Multimodal data integration to predict atrial fibrillation.

Area of Science:

• Cardiovascular disease research
• Biomarker discovery
• Predictive analytics in healthcare

Background:

• Atrial fibrillation (AF) risk prediction often uses clinical variables, polygenic risk scores, ECG, and plasma proteins.
• Comprehensive integration of these diverse data sources in a single study is limited.

Purpose of the Study:

• To assess the combined predictive power of clinical variables, polygenic risk scores, ECG, and plasma proteins for atrial fibrillation.
• To determine the most effective and parsimonious approach for AF risk prediction.

Main Methods:

• Utilized data from 8374 (Visit 3) and 3730 (Visit 5) Atherosclerosis Risk in Communities Study participants.
• Constructed clinical, polygenic, protein, and ECG risk scores.
• Evaluated prediction performance using logistic regression and Area Under the Curve (AUC).

Main Results:

• The addition of polygenic risk scores to clinical variables improved AUC for incident AF from 0.660 to 0.752 and for prevalent AF from 0.737 to 0.854.
• Further incorporating ECG and protein risk scores yielded modest AUC increases to 0.763 (incident) and 0.875 (prevalent).

Conclusions:

• A combination of clinical and polygenic risk scores provides the most effective and parsimonious method for AF prediction.
• ECG and protein risk scores offer limited additional predictive value beyond clinical and polygenic scores.