Assessment of Nocturnal Autonomic Cardiac Imbalance in Positional Obstructive Sleep Apnea. A Multiscale Nonlinear Approach

Positional obstructive sleep apnea (POSA) is a major phenotype of sleep apnea. Supine-predominant positional patients are frequently characterized by milder symptoms and less comorbidity due to a lower age, body mass index, and overall apnea-hypopnea index. However, the bradycardia-tachycardia pattern during apneic events is known to be more severe in the supine position, which could affect the cardiac regulation of positional patients. This study aims at characterizing nocturnal heart rate modulation in the presence of POSA in order to assess potential differences between positional and non-positional patients. Patients showing clinical symptoms of suffering from a sleep-related breathing disorder performed unsupervised portable polysomnography (PSG) and simultaneous nocturnal pulse oximetry (NPO) at home. Positional patients were identified according to the Amsterdam POSA classification (APOC) criteria. Pulse rate variability (PRV) recordings from the NPO readings were used to assess overnight cardiac modulation. Conventional cardiac indexes in the time and frequency domains were computed. Additionally, multiscale entropy (MSE) was used to investigate the nonlinear dynamics of the PRV recordings in POSA and non-POSA patients. A total of 129 patients (median age 56.0, interquartile range (IQR) 44.8–63.0 years, median body mass index (BMI) 27.7, IQR 26.0–31.3 kg/m2) were classified as POSA (37 APOC I, 77 APOC II, and 15 APOC III), while 104 subjects (median age 57.5, IQR 49.0–67.0 years, median BMI 29.8, IQR 26.6–34.7 kg/m2) comprised the non-POSA group. Overnight PRV recordings from positional patients showed significantly higher disorderliness than non-positional subjects in the smallest biological scales of the MSE profile (τ = 1: 0.25, IQR 0.20–0.31 vs. 0.22, IQR 0.18–0.27, p < 0.01) (τ = 2: 0.41, IQR 0.34–0.48 vs. 0.37, IQR 0.29–0.42, p < 0.01). According to our findings, nocturnal heart rate regulation is severely affected in POSA patients, suggesting increased cardiac imbalance due to predominant positional apneas.


Study Design and Sample Size
The proposed research was an ancillary study of the prospective observational ScreenOX study (NCT03295149), primarily aimed at assessing NPO as an abbreviated screening test for OSA at home. In order to ensure the statistical significance of the present secondary analysis in the context of POSA, a sample size was computed. The sample size was estimated using G*Power 3.1.9 (Düsseldorf, Germany) [31]. Differences in mean and standard deviation among POSA and non-POSA patients in previously reported HRV indices were used to measure the effect size [19]. A statistical power of 95% and a significance level of 0.05 were set, leading to a medium effect size equal to 0.45 and a minimum sample size of 216 patients.
In the ScreenOX study, a total of 320 eligible patients correctly completed both ambulatory PSG and simultaneous portable NPO. Regarding POSA, 87 studies (27.2%) were not consistent with the APOC criteria (<10% of sleep time in both non-supine and supine positions or a total AHI <5 events/h) and were removed from the study. Accordingly, 233 patients finally composed the population under study, which fit with the estimated minimum sample size. Table 1 shows the demographic and clinical characteristics of the population under study.

Heart Rate Modulation Assessment
Time, frequency and nonlinear analyses were conducted to thoroughly assess the PRV dynamics of POSA and non-POSA patients. Firstly, well-known pulse-to-pulse interval-based indices were computed [21]: (1) the average of the pulse-to-pulse interval (AVNN), which is a global estimate of the interbeat period (inverse of pulse rate); (2) the standard deviation of the pulse-to-pulse intervals (SDNN), which quantifies the degree of variability; and (3) the root mean square of the successive differences of pulse-to-pulse intervals (RMSSD), which accounts for vagal activity.
Despite some controversy [32,33], analysis in the frequency domain has been found to provide useful information on the modulation of heart rates by the autonomous nervous system. Accordingly, the conventional frequency bands in the framework of heart rate dynamics were characterized by the following indices [21,26]: (1) very low frequency (VLF) power (0.0033-0.04 Hz), which measures Entropy 2020, 22, 1404 5 of 17 rhythms linked to the influence of the vagal and renin-angiotensin system on the pulse rate; (2) low frequency (LF) power (0.04-0.15 Hz), which captures joint modulation of the pulse rate by sympathetic and parasympathetic branches of the autonomous nervous system; (3) high frequency (HF) power (0.15-0.40 Hz), which quantifies exclusively the influence of the parasympathetic nervous system; and (4) the low frequency to high frequency ratio (LF/HF), which measures the so-called sympathovagal balance.
In order to further characterize the modulating mechanisms of the pulse rate in the frequency domain, the widely known Shannon spectral entropy (SSE) was applied to the spectrum of PRV recordings. SSE parameterizes the shape of the power spectrum of a signal so that higher SSE values account for uniform distribution along frequencies (higher irregularity in the time domain), reflecting no dominance or influence of a particular system, whereas lower SSE values are representative of a condensed spectrum in a frequency band (lower irregularity or higher periodicity in the time domain), reflecting the higher influence of a particular system [28,34]. Accordingly, the spectral entropy was computed in the whole spectrum (SSE T ) and in the classic spectral bands of very low frequency (SSE VLF ), low frequency (SSE LF ), and high frequency (SSE HF ).
Multiscale entropy (MSE) was used to assess the nonlinear dynamics of PRV in POSA and non-POSA patients. Physiological control systems, such as cardiac modulation, are characterized by complex dynamics reflecting time-dependent fluctuations. MSE is aimed at quantifying the complexity of a time series and looking for changes in entropy along different time scales [35]. To characterize the dynamical structure of a physiological recording, different coarse-grained versions of the signal were composed and further analyzed. A new version of the signal in ever deeper time scales (the so-called coarse-grained versions) was composed by averaging the samples of the original time series within non-overlapping segments of a length τ, increasing the window length for each new coarse-grained version. For each τ (i.e., for each time scale), a single-scale entropy measure was computed for the corresponding coarse-grained sequence so that the MSE curve was obtained by plotting entropy as a function of τ [35,36]. In this study, the well-known sample entropy (SampEn) algorithm was used to estimate the entropy [37,38]. A maximum time scale τ = 14 was set to ensure a proper estimation of the SampEn in the highest time scale [28,37]. According to the original work by Costa et al. [36], signals showing larger entropy values for most time scales are more complex than signals reaching lower single-scale entropies in the same region.

Statistical Analysis
SPSS Statistics 24 (IBM Corp., Armonk, NY, USA) and Matlab R2020a (The MathWorks Inc., Natick, MA, USA) were used to carry out statistical analyses. Overall descriptive analyses were performed in terms of the median and the 25th-75th percentile. The Kolmogorov-Smirnov normality test confirmed that the variables under study did not follow a normal distribution. Accordingly, the non-parametric Mann-Whitney test was used to assess statistical differences among non-POSA and POSA patients for quantitative continuous variables. The chi-squared test was applied for the categorical ones. In the multiclass approach (non-POSA vs. APOC I vs. APOC II vs. APOC III), the non-parametric Kruskal-Wallis test was used to assess statistical differences among groups. In addition, the Mann-Whitney test was applied to inspect differences between each particular pair of patient groups. In this regard, Fisher's least significant difference procedure was applied to correct for multiple comparisons. All p-values <0.05 were considered statistically significant. Finally, linear association between the cardiac modulation indices and the polysomnographic variables was investigated using the non-parametric Spearman correlation index.

Results
A total of 129 patients were diagnosed as POSA according to APOC rules (37 APOC I, 77 APOC II and 15 APOC III), while the remaining 104 subjects comprised the non-POSA group. Table 1 shows the Entropy 2020, 22, 1404 6 of 17 demographics and sleep apnea severity distribution of both groups, as well as frequent comorbidities and medications that could affect heart rate modulation.
No significant differences were found in terms of age and gender, while POSA patients showed significantly lower body mass indices (BMIs) than non-POSA subjects (27.7 vs. 29.8 kg/m 2 ; p < 0.05). Regarding sleep apnea severity, moderate OSA was predominant among POSA patients (73.5% vs. 26.5%; p < 0.05), while the number of mild patients was significantly lower (37.7% vs. 62.3%; p < 0.05). Overall, severe patients were remarkably predominant in our sample, and there were no statistical differences between the groups under study regarding severe OSA. Finally, no significant differences were found in terms of common comorbidities and medications able to potentially influence HRV dynamics. Table 2 summarizes the polysomnographic variables for non-POSA and POSA patients. No statistically significant differences were found between both groups concerning sleep staging. In regard to respiratory event scoring, there were no statistical differences among groups, both for total AHI and for individual apnea and hypopnea indices. On the other hand, POSA patients showed significantly lower AHIs during the rapid eye movement (REM) stage than non-POSA subjects (28.2 vs. 41.9 events/h; p < 0.05), as well as significantly lower AHIs in the non-supine positions (10.9 vs. 31.0 events/h; p < 0.001). No statistical differences were found in terms of time sleeping in the supine position and AHI values while supine. Similarly, POSA and non-POSA patients showed no significant differences in the average duration of respiratory events.  Regarding the portable unattended NPO, Table 3 shows the average values of all the variables provided by the device for the groups under study. There were no significant differences between Entropy 2020, 22, 1404 7 of 17 groups concerning the oxygen desaturation indices of 3% and 4%. Regarding the hypoxemia measures, POSA patients showed significantly milder hypoxemia levels than non-POSA subjects in terms of the cumulative time with a saturation below 90% (CT90) (6.9% vs. 12.1%; p < 0.05) and minimum saturation (81.0% vs. 76.0%; p < 0.05). Similarly, POSA individuals showed slight but significantly higher average saturations than non-POSA subjects during desaturation events of both 3% (89.4% vs. 89.1%; p < 0.05) and 4% (88.8% vs. 88.1%; p < 0.05). Concerning the overall pulse rate, no statistical differences were found in terms of the average and minimum pulse rates, while POSA patients showed significantly lower maximum rates than non-POSA subjects (95.0 vs. 100.0 bpm; p < 0.05). Data are presented as a median [25th, 75th percentiles]; bpm = beats per minute; CT90 = cumulative time with saturation below 90%; ODI3 = oxygen desaturation index of 3%; ODI4 = oxygen desaturation index of 4%; PR AVG = average pulse rate; PR MAX = maximum pulse rate; PR MIN = minimum pulse rate; SpO2 AVG(inEv3%) = average blood oxygen saturation in desaturations >3%; SpO2 AVG(inEv4%) = average blood oxygen saturation in desaturations >4%; SpO2 AVG(noEv3%) = average blood oxygen saturation removing desaturations >3%; SpO2 AVG(noEv4%) = average blood oxygen saturation removing desaturations >4%; SpO2 MIN = minimum blood oxygen saturation; and TRT = total recording time. Table 4 summarizes the cardiac modulation indices from the long-term overnight PRV recordings of POSA and non-POSA patients. No significant differences were found between both groups regarding the conventional time and frequency domain indices. The SSE of the POSA individuals showed a slight but not significant trend toward higher irregularity in the whole spectrum (SSE T : 0.50 vs. 0.49; p = 0.056) and, particularly, in the low frequency band (SSE LF : 0.84 vs. 0.83; p = 0.062) compared with non-POSA subjects. In this way, Figure 1a illustrates the averaged power spectral content for both groups in the conventional frequency bands, where the power spectral density (PSD) curve for the POSA patients almost matched the one for the non-POSA subjects. On the contrary, nonlinear analysis by means of MSE yielded several indices able to properly parameterize the differences between the POSA and non-POSA individuals. Particularly, POSA patients showed significantly higher entropy (disorderliness) in the low time scales (τ ≤ 6) than the non-POSA subjects. Figure 2 shows the averaged MSE curve for each group under study. The curve for POSA patients is above the one for non-POSA subjects in all time scales, suggesting remarkably higher complexity in the overnight PRV recordings in the presence of POSA. It is important to point out that the entropy increased as the time scale also increased until a stability region was reached around τ = 10. This suggests that there was essential information beyond the original signal (τ = 1) for low time scales.      Table 5 shows the cardiac modulation indices for the non-POSA subjects and the three POSA categories, according to the APOC criteria. No differences among groups were found using conventional measures or SSE. In the same way, it can be observed in Figure 1b that the overnight spectral content of each group was very similar, with just slight visual differences in the lower frequency bands not leading to significant p-values. On the contrary, MSE analysis showed significant statistical differences among the groups in the lowest time scales (τ ≤ 2). A pair-wise post hoc analysis yielded significant differences between the non-POSA subjects and all the POSA groups for τ = 2, whereas differences between the non-POSA and APOC II and III groups were found for τ = 1. Figure 3 shows the MSE curves for all APOC groups. Table 5. PRV-derived cardiac indices among the POSA categories, according to the APOC criteria.   Table 5 shows the cardiac modulation indices for the non-POSA subjects and the three POSA categories, according to the APOC criteria. No differences among groups were found using conventional measures or SSE. In the same way, it can be observed in Figure 1b that the overnight spectral content of each group was very similar, with just slight visual differences in the lower frequency bands not leading to significant p-values. On the contrary, MSE analysis showed significant statistical differences among the groups in the lowest time scales (τ ≤ 2). A pair-wise post hoc analysis yielded significant differences between the non-POSA subjects and all the POSA groups for τ = 2, whereas differences between the non-POSA and APOC II and III groups were found for τ = 1. Figure 3 shows the MSE curves for all APOC groups. Data are presented as a median [25th, 75th percentiles]; nu = normalized units; AVNN = average of the pulse-to-pulse interval; HFn = normalized spectral power in the high frequency band; LF/HF = low frequency to high frequency ratio or sympathovagal balance; LFn = normalized spectral power in the low frequency band; OSA = obstructive sleep apnea; POSA = positional obstructive sleep apnea; PT = total signal power; RMSSD = root mean square of the successive differences of the pulse-to-pulse intervals; SampEnj = sample entropy in the scale τ = j; SDNN = standard deviation of the pulse-topulse interval; SSET = spectral entropy in the whole spectra; SSEVLF = spectral entropy in the very low frequency band; SSELF = spectral entropy in the low frequency band; SSEHF = spectral entropy in the high frequency band; VLFn = normalized spectral power in the very low frequency band. * Significant differences in non-POSA vs. APOC I subjects; ⁋ Significant differences in non-POSA vs. APOC II subjects; † Significant differences in non-POSA vs. APOC III subjects; ‡ Significant differences in APOC I vs. APOC II subjects; ╫ Significant differences in APOC I vs. APOC III subjects; ⸸ Significant differences in APOC II vs. APOC III subjects. Single-entropy values along the MSE curve for low scales (τ ≤ 6) showed moderate but significant correlation with the total AHI (SampEn4: 0.235, p < 0.001) and AHISUP (SampEn4: 0.224, p < 0.01), higher than the conventional time domain (AVNN: −0.201, p < 0.01 and −0.151, p < 0.05, respectively) and frequency domain (LF/HF: 0.183, p < 0.01 and 0.197, p < 0.01, respectively) indices. Overall, the SSEVLF     Data are presented as a median [25th, 75th percentiles]; nu = normalized units; AVNN = average of the pulse-to-pulse interval; HFn = normalized spectral power in the high frequency band; LF/HF = low frequency to high frequency ratio or sympathovagal balance; LFn = normalized spectral power in the low frequency band; OSA = obstructive sleep apnea; POSA = positional obstructive sleep apnea; PT = total signal power; RMSSD = root mean square of the successive differences of the pulse-to-pulse intervals; SampEnj = sample entropy in the scale τ = j; SDNN = standard deviation of the pulse-topulse interval; SSET = spectral entropy in the whole spectra; SSEVLF = spectral entropy in the very low frequency band; SSELF = spectral entropy in the low frequency band; SSEHF = spectral entropy in the high frequency band; VLFn = normalized spectral power in the very low frequency band. * Significant differences in non-POSA vs. APOC I subjects; ⁋ Significant differences in non-POSA vs. APOC II subjects; † Significant differences in non-POSA vs. APOC III subjects; ‡ Significant differences in APOC I vs. APOC II subjects; ╫ Significant differences in APOC I vs. APOC III subjects; ⸸ Significant differences in APOC II vs. APOC III subjects.  ta are presented as a median [25th, 75th percentiles]; nu = normalized units; AVNN = average of pulse-to-pulse interval; HFn = normalized spectral power in the high frequency band; LF/HF = frequency to high frequency ratio or sympathovagal balance; LFn = normalized spectral power he low frequency band; OSA = obstructive sleep apnea; POSA = positional obstructive sleep apnea; total signal power; RMSSD = root mean square of the successive differences of the pulse-to-pulse rvals; SampEnj = sample entropy in the scale τ = j; SDNN = standard deviation of the pulse-tose interval; SSET = spectral entropy in the whole spectra; SSEVLF = spectral entropy in the very low uency band; SSELF = spectral entropy in the low frequency band; SSEHF = spectral entropy in the h frequency band; VLFn = normalized spectral power in the very low frequency band. * Significant erences in non-POSA vs. APOC I subjects; ⁋ Significant differences in non-POSA vs. APOC II jects; † Significant differences in non-POSA vs. APOC III subjects; ‡ Significant differences in APOC s. APOC II subjects; ╫ Significant differences in APOC I vs. APOC III subjects; ⸸ Significant erences in APOC II vs. APOC III subjects.  Data are presented as a median [25th, 75th percentiles]; nu = normalized units; AVNN = average of the pulse-to-pulse interval; HFn = normalized spectral power in the high frequency band; LF/HF = low frequency to high frequency ratio or sympathovagal balance; LFn = normalized spectral power in the low frequency band; OSA = obstructive sleep apnea; POSA = positional obstructive sleep apnea; PT = total signal power; RMSSD = root mean square of the successive differences of the pulse-to-pulse intervals; SampEnj = sample entropy in the scale τ = j; SDNN = standard deviation of the pulse-topulse interval; SSET = spectral entropy in the whole spectra; SSEVLF = spectral entropy in the very low frequency band; SSELF = spectral entropy in the low frequency band; SSEHF = spectral entropy in the high frequency band; VLFn = normalized spectral power in the very low frequency band. * Significant differences in non-POSA vs. APOC I subjects; ⁋ Significant differences in non-POSA vs. APOC II subjects; † Significant differences in non-POSA vs. APOC III subjects; ‡ Significant differences in APOC I vs. APOC II subjects; ╫ Significant differences in APOC I vs. APOC III subjects; ⸸ Significant differences in APOC II vs. APOC III subjects.  Data are presented as a median [25th, 75th percentiles]; nu = normalized units; AVNN = average of the pulse-to-pulse interval; HFn = normalized spectral power in the high frequency band; LF/HF = low frequency to high frequency ratio or sympathovagal balance; LFn = normalized spectral power in the low frequency band; OSA = obstructive sleep apnea; POSA = positional obstructive sleep apnea; P T = total signal power; RMSSD = root mean square of the successive differences of the pulse-to-pulse intervals; SampEn j = sample entropy in the scale τ = j; SDNN = standard deviation of the pulse-to-pulse interval; SSE T = spectral entropy in the whole spectra; SSE VLF = spectral entropy in the very low frequency band; SSE LF = spectral entropy in the low frequency band; SSE HF = spectral entropy in the high frequency band; VLFn = normalized spectral power in the very low frequency band. * Significant differences in non-POSA vs. APOC I subjects;

Non-POSA (N = 104) APOC I (N = 37) APOC II (N = 77) APOC III (N = 15) p-Value
, presented as a median [25th, 75th percentiles]; nu = normalized units; AVNN = average of e-to-pulse interval; HFn = normalized spectral power in the high frequency band; LF/HF = uency to high frequency ratio or sympathovagal balance; LFn = normalized spectral power frequency band; OSA = obstructive sleep apnea; POSA = positional obstructive sleep apnea; l signal power; RMSSD = root mean square of the successive differences of the pulse-to-pulse ; SampEnj = sample entropy in the scale τ = j; SDNN = standard deviation of the pulse-toerval; SSET = spectral entropy in the whole spectra; SSEVLF = spectral entropy in the very low y band; SSELF = spectral entropy in the low frequency band; SSEHF = spectral entropy in the uency band; VLFn = normalized spectral power in the very low frequency band. * Significant es in non-POSA vs. APOC I subjects; ⁋ Significant differences in non-POSA vs. APOC II † Significant differences in non-POSA vs. APOC III subjects; ‡ Significant differences in APOC OC II subjects; ╫ Significant differences in APOC I vs. APOC III subjects; ⸸ Significant es in APOC II vs. APOC III subjects.  HF: 0.183, p < 0.01 and 0.197, p < 0.01, respectively) indices. Overall, the SSEVLF Significant differences in non-POSA vs. APOC II subjects; † Significant differences in non-POSA vs. APOC III subjects; ‡ Significant differences in APOC I vs. APOC II subjects; Data are presented as a median [25th, 75th percentiles]; nu = normalized units; AVNN = the pulse-to-pulse interval; HFn = normalized spectral power in the high frequency band low frequency to high frequency ratio or sympathovagal balance; LFn = normalized spec in the low frequency band; OSA = obstructive sleep apnea; POSA = positional obstructive sl PT = total signal power; RMSSD = root mean square of the successive differences of the puls intervals; SampEnj = sample entropy in the scale τ = j; SDNN = standard deviation of th pulse interval; SSET = spectral entropy in the whole spectra; SSEVLF = spectral entropy in th frequency band; SSELF = spectral entropy in the low frequency band; SSEHF = spectral entr high frequency band; VLFn = normalized spectral power in the very low frequency band. * differences in non-POSA vs. APOC I subjects; ⁋ Significant differences in non-POSA vs subjects; † Significant differences in non-POSA vs. APOC III subjects; ‡ Significant difference I vs. APOC II subjects; ╫ Significant differences in APOC I vs. APOC III subjects; ⸸ differences in APOC II vs. APOC III subjects.   Single-entropy values along the MSE curve for low scales (τ ≤ 6) showed moderate but significant correlation with the total AHI (SampEn 4 : 0.235, p < 0.001) and AHI SUP (SampEn 4 : 0.224, p < 0.01), higher than the conventional time domain (AVNN: −0.201, p < 0.01 and −0.151, p < 0.05, respectively) and frequency domain (LF/HF: 0.183, p < 0.01 and 0.197, p < 0.01, respectively) indices. Overall, the SSE VLF yielded the highest significant correlations with the total AHI (0.365, p < 0.001) and AHI SUP (0.350, p < 0.001). No scale from the MSE approach reached significant correlation with the AHI NSUP , whereas only the AVNN and SSE VLF yielded low (−0.164, p < 0.05) and moderate (0.217, p < 0.01) correlation, respectively.

Discussion
A thorough analysis of the overnight PRV signal of positional and non-positional patients was conducted using cardiac modulation indices from different complementary approaches. Particularly, to our knowledge, this is the first study that performed a multiscale nonlinear analysis to characterize changes in nocturnal heart rate modulation due to POSA. Our analyses showed significantly higher complexity in the PRV recordings from POSA patients, compared with subjects without positional influence. Interestingly, conventional time and frequency domain indices were not able to properly characterize these differences between non-POSA and POSA individuals concerning overnight cardiac modulation. On the contrary, multiscale nonlinear analysis captured the influence of positional apneas in nighttime long-term recordings, suggesting significantly higher cardiac imbalance linked to POSA.

The Characteristics of POSA in Our Sample
The prevalence of POSA in our sample (55.4%) was in the lower range of that reported in the literature (56-75%) [3,4,7]. In this regard, it is important to note that all sleep studies were conducted at home, and thus POSA was diagnosed based on ambulatory PSG in the present research. As in-laboratory PSG is known to increase the time sleeping in the supine position, potentially overestimating AHI severity [2,39,40], our study may reflect a more suitable analysis of POSA and its consequences. Moreover, the prevalence of POSA in our study matches that reported in recent works using portable devices for unattended sleep apnea testing at home [2,41]. Regarding the characteristics of positional patients, they showed a slightly lower age (non-significant) and significantly lower BMI, as reported in previous studies [12,42].
Our sample showed a lower prevalence of POSA in the milder patients (37.7% vs. 62.3%) and a higher prevalence in the moderate OSA group (73.5% vs. 26.5%), while no statistical differences were found in the overall severe OSA individuals (52.7% vs. 47.3%). On the contrary, it has been commonly reported that the prevalence of POSA decreases as the severity of OSA increases, using either APOC [4,43] or additional accepted criteria for POSA [2,3,6,7,44]. However, it is important to note that mild OSA was the minority class in our population under study. In addition, our high rate of POSA among severe patients agrees with a recent study focused on the analysis of POSA characteristics in the presence of severe OSA [3] and with similar studies using the APOC criteria [4]. It is also noticeable that the APOC II patients (i.e., patients who would decrease at least one category of severity if positional apneas were removed) were predominant (59.7%) among POSA individuals in our sample, probably due to the high median overall AHI, leading to the aforementioned higher prevalence of severe OSA in the population under study.
In the context of POSA, contradictory data can be found regarding respiratory disturbance indices. In the present study, no significant differences between POSA and non-POSA individuals were found for the total AHI, apnea index, or hypopnea index (Table 1), which suggests that the increased cardiac imbalance in POSA patients is not due to a higher OSA severity degree. This agrees with previous studies [2,19], while others reported statistical differences between positional and non-positional patients regarding the total AHI [3,13]. On the other hand, non-significant differences in terms of the AHI SUP , as well as a significantly lower AHI NSUP , have been consistently reported in the literature for POSA patients compared with non-POSA subjects [3,13,19], as in the present work. Regarding the AHI during REM, Joosten et al. pointed out the potential influence of REM sleep in the generation of obstructive events [45]. Nevertheless, we found a significantly higher AHI during REM in non-POSA subjects, as in the study by Oksenberg et al. [3]. Our findings concerning NPO indices matched those previously reported in the state of the art, showing no differences in terms of oxygen desaturation indices (ODIs), while average and minimum saturations were significantly lower and CT90 was significantly higher in non-positional patients [2,13,19].

MSE and POSA Categories
The MSE profile was similar in both the non-POSA and POSA groups under study (Figure 1), showing gradually increasing entropy values until a region of relative stability was reached. However, there was a marked shift toward higher entropy in the POSA patients compared with the non-POSA subjects. Particularly, the MSE curve of the POSA patients was consistently above that of the non-POSA group in all time scales, thus showing higher complexity. Moreover, higher statistical differences arose in the lower time scales (τ ≤ 6), where the influence of apneic events was higher as the coarse-graining procedure progressively removed the respiratory-related modulation of the heart rate [36]. Regarding the three POSA groups classified according to APOC criteria (Figure 2), a consistent behavior can be observed along the scales, with remarkably lower entropy values for the non-POSA subjects and a trend toward a higher entropy as the APOC group increases in POSA patients. No statistical differences were found among APOC categories I, II, and III. Nevertheless, our analyses showed significant differences in overnight cardiac modulation between the non-POSA patients and the three POSA groups in the smallest scales of the MSE profile (τ ≤ 2). MSE analysis revealed that the overnight cardiac dynamics of patients in APOC categories II and III fit with those of the APOC I group, demonstrating the convenience of categorizing those patients beyond the strictly positional ones (AHI NSUP < 5 events/h) as POSA because they can really benefit from positional therapy. The usefulness of MSE over traditional single-scale entropy measures applied only to the original signal maximizes in the time scale τ = 2, where the largest differences between groups are reached.

PRV Indices and Cardiac Dysfunction
It is admitted that airway obstructions are longer in the supine position, leading to deeper desaturations, longer arousals, and more severe brady-tachycardia changes compared with the lateral position [3,14,45,46]. In addition to the marked alterations in the heart rate, hypoxemia linked to the greater desaturations is related to a higher risk of cardiovascular disease and mortality [47,48]. However, POSA is commonly linked to fewer symptoms and milder disease states due to the lower overall AHI and BMI [6,9,12,13], probably underestimating the impact of positional apneas on cardiovascular regulation. In this regard, few studies investigated the potential influence of position-dependent apneas on cardiac diseases, and contradictory information exists. Favorable cardiovascular outcomes and less cardiovascular comorbidities have been suggested in positional patients [45,49,50]. Similarly, in a recent study by Byun et al. [19], POSA patients showed significantly higher parasympathetic activity (higher SDNN, RMSSD, and HF) than non-positional subjects, which had been related to reduced risk for cardiovascular disease and mortality [51,52]. In contrast, Kulkas et al. [53] reported significantly higher cardiovascular morbidity and mortality in POSA patients compared with non-positional subjects, particularly in the presence of severe OSA. Our analyses showed significantly higher randomness in the nocturnal cardiac modulation of POSA patients compared with non-POSA subjects. Higher irregularity (single-scale entropy) and complexity (multiscale entropy) are both frequently related to the greater adaptability of the autonomous nervous system and thus representative of healthier states [25,54]. However, it is important to note that entropy values are highly dependent on disease mechanisms, the conditions of recording (wakefulness vs. sleep, resting vs. exercise), and the characteristics of the time series (short-vs. long-term). As increased disorderliness of the HRV has been suggested as an independent risk factor for mortality [55], our findings are in line with the study by Kulkas et al., suggesting higher cardiac dysfunction in POSA patients. Similarly, higher entropy measures of HRV time series have been linked to increased diseased states, such as sick sinus syndrome (patient vs. healthy) [54], sleep apnea (OSA positive vs. OSA negative) [28], cardiac abnormalities (atrial fibrillation vs. healthy) [36], and overlap syndrome (COPD + OSA vs. COPD) [29]. In the same regard, Kabbach et al. recently found that COPD patients showed significantly higher variability during acute exacerbation than stable COPD patients [56]. Furthermore, in the same study, significantly higher parasympathetic activity (higher SDNN, RMSSD, and power in HF) was observed in HRV recordings from exacerbated patients, pointing out that in pathological conditions, cardiac autonomic imbalance is not exclusively associated with hyperactive sympathetic and simultaneously hypoactive parasympathetic systems, but also with increased vagal activity [51,56]. Accordingly, the higher parasympathetic activity observed by Byun et al. [19] in the HRV for POSA patients was not necessarily connected to a lower cardiovascular risk.
In the present study, the effect of POSA on cardiac regulation arose when nonlinear analysis was applied, while conventional indices were not able to detect overnight dysregulation. In this regard, it is important to point out that in the study by Byun et al., daytime short-term (5 min) HRV segments were analyzed [19], whereas overnight PRV recordings from long-term sleep studies were assessed in our research. Consequently, the influence of brady-tachycardia events during apneic episodes was actually present in our recordings. In this context, recurrent respiratory events superpose a quasi-periodic biological noise that alters autonomic cardiovascular dynamics [57]. Particularly, severe OSA has been found to introduce rhythmical fluctuations that hide the common modulation of the autonomous nervous system, notably affecting cardiac functioning [58]. Additionally, conventional indices, both in the time and in the frequency domain, are usually computed in short (≤5 min) segments to avoid non-stationary issues [21], while nonlinear MSE analyzes the entire signal, being able to capture the cumulative influence of positional apneas during the whole night beyond local or transient segments.

Limitations and Future Research
Some limitations should be taken into account. Firstly, there is not a standard criterion to define POSA. Several rules exist, and the literature in the context of POSA shows that there are contradictory findings concerning the characteristics and consequences of POSA potentially linked to the different criteria. Therefore, further research is needed to assess additional definitions of POSA in order to ensure the generalizability of our results. Regarding the use of PRV as a surrogate of HRV, recent studies highlighted the usefulness of PRV recordings in the assessment of autonomic cardiac function in different physiological conditions [59,60], while significant differences between both approaches were also reported [61,62] due to additional sources of variability not present in HRV [62]. Accordingly, further research is needed on the reliability of PRV analyses in the context of POSA. In addition, physiopathological mechanisms leading to the higher imbalance in overnight PRV modulation of POSA patients are not clear. The conventional CT90 and minimum saturation variables suggest higher hypoxemia in non-POSA individuals. Hypoxemia has been found to affect heart rate modulation. However, it is unknown whether nocturnal cardiac imbalance is more frequent or severe as the hypoxemia degree increases in sleep apnea patients, as well as the influence of the different OSA phenotypes on such an association. Therefore, novel measures of intermittent hypoxemia, such as the hypoxic burden, could be useful to further explain the effect of position-dependent apneic events on overnight PRV modulation. In the same regard, there is a trend toward a higher number of positional apneas in POSA patients, but a thorough analysis of individual positional apneas (event-based approach) is needed to definitely link POSA to increased cardiac dysfunction. Similarly, exhaustive research is necessary to assess whether the observed overnight imbalance becomes a continuous cardiac dysfunction in the long term. Finally, severe OSA was predominant in our sample, which could limit the generalizability of our results. Accordingly, particular analyses of overall mild and moderate OSA patients with and without positional dependence would be needed to ensure the general validity of our findings.

Conclusions
In summary, the overnight PRV recordings from POSA patients showed significantly higher complexity than those from individuals without positional dependence. This higher disorderliness points to an augmented nocturnal cardiac imbalance due to the cumulative effect of positional apneas during the whole night. Accordingly, our results suggest that POSA should not be categorized as a milder diseased state compared to non-positional sleep apnea. MSE has been found to be useful to characterize changes in nocturnal PRV modulation linked to predominant positional apneas, while conventional time and frequency domain cardiac indices were unable to detect differences between non-POSA and POSA patients.