AbstractBackgroundPeople with dementia (PwD) often become lost, which is commonly attributed to spatial disorientation, one of the earliest symptoms of dementia, particularly of the Alzheimer’s disease type. Spatial disorientation can limit a person's ability to navigate in an outdoor environment. As a result, PwD experience outdoor mobility decline, which, in turn, can have a negative impact on their cognitive functions. Thus, enabling safe outdoor mobility is important for dementia research. Artificial Intelligence (AI) methods in conjunction with Global Positioning System (GPS) data show great potential for supporting the outdoor mobility needs of PwD. The objective of this work is to evaluate the extent to which we can predict future destinations of PwD by learning from their past mobility habits.MethodEight cognitively‐intact older adults (CTL) and seven older adults with dementia completed four weeks of GPS recording. Each participant’s stops and trips were extracted from their trajectories. We determined the predictability of each participant’s mobility patterns using three approaches. First, we assumed each stop is visited with equal probability, thus, disregarding temporal aspects of travel. Next, we built on the previous approach by including stop visitation probability, thus, capturing the heterogeneity of visitations. Finally, to capture the full spatiotemporal characteristics of mobility, we added in‐depth temporal characteristics including the visitation frequency, the order in which the stops were visited, and the time spent at each stop.ResultRelying solely on the spatial dimension of mobility yielded no predictive power across the two groups. Adding the heterogeneity of visitation patterns, we observed an increase in the predictability power; PwD displayed a trend toward higher predictability compared to the CTLs, but the student’s t‐test did not reach statistical significance (0.833±0.085vs.0.768±0.029, t(13)=‐2.06, p=0.06; d=‐1.07). Finally, relying on full spatiotemporal characteristics, a 4‐week record of mobility patterns displayed 95% (SD=2%) and 92% (SD=1%) predictability among PwD and CTLs, respectively. This value was significantly higher among PwD, t(13) = ‐3.39, p<.01; d=‐1.75.ConclusionOur findings offer new perspectives on the predictive mobility models based on GPS data and AI that can be used to provide personalized assistance for outdoor navigation of people with dementia.
Read full abstract