Heart rate variability analysis in comorbid insomnia and sleep apnea (COMISA)

The present study analyzed 5,335 nocturnal PSGs from adults aged over 40 years, sourced from the Sleep Heart Health Study (SHHS) database (Clinical Trial Number: NCT00005275↗). The SHHS study was carried out according to the Declaration of Helsinki. In all patients, written information was provided, and the study was approved by the ethical review board at each site (https://sleepdata.org/datasets/shhs/files/m/browser/documentation/SHHS1_Protocol.pdf?inline=1↗). SHHS was a multicenter cohort study designed to evaluate cardiovascular and other consequences of sleep-disordered breathing. Detailed descriptions of the study design and methodology have been previously reported²⁷, and data are publicly accessible upon request at https://sleepdata.org/datasets/shhs↗. The original SHHS dataset comprises 5,804 subjects who underwent a baseline clinic visit between November 1995 and January 1998. From those 5,804 initial subjects, 469 participants were excluded due to missing information on insomnia symptoms or issues encountered during the processing stage (details provided below). This resulted in the final sample of 5,335 subjects. At the time of the clinic visit, patients underwent overnight PSG, during which electrocardiogram (ECG) signals were recorded at frequencies of 125–250 Hz. Sleep studies were annotated by medical sleep specialists following a strict standardized procedure²⁸, which subsequently allowed for the extraction of the AHI.

At the time the original SHHS study was conducted, a standardized measure of insomnia was not included. Nevertheless, aligning with previous works that used the SHHS dataset to investigate COMISA^24,25,29, the definition of Insomnia, COMISA, and OSA groups can be established based on current insomnia and COMISA diagnostic criteria⁹. In addition to the PSG studies, participants completed questionnaires regarding insomnia symptoms, sleep habits, and daytime functioning. Based on their answers, insomnia criteria were defined as self-reporting at least one nocturnal symptom more than 16 times per month, accompanied by daytime impairment^24,25,29. Nocturnal symptoms included difficulties initiating sleep, maintaining sleep, and/or waking up too early and being unable to resume sleep. Daytime impairments were characterized as feeling unrested for more than 5 days per month, feeling tired most of the time, and/or having little to no energy over the past month. OSA was defined as subjects with an AHI ≥ 15 e/h^24,25,29. Thus, a subject who met both insomnia and OSA criteria was classified as having COMISA (147 subjects). Subjects meeting only the insomnia criteria or the OSA criteria were placed in the Insomnia (190 subjects) or the OSA group (2,260 subjects), respectively. Subjects who did not meet the insomnia criteria and had an AHI < 15 e/h were included in the No-OSA group (2,738 subjects), which served as the reference population for this study. As explained, these criteria for establishing Insomnia, OSA, and COMISA groups have been employed in several works analyzing the SHHS database^24,25,29. Table 1 presents the demographic and relevant polysomnographic data of the study population.

In this study, we performed HRV analysis derived from ECG signals. As previously mentioned, ECG was acquired during PSG at frequencies of 125–250 Hz. The initial step in HRV analysis involved applying an R-peak detection algorithm based on the Hilbert transform, as proposed by Benitez et al.³⁰, which has been previously utilized in the OSA context^31–33. To address erroneous heartbeat detection, we applied an artifact rejection procedure to remove physiologically implausible peaks using the following criteria^31–33: (i) the duration of R-R intervals was constrained to between 0.33s and 1.5s, and (ii) the maximum allowed difference between two consecutive R-R intervals was 0.66s. By applying these criteria, we removed periods of signal disconnections (few heartbeats detected), as well as noisy segments where false peaks could be introduced, for example, by motion artifacts. This led us to obtain the normal-to-normal (N-N) intervals time series, which estimates the time between sinoatrial node heartbeats. Additionally, to ensure enough representative nocturnal data, we excluded subjects with preprocessed HRV recordings shorter than three hours^28,32.

In the original SHHS study, sleep stages were scored according to the guidelines developed by Rechtschaffen and Kales^28,34. Given the importance of distinguishing between wake and sleep stages in the context of insomnia, each individual sample throughout the entire recordings of each subject was assigned to either wake or sleep. HRV characterization was then conducted for the whole night (WN), as well as separately for wake and sleep stages. To complement our analysis, HRV characterization across non-rapid eye movement (NREM) and rapid eye movement (REM) sleep stages is also provided in the Supplementary Information.

Since the computed features did not fit homoscedasticity nor normality test, we applied the non-parametric Mann-Whitney U-test to evaluate differences between groups and to define specific frequency ranges, as mentioned earlier. The significance threshold for the study was set at p-values < 0.01 after applying FDR correction. As can be observed in the population distribution, there is a clear imbalance between the populations included in the COMISA and Insomnia groups compared to the No-OSA and OSA groups. To assess the influence of this imbalance on the statistical significance of the differences, a permutation test procedure was applied. A detailed description of this methodology is provided in the Supplementary Information, and the statistical results are shown in Supplementary Table S1.

Additionally, a descriptive analysis of the feature distributions for each group was represented using notched boxplots for all features where at least one statistically significant difference between two groups was observed.

The seek for specific frequency ranges resulted in the identification of three frequency bands that exhibited statistical differences between at least two groups. Figure 1 illustrates the results of this statistical analysis (Fig. 1a) and the behavior of the averaged PSDns for each group within the selected ranges (Figs. 1b-d). The first frequency band, spanning 0.0005–0.0045 Hz, corresponds to a range identified in our previous work that reflects macro-sleep structure³². Accordingly, we have renamed this band as BWMS. It can be appreciated that the Insomnia and No-OSA groups show higher activity within BWMS, while it is reduced in OSA and COMISA patients, especially near 0.0005 Hz (Fig. 1b). The second frequency band, covering 0.011–0.049 Hz, captures recurrence patterns between 20 and 90s, consistent with the periodicity of apneic events in adults³⁸. This spectral band corresponds to the OSA frequency range identified in our previous work³², but shifted to the specific apneic recurrence patterns in adults. As expected, activity in this range is higher for OSA and COMISA patients (Fig. 1c), so we have named it BWOSA. The third frequency band, a novel finding of this study, spans 0.071–0.11 Hz. Characterized by reduced activity in the COMISA group (Fig. 1d), we have designated this band as BWCOMISA.

Consequently, the spectral features computed during WN, wake, and sleep stages included RP_VLF, RP_LF, RP_HF, RP_BWMS, RP_BWOSA, RP_BWCOMISA, RP_BWRes, as well as the normalized power in LF (LFn).

Illustrative examples of normalized CWTs computed for a patient from each group (No-OSA, Insomnia, OSA, and COMISA) are depicted in Figs. 2, 3, 4 and 5. Figure 2 illustrates the time-frequency evolution for a No-OSA subject (Fig. 2a), alongside the wake-sleep hypnogram throughout the night (Fig. 2b). It can be observed that there is a higher power concentration at low frequencies (below 0.1 Hz), with additional activity within the HF range (0.15–0.4 Hz), corresponding to normal respiratory activity during sleep. Notably, there are few transitions from sleep to wake after sleep onset.

Figure 3 depicts the case of an Insomnia patient. A notable feature is the elevated sleep onset latency, exceeding two hours, during which power is mainly concentrated at very low frequencies (near 0.0005 Hz, Fig. 3a), corresponding to the BWMS frequency band. Once sleep begins, the power distribution becomes more similar to that of the No-OSA subject, with fewer sleep disruptions, though the disruptions that do occur are longer than those in the No-OSA subject. Additionally, there is evident respiratory activity within the HF range.

Figure 4 shows the overnight time-frequency evolution for an OSA patient. During sleep periods, power is concentrated below 0.05 Hz, corresponding to BWOSA frequencies, and the amount of power across the HF range is markedly lower compared to Insomnia and No-OSA subjects. Additionally, there are more frequent wake-sleep transitions after sleep onset.

Finally, Fig. 5 presents the HRV time-frequency distribution for a COMISA subject (Fig. 5a), along with the corresponding hypnogram (Fig. 5b). It took over an hour for the subject to fall asleep, with power spreading across the entire frequency range. Once sleep begins, power concentration is primarily focused on BWOSA frequencies. Additionally, there are several sleep disruptions, with the patient taking longer to return to sleep compared to the other cases. During these disruptions, power concentration shifts to lower frequencies than those observed during sleep.

Table 2 presents the p-values obtained from the comparisons between HRV features across the WN, representing overall HRV nocturnal behavior, as well as during wake and sleep periods. To complement these results and assess how the differences emerge, the boxplot distributions of the features showing any statistically significant differences across WN, wake, and sleep periods are depicted in Figs. 6, 7 and 8, respectively.

Regarding temporal features across WN, mHR was elevated, apparently due to insomnia effects (Fig. 6a), with statistically significant differences observed between both Insomnia and COMISA patients compared to the No-OSA group. Additionally, SDNNI was higher in the OSA group compared to both the Insomnia and No-OSA groups, but not significantly different from the COMISA population (Table 2; Fig. 6b). When evaluating frequency domain features, the specific HRV features developed in this study exhibited greater differences between groups than the classic HRV features. While RP_BWMS and RP_BWOSA showed the most pronounced differences, RP_VLF and RP_LF did not reach statistical significance between any groups (neither in WN, nor during sleep or wake periods). The differences in RP_BWMS and RP_BWOSA seem to arise as an OSA consequence, with notably reduced activity in BWMS and increased activity within BWOSA for OSA and COMISA groups compared to No-OSA and Insomnia patients (Fig. 6e and f). The No-OSA group displayed higher activity within HF and BWRes (Fig. 6h) bands, as well as reduced LFn compared to the OSA group, and lower activity across BWCOMISA (Fig. 6g) compared to both OSA and COMISA patients.

When examining features during wake, it can be observed that most of the significant differences observed across WN in the frequency domain disappeared, with temporal domain features becoming more relevant during this period. Specifically, reduced SDNNI was observed in the OSA and COMISA groups compared to the No-OSA group, as well as reduced pNN50 in OSA vs. No-OSA (Fig. 7a and c, respectively), while RMSSD was only altered in COMISA patients, showing lower values than the No-OSA group (Fig. 7b). In the frequency domain, the only features that reached statistically significant differences were reduced RP_BWMS and increased RP_BWOSA in the OSA group (but not in the COMISA group) compared to the No-OSA subjects (Fig. 7d and e).

Finally, regarding HRV features during sleep, the significance of elevated overnight mHR in Insomnia was confirmed, with increases observed in both the Insomnia and COMISA groups compared to the No-OSA population. This metric was also increased in the COMISA subjects compared to the OSA group, being the only feature in the study reaching statistically significant difference between these two groups (Table 2; Fig. 8a). SDNNI was also altered, showing lower values in Insomnia and higher values in OSA patients compared to the No-OSA group (Fig. 8b). On the other hand, features in the frequency domain regained relevance during these periods. Similar to WN, the most pronounced differences during sleep were observed for the specific features introduced in this study (RP_BWMS and RP_BWOSA). The same patterns as in WN were seen in RP_HF, LFn, RP_BWMS, RP_BWOSA, and RP_BWRes (Fig. 8e, f and h). The only differentiation in frequency domain features during sleep compared to WN was for RP_BWCOMISA, where the differences between OSA and No-OSA subjects disappeared, leaving the COMISA group as the only population showing altered BWCOMISA activity (Fig. 8g).

The approach followed revealed specific alterations in COMISA patients across the three evaluated periods. To support and contextualize the discussion of the major findings, Table 3 presents the main results that highlight the specific HRV patterns observed in the patients under study. Regarding WN features, the most significant alterations in COMISA patients were found in BWMS and BWOSA activity, thus allowing clear differentiation of their behavior from the HRV of Insomnia or No-OSA patients (see Tables 2 and 3, and Fig. 6). In our previous work, where we first identified ranges similar to BWMS and BWOSA in the context of pediatric OSA, we observed that BWMS activity decreased as OSA severity increased, with activity shifting to BWOSA frequencies, which capture sympathetic excitation due to recurrent apneic events³². Figures 2, 3, 4 and 5 illustrate that during wake periods, power is mainly concentrated around BWMS frequencies in all patient types. However, during sleep, there is a clear differentiation in power concentration between the No-OSA and Insomnia groups compared to OSA and COMISA patients. While individuals with an AHI < 15 e/h maintain power concentration at BWMS frequencies, patients meeting OSA criteria exhibit a shift in power distribution to BWOSA activity, resulting in the marked differences observed in BWMS and BWOSA (Table 2). Additionally, previous studies suggested that, while primary Insomnia patients have difficulty initiating sleep, the COMISA population is more likely to experience difficulties maintaining sleep^9,39. This pattern was also observed in our study (Table 1) as Insomnia patients have higher sleep latency, but COMISA and OSA patients experience greater awakenings and WASO. These differences in macro-sleep structure may also underlie the reduced BWMS activity in OSA and COMISA patients. Collectively, these results suggest that the primary overnight ANS dysfunction in COMISA patients arises as an OSA consequence, manifested as SNS activation during apneic events and sleep disruptions and leading to increased BWOSA and reduced BWMS activity. Furthermore, while reduced BWCOMISA and BWRes activity appear to be characteristic of OSA patients in comparison with the No-OSA group, only BWCOMISA reductions emerge in the COMISA population. Another feature that appears to play an important role in the COMISA population across WN is mHR. Similar to the main conclusions drawn from frequency features, an increased mHR reflects sympathetic excitation³⁵. However, in this case, the differences in mHR, observed when comparing the No-OSA group with the Insomnia and COMISA groups, suggest that the SNS alterations captured by the temporal analysis arise as a consequence of insomnia in the context of COMISA.

Regarding wake periods, the significance of frequency analysis in the COMISA population seems to diminish, as none of the frequency features reached statistically significant differences in this group. Interestingly, it is not the presence of differences but their absence that draws attention. While RP_BWMS and RP_BWOSA, which showed the most differentiation for COMISA patients across WN, did not reach statistically significant differences during wake periods, these features remain altered in the OSA population compared to the No-OSA group. Previous studies have demonstrated that sympathetic activation associated with apneic events persists during waking periods in OSA patients^14,40. This may imply that while the OSA group maintains altered HRV activity in these frequency ranges during wake periods, this is not the case for the COMISA population. Conversely, temporal features gain greater significance during wake periods, particularly when comparing HRV activity of the No-OSA group with that of the OSA and COMISA populations. Autonomic dysfunction during wake, evidenced by reduced SDNNI, appears in both OSA and COMISA groups, while PNS depression, indicated by reduced RMSSD, is only observed in COMISA patients. This finding highlights another specific ANS alteration in the context of COMISA.

Finally, the analysis across sleep periods confirmed the crucial role of mHR in evaluating the effects of insomnia and revealed that frequency analysis is essential for assessing the impact of OSA during these periods. Notably, while mHR did not show statistically significant differences during wakefulness, it emerged as the most relevant temporal feature during sleep, being the only feature across the entire study where differences between OSA and COMISA were observed. This can be attributed to the inherent sympathetic excitation during wakefulness, which increases heart rate in all groups compared to sleep periods⁴¹. However, Insomnia and COMISA patients experienced more wake periods throughout the night, hampering heart rate reduction during sleep, unlike the other groups. This elevated mHR is more pronounced in COMISA patients, as differences with the OSA group did not appear in the Insomnia population. The increased mHR during sleep is consistent with the findings of Spiegelharder et al.⁴², who reported a reduced wake-to-sleep HR decrease in insomnia patients. This phenomenon is attributed to a hyper-arousal state before sleep and the anxiety surrounding poor sleep, which is exacerbated by longer sleep latencies, acting as a significant stressor in individuals with insomnia^42,43. These observations highlight insomnia as a key contributor to the increased mHR in the COMISA population.

Relative to frequency analysis, the conclusions drawn from RP_BWMS and RP_BWOSA across the WN can be extended to sleep periods, underscoring the altered SNS activity due to recurring apneic events as the main ANS dysfunction in the COMISA population. However, a distinction emerges between WN and sleep when evaluating RP_BWCOMISA in OSA and COMISA patients. While both OSA and COMISA showed decreased BWCOMISA activity compared to the No-OSA group during WN, the differences with the OSA group did not reach statistical significance during sleep periods. As detailed in the Supplemental Information, specific impaired PNS modulation during NREM sleep periods in COMISA patients emerges as the main consequence of this effect. BWCOMISA reflects periodicities between 0.071 and 0.11 Hz. This comprises frequencies within LF (0.04–0.15 Hz), a spectral range associated with SNS and PNS activation, whereby parasympathetic activation has a rapid response time, and sympathetic activation occurs more slowly⁴⁴. Accordingly, it can be stated that BWCOMISA (0.071 to 0.11 Hz) reflects part of the periodic PNS modulation of HRV. It is known that during NREM, there is an intrinsic basal PNS activation^45,46. This could explain why No-OSA and OSA groups do not show statistically significant differences in BWCOMISA activity during NREM (see Supplemental Table S2), as OSA patients seem to retain the basal PNS function across this frequency range. In contrast, COMISA patients exhibit persistently reduced PNS activity even during NREM (see Supplemental Table S2 and Supplemental Figure S1), suggesting a deeper autonomic dysfunction. On the other hand, during REM, differences emerged in both OSA and COMISA compared to the No-OSA group (see Supplemental Table S2 and Supplemental Figure S2). However, as can be seen in Table 1, the time spent in NREM is considerably longer than in REM. As a result, when analyzing the complete sleep period, the cumulative effect of PNS impairment captured by BWCOMISA becomes statistically significant only in the COMISA group, distinguishing it as a specific HRV characteristic of this population during sleep.

To summarize, we have demonstrated that the combined effects of insomnia and OSA induce specific ANS dysfunction in the COMISA population, creating a distinct fingerprint that can be evaluated through time and frequency HRV analysis. This fingerprint is characterized by reduced RMSSD during wake periods, heightened mHR during sleep as a consequence of insomnia, and reduced RP_BWMS and RP_BWCOMISA together with increased RP_BWOSA activity during sleep due to OSA effects. These results underscore that the combined impact of both Insomnia and OSA in COMISA patients leads to greater ANS dysfunction than the presence of either sleep disturbance alone.

To the best of our knowledge, this is the first work where nocturnal HRV analysis is performed in the context of COMISA. However, some comparisons can be drawn with the unique previous study that analyzed HRV in the COMISA context during the pre-sleep period, conducted by Ma et al.²⁶. In that study, which also utilized the SHHS dataset, pre-sleep was defined as the interval between lights-off and sleep onset. When comparing their findings with our results during wake, which includes pre-sleep periods, both studies align in showing the highest SDNNI and lowest pNN50 in the No-OSA group, with these trends reversing in COMISA patients (Fig. 6). Similarly, they also observed no significant differences in HRV measures between Insomnia and COMISA patients (Table 2). Methodological differences between the studies, however, limit further comparisons. Since Ma et al.²⁶ did not analyze HRV during sleep, our findings on HRV features in COMISA patients throughout the night cannot be compared with their work or any other existing research. Nevertheless, as stated in the introduction section, ANS dysfunction caused by isolated insomnia and OSA has been extensively studied through HRV analyses, so our results can be discussed in this regard.

Concerning OSA, it is well-established that the cyclical variation of HR due to apneic events leads to ANS dysfunction, characterized by sympathetic predominance and reduced parasympathetic activity during sleep^19,41,47,48. Our study confirms this through the observed reduction in RP_HF and increase in LFn in the OSA population, alongside the rise in RP_BWOSA, which is devoted to capturing the recurrent SNS excitation resulting from apneic events³². Additionally, ANS dysfunction during wakefulness is reported in our findings through the reduced SDNNI in the OSA and COMISA groups, as well as the altered RP_BWMS and RP_BWOSA in OSA patients compared to the No-OSA population, corroborating previous research^40,41,47.

Regarding insomnia, despite more controversy about ANS impairment²¹, certain outcomes consistently appear in the literature. The most persistent finding is the increased mHR compared to healthy population^{21,22,43,49–51}. This increase in mHR due to insomnia effects has been confirmed in our study (see Table 2; Fig. 6a), with mHR being the feature showing the most significant differences between Insomnia and No-OSA groups. Additionally, as previously mentioned, the absence of differences in mHR during wakefulness, combined with increased mHR during sleep, aligns with the study by Spiegelhalder et al.⁴². Bonnet et al.⁴⁹ conducted one of the pioneering studies on HRV alterations due to insomnia, establishing increased SNS activity and a reciprocal reduction of PNS activity as classic alterations in Insomnia patients. Although some studies have corroborated these results to some extent^22,42,43, comprehensive replication has been inconsistent^21,22. In our research, besides mHR, the only feature showing ANS alterations in Insomnia patients compared to the No-OSA group was a reduction in SDNNI during sleep periods. This reduction due to insomnia has also been reported in the literature^43,49.

Certain limitations in this research deserve mention. First, although the overall sample size is substantial, there is a noticeable imbalance among the groups. Most participants belong to the No-OSA (n = 2,738) and OSA (n = 2,260) groups, while the target group, COMISA, comprises the fewest patients (n = 147). Despite the potential effects of group imbalance having been tested through the permutation test procedure followed, future research should aim to include a larger number of subjects, particularly in the Insomnia and COMISA groups, to achieve a more balanced population distribution, thus enhancing the robustness and generalizability of the findings reported here.

Several limitations refer to the diagnosis of insomnia and the information available in the dataset. Despite the criteria followed to establish the different groups aligned with previous research^24,25,29, insomnia diagnosis was based on sleep questionnaires, implying that it was self-reported. Although it is a well-established method⁶, the presence of symptoms such as feeling unrested or having little energy is subject to non-objective impressions. Moreover, those questionnaires were not standardized tools specifically developed for diagnosing insomnia, and the SHHS dataset lacks information on chronic insomnia (i.e., insomnia persisting for over three months)⁴. Additionally, due to the intrinsic limitations of the database, information on diagnosed mental health conditions, such as depression or anxiety, was not available. Furthermore, the population under study has a high incidence of cardiovascular risks. Taken together, all these aspects could further influence HRV activity across the night. Therefore, while the findings reported in this study provide valuable insights, they should be interpreted with caution. Accordingly, the replication in future studies using databases specifically designed to evaluate COMISA, with the criteria based on more objective methods, such as actigraphy or medical evaluation⁶, while addressing the aforementioned limitations, remains essential to increase the generalization of the specific HRV alterations identified in the COMISA population.

Also related to insomnia diagnosis, in the present work, we have combined patients presenting any of the insomnia subtypes (onset insomnia, maintenance insomnia, and early morning awakening insomnia)³⁹. Thus, despite the ANS dysfunction reported in this study in the Insomnia and COMISA populations, the evaluation of specific ANS dysfunction for each insomnia subtype, and its combination with OSA in the COMISA group, should be considered as a further step in future studies.

The analysis conducted in this study has been limited to characterizing HRV alterations over a single overnight ECG recording. Previous studies have demonstrated changes in HRV measures following the application of gold-standard treatments for insomnia^52,53 and OSA^19,41, consisting on cognitive behavioral therapy and continuous positive airway pressure, respectively. However, in the original SHHS study, despite including a follow-up group, no treatments for insomnia or OSA were administered, thus preventing the evaluation of treatment effects using this dataset. Consequently, assessing how treatments for both insomnia and OSA would impact the specific HRV alterations identified in COMISA patients represents a promising future research direction.

Finally, in the presence of respiratory efforts with the glottis closed, such as during the apneic events, the physiological interpretation of frequency ranges such as HF and LF classic frequency bands is hampered. Therefore, although the analysis of these frequency bands in the context of sleep apnea has been extensively performed, caution is needed when interpreting the results reached.