Automatic classifier based on heart rate variability to identify fallers among hypertensive subjects

Area of Science:

• Gerontology
• Biomedical Engineering
• Cardiology

Background:

• Accidental falls are a significant health concern for the elderly.
• Existing fall prediction technologies suffer from low specificity and high false positive rates.
• Autonomic nervous system (ANS) states influence balance control and may be linked to falls.

Purpose of the Study:

• To develop an automatic classifier for identifying fallers based on heart rate variability (HRV) analysis.
• To investigate the potential of HRV as an estimator of ANS states related to fall risk.
• To compare different data mining approaches for fall prediction using HRV.

Main Methods:

• Analysis of 24-hour electrocardiogram recordings from 168 cardiac patients (47 fallers, mean age 72 years).
• Linear and nonlinear HRV properties were analyzed in 30-minute segments.
• Feature selection using principal component analysis combined with the RUSBoost algorithm was employed.

Main Results:

• The hybrid RUSBoost algorithm achieved 80% specificity and 72% accuracy in identifying fallers.
• Sensitivity was limited to 51%, indicating a significant number of false negatives.
• The study demonstrated that certain ANS states associated with falls could be detected.

Conclusions:

• HRV analysis combined with advanced data mining offers a potential method for fall prediction in older adults.
• The findings suggest that while ANS dysfunction contributes to some falls, other factors are also involved.
• Further research is needed to improve sensitivity and comprehensively understand fall etiology.