Using sleep heart rate variability to investigate the sleep quality in children with obstructive sleep apnea

Background Obstructive sleep apnea (OSA) is associated with impaired sleep quality and autonomic dysfunction. Adenotonsillectomy significantly improves subjective and objective sleep quality in children with OSA. However, the postoperative changes in heart rate variability (HRV) indices (indicators of cardiac autonomic function) and their importance remain inconclusive in childhood OSA. This retrospective case series aimed to investigate the association of sleep HRV indices, total OSA-18 questionnaire score (a subjective indicator of sleep quality) and polysomnographic parameters (objective indicators of sleep quality), and effects of adenotonsillectomy on HRV indices, total OSA-18 questionnaire score and polysomnographic parameters in children with OSA. Methods Seventy-six children with OSA were included in baseline analysis, of whom 64 (84%) completed at least 3 months follow-up examinations after adenotonsillectomy and were included in outcome analysis. Associations between baseline variables, and relationships with treatment-related changes were examined. Results Multivariable linear regression models in the baseline analysis revealed independent relationships between tonsil size and obstructive apnea-hypopnea index (OAHI), adenoidal-nasopharyngeal ratio and very low frequency (VLF) power of HRV (an indicator of sympathetic activity), and normalized low frequency power (an indicator of sympathetic activity) and OAHI. The outcome analysis showed that adenotonsillectomy significantly improved standard deviation of all normal-to-normal intervals, and high frequency power, QoL (in terms of reduced total OSA-18 questionnaire score), OAHI and hypoxemia. Using a conceptual serial multiple mediation model, % change in OSA-18 questionnaire score and % change in VLF power serially mediated the relationships between change in tonsil size and % change in OAHI. Conclusions The improvement in OAHI after adenotonsillectomy was serially mediated by reductions in total OSA-18 questionnaire score and VLF power. These preliminary findings are novel and provide a direction for future research to investigate the effects of VLF power-guided interventions on childhood OSA.

Background: Obstructive sleep apnea (OSA) is associated with impaired sleep quality and autonomic dysfunction. Adenotonsillectomy significantly improves subjective and objective sleep quality in children with OSA. However, the postoperative changes in heart rate variability (HRV) indices (indicators of cardiac autonomic function) and their importance remain inconclusive in childhood OSA. This retrospective case series aimed to investigate the association of sleep HRV indices, total OSA-questionnaire score (a subjective indicator of sleep quality) and polysomnographic parameters (objective indicators of sleep quality), and e ects of adenotonsillectomy on HRV indices, total OSA-questionnaire score and polysomnographic parameters in children with OSA.
Methods: Seventy-six children with OSA were included in baseline analysis, of whom ( %) completed at least months follow-up examinations after adenotonsillectomy and were included in outcome analysis. Associations between baseline variables, and relationships with treatment-related changes were examined.
Results: Multivariable linear regression models in the baseline analysis revealed independent relationships between tonsil size and obstructive apnea-hypopnea index (OAHI), adenoidal-nasopharyngeal ratio and very low frequency (VLF) power of HRV (an indicator of sympathetic activity), and normalized low frequency power (an indicator of sympathetic activity) and OAHI. The outcome analysis showed that adenotonsillectomy significantly improved standard deviation of all normal-to-normal intervals, and high frequency power, QoL (in terms of reduced total OSA-questionnaire score), OAHI and hypoxemia. Using a conceptual serial multiple mediation model, % change in OSA-questionnaire score and % change in VLF power serially mediated the relationships between change in tonsil size and % change in OAHI.
. Materials and methods

. . Study participants
The Institutional Review Board of Chang Gung Medical Foundation approved this retrospective case series (No. 202200882B0). The requirement for written informed consent was waived because the current study was based on a secondary analysis of existing data. This study followed the World Medical Association's Declaration of Helsinki and the Strengthening the Reporting of Cohort Studies in Surgery guidelines (20).
We included consecutive children who underwent adenotonsillectomy for OSA at Chang Gung Memorial Hospital, Linkou Main Branch (Taoyuan, Taiwan) between March 1, 2017 and September 30, 2021. The inclusion criteria were: (1) age 5-12 years, and (2) obstructive AHI (OAHI) ≥ 2.0 events/h or obstructive apnea index (OAI) ≥ 1.0 events/h (21,22). The exclusion criteria were (1) patients with craniofacial, neuromuscular, or chronic inflammatory disorders (23,24), or (2) patients without available polysomnographic data. All the children underwent extracapsular tonsillectomy with tonsillar pillar suturing and adenoidectomy that aimed to improve the upper airway obstruction by the principal investigator (L-AL) in a single stage under general anesthesia (25). Children with follow-up polysomnographic data were included in outcome analysis ( Figure 1).

. . Clinical variables
Age, sex, body mass index (BMI), tonsil size, adenoidal-nasopharyngeal ratio (ANR) and evening blood pressure (BP) (2, 26), OSA-related quality of life, and polysomnographic parameters were recorded. All the clinical measurements were performed before and at least 3 months after adenotonsillectomy.
The tonsils were graded with a size scale from 1-4 (1: tonsils within the tonsillar; 2: tonsils visible outside the anterior pillars; 3: tonsils extending three-quarters of the way to the midline; 4: tonsils meeting at the midline) (27).
The ANR (distance from the point of maximal convexity of the adenoid shadow/the distance between the posterior border of the . /fpubh. . hard palate and the anteroinferior edge of the sphenobasioccipital synchondrosis) was measured on neck lateral view (28).
. . Sleep quality . . . Subjective measurement All parents evaluated their children's OSA-related quality of life using the Chinese version of the OSA-18 questionnaire (29), which includes 18 items grouped into 5 domains: sleep disturbance (4 items), physical suffering (4 items), emotional distress (3 items), daytime problems (3 items), and caregiver concerns (4 items). Each item was scored using a 7-point ordinal scale. The total score was calculated as the sum of the 18 items (overall range, 18-126) and has been shown to have excellent test-retest reliability (30).

. . . Objective measurement
All participants underwent full-night, in-laboratory polysomnography (Nicolet Biomedical Inc., Madison, WI, USA) (23). OAHI, OAI, arousal index, mean blood oxygen saturation (SaO 2 ), minimal SaO 2 , sleep stages and total sleep time were scored and manually verified by the study investigators (L-PC and Y-SH) using a standard approach of the American Academy of Sleep Medicine (31). For example, the AHI was calculated by dividing the sum of all apneas (defined as a ≥ 90% reduction in airflow for a duration of ≥ 2 consecutive breaths) and hypopneas (defined as a ≥ 30% reduction in airflow in association with electroencephalographic arousal or a ≥ 3% reduction in SpO 2 for a duration of ≥ 2 consecutive breaths) by the hours of total sleep time.

. . Sleep heart rate variability analysis
Electrocardiographic polysomnography signals were analyzed using HRV software (profusionSLEEP TM , version 4.5, build 502, Compumedics, Abbotsford, Australia). For artifact correction, automated annotations of electrocardiographic signals, such as loose leads, motion artifacts, and broken wires (32), were manually verified by trained technicians who had been certificated by the domestic board of the Taiwan Society of Sleep Medicine and shown substantial-to-almost perfect reliabilities in the scoring of respiratory events (intraclass correlation coefficients [ICCs] ranged from 0.66 to 0.98) (33). According to standard guidelines, timedomain indices, including standard deviation of all normal-tonormal (N-N) intervals (SDNN), number of pairs of adjacent N-N intervals differing by more than 50 ms in the entire recording divided by the total number of all N-N intervals (pNN50), and square root of the mean of the sum of the squares of differences between adjacent N-N intervals (RMSSD) were recorded. In addition, frequency-domain indices, including total power (0.0033-0.4 Hz), very low frequency (VLF) power (0.0033-0.04 Hz), low frequency (LF) power (0.04-0.15 Hz), normalized .

RMSSD ms
The square root of the mean of the sum of the squares of di?erences between adjacent N-N intervals.

. . Statistical analysis
Data were analyzed using SPSS version 25.0 (IBM Corp., Armonk, NY, USA) and GraphPad Prism 9.0 for Windows (Graph Pad Software Inc., San Diego, CA, USA). Changes in scores were calculated as postoperative minus preoperative values. Percentage change [(change in score/preoperative value) × 100] was calculated for variables of interest. Because all the children underwent extracapsular tonsillectomy, the change in tonsil size was equal to the negative value of tonsil size and used for further statistical analysis.
Using the Shapiro-Wilk test to examine normality, descriptive statistics were expressed as mean (standard deviation) for normally distributed continuous variables, median (interquartile range [IQR]) for skewed variables, and number (proportion) for categorical variables.
For continuous variables, the independent-samples t-test or Mann-Whitney U test was used to assess between-group changes; the paired-samples t-test or Wilcoxon signed-rank test was used to assess within-group changes as appropriate. Differences in categorical variables between two subgroups were analyzed using Fisher's exact test.
To facilitate comparisons with previous studies, linear regression models, or mediation and moderation analysis, nonnormally distributed data of reference studies were transformed to normal after estimation from the sample size (n), median (m), and the first (q 1 ) and third (q 3 ) quartiles (42,43). The sample standard deviation was estimated to be [(q 3 (42). In addition, non-normally distributed continuous variables were transformed to normal using a two-step approach: fractional rank and inverse-normal transformation (44). For comparisons with reference values, the one-sample t-test was applied.
Relationships between variables of interest were assessed using Pearson and Point-Biserial correlation tests as appropriate. Multivariable linear regression models, including all variables, with manual selection based on a probability of F < 0.05 were used to identify independent variables. The variance inflation factor of each predictor was calculated to adjust for intervariable relationships within the model. The regression model was repeated after removing all variables with a variance inflation factor ≥ 5 to reduce multicollinearity (45).
Conditional process analysis was performed to evaluate the mediators and moderators between changes in tonsil size/ANR . /fpubh. . Data are expressed as mean (standard deviation), median (interquartile range), or number (%). a Data were compared between participants included in the outcome analysis and those excluded using the independent-samples t-test, Mann-Whitney U test, or Fisher's exact test as appropriate.
ANR, adenoidal-nasopharyngeal ratio; BMI, body mass index; BP, blood pressure; bpm, beats per min; HF, high frequency; LF, low frequency; LF%, normalized LF power; N-N, normal-tonormal; OAHI, obstructive apnea-hypopnea index; OAI, obstructive apnea index; OSA, obstructive sleep apnea; pNN50, proportion of N-N50 divided by the total number of N-N intervals; REM, rapid eye movement; RMSSD, square root of the mean of the sum of the squares of di?erences between adjacent N-N intervals; SaO2, blood oxygen saturation; SDNN, standard deviation of all N-N intervals; TST, total sleep time; VLF, very low frequency.   baseline analysis (Figure 1), of whom 64 (84%) were included in the outcome analysis and 12 (16%) were not included due to no available follow-up polysomnography. All baseline variables were comparable between these two subgroups ( Table 2).

. . Sleep heart rate variability
Distributions of HRV indices in baseline analysis are summarized in Table 2. For comparing with previous studies, the HRV indices in this study (full-night), normal controls (full-night) (47), children with OSA (full-night) (14), children with moderate-to-severe OSA (N3 sleep) (13,17), and children with OSA/obesity (full-night) (48) are summarized in Table 3. Comparing with three representative full-night HRV studies (14,47,48), SDNN, total power and VLF power in the children with OSA were significantly higher than normal values (Figure 2).

. . Measurement reproducibility of sleep heart rate variability
To assess measurement reproducibility, we calculated ICCs using HRV indices measured at least 3 months apart in 12 patients with stable residual OSA after adenotonsillectomy (Table 4). Their variables of interest were comparable to the patients with altered OSA (Tables 5, 6). Most HRV measurements demonstrated moderate (N-N interval, SDNN, RMSSD, total power, LF%) or substantial (VLF power, LF power, LF/HF ratio) reproducibility. Further, the reproducibility of pNN50 and HF power were fair (41).

. . Associations between variables of interest at baseline
Nested data structure and significant correlations were found among the polysomnographic parameters, several clinical variables and HRV indices ( Figure 3). However, total OSA-18 questionnaire score was not associated with variables of interest. Using multivariable linear regression models (Table 7), male sex, OAHI and N3 sleep were independently associated with tonsil size, and systolic BP, OAHI and VLF power were independently associated with ANR. Furthermore, tonsil size, diastolic BP and LF% were independently correlated with OAHI. Table 7 summarizes the independent associations of other polysomnographic parameters with the variables of interest.

. . Changes in the variables of interest after adenotonsillectomy
The median follow-up period was 4 (IQR, 3-6) months. In outcome analysis, mean SaO 2 , minimal SaO 2 and rapid eye movement sleep significantly increased, and OSA-18 score, OAHI, OAI, arousal index and N1 sleep significantly reduced after adenotonsillectomy (Table 5).

. . Associations of percentage changes in the variables of interest
revealed significant associations with nested data structure ( Figure 4). Using multivariable linear regression models (Table 8), % changes in OSA-18 score, OAHI and HF power were independently associated with change in tonsil size. Age at diagnosis, male sex and % change in arousal index were independently associated with change in ANR, and change in tonsil size was independently correlated with % change in OAHI.

. . Mediation and moderation analyses
Consistent relationships between "tonsil size and OAHI" and "change in tonsil size and % change in OAHI" were observed. Mediation and moderation analyses were performed from change in tonsil size to % change in OAHI, especially with regards to HRV indices, and only a significant conceptual serial multiple mediation model was identified: change in tonsil size (independent variable), % change in OSA-18 score (first mediator), % change in VLF power (second mediator), and % change in OAHI (dependent mediator) ( Figure 5). The direct paths from change in tonsil size to % change in OAHI, change in tonsil size to % change in OSA-18 score, change in tonsil size to % change in VLF power, change in OSA-18 to % change in VLF power, and % change in VLF power to % change in OAHI were significant. In contrast, the direct paths from change in OSA-18 to % change in OAHI were not significant. The

. Discussion
This study is the first to report that OSA-related quality of life and VLF power were first and second mediators of the relationship between tonsil size and improvement in AHI using a conceptual serial multiple mediation model. Beyond providing important mechanistic insights, these results suggest that VLF power could be a new target for OSA therapy in children. For example, exercise training can decrease VLF power over time (49) and reduce AHI (50) in adults.
Our results confirmed the reproducibility of sleep HRV measurements at two time points. Most measures showed moderate or substantial reproducibility, except for pNN50 and HF power. The possible reason for this relatively lower reproducibility may be related to sleep stage and arousal index. To the best of our knowledge, no comprehensive reproducibility study has reported HRV measurements in children with OSA. Accordingly, the interpretations of sleep pNN50 and HF power should be made with caution in this population.
Using full-night HRV measurements, SDNN, total power, and VLF power in the children with OSA were significantly higher than normal values (Figure 2) (14,47,48). SDNN and total power represent total capacity of the regulation system, whilst VLF power represents sympathetic activity (36) ( Table 1). Although SNS and PNS activities both contribute to SDNN and total power, longterm recordings have revealed that SNS activity is more related to these indices (51). The transition between normal and pathological respiration can enhance SNS activity rather than PNS activity in adults with OSA (52). Additionally, the results of this and previous studies (13,14,53) suggest that sympathetic activity increases during sleep in children with OSA; however, SDNN in the 12 obese children with OSA in this study was comparable to normal values (47) (Table 3). This discrepancy may be explained by the patients' weight status, since childhood obesity is significantly related to low SDNN (54). Furthermore, this study and Isaiah's study (14) found that % changes in SDNN and total power were not related to % . /fpubh. .

FIGURE
Associations of polysomnographic variables with clinical variables and sleep heart rate variability indices. Data are summarized as Pearson's or Point-Biserial rho, as appropriate. Blank spaces mean two-sided P ≥ . . change in OAHI. Therefore, these changes could not be simply due to improvements in OAHI after adenotonsillectomy.
The baseline values and % changes in tonsil size and ANR were not consistently associated with most HRV indices. Despite increased ANR being related to decreased VLF power in children with OSA, the causal relationship between adenoid hypertrophy and reduced sympathetic activity could not be supported by the post-operative changes. Nevertheless, our findings suggested a positive relationship between change in tonsil size and % change in HF power of HRV (parasympathetic modulation). We hypothesize that tonsillectomy may directly injure or cause scar formation, thereby reducing function of the lingual branch of the hypoglossal nerve, interrupting baroreceptor signaling at the carotid sinus, influencing vagus nerve function, eventually resulting in decreased parasympathetic modulation and increased sympathetic activity of cardiac autonomic function during sleep. This condition may further interfere with the relationships between % change in SDNN or total power and % change in OAHI.
Our results demonstrated significant relationships between the change in tonsil size and % change in OAHI as well as relationships between the change in tonsil size and % change in OSA-18 score as previous studies (29,55). Tonsil size has been significantly associated with the change in OSA-18 score after tonsillectomy in children with sleep-disordered breathing (56). Although a change in AHI has been associated with a change in OSA-18 score (23), we found that this association was not independent in this study. In simple mediation and moderation models, % change in OSA-18 score neither mediated nor moderated the relationship between the Frontiers in Public Health frontiersin.org . /fpubh. . change in tonsil size and % change in OAHI. However, in serial mediation analysis, the relationship between the change in tonsil size and % change in OAHI was mediated by % change in OSA-18 score and % change in VLF power in serial analysis, and also by % change in VLF power alone ( Figure 5). VLF rhythm is a cardiac intrinsic rhythm which is essential for health and happiness (36). Even though there is currently no agreement on the physiological mechanisms responsible for activity within the VLF band, low VLF power has been associated with adverse outcomes and all-cause mortality (57,58). This band is generated by the stimulation of afferent sensory neurons in the heart (59). In animal models, stressful stimulation (60) and paradoxical sleep deprivation (61) have been shown to significantly reduce VLF power (60). In this study, the inverse relationship between % change in OSA-18 and % change in VLF power suggested that reduced OSA-specific stress and sleep disturbance may increase sleep VLF power.
However, VLF power is an independent predictor of AHI in humans (62). VLF power is significantly elevated during pathological respiration compared with normal respiration in OSA Frontiers in Public Health frontiersin.org . /fpubh. .

FIGURE
Associations of % changes in polysomnographic variables, % changes in clinical variables, and % changes in sleep heart rate variability indices. Blank spaces mean two-sided P ≥ . . patients (52). Therefore, it is reasonable that reduced AHI may contribute to a decrease in VLF power after adenotonsillectomy. Our mediation model also highlighted the possibility that changes in VLF power may influence changes in AHI in children with OSA. Sympathetic abnormalities were shown to precede the development of mild OSA in a cohort of adults with no known diagnosis of OSA (63). Although further direct evidence is warranted, these studies suggest that a reduction in VLF power may help to alleviate the AHI in children with OSA. Therefore, the mediation role of VLF power on the relationship between change in tonsil size and % change in AHI is of interest. Increasing exercise intensity can reduce awake VLF power (49), and morning exercise can increase sleep VLF power (64) in adults. Besides, exercise training (65) or aerobic exercise combined with resistance training can reduce AHI in adults. Therefore, VLF power is modifiable and may be a marker of therapeutic efficacy and a potential therapeutic target for OSA (66). However, in children with adenotonsillar hypertrophy, it may be unlikely that addressing the HRV independently will improve AHI unless there is a clear demonstration the neuromuscular tone is improved to the point that the tonsils do not medialize during sleep. Accordingly, future studies should focus on VLF power-lowering interventions and their effects on the severity of childhood OSA.

. . Strengths and limitations
Compared with previous studies (13,14,17,48), the greatest strengths of this investigation were the inclusion of a sample of representative and well-characterized pediatric OSA patients. Our results provided a preliminary yet comprehensive documentation of the relationships of HRV indices with clinical  variables and polysomnographic parameters before and after adenotonsillectomy, which showed some novel and interesting findings. However, limitations should be addressed. First, the HRV results may have been affected by certain psychophysiological changes (e.g., anthropometrics, lifestyle factors, acute or chronic diseases) other than adenotonsillectomy. However, the use of standardized, full-night, in-laboratory protocols with moderate-to-substantial reproducibility in most HRV indices among OSA patients with stable severity reduces this concern. Second, approximately half of our subjects received both adenotonsillectomy and medical treatment, and the heterogeneity of care may have had a confounding effect. Nevertheless, these interdisciplinary treatments are closer to real-world care for OSA, and a greater variability in AHI changes may contribute to better generalizability of this study. Third, 3 months may not be long enough to show cardiovascular changes, and studies with a longer follow-up period are warranted for this young population. Finally, in this study, there was no evidence of direct mediations of HRV on the relationship between adenotonsillectomy and AHI or AHI on the relationship between adenotonsillectomy and HRV indices (14), and this may be due to difficulties in measuring HRV across various sleep stages. Among school-age children, excessive body movements and parasomnia (67) make the researchers need 2-min epochs to analyze HRV and choose sleep periods free of respiratory events and movement artifacts (17). However, frequency-domain measurements, such as VLF power and LF/HF ratio, often require recording periods of at least 5 min (34). Therefore, measuring HRV across various stages in our study population is challenging. Nevertheless, averages may not be sensitive enough to detect sleep stage-specific mediating effects. In future studies, ultra-short-term HRV measurements during different sleep stages should be conducted to accurately assess the impact of nocturnal HRV changes on the management of OSA.

. Conclusion
In conclusion, we confirmed that analysis of electrocardiographic polysomnography signals is a reliable Frontiers in Public Health frontiersin.org . /fpubh. .

FIGURE
A serial multiple mediation model of the e ect of change in tonsil size. Data are summarized as β and standard errors. *P < . and ≥ . ; **P < . and ≥ . . Solid lines indicate significant paths, while a dotted line indicates a non-significant path. OAHI, obstructive apnea-hypopnea index; OSA, obstructive sleep apnea; VLF, very low frequency. method to measure HRV over 3 months in children with OSA. Adenotonsillectomy either reduced AHI or sympathetic activity during sleep. Improved OSA-specific quality of life and reduced sleep VLF power serially mediated the relationship between the change in tonsil size and % change in AHI. These findings suggest that HRV measurement may help monitor the sleep quality status and the disease burden of childhood OSA and many other venues. Our preliminary results also support applications of wireless HRV measurements with high-fidelity psychophysiology acquisition using edge computing in the patient's natural sleeping environment to overcome the highly obtrusive effects of visiting the sleep laboratory (68). This technology can potentially be a "platinum standard" of sleep studies instead of the traditional "gold standard" of in-laboratory polysomnography.

Data availability statement
The original contributions presented in the study are included in the article/supplementary material, further inquiries can be directed to the corresponding authors.

Ethics statement
The studies involving human participants were reviewed and approved by Institutional Review Board of Chang Gung Medical Foundation, Taoyuan, Taiwan. Written informed consent for participation was not provided by the participants' legal guardians/next of kin because: the current study was based on a secondary analysis of existing data. We provided a copy of the approval by the Institutional Review Board of Chang Gung Medical Foundation (No. 202200882B0), which approved the waiver of the participants' consent.