicine
• Artificial Intelligence
• Environmental Health

Background:

• Acute Coronary Syndromes (ACS) are a major global health concern, leading to significant mortality.
• Traditional Cardiovascular Risk Scores (CVRS) primarily use clinical data, often overlooking environmental influences like air pollution and climate.
• Integrating complex, time-series clinical and environmental data for cardiovascular risk prediction presents significant computational challenges.

Purpose of the Study:

• To develop a multimodal deep learning framework, TabulaTime, for enhanced prediction of ACS risk.
• To integrate clinical risk factors with time-series environmental data for improved cardiovascular health assessment.
• To address the limitations of traditional risk scores by incorporating environmental determinants.

Main Methods:

• Proposed TabulaTime, a multimodal deep learning framework for integrating clinical and environmental time-series data.
• Utilized PatchRWKV, a novel architecture for efficient extraction of complex temporal patterns with linear complexity.
• Employed attention mechanisms within the framework to enhance model interpretability.

Main Results:

• TabulaTime demonstrated a 20.5% improvement in prediction accuracy compared to the CatBoost model.
• Environmental data integration contributed a 10.1% gain in predictive performance.
• The PatchRWKV architecture outperformed existing state-of-the-art models, including MLP, CNN, RNN, and Transformer-based approaches.

Conclusions:

• TabulaTime offers a significant advancement in ACS risk prediction by effectively integrating multimodal data.
• The framework enhances personalized prevention strategies and strengthens public health initiatives against cardiovascular diseases.
• Highlighting key clinical and environmental predictors provides actionable insights for targeted interventions.