Association between sleep microarchitecture and cognition in obstructive sleep apnea

Abstract

Study objectives: Obstructive sleep apnea (OSA) increases the risk of cognitive impairment. Measures of sleep microarchitecture from EEG may help identify patients at risk of this complication.

Methods: Participants with suspected OSA (n = 1142) underwent in-laboratory polysomnography and completed sleep and medical history questionnaires, and tests of global cognition (Montreal Cognitive Assessment, MoCA), memory (Rey Auditory Verbal Learning Test, RAVLT) and information processing speed (Digit-Symbol Coding, DSC). Associations between cognitive scores and stage 2 non-rapid eye movement (NREM) sleep spindle density, power, frequency and %-fast (12-16Hz), odds-ratio product (ORP), normalized EEG power (EEGNP), and the delta:alpha ratio were assessed using multivariable linear regression (MLR) adjusted for age, sex, education, and total sleep time. Mediation analyses were performed to determine if sleep microarchitecture indices mediate the negative effect of OSA on cognition.

Results: All spindle characteristics were lower in participants with moderate and severe OSA (p ≤ .001, vs. no/mild OSA) and positively associated with MoCA, RAVLT, and DSC scores (false discovery rate corrected p-value, q ≤ 0.026), except spindle power which was not associated with RAVLT (q = 0.185). ORP during NREM sleep (ORPNREM) was highest in severe OSA participants (p ≤ .001) but neither ORPNREM (q ≥ 0.230) nor the delta:alpha ratio were associated with cognitive scores in MLR analyses (q ≥ 0.166). In mediation analyses, spindle density and EEGNP (p ≥ .048) mediated moderate-to-severe OSA's negative effect on MoCA scores while ORPNREM, spindle power, and %-fast spindles mediated OSA's negative effect on DSC scores (p ≤ .018).

Conclusions: Altered spindle activity, ORP and normalized EEG power may be important contributors to cognitive deficits in patients with OSA.

Keywords: cognition; mediation; obstructive sleep apnea; odds-ratio product; sleep microarchitecture; spindles.

Graphical Abstract

Figure 1.

Linear associations between cognitive scores and N2 spindle density (A-C), power (D-F), frequency (G-I), and percentage of fast spindles (J-L) in our primary analysis cohort of participants with ≥60 minutes of total sleep time. Standardized parameter estimates (β) and 95% CI provided in plots reflect the relationships between cognitive scores and spindle characteristics adjusting for age, sex, education (except MoCA), total sleep time, and OSA severity group. Abbreviations: DSC, Weschler Adult Intelligence Scale 4th Edition WAIS-IV digit symbol coding; MoCA, Montreal cognitive assessment; RAVLT, Rey auditory verbal learning test.

Figure 2.

Linear associations between cognitive scores and the odds-ratio product for total recording time (ORP_TRT; A-C), during wakefulness (ORP_WAKE; D-F), and during non-rapid eye movement sleep (ORP_NREM; G-I) in our primary analysis cohort of participants with ≥60 minutes of total sleep time. Standardized parameter estimates (β) and 95% CI provided in plots reflect the relationships between cognitive scores and ORP adjusting for age, sex, education (except MoCA), total sleep time, and OSA severity group. Associations between DSC and ORP_WAKE are reported for each group (F) because of a significant group-by-ORP_WAKE interaction (No/Mild OSA = solid line; Moderate OSA = dotted line; and Severe OSA = dashed line). Abbreviations: DSC, Weschler Adult Intelligence Scale 4th Edition WAIS-IV digit symbol coding; MoCA, Montreal cognitive assessment; RAVLT, Rey auditory verbal learning test.

Figure 3.

Mediation analyses assessing the effect of moderate/severe OSA on the Montreal cognitive assessment (MoCA) total score (A, spindle frequency included in the mediation model; B, spindles %-fast included in the mediation model) and the Weschler Adult Intelligence Scale 4th Edition (WAIS-IV) digit symbol coding (DSC) score (C) in our primary analysis cohort of participants with ≥60 minutes of total sleep time. Values are the mean effects and 95% bias-corrected and accelerated confidence intervals pooled across all 50 imputed data sets (10 000 bootstrap iterations per data set). Significant mediators are shown with dashed arrows while potential mediators whose 95% confidence interval just failed to exclude zero are shown with dotted arrows; nonsignificant mediators are shown with solid black arrows. Abbreviations: % Fast, percentage of spindles with a frequency of 12–16 Hz; BMI, body mass index; EEG_NP, overall normalized EEG power; ORP_NREM, odds-ratio product during non-rapid eye movement sleep; T90, percentage of total sleep time with oxyhemoglobin saturation <90%; and TST, total sleep time.