Abstract
Sensor-based remote health monitoring of persons living with dementia can be used to gain insights into their health and monitor the progression of their condition, with minimal intrusion. This helps minimize preventable hospital admissions. Existing approaches for detecting activity anomalies are challenged by several factors, such as lack of annotated datasets and interpretability. We propose and evaluate a solution based on the Matrix Profile, an ultra-fast distance-based anomaly detection algorithm, to detect unusual activity and onset of adverse health. Daily household movement data collected via passive infrared sensors are used to generate Contextual Matrix Profiles (CMPs), comparative maps of activity patterns aggregated by time. We create CMP-based multivariate anomaly detection models to generate a single daily normalized anomaly score for each patient and use these to discover digital biomarkers of anomalies. CMP-based models yield high sensitivity, low alert rate and excellent generalisation, when evaluated on data from 15 participant households collected by the UK Dementia Research Institute between August 2019 and July 2021. We have discovered early AM bathroom activity to be the prime cross-patient digital biomarker of urinary tract infections (UTI), which validates findings in literature that unusual bathroom activity is a clinically significant feature in UTI in dementia. To the best of our knowledge, our work is the first real-world study to adapt the CMP to continuous anomaly detection in healthcare. The CMP inherits the speed, accuracy and simplicity of the Matrix Profile. It provides configurability, ability to address noise in data, detect patterns, and easy explainability to clinical practitioners, along with better overall performance than state-of-the-art methods. It offers a clinically meaningful unsupervised anomaly detection technique not only for dementia, but also for wider healthcare, and other problem domains.