Data-Driven Forecasting of Agitation for Persons with Dementia: A Deep Learning-Based Approach.

Area of Science:

• Artificial Intelligence in Healthcare
• Neuroscience and Neurology
• Gerontology and Geriatric Care

Background:

• Millions diagnosed with dementia annually, with most cared for by family.
• Caregiver burden significantly impacts informal caregivers' quality of life.
• Unpredictable agitation episodes in persons with dementia (PwD) are a major stressor for caregivers.

Purpose of the Study:

• To develop a predictive model for agitation episodes in PwD.
• To reduce caregiver burden by enabling proactive management of agitation.
• To enhance the quality of life for familial caregivers of PwD.

Main Methods:

• Utilized Long Short-Term Memory (LSTM) deep learning models for forecasting.
• Predicted agitation events up to 30 minutes in advance.
• Handled missing data via imputation and addressed class imbalance through down-sampling.

Main Results:

• Achieved high prediction accuracy of 98.6% for agitation episodes.
• Demonstrated a recall (sensitivity) of 84.8% in predicting agitation.
• Validated model efficacy using real-world data from Alzheimer's disease caregiver-patient dyads.

Conclusions:

• Data-driven deep learning models are effective in predicting agitation in community-dwelling PwD.
• Forecasting agitation can significantly alleviate caregiver burden.
• This approach holds promise for improving caregiving experiences and patient well-being.