Exploring Factors Related to Social Isolation Among Older Adults in the Predementia Stage Using Ecological Momentary Assessments and Actigraphy: Machine Learning Approach

Abstract

Background: As the global population ages, the economic burden of dementia continues to rise. Social isolation-which includes limited social interaction and feelings of loneliness-negatively affects cognitive function and is a significant risk factor for dementia. Individuals with subjective cognitive decline and mild cognitive impairment represent predementia stages in which functional decline may still be reversible. Therefore, identifying factors related to social isolation in these at-risk groups is crucial, as early detection and intervention can help mitigate the risk of further cognitive decline.

Objective: This study aims to develop and validate machine learning models to identify and explore factors related to social interaction frequency and loneliness levels among older adults in the predementia stage.

Methods: The study included 99 community-dwelling older adults aged 65 years and above in the predementia stage. Social interaction frequency and loneliness levels were assessed 4 times daily using mobile ecological momentary assessment over a 2-week period. Actigraphy data were categorized into 4 domains: sleep quantity, sleep quality, physical movement, and sedentary behavior. Demographic and health-related survey data collected at baseline were also included in the analysis. Machine learning models, including logistic regression, random forest, Gradient Boosting Machine, and Extreme Gradient Boosting, were used to explore factors associated with low social interaction frequency and high levels of loneliness.

Results: Of the 99 participants, 43 were classified into the low social interaction frequency group, and 37 were classified into the high loneliness level group. The random forest model was the most suitable for exploring factors associated with low social interaction frequency (accuracy 0.849; precision 0.837; specificity 0.857; and area under the receiver operating characteristic curve 0.935). The Gradient Boosting Machine model performed best for identifying factors related to high loneliness levels (accuracy 0.838; precision 0.871; specificity 0.784; and area under the receiver operating characteristic curve 0.887).

Conclusions: This study demonstrated the potential of machine learning-based exploratory models, using data collected from mobile ecological momentary assessment and wearable actigraphy, to detect vulnerable groups in terms of social interaction frequency and loneliness levels among older adults with subjective cognitive decline and mild cognitive impairment. Our findings highlight physical movement as a key factor associated with low social interaction frequency, and sleep quality as a key factor related to loneliness. These results suggest that social interaction frequency and loneliness may operate through distinct mechanisms. Ultimately, this approach may contribute to preventing cognitive and physical decline in older adults at high risk of dementia.

International registered report identifier (irrid): RR2-10.1177/20552076241269555.

Keywords: actigraphy; aged; ecological momentary assessment; loneliness; machine learning; predementia; social interaction; social isolation.

Figure 1

Deep embedding used for clustering data into 3 clusters.

Figure 2

The classification of social interaction frequency and levels of loneliness of the 3 clusters. (A) Average graph of mobile EMA responses for the social interaction frequency of the 3 clusters. (B) Average graph of mobile EMA responses for the loneliness levels of 3 clusters. EMA: ecological momentary assessment.

Figure 3

The sequential steps involved data processing and exploratory modeling. EMA: Ecological Momentary Assessment.

Figure 4

Receiver operator characteristic (ROC) curve for each model’s analysis of low social interaction frequency and high levels of loneliness. (A) ROC curve of the low social interaction frequency exploratory model. (B) ROC curve of the high levels of loneliness exploratory model. GBM: Gradient Boosting Machine; LR: logistic regression; RF: random forest; XGB: Extreme Gradient Boosting.