Characterizing the Wandering Phenomenon in Dementia: a Multimodal Analysis using Wearable-Based Mobility Metrics

Wandering behavior represents one of the most challenging management aspects in people with dementia. Despite its clinical significance, a consensus definition remains elusive. Given its severe consequences, objective methods to detect and quantify Wandering are needed to support clinical assessment. In this thesis, the Wandering dimension was explored in 61 patients with dementia using a lower-back wearable sensor. Mobility was quantified through digital mobility biomarkers across domains such as gait, turn, diurnal patterns, sleep analysis and physical activity. The Cohen-Mansfield Agitation Inventory (CMAI) Item 12 has been used as a criterion for the Wandering phenomenon classification; adopting the behavioral frequency of several times per week as a threshold (CMAI Item 12 = 4) to distinguish the two groups. A sensitivity analysis of this criterion has been proposed relying on a review of clinical records, addressing Wandering-related keywords documented on the specific recording days. Wanderers exhibited significantly higher volumes of walking and turning and engaged in more long walking bouts compared to Non-Wanderers. Turns (>150°), reflecting repetitive back-and-forth movements within the clinical ward, emerged as a distinctive motor characteristic. Extracted metrics were integrated with clinical and neuropsychiatric assessments in a multivariate analysis to better characterize wandering. Sensor-derived physical activity parameters potentially showed additional discriminatory value over conventional clinical scores, and a combined classifier achieved good accuracy (AUC = 0.8) in detecting Wandering. This exploratory work highlights that episodic assessments and nursing questionnaires provide an incomplete picture of mobility and Wandering behavior. Thus, there is a compelling need for objective, continuous solutions for real-life assessments of mobility and Wandering detection.