Impact of Sleep Apnea on Perivascular Diffusivity and Cognition: A 4-Year Prospective Cohort Study

Rationale: Obstructive sleep apnea (OSA) may impair perivascular diffusivity, a proposed glymphatic biomarker, contributing to cognitive decline. Objectives: To investigate the longitudinal impact of OSA on cognitive performance and perivascular diffusivity measured by diffusion tensor imaging along the perivascular space (DTI-ALPS) and whether changes in diffusivity mediate the relationship between OSA and cognition. Methods: In a community-based cohort, 1,110 participants underwent polysomnography, brain DTI, and cognitive assessments at baseline and 4-year follow-up. Participants were categorized by apnea-hypopnea index (AHI) at baseline and further grouped longitudinally as OSA-free, improved OSA, stable OSA, or worsened OSA. DTI-ALPS index was calculated at both time points to assess perivascular diffusivity. Multivariable models were adjusted for age, sex, education, body mass index, smoking, alcohol use, hypertension, diabetes, total sleep time, and sleep efficiency. Longitudinal models (i.e., Δ-values) also included baseline outcome values. Measurements and Main Results: Higher AHI was consistently associated with lower DTI-ALPS index. Each 1-event-per-hour increase in AHI corresponded to a 0.008 decrease in DTI-ALPS index at baseline and a 0.007 decrease over 4 years (both P < 0.001), with the worsened OSA group showing the greatest reduction. Increases in DTI-ALPS index were associated with improvements in visual memory, including immediate recall (β = 0.401), delayed recall (β = 0.351), and recognition (β = 0.145; all P < 0.001). ΔDTI-ALPS mediated the relationship between ΔAHI and changes in immediate visual recall (β = -0.015), delayed recall (β = -0.021), and recognition (β = -0.016; all P < 0.001). Conclusions: OSA severity was associated with altered perivascular diffusivity and related changes in memory, suggesting a potential pathway linking OSA to cognitive impairment via perivascular fluid dynamics.