Driving the neural exposome: Latent mobility states from naturalistic GPS data in older adults

INTRODUCTION: Naturalistic driving provides real-world behavioral indicators of early cognitive and functional changes. This study integrated naturalistic driving GPS trajectories collected from in-vehicle sensors with points of interest (POIs) to quantify daily environmental engagement among older adults who were cognitively normal at enrollment. METHODS: Data from 438 participants enrolled in the Driving Real-world in-Vehicle Evaluation System Project were used to generate daily POI share vectors and model latent engagement patterns using a logistic-normal hidden Markov model (HMM). Thirteen latent states described distinct modes of environmental interaction. From each participant's inferred state sequence, we derived mobility features - state occupancy, dwell time, transition entropy, and self-transition probability - and examined their differences across clinical status groups and associations with Preclinical Alzheimer's Cognitive Composite (PACC) performance. RESULTS: Transition entropy and several state-specific occupancy and dwelling metrics differed across clinical groups, but none of the mobility features were significantly associated with PACC scores. DISCUSSION: Mobility-derived behavioral features differentiate clinical status groups and may reflect early functional changes preceding cognitive decline.