[Paper Review] An attention model to analyse the risk of agitation and urinary tract infections in people with dementia

Honglin Li, Roonak Rezvani+6 more|arXiv (Cornell University)|Jan 18, 2021

Machine Learning in Healthcare36 references4 citations

TL;DR

This study proposes a deep learning model combining attention mechanisms and rationalization to analyze in-home sensor data for early detection of agitation and urinary tract infections (UTIs) in people with dementia. The model processes time-series data to identify critical patterns, achieving 91% recall and 83% precision, and provides interpretable predictions by highlighting relevant time-steps and features for clinical decision-making.

ABSTRACT

Behavioural symptoms and urinary tract infections (UTI) are among the most common problems faced by people with dementia. One of the key challenges in the management of these conditions is early detection and timely intervention in order to reduce distress and avoid unplanned hospital admissions. Using in-home sensing technologies and machine learning models for sensor data integration and analysis provides opportunities to detect and predict clinically significant events and changes in health status. We have developed an integrated platform to collect in-home sensor data and performed an observational study to apply machine learning models for agitation and UTI risk analysis. We collected a large dataset from 88 participants with a mean age of 82 and a standard deviation of 6.5 (47 females and 41 males) to evaluate a new deep learning model that utilises attention and rational mechanism. The proposed solution can process a large volume of data over a period of time and extract significant patterns in a time-series data (i.e. attention) and use the extracted features and patterns to train risk analysis models (i.e. rational). The proposed model can explain the predictions by indicating which time-steps and features are used in a long series of time-series data. The model provides a recall of 91\% and precision of 83\% in detecting the risk of agitation and UTIs. This model can be used for early detection of conditions such as UTIs and managing of neuropsychiatric symptoms such as agitation in association with initial treatment and early intervention approaches. In our study we have developed a set of clinical pathways for early interventions using the alerts generated by the proposed model and a clinical monitoring team has been set up to use the platform and respond to the alerts according to the created intervention plans.

Motivation & Objective

• To address the challenge of early detection of agitation and UTIs in people with dementia, which are major causes of hospitalization and distress.
• To overcome limitations of traditional machine learning models in handling noisy, high-dimensional time-series sensor data from in-home monitoring.
• To develop a model that not only improves prediction accuracy but also provides interpretable explanations for clinical decisions.
• To integrate the model into a real-world clinical pathway with a monitoring team for timely, person-centered interventions.
• To reduce unplanned hospitalizations by enabling early, targeted clinical responses based on automated alerts from sensor data.

Proposed method

• The model employs a dual-architecture framework: an attention mechanism to filter and prioritize informative time-steps in long-term sensor data, and a rationalization block to explain predictions by identifying key features and temporal segments.
• The attention mechanism dynamically weights input features across time, focusing on clinically relevant patterns such as activity changes or irregular movement patterns.
• A rationalization block generates interpretable explanations by highlighting which sensor inputs and time intervals contributed most to the prediction, enhancing clinical trust and transparency.
• The model uses focal loss to address class imbalance in clinical datasets, improving performance on rare but critical events like UTIs and severe agitation.
• The system integrates data from multiple in-home sensors (e.g., motion, door sensors, environmental conditions) to create a comprehensive view of daily behavior and health status.
• Predictions are used to trigger alerts in a clinical monitoring pathway, where trained staff validate symptoms and initiate non-pharmacological or medical interventions.

Experimental results

Research questions

• RQ1 Can an attention-based deep learning model effectively detect early signs of agitation and UTIs in people with dementia using in-home sensor data?
• RQ2 How does the integration of attention and rationalization mechanisms improve prediction performance and interpretability compared to traditional machine learning models?
• RQ3 To what extent can the model’s explanations support clinical decision-making and timely intervention in real-world care settings?
• RQ4 How effective is the model in handling imbalanced clinical datasets, particularly for rare but critical events like UTIs?
• RQ5 What is the impact of the model’s alerts on reducing hospitalizations and improving care outcomes in a community-dwelling dementia population?

Key findings

• The proposed model achieved a recall of 91% and precision of 83% in detecting the risk of agitation and UTIs using in-home sensor data.
• The attention mechanism successfully filtered out redundant and noisy data, focusing on clinically relevant time-steps and features in long-term time-series data.
• The rationalization block provided interpretable explanations by identifying specific sensor inputs and temporal patterns that contributed to predictions, enhancing clinical trust.
• The use of focal loss significantly improved model performance on imbalanced datasets, particularly for rare events such as UTIs.
• The model was successfully integrated into a clinical pathway where alerts triggered timely responses, including home visits and GP referrals, reducing the risk of delayed treatment.
• The system demonstrated potential for reducing unplanned hospitalizations by enabling early, person-centered interventions before symptom escalation.