Serum neurofilament light reflects cognitive dysfunctions in children with obstructive sleep apnea

Abstract

Background: In children, obstructive sleep apnea (OSA) can cause cognitive dysfunctions. Amyloid-beta and tau are elevated in OSA. Neurofilament light (NfL) is a marker of neuro-axonal damage, but there are no reports of NfL for OSA. The objective was to investigate the serum levels of NfL and tau in children with or without OSA and explore their relationship with cognitive dysfunctions caused by OSA.

Methods: This retrospective case-control study included children diagnosed with adenoid tonsil hypertrophy from July 2017 to September 2019 at the Second Affiliated Hospital of Xi'an Jiaotong University. Correlations between cognitive scores and tau and NfL were examined.

Results: Fifty-six OSA and 49 non-OSA children were included. The serum NfL levels were higher in the OSA group (31.68 (27.29-36.07) pg/ml) than in the non-OSA group (19.13 (17.32-20.95) pg/ml) (P < 0.001). Moreover, NfL was correlated with the course of the disease, apnea-hypopnea index (AHI), obstructive apnea index (OAI), obstructive apnea-hypopnea index (OAHI), average oxygen saturation (SaO₂), respiratory arousal index (RAI), and cognitive dysfunctions evaluated by the Chinese Wechsler Intelligence Scale for Children (C-WISC) (all P < 0.05). The area under the receiver operating characteristics curve (AUC) of NfL was 0.816 (95%CI: 0.736-0.897). Multiple regression analysis revealed that NfL was significantly associated with verbal intelligence quotient (VIQ), performance intelligence quotient (PIQ) and full-scale intelligence quotient (FIQ) (P < 0.001, respectively).

Conclusions: Serum NfL levels are associated with the severity of cognitive dysfunctions in children diagnosed with adenoid tonsil hypertrophy and might be a candidate noninvasive, objective marker to identify cognitive dysfunctions in children with OSA.

Keywords: Children; Cognitive dysfunctions; Neurofilament proteins; Obstructive sleep apnea; Tau proteins.

Fig. 1

Summary of patient inclusion. Two hundred thirty-two children were diagnosed with adenotonsillar hypertrophy, and 75 children were excluded: 6 with genetic syndromes associated with cognitive disabilities and psychiatric conditions, 28 cases with a history of sedatives or corticosteroids, 8 with neurological abnormalities such as traumatic brain injury and cerebral palsy, and 33 with visual or hearing disorders. The 157 included children all underwent polysomnography (PSG), and 65 were diagnosed with obstructive sleep apnea (OSA). After checking the available serum, 56 OSA and 49 non-OSA children were included in the final analysis

Fig. 2

Comparison of serum neurofilament light (NfL) and tau levels between the non-OSA and OSA groups. a) The serum NfL levels were high in the OSA group (31.68 (27.29–36.07) pg/ml) compared with the non-OSA group (19.13 (17.32–20.95) pg/ml) (P < 0.001). b) There were no significant differences in serum tau levels between the two groups (P = 0.172)

Fig. 3

Multivariate correlations with neurofilament light (NfL), tau and clinical parameters. The numerical value of each cell is the spearman correlation coefficient. The baseline (r) value (white) is 0, the maximum value (red) is 1, the light green is -0.3, and the dark green is below -0.3. **P < 0.01, *P < 0.05

Fig. 4

Diagnostic performance of neurofilament light (NfL). Area under the curve (AUC) = 0.816 (95% confidence interval: 0.736–0.897), P < 0.001

Fig. 5

Least absolute shrinkage and selection operator (LASSO) regression analyses for VIQ (a), PIQ (b) and FIQ (c). The left panels show the selection of the optimal parameters (lambda) in the LASSO model used the minimum criterion of tenfold cross-validation. Partial likelihood deviation curves were plotted by lambda. Two dotted vertical lines were drawn at the optimal values at the minimum criterion (right) and the 1-SE criterion (left). The right panels show that LASSO coefficient profiles of 17 features. A coefficient profile plot was produced against the log lambda sequence. Vertical lines were drawn at the value selected using tenfold cross-validation, where optimal lambda resulted from the features with non-zero coefficients