Introduction

Sleep-disordered breathing (SDB) is a common co-morbidity in patients with heart failure [13] and is associated with an increased risk of cardiac events independently of established risk factors [46]. Contrary to findings in other populations, patients with heart failure and SDB most often do not experience subjective symptoms such as daytime sleepiness [7, 8]. Therefore, screening for and diagnosis of SDB has to rely almost exclusively on objective testing [9].

The standard approach for objective assessment of SDB is in-hospital, technician-attended polysomnography (PSG) [10]. The use of this technique, however, is limited by its high cost and limited accessibility; therefore, unattended portable PSG and cardiorespiratory polygraphy are often used in heart failure patients as a lower-cost and more accessible alternative, as evidenced by recent epidemiological [1, 3], clinical [8, 11, 12] and prognostic studies [6, 13].

Typically, cardiorespiratory polygraphs use the same respiratory signals as portable PSG (i.e. nasal/oronasal airflow, chest wall and abdominal movements, oxygen saturation) but do not incorporate EEG, EOG and EMG monitoring. Therefore, they are much cheaper than the portable PSG and allow a marked reduction of both electrode placement time and scoring time. The price to pay for these resource-saving benefits is the lack of sleep staging and arousal scoring. As a consequence, the severity of SDB as quantified by the apnea/hypopnea index (AHI) may be underestimated, since it is computed as an average number of respiratory events per hour of monitoring time [10, 14], instead of per hour of sleep time, as that occurs for portable PSG. Such underestimation might be particularly relevant in patients with heart failure, owing to their reduced sleep efficiency [7, 15]. Moreover, cardiorespiratory polygraphy might further underestimate the AHI due to the inability of scoring hypopneas on the basis of arousal occurrence [16].

Thus, the purpose of this study was to compare the two monitoring systems in order to assess the impact of lack of sleep staging and arousal scoring on the assessment of SDB severity in heart failure patients.

In designing our study, we considered either to let the patients use the two monitoring systems on two different nights or to use them simultaneously during the same night. Both approaches, however, were deemed not to be appropriate because, in the first case, night-to-night changes due to biological, technical and environmental factors would have acted as confounders and, in the second case, because the huge number of electrodes, sensors and lead wires would have been too intrusive for the patient. Therefore, starting from the observation that cardiorespiratory polygraphs use the same sensors to record respiratory signals as portable PSG, we used a single portable polysomnograph during a single night and scored each recording twice, once using all acquired signals (portable PSG mode), and the second time after exclusion of EEG, EOG and EMG channels (cardiorespiratory polygraphy mode).

Methods

Subjects

Seventy-five clinically stable, optimally treated, moderate-to-severe heart failure patients consecutively referred to our heart failure unit for periodic follow-up evaluation were recruited for the study. Exclusion criteria were the following: (1) chronic obstructive pulmonary disease; (2) cardiac surgery, percutaneous coronary intervention, myocardial infarction or unstable angina within 3 months prior to study; (3) history of previous stroke; (4) peripheral or central nervous system disorders; and (5) current use of therapy for obstructive sleep apnea (OSA)/central sleep apnea (CSA). The study was approved by the local ethical committee, and all subjects gave their written informed consent before participation.

Protocol

All patients underwent an unattended, in-hospital nocturnal recording using the portable polysomnograph Embla Titanium™ (Embla, Thornton, CO, USA) while sleeping in their own hospital beds. This device is compliant with recommended technical specifications for PSG recording [16] and portable monitoring [14]. We recorded the following: ECG, thoraco-abdominal movements by uncalibrated inductance plethysmography, nasal airflow by nasal pressure transducer, oronasal airflow by thermistor, oxygen saturation by pulse oximetry with a finger probe, EEG (derivations: O1-M2, O2-M1, C3-M2, C4-M1, F3-M2, F4-M1), EOG (derivations: E1-M2, E2-M1), EMG (chin and anterior tibialis muscles) and body position.

Each recording was scored twice by the same expert scorer, once using all acquired signals (portable PSG mode) and, after ≥2 weeks, using only thoraco-abdominal movements, nasal airflow, oronasal airflow, oxygen saturation and body position (cardiorespiratory polygraphy mode). Thus, except for intra-scorer variability, we obtained the same SDB scoring that the scorer would have performed if a cardiorespiratory polygraph (e.g. Embletta; Embla, Thornton, CO, USA) had been used.

Signal analysis

Only studies fulfilling the following criteria were included in the analysis: (1) a minimum of 4 h of scorable EEG data collected on at least three channels (one frontal, one central and one occipital) and at least one scorable central EEG channel collected for the entire recording period and (2) at least nasal pressure + two inductance plethysmography channels + pulse oximetry collected for the entire recording period. Scoring of sleep and respiratory events was performed according to the American Academy of Sleep Medicine guidelines [16], using the Somnologica software version 5.1.0 (Embla, Thornton, CO, USA). The start time and end time of both scoring procedures were set at 2300 and 0600 hours, these being, respectively, the time at which lights are turned off in the ward and the time of first daily administration of diuretics by nurses (lights on).

Out of the two criteria proposed by the American Academy of Sleep Medicine for scoring a hypopnea, we chose that requiring a drop in the peak-to-peak nasal pressure by ≥50 % of baseline, lasting ≥10 s and associated with a ≥3 % desaturation or an arousal [16]. Since arousals could only be detected by portable PSG, this choice allowed us to provide “worst case” results. Following a common practice, we classified hypopneas as central or obstructive if, respectively, (1) thoraco-abdominal movements were in-phase and there was no evidence of airflow limitation on nasal pressure or snoring and (2) thoraco-abdominal movements were out-of-phase and/or there was evidence of airflow limitation on nasal pressure or snoring [6, 17, 18]. Mixed events were scored when in the initial portion of the event the inspiratory effort met the criteria for central events, while in the second portion met the criteria for obstructive events.

Intra-scorer reliability for the identification of respiratory events was previously assessed in a random sample of 20 patients [19]. The intra-class correlation coefficient (ICC) was ≥0.99 for all SDB indexes, indicating excellent intra-scorer reliability.

Quantification of SDB severity

The severity of SDB was quantified using standard measures expressed both in absolute and relative units. Absolute measures included the total number of respiratory events and the number of central and obstructive events. Relative measures included the AHI, central AHI and obstructive AHI. According to the two scoring modalities, namely, portable PSG mode and cardiorespiratory polygraphy mode, all indexes were computed relative to sleep time and monitoring time, respectively. Monitoring time was defined as recording time minus movement time and standing position time.

Absolute measures of SDB severity obtained from cardiorespiratory polygraphy were also recomputed during wake time periods. To this purpose, we developed a dedicated software that merged sleep scoring from portable PSG analysis with respiratory event scoring from cardiorespiratory polygraphy analysis.

Quantification of SDB severity was also carried out categorising the AHI according to the standard cut-offs of the following: (1) AHI <5/h (no SDB), (2) 5/h ≤ AHI < 15/h (mild SDB), (3) 15/h ≤ AHI < 30/h (moderate SDB) and (4) AHI ≥ 30/h (severe SDB) [20]. Moreover, following a common practice, the AHI was also dichotomised according to the cut-offs of the following: ≥5/h (clinically significant SDB [1, 4]) and ≥15/h (moderate-to-severe SDB [1, 5]). Patients with an AHI of ≥5/h were further grouped into those with dominant CSA if >50 % of respiratory events were central or mixed in nature, and those with dominant OSA if obstructive events were ≥50 %.

Statistical methods

The agreement between continuous measures of SDB severity was assessed by Bland–Altman analysis [21]. Since between-method differences showed a non-normal distribution, the systematic bias (i.e. constant offset) and limits of agreement between portable PSG and cardiorespiratory polygraphy were estimated by computing, respectively, the median and the 5th and 95th percentiles of the differences (non-parametric approach) [21]. The latter represent the range within which most differences are expected to lie. Pairwise comparisons were performed using the Wilcoxon signed-rank test. A p value of <0.05 was considered statistically significant, and all tests were two-sided.

The agreement in the classification of SDB severity according to categorised AHI was assessed by computing the percentage of patients concordantly classified by the two monitoring systems and by computing the Cohen's kappa coefficient of agreement with 95 % confidence interval.

Descriptive statistics are reported as median (lower quartile, upper quartile) and mean ± SD for, respectively, non-normally and normally distributed data. All statistical analyses were carried out using the SAS/STAT statistical package, release 9.2 (SAS Institute, Inc., Cary, NC, USA).

Results

Out of the 75 overnight recordings carried out in the study, eight had to be excluded due to failure to meet the criteria for signal quality in respiratory (six recordings) or EEG signals (two recordings). This led to a final sample of 67 patients. Their demographic and clinical characteristics are summarised in Table 1. The time elapsed between the two scoring procedures (portable PSG mode and cardiorespiratory polygraphy mode) was (median (lower quartile, upper quartile)) 64 (31, 150) days. As reported in Table 2, the assessment of SDB severity according to cardiorespiratory polygraphy was based on the analysis of a period of the night that was 24 ± 18 % larger than that analysed using portable PSG.

Table 1 Patients' demographic and clinical characteristics
Table 2 Descriptive statistics for sleep time, monitoring time and sleep architecture

Bland–Altman plots for absolute and relative measures of SDB severity are shown in Fig. 1. Descriptive and agreement statistics are reported in Table 3. The total number of respiratory events detected using cardiorespiratory analysis showed a systematic positive bias relative to portable PSG. The extra events were almost entirely of central origin. The number of arousal-related hypopneas detected using portable PSG was very low.

Fig. 1
figure 1

Bland–Altman plots of the difference (Δ) between cardiorespiratory polygraphy and portable PSG against the mean of the two measurements, in the quantification of SDB severity expressed both in absolute and relative units. Solid lines and dashed lines represent, respectively, the systematic bias between the two methods (estimated by the median of the differences) and the 90 % limits of agreement (estimated by the 5th and 95th percentiles of the differences). From top to bottom, a total number of respiratory events (apneas and hypopneas), b total number of central events, c total number of obstructive events, d apnea/hypopnea index (AHI), e central AHI, f obstructive AHI

Table 3 Descriptive and agreement statistics for absolute and relative measures of sleep-disordered breathing severity according to portable polysomnography (PSG) and cardiorespiratory polygraphy

The AHI estimated by cardiorespiratory polygraphy showed a negligible negative bias relative to portable PSG (Table 3, 5th–7th row). As shown by limits of agreement, the distribution of between-method differences was skewed towards negative values.

Tables 4 and 5 show the relationship between portable PSG and cardiorespiratory polygraphy in the classification of patients according to standard cut-offs of the AHI. Using the cut-offs of AHI < 5/h, 5/h ≤ AHI < 15/h, 15/h ≤ AHI < 30/h and AHI ≥ 30/h, 58 patients (87 %) were concordantly classified by the two systems with a weighted kappa statistic of 0.89 (95 % confidence interval (CI) 0.82, 0.96). Of note, disagreement always occurred between contiguous classes. Using the cut-offs of AHI ≥ 5/h and AHI ≥ 15/h, 97 and 94 % of the patients were concordantly classified by the two systems, with a kappa statistic of 0.90 (95 % CI 0.76, 1.0) and 0.88 (95 % CI 0.77, 0.99), respectively. The dominant form of SDB was concordantly identified by the two monitoring systems in 100 % of the patients.

Table 4 Relationship between portable PSG and cardiorespiratory polygraphy in the classification of patients according to the severity of SDB based on standard cut-offs of the AHI: (1) no SDB (AHI < 5/h), (2) mild (5/h ≤ AHI < 15/h), moderate (15/h ≤ AHI < 30/h) and (4) severe (AHI ≥ 30/h)
Table 5 Relationship between portable PSG and cardiorespiratory polygraphy in the classification of SDB severity according to dichotomised AHI and dominant SDB (dominant SDB has been computed only in patients with an AHI of ≥5/h (see “Methods”))

When absolute measures of SDB severity derived from cardiorespiratory polygraphy were recomputed during wake time periods, we found that the total number of respiratory events was (median (lower quartile, upper quartile)) 9 (3,18) events. During the same period, the number of central events was 6 (1, 14), while the number of obstructive events was 0 (0, 1). Comparing these results with the forth column of Table 3 (lines 1–3), it clearly appears that the number of extra events detected by cardiorespiratory polygraphy was almost equal to the number of events detected by the same system during wake time.

Discussion

Portable PSG and cardiorespiratory polygraphy are increasingly being used in the assessment of SDB in heart failure patients. Due to the lack of EEG, EOG and EMG monitoring, scoring of SDB using cardiorespiratory polygraphy may potentially produce different results as compared to portable PSG. In this study, we have addressed this issue by comparing the two scoring methods. Our results indicate that (1) a larger number of respiratory events are detected using cardiorespiratory polygraphy, the extra events being, for the vast majority, central in origin; (2) the AHI estimated by cardiorespiratory polygraphy has a negligible negative bias relative to portable PSG; and (3) there is good agreement between the two approaches in the classification of SDB severity according to standard cut-offs of the AHI.

Agreement between monitoring systems

The total number of respiratory events detected by cardiorespiratory polygraphy showed a systematic positive bias with respect to portable PSG and a high upper limit of agreement, indicating that cardiorespiratory polygraphy discloses the occurrence in some patients of a large number of apneas and hypopneas that go unnoticed using the other system. An interesting finding of this study was that the vast majority of extra events detected by cardiorespiratory polygraphy were of central origin and occurred during wake time. This is not surprising since Cheyne–Stokes respiration may also occur during wakefulness [22, 23] due to mechanisms which are not state specific [24].

Although in other patient populations replacing sleep time with the larger monitoring time in the denominator of the AHI is known to lead to a reduction of the AHI [10, 14], in our study, this effect was mostly offset by the simultaneous increase of the numerator, namely, the total number of respiratory events. As a result, the bias between cardiorespiratory polygraphy and portable PSG was negligible, and a good agreement was found between the two monitoring systems when the patients were classified according to standard cut-offs of the AHI. Indeed, not only were about 90 % of the patients concordantly classified by the two systems, but misclassification occurred between contiguous classes, which is an almost unavoidable problem of any classification system.

In a previous study on the night-to-night repeatability of indices of sleep apnea severity measured by the same cardiorespiratory polygraph in the same population of patients, we found much larger limits of agreement and significantly lower classification agreement between the two nights than those found in the present study between the two monitoring systems [19]. In another similar study, Vazir and co-workers [25] reported changes in the AHI through four consecutive nights that were often larger than the largest differences between portable PSG and cardiorespiratory polygraphy observed in our study. Similar findings were also observed in patients with OSA using standard PSG [26]. Therefore, the agreement between the two monitoring systems appears to be much better than the agreement between two consecutive nights using the same type of monitor.

Methodological considerations

All recordings for this study were performed in the hospital setting. Patients were sleeping in their bed, and being the ward a rehabilitation ward, the rooms were quiet during the night. Recordings were usually taken 4–5 days after hospital admission; therefore, patients were accustomed to the sleeping accommodation and environment. Nurses did not intervene to optimise signal quality during nocturnal recordings. Duration of sleep was similar to that previously observed in the home setting [8]. Thus, it is unlikely that this hospital setting would have influenced the results of the recordings as compared to the home setting.

We excluded two patients due to failure to meet quality criteria for EEG signals. Since the occurrence of signal loss or high-level noise in EEG channels was unrelated to patient's clinical characteristics and to the breathing pattern, this exclusion did not cause selection bias in the results from cardiorespiratory monitors.

Study limitations

There is not yet sufficient evidence to guide the use of portable monitors—both PSG and cardiorespiratory—in asymptomatic individuals such as heart failure patients [14]. However, due to the need of screening all patients, portable monitoring represents the only feasible approach in the majority of cardiology departments involved in the evaluation and treatment of SDB. Most departments use cardiorespiratory polygraphs, while others use portable PSG monitors or both. Hence, knowing whether the two monitoring systems provide comparable results in the assessment of SDB is of clinical and practical relevance. Moreover, several investigations on epidemiological, clinical and prognostic aspects of sleep-disordered breathing in patients with heart failure have been based on either portable PSG or cardiorespiratory polygraphy in the absence of any previous comparison between the different approaches [1, 3, 6, 8, 1113]. Therefore, knowing the relationship between the data obtained from portable PSG and data obtained from cardiorespiratory polygraphy is necessary to interpret and compare published studies. Thus, we believe that our study addresses a knowledge gap which is relevant to both clinicians and researchers involved in the assessment of SDB in heart failure patients.

The observed differences between the two scoring procedures were unavoidably affected by intra-scorer variability. In order to reduce as much as possible this unwanted source of data variation, scoring of all recordings was blindly performed by an expert scorer with excellent intra-scorer reliability [19]. However, we cannot exclude that better/worse results (i.e. reduction/increase of limits of agreement) would have been obtained if another scorer with lower/higher intra-scorer variability had analysed the sleep studies.

The start time and end time of the scoring procedure were set at 2300 and 0600 hours, these being, respectively, the time at which lights are turned off in the ward and the time of first daily administration of diuretics by nurses (lights on). It might be argued that this strict time period is unlikely to be representative of how portable monitors are used at home. However, since the time 0600 hours was determined by the medical prescription for taking diuretics, this was presumably also the “lights-on” time of patients at home. As for 2300 hours, it is reasonable to think that this time was close to the actual “lights-off” time of patients at home because (1) patients were not old (mean age, 59 ± 8 years), and (2) no patient had acute or advanced heart failure and was therefore in a seriously ill condition requiring special hospital care.

According to current guidelines for scoring respiratory events, classification of a hypopnea as central, obstructive or mixed should not be performed without a quantitative assessment of ventilatory effort by oesophageal manometry, calibrated inductance plethysmography or diaphragmatic/intercostal EMG [16]. Yet, due to the obvious difficulties involved in the use of these techniques, almost all studies carried out so far on SDB in heart failure patients, including clinical trials on SDB treatment [17], have used this definition to discriminate between the two types of hypopnea [6, 18].

Conclusions

In summary, the results of this study show that, in patients with heart failure, measurements of the AHI obtained from cardiorespiratory polygraphy have a negligible negative bias with respect to portable PSG, and the variability of the differences is lower than that expected due to night-to-night variation. This translates into a high degree of agreement between the two approaches in the classification of SDB severity. These findings suggest that cardiorespiratory polygraphy may be used as an alternative to portable PSG in the assessment of SDB in heart failure patients.