Frequency-domain Index of Oxyhemoglobin Saturation from Pulse Oximetry for Obstructive Sleep Apnea Syndrome

Journal of Medical and Biological Engineering, 32(5): 343-348 343 Frequency-domain Index of Oxyhemoglobin Saturation from Pulse Oximetry for Obstructive Sleep Apnea Syndrome Liang-Wen Hang 1,2 Chen-Wen Yen 3 Chen-Liang Lin 4,* 1 Sleep Medicine Center, Department of Internal Medicine, China Medical University Hospital, Taichung 404, Taiwan, ROC 2 Department of Healthcare Administration, Asia University, Taichung 413, Taiwan, ROC 3 Department of Mechanical and Electro-Mechanical Engineering, National Sun Yat-Sen University, Kaohsiung 804, Taiwan, ROC 4 Biomedical Technology and Device Research Labs, Industrial Technology Research Institute, Hsinchu 310, Taiwan, ROC Received 20 Jul 2011; Accepted 18 Oct 2011; doi: 10.5405/jmbe.978 Abstract This study proposes a frequency-domain index of oxyhemoglobin saturation. We evaluate its reliability and compare with those of previous frequency-domain indices for screening the obstructive sleep apnea syndrome (OSAS). Patients diagnosed with OSAS by standard polysomnography were recruited Hospital Centre. Overall, 257 patients are used for training and 279 patients are used for validation. The presence of OSAS is defined as apnea-hypopnea index (AHI) > 5/h. Moderate and severe OSAS are defined as AHI > 15/h and AHI > 30/h, respectively. The proposed oxyhemoglobin index outperforms previously proposed frequency-domain indices. For predicting the severity of OSAS, the proposed index has a sensitivity/specificity of 74.3% / 96.1% for AHI > 15/h and 87.8% / 92.4% for AHI > 30/h. The results reveal that the proposed oxyhemoglobin index has a significantly smaller standard error of AHI than those of previous indices. Keywords: Pulse oximetry, Obstructive sleep apnea syndrome (OSAS), Polysomnography (PSG), Apnea-hypopnea index (AHI) 1. Introduction Obstructive sleep apnea syndrome (OSAS) is a growing health concern that affects up to 10% of middle-aged men [1]. Cardiovascular and neuropsychological morbidity has been demonstrated in untreated sleep apnea [2-4]. The gold standard for a definitive diagnosis of OSAS is in-laboratory polysomnography (PSG). However, this approach is expensive, time-consuming, and labor-intensive. The development of a simple screening tool could help the identification of patients with high probabilities of OSAS and thus allow resources to be focused on candidates likely to be afflicted with the disorder [5]. Oximetry is a commonly used American Academy of Sleep Medicine Type 4 monitoring technique for screening sleep apnea in the home [6]. Oxyhemoglobin indices from pulse oximetry have been used to screen and predict sleep apnea severity [7-11]. The lack of airflow during apneic or hypopnea periods may lead to recurrent episodes of hypoxemia that can be detected as fluctuations in oxyhemoglobin saturation (SpO 2 ) [12]. Spectral analysis is one of several methods for conducting variability analysis [13]. Studies have utilized spectral analysis * Corresponding author: Chen-Liang Lin Tel: +886-3-5919131; Fax: +886-3-5912444 E-mail: chenlianglin@itri.org.tw and its graphical presentation, the periodogram, to examine the variability of SpO 2 during sleep [14]. Previous studies reported a spectrum peak between period boundaries of 30 to 70 s which has been proposed as a screening marker for sleep apnea [15,16]. Absolute frequency regions are used by previously proposed frequency-domain indices; therefore, individual variance is not considered. The present study proposes a frequency-domain index and evaluates its reliability and compare with those of previously proposed frequency-domain oxyhemoglobin indices for predicting the severity of OSAS using automated digital analysis. 2. Materials and methods 2.1 Study subjects Patients diagnosed with OSAS by standard PSG were recruited from the China Medical University Hospital Centre. The subjects in the training set (257 patients) were enrolled in 2004, and those in the validation set (279 patients) were collected in 2005. Clinical data were collected retrospectively. Table 1 shows the clinical features and PSG results. All subjects were assessed with the Epworth Sleepiness Scale (ESS) before the PSG test. Periodic limb movements (PLM) were determined for periodic, jerking leg movements by

344 J. Med. Biol. Eng., Vol. 32 No. 5 2012 bilateral leg electromyograms (EMGs). Periodic limb movement disorder (PLMD) was defined as PLM > 5/h. Inclusion criteria were being older than 16 years and being diagnosed with OSAS with AHI > 5 / h. Exclusion criteria were chronic obstructive pulmonary disease, chronic chest wall disease, and having medical data with a total recording time less than three hours. The study was approved by the Medical Research Ethics Committee of the China Medical University Hospital (DMR 96-IRB-17). Table 1. Clinical characteristics of subjects and PSG results for training and validation sets. Clinical characteristic Training set (n = 257) Validation set (n = 279) P value Mean ± SD Mean ± SD Age (years) 42.7 ± 12.3 44.6 ± 12.4 NS BMI (kg/m 2 ) 26.3 ± 3.9 26.8 ± 4.4 NS ESS 8.9 ± 4.7 9.4 ± 5.0 NS AHI (h -1 ) 37.6 ± 23.6 36.7 ± 27.0 NS Male / Female 209/48 233/46 NS Presence of PLMD 1 1 NS BMI: Body mass index; ESS: Epworth Sleepiness Scale; AHI: Apnea-hypopnea index; PLMD: Periodic limb movement disorder Data are presented as mean ± SD or No. P values are between the training and validation sets. 2.2 Polysomnographic study PSG data were recorded using a computerized PSG system (Alice 4, Healthdyne Technologies, Atlanta, Georgia, USA). This system includes a standardized montage of twochannel electroencephalograms (EEGs; C4/A1, C3/A2), bilateral electro-oculograms (EOGs), submental EMGs, bilateral leg EMGs, and electrocardiograms (ECG). Oxyhemoglobin saturation was recorded using a finger-probe oximetry unit. The sampling rate of the oximeter was 1 Hz. For pulse rates below 112 beats per min (BPM), the averaging calculation was based on a four beat exponential average for SpO 2. For pulse rates above 112 BPM, the averaging was doubled, and then re-doubled above 225 BPM. Airflow was measured using oronasal pressure, and respiratory effort was assessed by inductance plethysmography. The stored data were digitized for computer analysis by data analysis software (Matlab; MathWorks Inc, USA). Artifacts were removed by eliminating all changes of oxygen saturation between consecutive sampling intervals of > 4%/s, and any oxygen saturation < 50% by an automated algorithm [7]. The raw data were reviewed by an experienced doctor and separately scored by a sleep technician certified by the Taiwan Sleep Medicine Society. Sleep stages were scored according to the criteria of Rechtschaffen and Kales [17]). Arousals were defined as episodes lasting 3 s or longer in which there was a return of alpha activity associated with a discernible increase in EMG activity. Apnea was defined as a cessation of oronasal airflow for a minimum of 10 s. Hypopnea was defined as a reduction of oronasal airflow to 50% or less of the value prevailing during a preceding period of normal breathing, for at least 10 s, and associated with 4% oxyhemoglobin desaturation and (or) EEG arousal [17,18]. 2.3 Methods The spectrum was calculated using the fast Fourier transformation (FFT) method. To mitigate the influence of oxyhemoglobin saturation artifacts and spectral leakage due to discontinuities that appear at the beginning and the end of the signal, an overlapping moving window method was applied. Each window size was 30 s, and shifted by 1 s each time. In order to increase frequency resolution, each window was zeropadded to use 1024-point FFT. The bandwidth of each window was defined as the frequency at which the accumulated energy from 0 Hz was 95% of the total energy. The bandwidths of all windows were recorded and sorted. Finally, the following four indices were calculated: the bandwidth mean of all windows (BW100), the bandwidth mean of the top 50% of windows (BW50), the bandwidth mean of the top 30% of windows (BW30), and the bandwidth mean of the top 10% of windows (BW10). For previous frequency-domain indices, if a peak in the spectrum between the period boundaries of 30 and 70 s was observed, the subject was considered to have OSAS, as shown in Fig. 1. The following four indices were obtained, as shown in Fig. 1 [15,16], where the periodogram is an estimate of the power spectral density (PSD) of a signal: the total area of the periodogram (S total ), the ratio of the area enclosed in the periodogram within the period of 30 to 70 s (S 30-70 ), the ratio of the area enclosed in the periodogram within the period of 30 to 70 s with respect to the total area of the periodogram (S), and the peak amplitude of the periodogram in the period of 30 to 70 s (PA). The accuracy and reliability of these oximetry indices for the prediction of the severity of OSAS were analyzed. Based on the best standard error of estimate (SEE) between AHI and the proposed and previously proposed frequency-domain indices, the sensitivity and specificity of the screening of moderate (AHI 15/h) and severe (AHI 30/h) OSAS patients were determined. The sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) are reported. A receiver operating characteristic (ROC) curve was constructed and the area under the curve (AUC) was calculated. ROC analysis relates sensitivity and 1-specificity. For the ROC curve, the point with the largest sum of sensitivity and specificity was chosen as a threshold. In addition, Cohen s kappa coefficient is also reported. Differences in means of continuous variables were assessed with Student s t-test. All data are reported as mean ± standard deviation (SD). A two-tailed value of P < 0.05 was considered significant. 3. Results From PSG, in the training set, AHI 15/h was confirmed in 206 (80.2 %) of 257 and 139 (54.1 %) of 257 patients above 30/h. Fifty-one patients (60.8 % male and 39.2% female), 67 patients (80.6 % male and 19.4% female), and 139 patients (89.2 % male and 10.8 % female) had 5/h AHI < 15/ h, 15/h AHI < 30/h, and AHI 30 /h, respectively. All of the

Oxyhemoglobin Saturation index for OSAS 345 (a) oxyhemoglobin indices BW10 and S 30-70, respectively, for various OSAS thresholds. In the group with AHI 30/h, the AUC values of BW10 and S 30-70 are 0.96 and 0.91, respectively. For AHI 15/h, the AUC values of these 2 indices are 0.93 and 0.84, respetively. The results show that the performance of the two kinds of oxyhemoglobin index was better when the OSAS threshold was AHI 30/h. The results for the validation set and the training set are consistent, as shown in Tables 3 and 4. For the validation set, the proposed indices outperform the previously proposed frequency-domain indices. In the group of AHI 30/h, the AUC values of BW10 and S 30-70 are 0.94 and 0.90, respectively. For AHI 15/h, the AUC values of these 2 indices are 0.91 and 0.83, respectively. The screening performance for AHI 30/h was better than that for AHI 15/h. Tables 3 and 4 show the AUC of the ROC curve and the optimal cutoff of the ROC curve for various OSAS thresholds for the training set and validation set. (b) Figure 1. Diagram of frequency-domain indices. (a) Typical spectrum of an OSAS subject. (b) Typical spectrum of a normal subject. (a) SEE, linear regression parameters, correlation, and significance levels between AHI and the indices are shown in Table 2. Figures 2 shows plots of AHI versus BW10 and AHI versus S 30-70, respectively. The BW10 is a better index with a smaller SEE among the proposed indices and S 30-70 is a better index with a smaller SEE among the previously proposed indices. Table 2. Linear regression parameters between AHI and oxyhemoglobin frequency-domain indices for training set. Index SEE Slope Intercept Correlation P value BW100 12.53 23688.846-1374.558 0.85 < 0.001 BW50 12.58 9347.718-532.793 0.85 < 0.001 BW30 12.77 5918.172-331.105 0.84 < 0.001 BW10 11.03 3418.094-187.413 0.88 < 0.001 S 30-70 18.71 823.387-158.786 0.61 < 0.001 S total 19.84 5.205E-05-104.518 0.54 < 0.001 Pa 19.32 0.003 23.663 0.57 < 0.001 S 18.74 1.375E-04-52.335 0.61 < 0.001 SEE: standard error of estimate. Correlation and P values are between the indices and AHI. The proposed indices outperform the previously proposed frequency-domain indices. Figures 3 shows the ROC curves for (b) Figure 2. Plot of AHI versus oxyhemoglobin indices for training set. (a) BW10 and (b) S 30-70.

346 J. Med. Biol. Eng., Vol. 32 No. 5 2012 4. Discussion (a) (b) Figure 3. ROC curves of oxyhemoglobin indices for OSAS screening for training learning set. (a) BW10 and (b) S 30-70. Table 3. AUC of ROC curve for oxyhemoglobin indices and optimal cutoff of ROC curve for various OSAS thresholds for training set (n = 257). AHI cutoff of 30/h AHI cutoff of 15/h Oxyhemoglobin index BW10 S 30-70 BW10 S 30-70 AUC 0.96 0.91 0.93 0.84 Kappa 0.80 0.73 0.51 0.40 Sensitivity (%) 87.8 81.3 74.3 63.1 Specificity (%) 92.4 91.5 96.1 98.0 PPV (%) 93.1 91.9 98.7 99.2 NPV (%) 86.5 80.6 48.0 39.7 AUC: the area under the ROC curve; ROC curve: receiver operating characteristic curve; Kappa: Cohen s kappa coefficient; PPV: positive predictive value; NPV: negative predictive value Table 4. AUC of the ROC curve for oxyhemoglobin indices and optimal cutoff of ROC curve for variouis OSAS thresholds for validation set (n = 279). AHI cutoff of 30/h AHI cutoff of 15/h Oxyhemoglobin index BW10 S 30-70 BW10 S 30-70 AUC 0.94 0.90 0.91 0.83 Kappa 0.73 0.70 0.61 0.48 Sensitivity (%) 73.3 80.7 78.9 61.3 Specificity (%) 99.3 88.2 92.0 98.7 PPV (%) 99.0 86.5 96.4 99.2 NPV (%) 79.9 83.0 61.6 48.4 AUC: the area under the ROC curve; ROC curve: receiver operating characteristic curve; Kappa: Cohen s kappa coefficient; PPV: positive predictive value; NPV: negative predictive value. Spectral analysis of SpO 2 during sleep is an effective approach for screening sleep apnea [15,16]. This study proposed a frequency-domain index for predicting the severity of OSAS automatically, and compared the index to frequency-domain indices. The results indicate that the proposed index has a significantly stronger correlation and better screening performance than those of previously proposed frequencydomain indices and that the training set and validation set produce consistent results. PSG is the gold standard for a definitive diagnosis of OSAS. Oxyhemoglobin saturation data were collected as part of a standard laboratory PSG using pulse oximetry. Therefore, pulse oximetry can reduce sleep laboratory efforts in screening severe OSAS. Previous studies used spectral analysis to evaluate the power density of SpO 2 by performing the fast Fourier transform on the oximetry signal to obtain four frequency-domain indices [15,16]. These previously proposed frequency-domain indices have proved effective in screening OSAS. The presence of a peak has a sensitivity of 78.0% and a specificity of 89.0%. Using a threshold of 0.15 for the ratio of the area enclosed in the 30 to 70 s peak to the total area of the spectrum, the sensitivity was 91.0% and the specificity was 67.0%. The performance of these indices is better in the previous studies. It might be caused by the different database. In previous study, the AHI values of the OSAS and non-osas groups were quite different: 40.1 23.0 /h and 2.2 2. 7 /h, respectively [15,16]. Since the oxygen saturation data recorded in sleep is a time series, this study also tested the time-domain index called the Delta index. The Delta index measures the average of absolute differences of oxyhemoglobin saturation between successive 12-s intervals [7,19]. For a Delta index threshold of 0.6, the sensitivity of oximetry for screening OSAS was 98.0% and the specificity was 46.0%. With a Delta index threshold of 0.4, the sensitivity of oximetry for screening OSAS was 88.0% and the specificity was 70.0%. In this study, for predicting the severity of OSAS, when the Delta index threshold was at 1.04 and 1.21, the sensitivity/specificity were 70.9%/92.2% for AHI > 15/h and 84.9%/92.4% for AHI > 30/h, respectively. The difference may be due to the different population, in terms of age and greater obesity, in Magalang [7] and Levy [19] studies (age: 56.0 and 48.9 versus 42.7 here, and BMI: 32.0 and 32.3 versus our 26.3 here). In addition, the results show that the proposed frequency-domain index has better screening performance of OSAS than the Delta index. The BW10 index has the best screening performance of all BW indices, perhaps due to it being able to focus on the fluctuant parts of nocturnal oxyhemoglobin saturation with more specificity. The variation of oxyhemoglobin saturation is usually higher during respiratory events, with the bandwidth being correspondingly higher. Oxyhemoglobin saturation is an indirect parameter that reflects the consequences of breathing. Not all respiratory events are accompanied by oxyhemoglobin desaturation [20], and this is one of the major limitations of pulse oximetry. Due to its ease

Oxyhemoglobin Saturation index for OSAS 347 of application, pulse oximetry with more sensitivity indices could be an effective tool for clinical use. However, repeat oximetry is required if the sleep efficiency of the subject is low or clinical suspicion is high [15]. 5. Conclusion A frequency-domain oxyhemoglobin index for predicting the presence and severity of OSAS was proposed. The relative utilities of the proposed index and previous frequency-domain indices in the screening of OSAS with a large number of subjects were systematically compared. The results reveal that the proposed frequency-domain oxy-hemoglobin index provides high screening performance. Clinically, pulse oximetry can be a helpful abbreviated testing modality for moderate to severe OSAS. Acknowledgements This study was supported by a grant from China Medical University Hospital (DMR-97-029). References [1] T. Young, M. Palta, J. Dempsey, J. Skatrud, S. Wever and S. Badr, The occurrence of sleep-disordered breathing among middle-aged adults, N. Engl. J. Med., 328: 1230-1235, 1993. [2] P. Lavie, P. Herer and V. Hoffstein, Obstructive sleep apnea syndrome as a risk factor for hypertension: population study, Br. Med. J., 320: 479-482, 2000. [3] P. Lanfranchi and V. A. Somers, Obstructive sleep apnea and vascular disease, Respir. Res., 2: 315-319, 2001. [4] A. Malhotra and D. P. White, Obstructive sleep apnea, Lancet, 360: 237-245, 2002. [5] F. Roche, J. Gaspoz, I. Court-Fortune, P. Minini, V. Pichot, D. Duverney, F. Costes, J. Lacour and J. Barthelemy, Screening of obstructive sleep apnea syndrome by heart rate variability analysis, Circulation, 100: 1411-1415, 1999. [6] C. Iber, S. Ancoli-Israel, A. L. Chesson Jr. and S. F. Quan, AASM Manual for the Scoring of Sleep and Associated Events, Westchester, IL: American Academy of Sleep Medicine, 2007. [7] U. J. Magalang, J. Dmochowski, S. Veeramachaneni, A. Draw, M. J. Mador, A. El-Solh and B. J. Grant, Prediction of the apnea-hypopnea index from overnight pulse oximetry, Chest, 124: 1694-1701, 2003. [8] J. Vazquez, W. H. Tsai, W. W. Flemons, A. Masuda, R. Brant, E. Hajduk, W. A. Whitelaw and J. E. Remmers, Automated analysis of digital oximetry in the diagnosis of obstructive sleep apnea, Thorax, 55: 302-307, 2000. [9] E. Chiner, J. Signes-Costa, J. M. Arrieiro, J. Marco, I. Fuentes and A. Sergado, Nocturnal oximetry for the diagnosis of the sleep apnea hypopnea syndrome: A method to reduce the number of polysomnographies? Thorax, 54: 968-971, 1999. [10] S. Gyulay, L. G. Olson, M. J. Hensley, M. T. King, K. M. Allen and N. A. Saunders, A comparison of clinical assessment and home oximetry in the diagnosis of obstructive sleep apnea, Am. Rev. Respir. Dis., 147: 50-53, 1993. [11] R. Golpe, A. Jimenez, R. Carpizo and J. M. Cifrian, Utility of home oximetry as a screening test for patients with moderate to severe symptoms of obstructive sleep apnea, Sleep, 22: 932-937, 1999. [12] L. J. Epstein and G. R. Dorlac, Cost-effectiveness analysis of nocturnal oximetry as a method of screening for sleep apnea-hypopnea syndrome, Chest, 113: 97-103, 1998. [13] A. J. Seely and P. T. Macklem, Complex systems and the technology of variability analysis, Crit. Care, 8: R367-R384, 2004. [14] C. C. Hua and C. C. Yu, Smoothed periodogram of oxyhemoglobin saturation by pulse oximetry in sleep apnea syndrome: an automated analysis, Chest, 131: 750-757, 2007. [15] C. Zamarron, P. V. Romero, J. R. Rodriguez and F. Gude, Oximetry spectral analysis in the diagnosis of obstructive sleep apnoea, Clin. Sci., 97: 467-473, 1999. [16] C. Zamarron, F. Gude, J. Barcala, J. R. Rodriguez and P. V. Romero, Utility of oxygen saturation and heart rate spectral analysis obtained from pulse oximetric recordings in the diagnosis of sleep apnea syndrome, Chest, 123: 1567-1576, 2003. [17] A. Reschtaschaffen and A. Kales, A Manual of Standardized Terminology Scoring System for Sleep Stages of Human Subjects, Washington DC: Public Health Service, US Government Printing Office, 1968. [18] The Report of an American Academy of Sleep Medicine Task Force, Sleep-related breathing disorders in adults: recommendations for syndrome definition and measurement techniques in clinical research, Sleep, 22: 667-689, 1999. [19] P. Levy, J. L. Pepin, C. Deschaux-Blanc, B. Paramelle and C. Brambilla, Accuracy of oximetry for detection of respiratory disturbances in sleep apnea syndrome, Chest, 109: 395-399, 1996. [20] I. Ayappa, B. S. Rappaport, R. G. Norman and D. M. Rapoport, Immediate consequences of respiratory events in sleep disordered breathing, Sleep Med., 6: 123-130, 2005.

348 J. Med. Biol. Eng., Vol. 32. No. 5 2012