screening for obstructive sleep apnea by cyclic...

Screening for Obstructive Sleep Apnea by Cyclic Variationof Heart Rate

Junichiro Hayano, MD; Eiichi Watanabe, MD; Yuji Saito, MD; Fumihiko Sasaki, MD;Keisaku Fujimoto, MD; Tetsuo Nomiyama, MD; Kiyohiro Kawai, MS;

Itsuo Kodama, MD; Hiroki Sakakibara, MD

Background—Despite the adverse cardiovascular consequences of obstructive sleep apnea, the majority of patients remainundiagnosed. To explore an efficient ECG-based screening tool for obstructive sleep apnea, we examined the usefulnessof automated detection of cyclic variation of heart rate (CVHR) in a large-scale controlled clinical setting.

Methods and Results—We developed an algorithm of autocorrelated wave detection with adaptive threshold (ACAT). Thealgorithm was optimized with 63 sleep studies in a training cohort, and its performance was confirmed with 70 sleepstudies of the Physionet Apnea-ECG database. We then applied the algorithm to ECGs extracted from all-nightpolysomnograms in 862 consecutive subjects referred for diagnostic sleep study. The number of CVHR per hour (theCVHR index) closely correlated (r�0.84) with the apnea-hypopnea index, although the absolute agreement with theapnea-hypopnea index was modest (the upper and lower limits of agreement, 21 per hour and �19 per hour) withperiodic leg movement causing most of the disagreement (P�0.001). The CVHR index showed a good performance inidentifying the patients with an apnea-hypopnea index �15 per hour (area under the receiver-operating characteristiccurve, 0.913; 83% sensitivity and 88% specificity, with the predetermined cutoff threshold of CVHR index �15 perhour). The classification performance was unaffected by older age (�65 years) or cardiac autonomic dysfunction (SDof normal-to-normal R-R intervals over the entire length of recording �65 ms; area under the receiver-operatingcharacteristic curve, 0.915 and 0.911, respectively).

Conclusions—The automated detection of CVHR with the ACAT algorithm provides a powerful ECG-based screeningtool for moderate-to-severe obstructive sleep apnea, even in older subjects and in those with cardiac autonomicdysfunction. (Circ Arrhythm Electrophysiol. 2011;4:64-72.)

Key Words: diagnosis � electrocardiography � heart rate � sleep apnea

Obstructive sleep apnea (OSA) is a syndrome characterizedby repeated partial or complete obstruction of the upper

airway during sleep, resulting in intermittent hypoxia and tran-sient repetitive arousals from sleep. Recent prospective studieshave reported that OSA is associated with an increased risk offatal and nonfatal cardiovascular events, and this risk may bereduced by continuous positive airway pressure treatment.1–3 Inpatients with OSA, the repeated apneic episodes elicit increasedsympathetic activation,4,5 surge in blood pressure,4,5 and cardiacarrhythmias.6–10 Concordant with this, the circadian distributionof sudden cardiac death among patients with OSA has a peakduring the sleeping hours.11 OSA is also associated with hyper-coagulability, vascular oxidative stress, systemic inflammation,and endothelial dysfunction.12 These facts indicate that diagnosisof OSA is essential in cardiovascular clinical practice. Definite

diagnosis of OSA requires all-night laboratory polysomno-graphic examination, which is hard to perform in many patientsbecause of its substantial cost and inconvenience and thepotential risk of serious cardiovascular events during monitor-ing. The introduction of simple and efficient screening methodsis necessary to identify patients with a high probability of OSA.

Clinical Perspective on p 72For this purpose, ambulatory ECG monitoring during sleep

may be promising. Episodes of OSA are accompanied by acharacteristic heart rate pattern, known as cyclic variation ofheart rate (CVHR), which consists of bradycardia duringapnea followed by abrupt tachycardia on its cessation.13

Several earlier studies have demonstrated that this pattern canbe used to detect OSA and suggested that the analysis ofHolter ECG during sleep may be used as a screening tool for

Received June 4, 2010; accepted October 25, 2010.From the Department of Medical Education (J.H.), Nagoya City University Graduate School of Medical Sciences, Nagoya, Japan; the Division of

Cardiology (E.W.), Department of Internal Medicine and the Department of Respiratory Medicine (Y.S., H.S.), Fujita Health University School ofMedicine, Toyoake, Japan; Takaoka Clinic (F.S.), Nagoya Japan; the Department of Biomedical Laboratory Sciences (K.F.), Shinshu University Schoolof Health Sciences and the Department of Preventive Medicine and Public Health (T.N.), Shinshu University School of Medicine, Matsumoto, Japan;Suzuken Company Limited (K.K.), Nagoya, Japan; and Nagoya University (I.K.), Nagoya, Japan.

Correspondence to Junichiro Hayano, MD, Department of Medical Education, Nagoya City University Graduate School of Medical Sciences, 1Kawasumi Mizuho-cho Mizuho-ku, Nagoya 467-8601, Japan. E-mail [email protected]

© 2011 American Heart Association, Inc.

Circ Arrhythm Electrophysiol is available at http://circep.ahajournals.org DOI: 10.1161/CIRCEP.110.958009

64

by guest on July 10, 2018http://circep.ahajournals.org/

Dow

nloaded from

http://circep.ahajournals.org/

OSA.14–17 However, most of these earlier studies were basedon observations in a limited number of subjects (�150),consisting of typical OSA patients and normal subjects, andtheir primary outcome was classification performance be-tween them. The CVHR has been reported to decrease amongpatients with autonomic dysfunction.13 Heart rate variationssimilar to CVHR also accompany periodic leg movements(PLMs) during sleep.18,19 Moreover, classification perfor-mance depends on the definition of OSA (ie, cutoff level forseverity) and the proportion of patients (pretest probability).To elucidate the usefulness of ECG-based screening for OSA,the relationship between CVHR and OSA severity must bestudied quantitatively in controlled clinical settings.

In the present study, we examined whether the automatedCVHR detection with ECG during sleep provides a usefulmarker for OSA screening. We developed a new automatedalgorithm, an autocorrelated wave detection with adaptivethreshold (ACAT), for detecting individual CVHR events.After training the ACAT algorithm in independent samples,we applied the algorithm to ECG recordings during sleep in1193 consecutive subjects referred for a diagnostic polysom-nographic examination. We examined the correlation andagreement between the apnea-hypopnea index (AHI) and theCVHR as well as the ability of ACAT to identify patientswith moderate-to-severe OSA.

MethodsThe protocol of the present study was approved by the InstitutionalReview Board of the Fujita Health University, Toyoake, Aichi,Japan, and the Ethics Review Committee of the Nagoya CityUniversity Graduate School of Medical Sciences, Nagoya, Japan.

SubjectsWe studied the all-night polysomnographic recordings of 1193consecutive subjects (the study cohort) referred to the Sleep Labo-ratory of the Fujita Health University Hospital between January 2005and December 2008 for diagnostic evaluation of suspectedsleep-disordered breathing. Subjects were excluded if they (1)were �16 year of age, (2) had an implanted pacemaker, or (3) hadpersistent atrial fibrillation. Data were also excluded if the totallength of analyzable ECG in the polysomnographic recording was�360 minutes.

We used 2 additional databases obtained from independent all-night polysomnograms. To refine our newly developed automatedCVHR detection algorithm (ACAT), we used polysomnograms in 63subjects (the training cohort). They were randomly selected from 364(287 male and 77 female) subjects who were examined at the SleepLaboratory of the Fujita Health University Hospital between January2004 and December 2004 for a diagnostic evaluation of suspectedsleep-disordered breathing. From the polysomnograms, a databasewas generated by extracting ECG, oxyhemoglobin saturation bypulse oximetry (SpO2), and oronasal airflow signals. The trainingcohort consisted of 55 men and 8 women between 18 and 86 yearsof age, with a body mass index of 26�6 (19 to 46) kg/m2 and an AHIof 26�22 (0 to 91) per hour. All subjects were in sinus rhythm andfree from PLM, and the length of recording was 454 to 596 minutes.

The other database we used was the open-access Physionet SleepApnea-ECG database (http://www.physionet.org/physiobank/database/apnea-ecg/), with which the methodological performance of theACAT algorithm was tested. This database was generated for theComputers in Cardiology Challenge 2000.14 It consists of 70single-lead ECG signals extracted from polysomnograms, with alength of 401 to 578 minutes, digitized at a sampling frequency of100 Hz. Each recording has the annotation for each minute as tothe presence or absence of sleep apnea during that minute. The

subjects are 57 men and 13 women between 27 and 63 years ofage, with a body mass index of 28�6 (19 to 45) kg/m2 and an AHIof 28�28 (0 to 91) per hour.

Polysomnographic DataThe polysomnographic recordings of the study and training cohortswere obtained using the Alice 3 and Alice 4 Diagnostic Sleep System(Philips Respironics, The Netherlands). The sleep stages, respiratoryevents, and PLM were scored using the standard diagnostic criteriaof the American Academy of Sleep Medicine20 and a publishedcriterion21 by registered polysomnogram technicians at the SleepLaboratory of the Fujita Health University Hospital. Briefly, apneaand hypopnea were defined as the cessation or reduction of respira-tion for �10 seconds caused by either airway obstruction (obstruc-tive apnea) or lack of respiratory effort (central apnea) that wasassociated with at least 1 of the following: (1) �50% airflowreduction, (2) �3% oxygen desaturation, and (3) subsequent arousal.The AHI was determined for each polysomnographic recording asthe mean number of apneas and hypopneas per hour of time in bed(TIB). The PLMs were defined as a series of �4 distinct legmovements, with an individual length of 0.5 to 5 seconds and anintermovement interval of 4 to 90 seconds, detected from the legelectromyographic recording. PLMs with a close temporal relation toapneas or hypopneas were not interpreted as PLMs. The PLM indexwas determined as the mean number of PLMs per hour of TIB.

In the present study, subjects with an AHI �15 per hour wereconsidered to have moderate-to-severe OSA and those with a PLMindex �10 per hour were considered to have significant PLM.

ECG AnalysisThe single-lead digital ECG signals were extracted from the polysom-nograms at a sampling frequency of 100 Hz. The ECG signals werescanned on a personal computer using a customized beat-detection andbeat-classification algorithm that identified all QRS complexes andlabeled the beats as normal, ventricular ectopic, supraventricular ec-topic, and artifact. The results were reviewed, and all errors in beatdetection and classification were corrected interactively on the computerscreen by expert technicians of Holter ECG, who were blinded to thesubjects’ polysomnographic diagnosis and other characteristics.

To evaluate the cardiac autonomic nervous function, the SD ofnormal-to-normal R-R intervals over the entire length of recording(SDNN) was calculated. In this study, subjects with an SDNN �60ms were considered to have cardiac autonomic dysfunction. For thepurpose of CVHR detection, the R-R interval time series wereinterpolated with a horizontal-step function using only N-N intervalsand resampled at 2 Hz.

Automated CVHR DetectionFor the automated CVHR detection, we developed a new algorithmby the method of ACAT (Patent application No. 2010–51387,Japan). This algorithm has been incorporated into a Holter ECGscanner (Cardy Analyzer, Suzuken Co, Ltd, Nagoya, Japan).

The ACAT algorithm is a time-domain method that uses onlyinterbeat interval data. The algorithm detects the CVHR as cyclicand autocorrelated dips in smoothed interbeat interval time series anddetermines the temporal position of the individual dips comprisingthe CVHR (Figure 1 and Figure 2). The processes of the ACATalgorithm were as follows: Interbeat interval time series weresmoothed by second-order polynomial fitting, and all dips in thesmoothed trend with widths between 10 and 120 seconds anddepth-to-width ratios of �0.7 ms/s were detected. Also, the upperand lower envelopes of the interbeat interval variations were calcu-lated as the 95th and 5th percentile points, respectively, within asifting window with a width of 130 seconds. Then, the dips that metthe following criteria were considered CVHR: (1) a depth �40% ofthe envelope range at that point (adaptive threshold), (2) interdipintervals (cycle length) between 25 and 130 seconds, (3) a wave formsimilar to those of the 2 preceding and 2 subsequent dips with a meanmorphological correlation coefficients �0.4 (autocorrelated wave),

Hayano et al Automated ECG Detection of Obstructive Sleep Apnea 65


Dow

nloaded from


and (4) 3 cycle lengths between 4 consecutive dips that meet thefollowing equivalence criteria:

�3�2l1/s)(3�2l2/s)(3�2l3/s)�0.8,

where l1, l2, and l3 are 3 consecutive cycle lengths ands�(l1�l2�l3)/3.

In the present study, we counted the number of CVHR as thenumber of dips comprising the CVHR and calculated the CVHRindex as the mean number of CVHRs per hour of TIB.

ProtocolThe parameters of the ACAT algorithm were optimized using thedata from the training cohort so that the best correlation and

agreement were obtained between the AHI and the CVHR index.The cutoff threshold for the CVHR index was also determined in thetraining cohort as that corresponding with the highest average ofsensitivity and specificity to identify subjects with moderate-to-severe OSA. The cutoff threshold was used as the predeterminedcutoff threshold in the analysis that followed. We then applied theACAT algorithm to the Physionet Apnea-ECG database to test themethodological performance of the ACAT algorithm and tocompare it with the ECG-based OSA detection algorithms re-ported in earlier studies. The optimized ACAT algorithm wasthen used to detect CVHR in the study cohort. The CVHRdetection was performed with the automated ACAT algorithm bya technician who was blinded to the subjects’ polysomnographicdiagnosis and other characteristics.

Figure 1. Algorithm of autocorrelated wave detec-tion with adaptive threshold. The algorithm detectsthe temporal positions of CVHR in the interbeatinterval time series as the cyclic and autocorre-lated dips that meet 4 specific criteria. See text fordetails.

66 Circ Arrhythm Electrophysiol February 2011


Dow

nloaded from


Statistical AnalysisThe correlation and agreement between the AHI and the CVHRindex were evaluated with the Pearson product-moment correlationcoefficient and Bland and Altman limits of agreement (95% confi-dence limits of overestimations and underestimations),22 respec-tively. The latter was calculated as the mean difference �1.96 SD ofthe differences between the AHI and the CVHR index. The signif-icance of the difference in the levels of agreement between groupswas evaluated by comparing the SD of the differences with an F test.The classification performance of the CVHR index in identifying thesubjects with moderate-to-severe OSA was evaluated by receiver-

operating characteristic (ROC) curve analysis. The significance ofthe difference in classification performance between groups wasevaluated by comparing the areas under the ROC curve (AUC) withthe Hanley and McNeil method.23 The classification performancewith a cutoff threshold was evaluated according to its sensitivity,specificity, positive predictive accuracy, and negative predictiveaccuracy. We considered a probability value �0.05 significant.

ResultsOptimization of ACAT Algorithm in theTraining CohortFigure 2 shows the strips of the R-R interval, SpO2, respira-tion (oronasal airflow), and the positions of CVHRs detectedby the ACAT algorithm in 3 representative subjects withOSA in the training cohort. The ACAT algorithm reliablydetected the temporal positions of individual CVHRs. Itshowed robustness to the changes in the amplitude andfrequency of the CVHRs (Figure 2A), provided good detec-tion even when the CVHR amplitude was very low (Figure2B), and offered good discriminatory power from the physi-ological very-low-frequency component of heart rate vari-ability (Figure 2C).

Figure 3 shows the relationship between the AHI and theCVHR index in the training cohort. The CVHR index closelycorrelated with AHI (r�0.95, P�0.001) and the Bland andAltman plot showed that the upper and lower limits of theagreement were 13 and �13 per hour, respectively. The ROCcurve of the CVHR index for detecting subjects withmoderate-to-severe OSA had an AUC of 0.913 (standarderror, 0.011). The ROC curve analysis also indicated that thehighest average of sensitivity (85%) and specificity (81%)was obtained with a cutoff threshold of CVHR index �15 perhour.

Detection Performance in the PhysionetApnea-ECG DatabaseAmong the 70 subjects of the Physionet Apnea-ECG data-base, the correlation coefficients between the AHI and theCVHR index was 0.91, and the upper and lower limits ofagreement were 22 and �23, respectively. The AUC of theROC curve analysis for the performance in identifying thesubjects with moderate-to-severe OSA was 0.979 (standarderror, 0.0127). With the cutoff threshold of CVHR index �15per hour, the sensitivity, specificity, positive predictive accu-racy, and negative predictive accuracy were 90%, 100%,100%, and 88%, respectively. The presence or absence ofCVHR in each minute agreed with the minute-by-minuteannotation for sleep apnea for 83% of the total minutes.

Application to the Study CohortOf 1193 eligible subjects, 319 (27%) were excluded becauseof poor ECG signal quality or an insufficient length (�360minutes) of analyzable ECG. An additional 12 subjects wereexcluded because of atrial fibrillation during the polysomno-graphic recordings. Consequently, 862 subjects were in-cluded in the study cohort. Their characteristics are shown inTable 1. The 331 excluded subjects did not differ signifi-cantly from the 862 included subjects in age (48�16 year),sex (female, 22%), body mass index (27�6 kg/m2), TIB

Figure 2. Performance of the ACAT algorithm in detecting vari-ous types of CVHR. Graphs showing the time series of R-Rinterval (RRI), oxygen saturation of pulse oximetry (SpO2),oronasal air flow (Resp), and the temporal positions of CVHRdetected by the ACAT algorithm in 3 representative subjectswith CVHR showing the changes in amplitude and frequency(A), CVHR with a very low amplitude (B), and CVHR existingwith the physiological very-low-frequency component of heartrate variability (C).



Dow

nloaded from


(494�25 minutes), AHI (15�20 per hour), or PLM index(5�15 per hour).

The CHVR index obtained by the ACAT algorithm closelycorrelated with the AHI in the study cohort (r�0.84,P�0.001), although the absolute agreement was modest(upper and lower limits of agreement, 21 and �19 per hour,respectively; Figure 4A). The ROC curve analysis, however,showed a good performance in identifying the subjects withmoderate-to-severe OSA with an AUC of 0.913 (standard error,0.0108) and indicated that the predetermined cutoff threshold ofCVHR index �15 per hour was useful (Figure 5).

To determine the factors affecting the relationship andagreement between the AHI and the CVHR index and theclassification performance, we compared subgroups of sub-jects (Tables 2 and 3). Although the correlation coefficientwas lower in the older subjects (age �65 years) than in theyounger subjects (P�0.001), neither the limits of agreementnor the classification performance were affected. The corre-lation, agreement, and classification performance were unaf-fected by the presence of cardiac autonomic dysfunction

(SDNN �65 ms). The presence or coexistence of centralsleep apnea (central sleep apnea index �1 per hour) slightlyincreased the limits of agreement (P�0.001), toward anoverestimation of AHI, although it did not reduce the classi-fication performance. Among the subjects with a PLM index�10 per hour, the correlation coefficient was decreased, thelimit of agreement was increased toward an overestimation ofAHI and the classification performance was reduced(P�0.001 for all).

PLM episodes were frequently associated with the cyclicchanges in heart rate, many of which were detected by theACAT algorithm as CVHR (Figure 6). Among 150 (17%)subjects with a PLM index �10 per hour, 67 (45%) had nosignificant OSA (AHI �5 per hour). In these patients withPLM alone, 5158 episodes of CVHR were detected. Theseepisodes were shorter in duration (dip width; 30�11 seconds)and in cycle length (44�16 seconds) than the 66 760 episodesof CVHR (36�12 and 54�17 seconds, respectively) ob-served in 227 subjects with increased OSA (AHI �15 perhour) and no significant PLM (PLM index �5 per hour;P�0.0001 for both). Their distributions, however, overlappedconsiderably (Figure 7). Concordant with this, the removal ofsubjects with a PLM index �10 per hour (n�150) resulted ina substantial improvement in the AHI estimation with theCVHR index (Figure 4B).

DiscussionThis is the first study to examine the ability of ECG-basedautomated CVHR detection to estimate the presence andseverity of OSA in a large-scale clinical setting. Among the862 subjects referred for a polysomnographic examination,we observed that the CVHR index obtained by the ACATalgorithm closely correlated with the AHI (r�0.84) andshowed a good performance in identifying patients withmoderate-to-severe OSA. Although the correlation betweenthe AHI and the CVHR index was reduced in older subjects(�65 years), the classification performance was maintainedeven in older subjects (�65 years) and in subjects with

Figure 3. Scatter graph with the regression line and Bland andAltman plot for the relationship between the AHI and the CVHRindex obtained with the optimized ACAT algorithm in the train-ing cohort. Horizontal solid line and dashed lines in the Blandand Altman plot indicate the mean difference and the upper andlower limits of agreement, respectively.

Table 1. Characteristics of Subjects in the Study Cohort

n 862

Age, y 49�15 (16–83)

Female, n (%) 154 (18)

BMI, kg/m2 27�5 (16–47)

TIB, min 493�24 (418–564)

AHI, per hour 15�19 (0–110)

AHI �15 per hour, n (%) 280 (33)

4%ODI, per hour 13�18 (0–104)

Minimum SpO2, % 86�5 (54–93)

PLM index, per hour 6�15 (0–98)

PLM index �10 per hour, n (%) 150 (17)

Heart rate, bpm 62�9 (45–89)

SDNN, ms 85�27 (26–206)

SDNN �65 ms, n (%) 206 (24)

Data represent n (%) for categorical variables and mean�SD (range) forcontinuous variables.

BMI indicates body mass index; ODI, oxygen desaturation index.



Dow

nloaded from


cardiac autonomic dysfunction (SDNN �60 ms). We ob-served, however, that the absolute agreement between theAHI and the CVHR index was modest and that the presenceof PLM during sleep resulted in an overestimation of AHI.Given the adverse consequences of OSA, developing a simpleand efficient screening tool for OSA is an urgent issue. Ourobservations are important for determining the usefulness andlimitations of ECG-based screening for OSA.

Recently, several authors have developed algorithms forthe automated ECG detection of OSA. Khandoker et al15 useda machine learning technique for automated recognition ofOSA from wavelet analysis of R-R intervals and ECG-derived respiratory signal. They developed a classificationalgorithm using 83 training sets of sleep studies and appliedit to 42 test studies selected from the Physionet Apnea-ECGdatabase. The algorithm correctly recognized 24 of 26 OSA

subjects and 15 of 16 non-OSA subjects. Mendez et al16

compared the empirical mode decomposition and the waveletanalysis for detecting OSA from the ECG signal. Using 25training sets and 25 test sets of sleep studies generated fromthe Physionet Apnea-ECG database, they reported 85%accuracy for the empirical mode decomposition and 89%accuracy for wavelet analysis in classifying minute-by-minute apnea/nonapnea periods; both methods perfectly dis-criminated OSA patients from normal subjects. In the present

Figure 4. Scatter graphs with the regres-sion line and Bland and Altman plot forthe relationship between the AHI and theCVHR index in the study cohort. A andC, Results obtained from the total sub-jects (n�862); B and D, results obtainedfrom subjects without PLM (PLM index�10 per hour; n�712).

Figure 5. Plot of cumulative frequencies for subjects with andwithout moderate-to-severe OSA (AHI �15 per hour) versus cri-terion values for the CVHR index in the study cohort. Dottedlines indicate ranges of the 95% confidence interval. Verticaldashed line indicates the predetermined cutoff threshold. NPA,negative predictive accuracy; PPA, positive predictive accuracy.

Table 2. Correlation and Agreement Between the AHI and theCVHR Index in the Subjects Grouped by Age and the Presence ofCardiac Autonomic Dysfunction, Central Sleep Apnea, and PLM

n RSD of

Difference*

Limits ofAgreement†

Upper Lower

Age �65 years 717 0.84 10 21 �19

Age �65 years 145 0.76‡ 11 25 �19

SDNN �65 ms 656 0.84 11 23 �19

SDNN �65 ms 206 0.81 10 20 �17

CAI �1 per hour 817 0.82 10 22 �18

CAI �1 per hour 45 0.88 12§ 25 �20

PLM index �10per hour

712 0.86 9 18 �17


150 0.64‡ 11§ 27 �15

CAI indicates central apnea index.*SD of the differences between the CVHR index and the AHI.†Upper and lower limits of agreement of Bland and Altman plot (mean

difference�1.96 SD of the differences).‡Correlation coefficient significantly different from that of the rest of the

subjects (P�0.001).§SD significantly different from that of the rest of the subjects, determined

with F test (P�0.001).



Dow

nloaded from


study, we observed comparable performances for the inde-pendently optimized ACAT algorithm in the PhysionetApnea-ECG database. Only a few studies, however, havebeen performed in clinical settings. Roche et al17 proposed anincrease in the relative power of very-low-frequency compo-nent (0.01 to 0.05 Hz) of interbeat interval increment(%VLFI) as a marker for OSA. Among a sample of 150patients referred to a university hospital for clinically sus-pected OSA, the authors reported an AUC of 0.70 foridentifying the patients with an AHI �15 per hour (n�100)and 64% sensitivity and 69% specificity with using %VLFI�4% as the cutoff threshold.

The ACAT algorithm is unique among reported algo-rithms. Unlike the other algorithms that report OSA as thesegment of time (typically 1 minute) that includes apnea/hypopnea episode(s), the ACAT algorithm provides thetemporal position of each CVHR episode. This feature of the

ACAT algorithm allowed us to predict the severity of OSAdirectly from the CVHR index. Moreover, the ACAT algo-rithm furnishes a time-local data adaptability with the localwidth of the envelopes of interbeat interval fluctuations as thereference for determining the amplitude threshold for CVHR

Table 3. Classification Performance of the CVHR Index for Detecting Moderate-to-SevereSleep Apnea in the Subjects Grouped by Age and the Presence of Cardiac AutonomicDysfunction, Central Sleep Apnea, and PLM

NAHI �15 perHour, n (%)

AUC of ROCCurve (SE) Sensitivity* % Specificity* % PPA* % NPA* %

Age �65 years 717 236 (33) 0.914 (0.0114) 82 88 77 91

Age �65 years 145 44 (30) 0.915 (0.0298) 91 86 74 96

SDNN �65 ms 656 228 (35) 0.912 (0.0121) 83 87 77 91

SDNN �65 ms 206 52 (25) 0.911 (0.0262) 83 91 75 94

CAI �1 per hour 817 255 (31) 0.908 (0.0115) 83 88 76 92

CAI �1 per hour 45 25 (56) 0.969 (0.0201) 88 90 92 86


712 238 (33) 0.936 (0.0105) 83 94 88 92


150 42 (28) 0.799 (0.0390)† 86 61 46 92

CAI indicates central apnea index; NPA, negative predictive accuracy; PPA, positive predictive accuracy; and SE,standard error.

*Performance in detecting subjects with AHI �15 per hour with the predetermined cutoff threshold of CVHR index�15 per hour.

†AUC significantly different from that for PLM index �10 per hour (P�0.001).

Figure 6. CVHR accompanying PLM. Graphs showing the timeseries of R-R interval (RRI), oxygen saturation of pulse oximetry(SpO2), oronasal air flow (Resp), tibial electromyogram (EMGleg),and the temporal positions of CVHR detected by the ACATalgorithm in a representative subject with PLM. Respiratorytrace and SpO2 show no apnea-hypopnea, whereas RRI showscyclic variations.

Figure 7. Distributions of the duration (dip width; A) and cyclelength (B) of CVHR. PLM alone accounts for data for 5158 epi-sodes of CVHR detected in 67 subjects with PLM index �10per hour and AHI �5 per hour. OSA alone accounts for data for66 760 episodes of CVHR detected in 227 subjects with AHI�15 per hour and PLM index �5 per hour.



Dow

nloaded from


detection. This feature may contribute to the good classifica-tion performance in older (�65 years) subjects and those withcardiac autonomic dysfunction. These factors have beenreported to be associated with decreased CVHR amplitude13

and consequently, increased failure of detection.24

The Bland and Altman plot revealed that the absoluteagreement between the AHI and the CVHR index wasmodest, with a tendency to overestimate in patients withcentral apnea and in those with PLM. The CVHRs accompa-nying the sleep apneas driven by respiratory control dysfunc-tion are known to show highly predictable and almostmetronomic patterns of oscillations, whereas those accompa-nying sleep apneas driven by anatomic dysfunction of theupper airway show more variable and irregular patterns ofoscillations.25–27 Because of these features, the episodes ofcentral sleep apneas may be detected more consistently withthe ACAT algorithm than pure OSA episodes; this may haveled to the tendency to overestimate in subjects with centralapnea. In contrast, PLM episodes have been known to beaccompanied by autonomic activations and heart rate changesconsistent with CVHR.18,19 Although the CVHR associatedwith PLM has been reported to have a briefer duration, amore triangular shape, and a shorter cycle length than thoseassociated with OSA,24 our observations suggest that both theduration (dip width) and the cycle length of CVHR associatedwith PLM and that associated with OSA may overlapconsiderably (Figure 7). To maintain the sufficient sensitivityto the CVHR associated with OSA, the optimized ACATalgorithm detected a substantial part of the CVHR associatedwith PLM, which may presents a possible methodologicalimprovement for future studies.

Future studies must also address 2 important issues. First,in the present study, we studied subjects who were referred tothe sleep laboratory of the university hospital for diagnosticpolysomnographic examination. The performance of algo-rithm when applied to different populations, particularlythose with a low pretest probability of OSA, is unclear. Giventhe needs for the OSA screening in general practice andhealth promotion, studies in the general population seemimportant. Second, we studied the ECG signal extracted frompolysomnographic recordings obtained in the sleep labora-tory. OSA detection may have been affected by the recordingenvironment. To confirm the clinical usefulness of themethod, appropriately controlled diagnostic studies compar-ing the CVHR index obtained with ambulatory ECG moni-toring and the results of polysomnographic examination arealso needed.

OSA is associated with increased risk for cardiovascularmorbidity and mortality,1,2 and it may relate to importantcardiovascular pathophysiologic mechanisms, including sym-pathetic activation,4,5 hypercoagulability, vascular oxidativestress, systemic inflammation, and endothelial dysfunction.12

Despite the adverse cardiovascular consequences of OSA, themajority of patients remain undiagnosed.28 The present studyindicates that automated CVHR detection with the ACATalgorithm provides a powerful ECG marker for screeningmoderate-to-severe OSA even in older subjects and thosewith cardiac autonomic dysfunction. Although the challengeof handling PLM remains, our observations suggest the

possibility that routine ambulatory ECG monitoring may beused as a screening tool for OSA by incorporating theautomated algorithm into Holter scanners.

Sources of FundingThis research was supported by a Grant-in-Aid for ScientificResearch (C) from the Japan Society for the Promotion of Science(JSPS), Japan and by a Research Grant (20B-7) for Nervous andMental Disorders from the Ministry of Health, Labor and Welfare,Japan.

DisclosuresDr Hayano has applied for a patent on the ACAT algorithm (No.2010–51387, Japan), which has been licensed to Suzuken CompanyLimited. K. Kawai is an employee of Suzuken Company Limited. DrKodama is an advisor to Suzuken Company Limited.

References1. Marin JM, Carrizo SJ, Vicente E, Agusti AG. Long-term cardiovascular

outcomes in men with obstructive sleep apnoea-hypopnoea with orwithout treatment with continuous positive airway pressure: an observa-tional study. Lancet. 2005;365:1046–1053.

2. Young T, Finn L, Peppard PE, Szklo-Coxe M, Austin D, Nieto FJ,Stubbs R, Hla KM. Sleep disordered breathing and mortality:eighteen-year follow-up of the Wisconsin sleep cohort. Sleep. 2008;31:1071–1078.

3. Cassar A, Morgenthaler TI, Lennon RJ, Rihal CS, Lerman A. Treatmentof obstructive sleep apnea is associated with decreased cardiac death afterpercutaneous coronary intervention. J Am Coll Cardiol. 2007;50:1310–1314.

4. Leuenberger U, Jacob E, Sweer L, Waravdekar N, Zwillich C, SinowayL. Surges of muscle sympathetic nerve activity during obstructive apneaare linked to hypoxemia. J Appl Physiol. 1995;79:581–588.

5. Somers VK, Dyken ME, Clary MP, Abboud FM. Sympathetic neuralmechanisms in obstructive sleep apnea. J Clin Invest. 1995;96:1897–1904.

6. Guilleminault C, Connolly SJ, Winkle RA. Cardiac arrhythmia and con-duction disturbances during sleep in 400 patients with sleep apneasyndrome. Am J Cardiol. 1983;52:490–494.

7. Shepard JW Jr, Garrison MW, Grither DA, Dolan GF. Relationship ofventricular ectopy to oxyhemoglobin desaturation in patients withobstructive sleep apnea. Chest. 1985;88:335–340.

8. Fichter J, Bauer D, Arampatzis S, Fries R, Heisel A, Sybrecht GW.Sleep-related breathing disorders are associated with ventricular arrhyth-mias in patients with an implantable cardioverter-defibrillator. Chest.2002;122:558–561.

9. Mehra R, Benjamin EJ, Shahar E, Gottlieb DJ, Nawabit R, Kirchner HL,Sahadevan J, Redline S. Association of nocturnal arrhythmias with sleep-disordered breathing: the Sleep Heart Health Study. Am J Respir CritCare Med. 2006;173:910–916.

10. Monahan K, Storfer-Isser A, Mehra R, Shahar E, Mittleman M, RottmanJ, Punjabi N, Sanders M, Quan SF, Resnick H, Redline S. Triggering ofnocturnal arrhythmias by sleep-disordered breathing events. J Am CollCardiol. 2009;54:1797–1804.

11. Gami AS, Howard DE, Olson EJ, Somers VK. Day-night pattern ofsudden death in obstructive sleep apnea. N Engl J Med. 2005;352:1206–1214.

12. Shamsuzzaman AS, Gersh BJ, Somers VK. Obstructive sleep apnea:implications for cardiac and vascular disease. JAMA. 2003;290:1906–1914.

13. Guilleminault C, Connolly S, Winkle R, Melvin K, Tilkian A. Cyclicalvariation of the heart rate in sleep apnoea syndrome: mechanisms, andusefulness of 24 h electrocardiography as a screening technique Lancet.1984;1:126–131.

14. Penzel T, McNames J, Murray A, de Chazal P, Moody G, Raymond B.Systematic comparison of different algorithms for apnoea detection basedon electrocardiogram recordings. Med Biol Eng Comput. 2002;40:402–407.

15. Khandoker AH, Palaniswami M, Karmakar CK. Support vector machinesfor automated recognition of obstructive sleep apnea syndrome from ECGrecordings. IEEE Trans Inf Technol Biomed. 2009;13:37–48.



Dow

nloaded from


16. Mendez MO, Corthout J, Van Huffel S, Matteucci M, Penzel T, CeruttiS, Bianchi AM. Automatic screening of obstructive sleep apnea from theECG based on empirical mode decomposition and wavelet analysis.Physiol Meas. 2010;31:273–289.

17. Roche F, Celle S, Pichot V, Barthelemy JC, Sforza E. Analysis of theinterbeat interval increment to detect obstructive sleep apnoea/hypopnoea.Eur Respir J. 2007;29:1206–1211.

18. Guggisberg AG, Hess CW, Mathis J. The significance of the sympatheticnervous system in the pathophysiology of periodic leg movements insleep. Sleep. 2007;30:755–766.

19. Sforza E, Pichot V, Barthelemy JC, Haba-Rubio J, Roche F. Cardio-vascular variability during periodic leg movements: a spectral analysisapproach. Clin Neurophysiol. 2005;116:1096–1104.

20. Sleep-related breathing disorders in adults: recommendations forsyndrome definition and measurement techniques in clinical research: thereport of an American Academy of Sleep Medicine Task Force. Sleep.1999;22:667–689.

21. Coleman RM, Bliwise DL, Sajben N, Boomkamp A, de Bruyn LM,Dement WC. Daytime sleepiness in patients with periodic movements insleep. Sleep. 1982;5(Suppl 2):S191–S202.

22. Bland JM, Altman DG. Statistical methods for assessing agreementbetween two methods of clinical measurement. Lancet. 1986;1:307–310.

23. Hanley JA, McNeil BJ. A method of comparing the areas under receiveroperating characteristic curves derived from the same cases. Radiology.1983;148:839–843.

24. Stein PK, Duntley SP, Domitrovich PP, Nishith P, Carney RM. A simplemethod to identify sleep apnea using Holter recordings. J CardiovascElectrophysiol. 2003;14:467–473.

25. Badr MS. Central sleep apnea. Prim Care. 2005;32:361–374.26. Leung RS, Floras JS, Lorenzi-Filho G, Rankin F, Picton P, Bradley TD.

Influence of Cheyne-Stokes respiration on cardiovascular oscillations inheart failure. Am J Respir Crit Care Med. 2003;167:1534–1539.

27. Thomas RJ, Mietus JE, Peng CK, Gilmartin G, Daly RW, Goldberger AL,Gottlieb DJ. Differentiating obstructive from central and complex sleepapnea using an automated electrocardiogram-based method. Sleep. 2007;30:1756–1769.

28. Young T, Evans L, Finn L, Palta M. Estimation of the clinicallydiagnosed proportion of sleep apnea syndrome in middle-aged men andwomen. Sleep. 1997;20:705–706.

CLINICAL PERSPECTIVERecent prospective studies have reported that obstructive sleep apnea (OSA) is associated with an increased risk ofcardiovascular events and this risk may be reduced by treatment. In patients with OSA, the repeated apneic episodes elicitincreased sympathetic activation, surge in blood pressure, and cardiac arrhythmias. Concordantly, the circadian distributionof sudden cardiac death among OSA patients has a peak during the sleeping hours. Diagnosis of OSA, however, requirespolysomnographic examination, which is hard to perform in many patients. Episodes of OSA are accompanied by acharacteristic heart rate pattern, known as cyclic variation of heart rate (CVHR). Detection of CVHR by ambulatory ECGhas long been suggested as a useful method for OSA screening, but no large-scale clinical study has been performed. Inthis study, we examined a new automated ECG algorithm called ACAT for its accuracy in detecting OSA. The ACAT wasdeveloped with 63 sleep studies in a training cohort, and its performance was confirmed with a published database(Physionet Apnea-ECG). It was then applied to 862 consecutive patients referred for diagnostic sleep study. The numberof CVHR detected by the ACAT closely correlated (r�0.84) with the apnea-hypopnea index. The CVHR �15 episodesper hour identified the patients with apnea-hypopnea index �15 with 83% sensitivity and 77% positive predictive accuracy.The classification performance was unaffected by older age (�65 years) or cardiac autonomic dysfunction. Our studyindicates that the automated detection of CVHR with the ACAT algorithm provides a powerful ECG-based screening toolfor moderate-to-severe OSA.



Dow

nloaded from


Nomiyama, Kiyohiro Kawai, Itsuo Kodama and Hiroki SakakibaraJunichiro Hayano, Eiichi Watanabe, Yuji Saito, Fumihiko Sasaki, Keisaku Fujimoto, Tetsuo

Screening for Obstructive Sleep Apnea by Cyclic Variation of Heart Rate

Print ISSN: 1941-3149. Online ISSN: 1941-3084 Copyright © 2010 American Heart Association, Inc. All rights reserved.

Avenue, Dallas, TX 75231is published by the American Heart Association, 7272 GreenvilleCirculation: Arrhythmia and Electrophysiology

doi: 10.1161/CIRCEP.110.9580092011;4:64-72; originally published online November 12, 2010;Circ Arrhythm Electrophysiol.

http://circep.ahajournals.org/content/4/1/64World Wide Web at:

The online version of this article, along with updated information and services, is located on the

http://circep.ahajournals.org//subscriptions/

is online at: Circulation: Arrhythmia and Electrophysiology Information about subscribing to Subscriptions:

http://www.lww.com/reprints Information about reprints can be found online at: Reprints:

document. Answer

Permissions and Rights Question andunder Services. Further information about this process is available in thepermission is being requested is located, click Request Permissions in the middle column of the Web pageClearance Center, not the Editorial Office. Once the online version of the published article for which

can be obtained via RightsLink, a service of the CopyrightCirculation: Arrhythmia and Electrophysiologyin Requests for permissions to reproduce figures, tables, or portions of articles originally publishedPermissions:


Dow

nloaded from

http://circep.ahajournals.org/content/4/1/64

http://www.ahajournals.org/site/rights/

http://www.ahajournals.org/site/rights/

http://www.lww.com/reprints

http://circep.ahajournals.org//subscriptions/


screening for obstructive sleep apnea by cyclic...

Documents