Naturalistic Driving Engineered Trajectories Forecast Future Cognitive Decline

Early population-based screening for Alzheimer's disease remains limited by the scalability and accessibility of conventional biomarker testing. We tested whether passive naturalistic driving changes predict progression from cognitively normal status to sustained mild cognitive impairment.

Season-adjusted monthly driving data from 298 older adults, including 25 progressors, were summarized into within-person trajectory features capturing drift and instability in cornering and destination diversity relative to a six-month baseline. A GEE logistic regression was evaluated under leave-one-out cross-validation with a paired-consecutive flagging rule. Landmark survival analyses at 12 and 24 months assessed prognostic value. Five held-out clinical subgroups (n = 82 evaluable), plasma pTau217, MoCA trajectories, and depression scores provided independent post hoc validation.

The model detected 19/25 progressors (AUC=0.752, 95%CI: 0.627-0.863; 76% sensitivity; 75% specificity) with an 18-month median lead time before sustained CDR conversion. Adding static baseline driving features reduced discrimination (AUC=0.702), supporting that within-person driving change, not absolute level, carries a predictive signal. Driving markers stratified risk at 12 months (HR = 6.02, 95%CI: 2.35-15.41) and 24 months (HR = 13.38, 95%CI: 3.03-59.04). Signals converged or were directionally consistent with higher pTau217 among detected progressors and clinically ambiguous single-endpoint cases, whereas flagged stable participants showed MoCA decline without pTau217 enrichment. Depressive symptoms did not explain the driving signal (Spearman rho=0.05, p=0.53). The sustained conversion rate (12.3/1,000 person-years) matched epidemiological expectations.

Scalable driving-trajectory features can detect preclinical functional decline with clinically meaningful lead time, without demographic, genetic, or cognitive data, and may serve as an objective neurobehavioral screening tool.