Sleep and Breathing

, Volume 14, Issue 2, pp 109–114

Diagnostic performance of single airflow channel recording (ApneaLink) in home diagnosis of sleep apnea

Authors

    • Ruhrlandklinik, Department of Respiratory Medicine, Faculty of MedicineUniversity of Duisburg-Essen
  • Yi Wang
    • Ruhrlandklinik, Department of Respiratory Medicine, Faculty of MedicineUniversity of Duisburg-Essen
  • Gerhard Weinreich
    • Ruhrlandklinik, Department of Respiratory Medicine, Faculty of MedicineUniversity of Duisburg-Essen
  • Helmut Teschler
    • Ruhrlandklinik, Department of Respiratory Medicine, Faculty of MedicineUniversity of Duisburg-Essen
    • Graduate School of Biomedical EngineeringUniversity of New South Wales
Original Article

DOI: 10.1007/s11325-009-0290-2

Cite this article as:
Ragette, R., Wang, Y., Weinreich, G. et al. Sleep Breath (2010) 14: 109. doi:10.1007/s11325-009-0290-2

Abstract

Background

ApneaLink is a novel single-channel screening device for sleep apnea detection which is based on pressure-transduced measurement of oronasal airflow, summarised as respiratory disturbance index per hour of recording time (RDIApneaLink). We tested ApneaLink's diagnostic performance in a patient population with high prevalence of sleep apnea.

Methods

ApneaLink was applied simultaneously with in-laboratory polysomnography (PSG) (n = 102, 24 female, age 54.7 years) and sequentially with PSG in the unattended home setting (n = 131, 37 female, age 59.1 years). Predictive values were computed for apnea-defining thresholds of apnea–hypopnea index (AHI) ≥ 5/h, ≥10/h, ≥15/h. Night-to-night variability (NNV) was assessed over three consecutive nights (n = 55, 10 female, age 48.9 years).

Results

RDIApneaLink correlated well with apnea–hypopnea index on PSG (PSGAHI) on simultaneous (r = 0.98, bias −0.7) and unattended home application (r = 0.95, bias −0.6). Predictive values were highest at AHI ≥ 10/h (in-laboratory: sensitivity 91.1%, specificity 87.5%, LR+ (positive likelihood ratio) 7.4, LR− 0.1; home: sensitivity 80%, specificity 85.5%, LR+5.5, LR− 0.2). NNV was low \( \left( {V = 0.58 \pm 0.44,{\text{range}}\,0 - 1.69} \right) \).

Conclusion

ApneaLink is an accurate screening tool for sleep apnea in a population with high prevalence of the disorder.

Keywords

ApneaLinkSleep-disordered breathingSleep apneaObstructive sleep apneaScreening

Introduction

Sleep apnea is common in men and women over the age of 30 years [1]. It causes non-restorative sleep and excessive daytime sleepiness [2]. Because of its association with hypertension [3, 4], heart disease [5, 6], stroke [7], congestive heart failure [8] and accident risk [9], sleep apnea is of considerable public health concern [10]. Diagnosis, therefore, is critical. Standard approach is in-laboratory, technician-attended polysomnography (PSG) [11], which monitors sleep, respiration, cardiac rhythm, snoring, limb movements and body position. PSG, however, is labour-intensive and limited in its availability. Many patients with sleep apnea go undiagnosed [12]. In search of alternative modes of diagnosis, portable monitors have been developed [13]. They allow for unattended recording of a limited number of biosignals in the patient's home, thereby increase or reduce the likelihood of sleep apnea and facilitate triage for definitive PSG. Portable monitoring has proved useful in the detection of sleep apnea in the attended setting [14, 15]. Utility in the unattended setting, particularly with use of less than four biosignals, remains unsettled [16, 17].

Recently, a single-channel recording device using a pressure-transduced airflow sensor (ApneaLink) was introduced. It demonstrated excellent agreement with PSG in the experimental setting [18]. Clinical application in a random diabetic outpatient population, however, showed reliable identification of sleep apnea in the moderate to severe range (apnea–hypopnea index (AHI) ≥ 15/h) only [19].

We studied ApneaLink in a sleep laboratory patient cohort with high pre-test probability for sleep apnea. Our aim was (1) to expand on experimental pilot data, including sources of recording error, (2) to assess impact of night-to-night variability, and (3) to assess predictive performance in the unattended home setting.

Methods

Study design and subjects

The study was conducted in three parts and with three sets of patients. Part I (July to October 2003) compared ApneaLink with simultaneous PSG in the in-laboratory, technician-attended setting. One hundred and two patients were recruited consecutively from the sleep lab (Table 1). Part II (January 2005 to February 2006) applied ApneaLink in the unattended home setting and compared it with sequential in-laboratory PSG, obtained within 8 weeks. One hundred and thirty-one patients were recruited from the sleep centre's day clinic (Table 1) following presentation with a variety of sleep-related complaints and comorbid conditions. Part III (January 2005 to May 2006) tested night-to-night variability of ApneaLink recordings on three consecutive nights in patients randomly recruited from a respiratory rehabilitation centre (Table 1). Patients with pre-existing diagnosis of sleep apnea, symptomatic cardio-respiratory disease, home oxygen or home ventilation therapy were generally excluded. An institutional review board approved the study, and subjects provided informed consent prior to participation.
Table 1

Demographic data for study parts I and II and III

Demographic characteristics

Results (part I)

Results (part II)

Results (part III)

Number of patients

102

131

55

Age (years)

54.7 ± 13.3

59.1 ± 14.6

48.9 ± 10.4

Sex, M/F

78/24

95/36

45/10

BMI (kg/m2)

29.5 ± 5.1

28.0 ± 4.4

29.2 ± 4.9

Hypertension

46

57

15

Coronary heart disease

9

9

4

Congestive heart failure

3

10

1

Chronic obstructive bronchitis

15

41

6

Asthma

6

21

18

Diabetes mellitus

7

6

2

Materials

Standard overnight polysomnography was performed in the sleep laboratory using 12-channel technique (two electroencephalograms (EEG), two electroocculograms, one submental EEG, bilateral leg electrodes, ECG, nasal pressure canula, thoracic and abdominal inductive plethysmography, snoring sensor) and digitalised recording equipment (Embla, Flaga, Reykjavik, Island). Respiratory events were counted per hour of sleep and summarised as the apnea–hypopnea index (PSGAHI). Data were scored manually according to published guidelines [20, 21].

ApneaLink transmits oronasal airflow via a nasal canula (adult nasal pressure canula; 21/53 cm, Medcare, Munich) to a differential pressure transducer attached to the front of the patient's chest. Flow measurements are digitalised for storage and downloaded to a computer at a later time. Analysis of the flow signal is fully automated but may be reviewed and rescored by the clinician. Sequences of poor recording quality are automatically removed from analysis (9.8% in present study). Recordings were not edited in the present study. Default settings defined apneas as a decline in airflow of >90% for 10–100 s, hypopneas as a decline in airflow by 50–90% for 10–100 s. Apneas and hypopneas were counted per hour of recording time and summarised as the respiratory disturbance index (RDI). Snoring and inspiratory airflow limitations were not analysed.

Statistical analysis

Agreement between ApneaLink-generated RDI and PSG-generated AHI was evaluated using Spearman's rank-order correlation analysis, Bland–Altman plots of mean differences and calculations of sensitivities, specificities, positive predictive values (PPV), negative predictive values (NPV) and positive and negative likelihood ratios (LR). Receiver operating characteristics curves were computed to establish optimal performance at sleep apnea-defining thresholds of PSGAHI ≥ 5/h, ≥10/h, and ≥15/h. ApneaLink variability was calculated as follows: individual RDI range was divided by the mean of the three study nights. Differences in RDI were assessed using the Kruskal–Wallis test. Data are presented as mean ± SD. Statistical significance was accepted at p < 0.05.

Results

ApneaLink compared with simultaneous PSG (part I)

PSG (total sleep time 331 ± 86 min) identified sleep apnea in 82 of 102 (80.4%), 62 of 102 (60.8%) and 50 of 102 (49.0%) patients using defining thresholds of AHI ≥ 5/h, ≥10/h and ≥15/h, respectively. ApneaLink identified sleep apnea in 87 of 102 (85.3%), 62 of 102 (60.8%) and 52 of 102 (51.0%) patients using defining thresholds of RDI ≥ 5/h, ≥10/h and ≥15/h. Correlation analysis and Bland–Altman plots (Fig. 1a–c) showed excellent agreement between AHIPSG and RDIApneaLink (total recording time, 366 ± 97 min). Systematic bias was negligible (−0.7). Analysis of individual apnea and hypopnea indices demonstrated overestimation of hypopneas (r = 0.72, bias = −2.6/h) and underestimation of apneas (r = 0.96, bias = 2.7/h). Predictive values for sleep apnea were computed using thresholds of AHI ≥ 5/h, ≥10/h and ≥15/h (Table 2).
https://static-content.springer.com/image/art%3A10.1007%2Fs11325-009-0290-2/MediaObjects/11325_2009_290_Fig1_HTML.gif
Fig. 1

Comparison of ApneaLink and PSG in simultaneous recordings. a Correlation analysis between AHIPSG and RDIApneaLink. b Enlargement of area AHI/RDI = 0–20/h. c Bland–Altman plot

Table 2

Comparison of ApneaLink and PSG in simultaneous recordings

AHI (1/h)

Sensitivity (%)

Specificity (%)

PPV (%)

NPV (%)

LR+

LR−

5

93.9

50.0

88.5

66.7

1.9

0.1

10

91.9

87.5

91.9

87.5

7.4

0.1

15

92.0

88.5

88.5

92.0

8.0

0.1

Predictive values for sleep apnea at different AHI thresholds

PPV positive predictive value, NPV negative predictive value, LR+ positive likelihood ratio, LR− negative likelihood ratio

ApneaLink home study compared with sequential PSG (part II)

PSG (total sleep time 312 ± 74 min) identified sleep apnea in 98of 131 (72.4%), 70 of 131 (53.0%) and 52 of 131 (39.4%) patients using thresholds of AHI ≥ 5/h, ≥10/h and ≥15/h, respectively. ApneaLink identified sleep apnea in 98 of 131 (72.4%), 65 of 131 (49.6%) and 50 of 131 (38.2%) patients using defining thresholds of RDI ≥ 5/h, ≥10/h and ≥15/h. PSG compared well with ApneaLink (total recording time 377 ± 111 min) with regard to total RDI vs. AHI (Fig. 2a–c) but less with regard to apnea and hypopnea indices. Analysis of individual indices demonstrated systematic overestimation of hypopneas (r = 0.58, bias = 1.2/h) and underestimation of apneas (r = 0.75, bias = −1.4/h). ApneaLink misclassified as false-positive three patients with periodic leg movements (RDI 13.0 ± 4.6/h vs. AHI 1.7 ± 1.2/h). It was negative in three patients with supine apnea identified on PSG (RDI 5.7 ± 3.1/h vs. AHI 36.0 ± 10.1/h, threshold AHI ≥ 10/h). Predictive values, summarised for different sleep apnea-defining thresholds, were highest at AHI ≥ 10/h (Table 3).
https://static-content.springer.com/image/art%3A10.1007%2Fs11325-009-0290-2/MediaObjects/11325_2009_290_Fig2_HTML.gif
Fig. 2

Comparison of home ApneaLink versus sequential PSG recordings. a Correlation analysis between AHIPSG und RDIApneaLink. b Enlargement of area AHI/RDI = 0–20/h. c Bland–Altman plot

Table 3

Comparison of unattended home ApneaLink to in-lab PSG

AHI/RDI (1/h)

Sensitivity (%)

Specificity (%)

PPV (%)

NPV (%)

LR+

LR−

5

91.8

76.5

91.8

76.5

3.9

0.1

10

80.0

85.5

86.2

79.1

5.5

0.2

15

73.1

84.7

76.0

82.5

4.8

0.3

Predictive values for sleep apnea at different AHI thresholds

PPV positive predictive value, NPV negative predictive value, LR+ positive likelihood ratio, LR− negative likelihood ratio

Night-to-night variability of ApneaLink home study (part III)

Thirty-seven of 55 (67.3%) patients had recordings consistent with sleep apnea (RDI ≥ 5/h). Findings were mild (RDI 5–15/h) in 18 patients (32.8%), mild–moderate (RDI 15–30/h) in 15 (27.3%) and severe (RDI < 30/h) in four patients (7.3%). Total RDI was 13.1 ± 13.3/h on the first, 11.0 ± 9.1/h on the second and 11.4 ± 10.0/h on the third night (p non-significant). Discordant results (findings consistent with sleep apnea on one night but normal on two nights) were found in 13 of 55 patients (23.6%) at threshold RDI ≥ 5/h, in 11 of 55 patients (20.0%) at threshold RDI ≥ 10/h, and in eight of 55 patients (14.5%) at RDI ≥ 15/h. Night-to-night variability (Fig. 3) was quantified by the variability factor (V), a measure of fluctuation of RDI around the mean. V was 0.58 ± 0.44 (range 0–1.69) for the group as a whole and did not differ among severity subgroups.
https://static-content.springer.com/image/art%3A10.1007%2Fs11325-009-0290-2/MediaObjects/11325_2009_290_Fig3_HTML.gif
Fig. 3

Night-to-night variability of ApneaLink recordings on three consecutive nights

Discussion

The present study showed that ApneaLink accurately identified sleep apnea in patients with high pre-test probability for the disorder. Diagnostic performance, when compared with PSG, was excellent in the attended in-laboratory and good in the unattended home setting. At defining thresholds of AHI ≥ 10/h, the device correctly identified sleep apnea over the wide range of AHI and minimised false-positive results. Our study, therefore, confirms and improves upon the clinical performance data previously published [18, 19].

Using predictive criteria of sensitivity and specificity >80%, Erman et al. [19] in a randomly selected diabetic outpatient population had found acceptable predictive performance in the moderate to severe disease range only (AHI > 15/h). Our study, at comparable prevalence rates for sleep apnea, demonstrated excellent predictive performance at AHI 5–10/h, the mild to moderate disease range. The difference is likely explained on the basis of detailed, physician-directed and individualised instructioning offered to patient and partner (including supervised device montage and trial recording, 10–15 min) in the present study, which resulted in good quality recordings, so that no study was repeated or eliminated from analysis. Conceivably, periodic leg movements, which are common in the diabetic patients, may also have contributed to a higher rate of false-positive ApneaLink studies in the Erman study (discussed below).

ApneaLink's improved diagnostic accuracy is principally based on two technical adaptions that enhance sensitivity and specificity of respiratory event recognition: use of pressure-transduced oronasal airflow sensors identical to those used in polysomnography equipment, allowing for linear flow registration across the flow range [21], and integration into the automated analysis of an algorithm for elimination of poor quality recording sequences which reduces the amount of erroneous recording time. Together, these adaptions lead to near-identical flow and event registration between PSG and ApneaLink, as demonstrated in the experimental setting [18]. A small trend towards overreading of hypopneas or misclassification of apneas as hypopneas is well described with pressure canulas [22] and did not affect composite RDI. Visual confirmation showed event misclassification to be most likely at points of abrupt changes of signal amplitude, as seen with sighs, arousals or limb movements. As such, periodic leg movements, because of the associated increase in respiratory amplitude, were misclassified as hypopneas in the three cases involved. Duplicate event registration was also possible, particularly when signal amplitudes fluctuated more than 20% during prolonged hypopneas or following periods of low signal amplitude. Event underscoring, by contrast, was rare with the device's algorithm for elimination of poor signal sequences.

Despite the improvements in event recognition and classification, diagnostic performance was not as good in the unattended home as in the sleep laboratory. Sensitivity and negative predictive value were primarily affected. They dropped approximately 10% at AHI ≥ 10/h, consistent with lower case-finding capacity in the home. The finding is not unexpected given the well-recognised differences in sleep quality between home and sleep laboratory, in time spent in supine position, in amount of REM sleep and in recording beyond sleep time [23].

Night-to-night variability in the manifestation of sleep apnea is another well-recognised factor affecting concordance between nonsynchronous recordings [24, 25]. Our study showed fluctuations in RDI of approximately 50% around the mean on recordings of three consecutive nights. Such variability in disease manifestation will most significantly affect case recognition in the mild to moderate disease range and underscores the need for definitive (PSG) studies in the setting of a negative ApneaLink study but with clinical findings suspicious of sleep apnea or in the case of a mildly positive ApneaLink study but lacking clinical findings consistent with the disorder.

At predictive values above 80%, ApneaLink out-performed conventional type IV single or dual [26, 27] and many conventional type III multichannel recording devices [28, 29] already in use for screening purposes. Screening utility of ApneaLink was emphasised by its ease of use (no recording had to be repeated) and low cost (~30 Euro/night). Together, they facilitate device application in a variety of clinical settings in which polysomnography is unavailable, impractical or unaffordable. Given the excellent agreement between ApneaLink and PSG in moderate to severe sleep apnea, initiation of therapy based on a screening study that is complemented by consistent clinical finding seems justified in these cases and in the context of exploding health care costs. As with most devices, screening accuracy was better in terms of case finding (high sensitivity and high positive predictive value) than case elimination (high specificity and negative predictive value). Confirmation with PSG, therefore, continues to be important when screening results show mild sleep apnea, are discordant with clinical presentation or are derived from study populations with a low prevalence of sleep apnea.

Copyright information

© Springer-Verlag 2009