Machine learning tool feasible for detecting agitation in dementia

Key takeaways:

• The model, along with a traffic-light system, boosted sensitivity and specificity of agitation recognition.
• During agitation intervals, awake time and illumination were higher while respiratory rate was lower.

An interpretive machine learning model may be a feasible method for measuring agitation in dementia as well as in predicting certain environmental factors that are likely to contribute to the condition, according to a study.

“Current clinical practice for managing agitation in [people living with dementia] primarily relies on pharmaceutical interventions,” Marirena Bafaloukou, BSc, a doctoral candidate and research assistant in machine learning for clinical data analysis at Imperial College London, and colleagues wrote in eClinical Medicine. “Identification and appropriate treatment of agitation are essential for mitigating associated risks, lowering health care expenses and alleviating carer strain.

Agitation in dementia is usually identified by resource-challenging subjective clinical scales and direct patient observation, but clinical applications for data-driven methodology are limited in scope, Bafaloukou and colleagues wrote.

As such, the researchers sought to examine how real-world factors interact with and influence dementia-related agitation within home settings. They also attempted to apply an interactive, machine-learning monitoring tool to identify the conditions under which agitation is present in affected persons.

The Minder study, to date, has recruited 127 individuals who were aged 50 years or older, had a clinical diagnosis of dementia or mild cognitive impairment, along with either current or prior treatment in a psychiatric unit.

Psychometric and cognitive assessment tools were administered during 3-month visits within the study, while additional weekly behavioral monitoring questionnaires were distributed to participants or partners to track the presence or absence of participants’ behavioral symptoms based on established definitions of agitation.

For their proof-of-concept venture, the researchers utilized longitudinal data amounting to 32,896 person-days from 63 persons living with dementia collected using in-home monitoring devices over 512 weeks between December 2020 and March 2023. They incorporated a traffic-light stratification system to judge agitation probability and utilized the SHapley Additive exPlanations (SHAP) framework to enhance interpretability. From these data, Bafaloukou and colleagues designed an interactive tool that allows for individualized, non-pharmacological interventions based on living conditions.

The performance of multiple machine learning models, including Light Gradient-Boosting Machine (LightGBM), Adaptive Boosting (ADABoost) and Support Vector Machine (SVM) were compared in their ability to identify agitation in the study population over an 8-day period. Among the principal features for identifying agitation were sleep/wake cycles, respiration, temperature regulation and indoor lighting.

According to the results, LightGBM recorded the best agitation recognition performance, with a sensitivity of 71.32%±7.38 and specificity of 75.28%±7.38, while incorporating the traffic-light system increased specificity to 90.3%±7.55.

The researchers further found, via data analysis of their interactive tool, that sleep, light and temperature adjustments were the most feasible in-home interventions for the study group.

During weeks when agitation was present, data show, participants’ average awake ratio was higher than in non-agitation weeks, while a similar correlation was found between agitation presence and level of interior illumination; however, this correlation was inverse with respect to respiratory rate, which was lower during non-agitation intervals.

“By employing an agitation monitoring model in real-world settings, we could enhance the detection of missed agitation instances,” Bafaloukou and colleagues wrote. “Ultimately improving patient care and contributing to the development of more precise definitions for agitation episodes.