How Naturalistic Driving Compares to Aβ‐related CSF and Plasma Biomarkers in the Detection of Amyloid PET Positivity: A Machine Learning Approach

Abstract Background Identifying individuals with Alzheimer disease (AD) pathology a decade or more before the onset of dementia symptoms can be achieved with amyloid positron emission tomography (PET), cerebrospinal fluid (CSF) biomarkers, or blood biomarkers. This study compares driving behavior with CSF and blood biomarkers to identify amyloid deposition in AD, as defined by the “gold standard” PET‐amyloid. Method We used commercial‐in‐vehicle‐GPS dataloggers to study naturalistic driving behaviors among 101 cognitively normal older drivers (aged 65+). All participants had blood, CSF, and PET biomarker data. The cohort included 37 individuals with and 64 without PET‐amyloid positivity. Driving variables included total number of trips, number of trips with a distance less than 1‐mile, trips in evening rush hour, trips with hard acceleration, speeding events, and average speed. Three Multi‐layer Perceptron (MLP) classifiers were trained with three sets of inputs: (1) driving variables, (2) driving variables and age, and (3) driving variables, age, and APOE ε4 status. Additionally, plasma Aβ42/Aβ40 < 0.1013, CSF Aβ42/Aβ40 < 0.0673, or Amyloid Probability Score (APS) (modeled score incorporating plasma Aβ42/40, age, and APOE ε4 status) > 15 were used to detect PET‐amyloid positivity. The driving models were trained on 80% of the data and all biomarkers were evaluated on the remaining 20% of the data. Specificity (Spe), sensitivity (Sen), F1‐score (F1), and area under the receiver operating curve (AUC) were compared across the three driving models and three fluid‐based biomarkers. Result For predicting PET‐amyloid positivity, CSF Aβ42/Aβ40 achieved the highest performance (Spe=0.96, Sen=0.98, F1=0.97) followed by the MLP model with driving variables, age, and APOE ε4 status (Spe=0.93, Sen=0.91, and F1=0.92), MLP model with driving variables and age (Spe=0.88, Sen=0.85, and F1=0.86), APS (Spe=0.82, Sen=0.84, and F1=0.81), plasma Aβ42/Aβ40 (Spe=0.79, Sen=0.80, and F1=0.76), and MLP model with driving variables (Spe=0.71, Sen=0.72, and F1=0.71). The AUC from driving variables was 0.800, and improved by adding age to 0.933, and age and APOE ε4 status to 0.977. Furthermore, AUC scores were 0.986, 0.911, and 0.873 for CSF Aβ42/Aβ40, APS, and plasma Aβ42/Aβ40, respectively. Conclusion Driving is a useful neurobehavioral marker in predicting PET‐amyloid positivity among cognitively normal older drivers.

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.