Integrative Identification of Frailty Status in Elderly Patients with Atrial Fibrillation Using a Noise-Resilient...

Area of Science:

• Gerontology
• Cardiology
• Artificial Intelligence

Background:

• Frailty is prevalent and underdiagnosed in elderly patients with atrial fibrillation (AF), negatively impacting outcomes.
• Traditional frailty assessments are subjective and time-consuming.
• Machine learning (ML) combined with multimodal data offers automated frailty detection potential.

Purpose of the Study:

• To develop a noise-resilient ML framework for identifying frailty in elderly AF patients using ECG and clinical data.
• To compare traditional and deep learning models for frailty detection under varying signal conditions.

Main Methods:

• Retrospective study of 110 elderly AF patients (≥65 years).
• Frailty assessed using Fried Phenotype and Clinical Frailty Scale.
• Multimodal data (180+ ECG features, clinical parameters) used to train five ML models (including SiamAF, CNN, LSTM-attention).
• Model performance evaluated using accuracy, F1-score, ROC-AUC, Brier score; robustness tested with synthetic noise.

Main Results:

• Frailty and pre-frailty prevalence were 41.82% and 34.55%.
• The SiamAF model achieved the highest performance (accuracy 90%, F1-score 89.75%, ROC-AUC 93%), demonstrating noise robustness.
• Deep learning models showed strong performance (ROC-AUC > 90%).
• Key predictors included NN interval variability, corrected QT interval, NT-proBNP, and grip strength.

Conclusions:

• Multimodal ML effectively identifies frailty status in elderly AF patients.
• The SiamAF model shows promising discriminatory performance and noise resilience.
• Findings are hypothesis-generating, requiring external validation in larger, multicenter AF populations.