Improving diagnostic ability of blood oxygen saturation from overnight pulse oximetry in obstructive sleep apnea detection by means of central tendency measure Daniel A ´ lvarez a, * , Roberto Hornero a , Marı ´a Garcı ´a a , Fe ´lix del Campo b , Carlos Zamarro ´n c a E.T.S.I. de Telecomunicacio ´n, University of Valladolid, Camino del Cementerio s/n, 47011 Valladolid, Spain b Hospital del Rı ´o Hortega, Servicio de Neumologı ´a, c/Cardenal Torquemada s/n, 47010 Valladolid, Spain c Hospital Clı ´nico Universitario, Servicio de Neumologı ´a, Travesı ´a de la Choupana s/n, 15706 Santiago de Compostela, Spain Received 18 October 2006; received in revised form 29 May 2007; accepted 12 June 2007 Artificial Intelligence in Medicine (2007) 41, 13—24 http://www.intl.elsevierhealth.com/journals/aiim KEYWORDS Central tendency measure; Nonlinear methods; Blood oxygen saturation; Oximetry; Obstructive sleep apnea Summary Objectives: Nocturnal pulse oximetry is a widely used alternative to polysomnogra- phy (PSG) in screening for obstructive sleep apnea (OSA) syndrome. Several oximetric indexes have been derived from nocturnal blood oxygen saturation (SaO 2 ). However, they suffer from several limitations. The present study is focused on the usefulness of nonlinear methods in deriving new measures from oximetry signals to improve the diagnostic accuracy of classical oximetric indexes. Specifically, we assessed the validity of central tendency measure (CTM) as a screening test for OSA in patients clinically suspected of suffering from this disease. Materials and methods: We studied 187 subjects suspected of suffering from OSA referred to the sleep unit. A nocturnal pulse oximetry study was applied simulta- neously to a conventional PSG. Three different index groups were compared. The first one was composed by classical indexes provided by our oximeter: oxygen desaturation indexes (ODIs) and cumulative time spent below a saturation of 90% (CT90). The second one was formed by indexes derived from a nonlinear method previously studied by our group: approximate entropy (ApEn). The last one was composed by indexes derived from a CTM analysis. * Corresponding author. Tel.: +34 983 423000x5589; fax: +34 983 423661. E-mail address: [email protected](D. A ´ lvarez). 0933-3657/$ — see front matter # 2007 Elsevier B.V. All rights reserved. doi:10.1016/j.artmed.2007.06.002
Transcript
Artificial Intelligence in Medicine (2007) 41, 13—24
http://www.intl.elsevierhealth.com/journals/aiim
Improving diagnostic ability of blood oxygensaturation from overnight pulse oximetry inobstructive sleep apnea detection by meansof central tendency measure
Daniel Alvarez a,*, Roberto Hornero a, Marıa Garcıa a,Felix del Campo b, Carlos Zamarron c
a E.T.S.I. de Telecomunicacion, University of Valladolid, Camino del Cementerio s/n,47011 Valladolid, SpainbHospital del Rıo Hortega, Servicio de Neumologıa, c/Cardenal Torquemada s/n,47010 Valladolid, SpaincHospital Clınico Universitario, Servicio de Neumologıa, Travesıa de la Choupana s/n,15706 Santiago de Compostela, Spain
Received 18 October 2006; received in revised form 29 May 2007; accepted 12 June 2007
Objectives: Nocturnal pulse oximetry is a widely used alternative to polysomnogra-phy (PSG) in screening for obstructive sleep apnea (OSA) syndrome. Several oximetricindexes have been derived from nocturnal blood oxygen saturation (SaO2). However,they suffer from several limitations. The present study is focused on the usefulness ofnonlinear methods in deriving new measures from oximetry signals to improve thediagnostic accuracy of classical oximetric indexes. Specifically, we assessed thevalidity of central tendency measure (CTM) as a screening test for OSA in patientsclinically suspected of suffering from this disease.Materials and methods: We studied 187 subjects suspected of suffering from OSAreferred to the sleep unit. A nocturnal pulse oximetry study was applied simulta-neously to a conventional PSG. Three different index groups were compared. The firstone was composed by classical indexes provided by our oximeter: oxygen desaturationindexes (ODIs) and cumulative time spent below a saturation of 90% (CT90). Thesecond one was formed by indexes derived from a nonlinear method previouslystudied by our group: approximate entropy (ApEn). The last one was composed byindexes derived from a CTM analysis.
Results: For a radius in the scatter plot equal to 1, CTM values corresponding to OSApositive patients (0.30 � 0.20, mean � S.D.) were significantly lower ( p� 0.001)than those values from OSA negative subjects (0.71 � 0.18, mean � S.D.). CTM wassignificantly correlated with classical indexes and indexes from ApEn analysis. CTMprovided the highest correlation with the apnea—hipopnea index AHI (r = �0.74,p < 0.0001). Moreover, it reached the best results from the receiver operatingcharacteristics (ROC) curve analysis, with 90.1% sensitivity, 82.9% specificity, 88.5%positive predictive value, 85.1% negative predictive value, 87.2% accuracy and anarea under the ROC curve of 0.924. Finally, the AHI derived from the quadraticregression curve for the CTM showed better agreement with the AHI from PSG thanclassical and ApEn derived indexes.Conclusion: The results suggest that CTM could improve the diagnostic ability of SaO2
signals recorded from portablemonitoring. CTM could be a useful tool for physicians inthe diagnosis of OSA syndrome.# 2007 Elsevier B.V. All rights reserved.
1. Introduction
The obstructive sleep apnea (OSA) syndrome is char-acterized by repetitive reduction or cessation ofairflow due to partial or complete airway obstruction[1]. This disease is usually associated with hypoxe-mia, bradycardia, arousals and fragmented sleep [2].Nowadays,OSA is themost common respiratory refer-ral in many sleep centers [3]. The estimated OSAprevalence varies from 1 to 5% of adult men inwestern countries [4]. OSA is associated with condi-tions that are responsible for the most importantcauses of mortality in adults: hypertension and car-diovascular and cerebrovascular diseases. Severalneurobehavioral morbidities, which are of poten-tially great public health and economic importance,are linked with OSA [4]. The major behavioral symp-toms include excessive daytime sleepiness (EDS),neurocognitive deficits like impairments in concen-tration andmemory, and psychological problems likedepression or personality changes [5]. Individualswith OSA are dangerous drivers with an increasedrisk of being involved in road and work accidents [3].
The standard diagnostic test for OSA syndrome isovernight polysomnography (PSG) [6], consisting inthe recording of neurophysiological and cardiore-spiratory signals subsequently used to analyze sleepand breathing. The apnea—hypopnea index (AHI)derived from the PSG is then used to diagnose thedisease. Portable monitoring has been proposed as asubstitute for PSG in the diagnostic assessment ofpatients with suspected sleep apnea [7]. Due to itsnoninvasive nature and simplicity, nocturnal pulseoximetry is widely used in many medicine areas todetermine patient’s blood oxygen saturation (SaO2)and heart rate. The lack of airflow during apneicperiods can lead to recurrent episodes of hypoxemiathat can be detected on oximetry as fluctuations inSaO2 records [8].
Several quantitative indexes derived from noctur-nal oximetry have been developed to diagnose OSA.The most frequently used by physicians include oxy-gen desaturation indexes (ODIs), which measure thenumber of dips in the SaO2 signal below a certainthreshold [9—11], and the cumulative time spentbelow a certain saturation level (CT) [12,13]. How-ever, these indexes have significant limitations. Ingeneral, CT indexes did not achieve high diagnosticaccuracies [13,14]. On the other hand, there is not auniversally accepted definition for oxygen desatura-tion.Moreover, there isnotaconsensusona thresholdto diagnose OSA based on ODIs [14,15]. Furthermore,correlation between oximetry indexes and AHI is nothigh [13]. In previous studies [16—18], our group hasshown that nonlinear analysis could provide usefulinformation in the diagnosis of OSA syndrome. Aregularity measure from SaO2 signals obtained apply-ing approximate entropy (ApEn) improved the diag-nostic accuracy of classical oximetric indexes [16].ApEn was also applied to heart rate signals fromnocturnal oximetry, obtaining promising results[17]. Moreover, additional nonlinear methods, cen-tral tendency measure (CTM) and Lempel—Ziv (LZ)complexity, were applied to SaO2 records [18]. Theresults suggested that both CTM and LZ complexitycould help physicians in screening for OSA syndrome.Particularly, a variability measure by means of theCTM provided the best diagnostic accuracy. The pre-sent study intended to go more deeply into theusefulness of the CTM to diagnose OSA. We assessedits advantages over classical oximetric indexes andother nonlinear methods: it is a simple parameter toestimate the signal variability with a low computa-tional cost [19]. Furthermore, we studied thechanges in the diagnostic accuracy when using dif-ferent values of the input parameters.
Variability measures of ECG allow to distinguishbetween normal and chronic heart failure subjects
Improving diagnostic ability of blood oxygen saturation using CTM 15
[19]. Some authors have applied nonlinear methodsover different respiratory patterns to study sleepstages and the coordination between brain and lungs[20] or panic disorder [21].
In the present study, we applied CTM looking fordifferences in variability between OSA positive andOSA negative patients. Our study is aimed to esti-mate the overall variability of each overnight oxi-metric recording by means of CTM in order to assessits utility in OSA diagnosis. We assessed its useful-ness to physicians in screening for OSA syndrome,comparing it with classical oximetric indexes andwith ApEn analysis. We studied three different indexgroups: classical indexes, regularity indexes fromApEn analysis and variability indexes from CTManalysis.
2. Subjects and signals
A total of 187 patients (147 males and 40 females)suspected of having OSAwere studied. Patients havea mean � standard deviation (S.D.) age of57.97 � 12.84 years and a body mass index (BMI)of 29.54 � 5.51 kg/m2. All subjects presented day-time hypersomnolence, loud snoring, nocturnalchoking and awakenings, or apneic events (or allfour symptoms) reported by the subject or a bedmate. Sleep studies were carried out usually frommidnight to 08:00 a.m. in the Sleep Unit of HospitalClınico Universitario in Santiago de Compostela,Spain. The Review Board on Human Studies at thisinstitution approved the protocol, and all subjectsgave their informed consent to participate in thestudy.
2.1. Conventional polysomnography
All patients underwent overnight PSG (UltrasomNetwork, Nicolet, Madison, WI, USA) which includedelectroencephalogram (EEG), electrocardiogram(ECG), electrooculogram (EOG), chin electromyo-gram (EMG), measurement of chest wall movement
Table 1 Demographic and clinical features of the subjects
Data are presented as mean � S.D. unless otherwise indicated. OSApatients without obstructive sleep apnea.
and airflow measurement (three-port thermistor).The PSG register was analyzed over periods of 30 sduring sleep phases I—IV and rapid eye movement,according to Rechtschaffen and Kales rules [22].Apnea was defined as the cessation of airflow formore than 10 s and hypopnea as the reduction ofrespiratory flow for at least 10 s accompanied by a4% or more decrease in the saturation of hemoglobin[23—26]. The average AHI was calculated for hourlyperiods of sleep. According to the American Acad-emy of Sleep Medicine Task Force criteria [27], anAHI greater than or equal to 10 events per hour (e/h)of sleep was considered as diagnosis of OSA. If thesubject had less than 3 h of total sleep, the sleepstudy was repeated.
After the PSG, a conventional spirometry study(Collins spirometer) was carried out. Chronicobstructive pulmonary disease (COPD) was definedas a disease state characterized by airflow limitationthat is not fully reversible. The airflow limitation isusually both progressive and associated with anabnormal inflammatory response of the lungs to nox-ious particles or gases [28]. The spirometry showedthat 42 patients had COPD (mean age of 62.26� 13.65 years and a BMI of 29.66 � 17.31 kg/m2).Moreover, 9 of the 42 patients with COPD (21.8%)presented respiratory failure.According to the globalinitiative for chronic obstructive lung disease (GOLD)consensus [28], 22 (52.4%) of these subjects could beclassified as mild COPD patients, 14 (33.3%) as mod-erate COPD patients and 6 (14.3%) as severe COPDpatients.
Table 1 summarizes the demographic and clinicalfeatures of the subjects under study, as well as thegroups derived from the PSG diagnosis. The OSApositive group consisted of 111 patients (59.4%)diagnosed as OSA according to an AHI � 10 e/h(40.07 � 19.64 e/h), whereas the remaining 76 sub-jects (40.6%) made up the OSA negative group(2.04 � 2.36 e/h). In the OSA positive group, therewere men and women between 28 and 81 years(58.30 � 12.88 years) and with BMI between 20.57and 46.51 kg/m2 (30.45 � 4.92 kg/m2). The OSA
positive: patients with obstructive sleep apnea. OSA negative:
16 D. Alvarez et al.
negative group consisted of subjects between 21and 79 years (57.57 � 12.87 years) and withBMI between 19.53 and 42.19 kg/m2 (28.42 �6.02 kg/m2).
2.2. Overnight pulse oximetry
An overnight pulse oximetry analysis was carried outsimultaneously to the conventional PSG study.Recording of SaO2 was carried out using a Criticare504 oximeter (CSI, Waukesha, WI, USA). SaO2 andheart rate were both simultaneously recorded usinga dual wavelength-based finger probe with a sam-pling frequency of 0.2 Hz (one sample every 5 s). Inthis study only the SaO2 signals were used. Althoughnew oximeters work at higher sampling frequencies,the study carried out by Warley et al. [29] showedthat this sampling frequency provides reasonableresolution in SaO2 variability. Moreover, oximetrysignals recorded at higher sampling frequencies aresubsequently averaged to obtain usually one sampleevery 12 s [7,14,30]. There is some underestimationof the peak SaO2 in recovery post-apnea, but thesignal shape and variability were preserved [29].
The SaO2 signals were saved to separate files andprocessed off-line. Artifacts due to poor contactfrom the finger probe, patient movements or badregional circulation, were removed by visual inspec-tion of SaO2 signals, discarding data showing dropsto zero. Fig. 1 displays three common oximetricrecordings. Fig. 1(a) depicts a common OSA negativesubject. Fig. 1(b) shows a SaO2 record with clearlymarked desaturations, corresponding to an appar-ent OSA positive subject. Fig. 1(c) illustrates moreexactly the difficulty in the diagnosis of the disease.It shows the SaO2 record for an uncertain OSApositive patient. In this case, dips in SaO2 are notso extreme and the diagnosis by visual inspection isnot evident.
Figure 1 SaO2 records from nocturnal oximetry for (a) acommon OSA negative subject, (b) an apparent OSA posi-tive patient and (c) an uncertain OSA positive subject.
3. Methods
3.1. Classical oximetry indexes
Our oximeter provided the following indexes: oxy-gen desaturation indexes of 4% (ODI4), 3% (ODI3), 2%(ODI2) and cumulative time spent below a satura-tion of 90% (CT90). The number of falls in each SaO2
record greater than or equal to 4, 3 and 2% werecomputed from baseline. Baseline was set initiallyas the mean level in the first 3 min of recording [31].These oxygen desaturation indexes were computedper hour of recording. CT90 was calculated as thepercentage of time during which the SaO2 registerwas below 90%.
3.2. Approximate entropy
ApEn is a family of statistics introduced as a quan-tification of regularity in sequences and time seriesdata, initially motivated by applications to rela-tively short and noisy data sets [32]. ApEn evaluatesboth dominant and subordinant patterns in the data,and discriminates series for which clear featurerecognition is difficult [32]. Several properties ofApEn make it highly suitable for biomedical timeseries analysis: ApEn is almost unaffected by low-level noise; it is robust to outliers, scale invariantand model independent; it is applicable to timeseries with at least 50 data points, with good repro-ducibility; it can be applied to discriminate generalclasses of correlated stochastic processes, as well asnoisy deterministic systems, providing finite valuesfor both stochastic and deterministic processes [32].
ApEn assigns a non-negative number to a timeseries, with larger values corresponding to greaterrandomness or irregularity in the data [32]. Thealgorithm applied to compute ApEn can be seen indetail in [16] and [17]. Briefly, ApEn measures thelogarithmic likelihood that runs of patterns that areclose (within r) for m contiguous observationsremain close (within the same tolerance r) on sub-sequent incremental comparisons (pattern lengthm + 1) [32]. Two input parameters,m and r, must befixed to compute ApEn(m, r): m is the length ofcompared runs, and r is effectively a filter [33]. Toensure appropriate comparisons betweendata sets,all input parametersm, r andNmust be the same foreach data set [33,34]. No guidelines exist for opti-mizing the m and r values. However, Pincus sug-gested parameter values ofm = 1 orm = 2 andwith ra fixed value between 0.1 and 0.25 times the S.D. ofthe original time series [32]. Multiple previous stu-dies have demonstrated that these input para-meters produce good statistical reproducibilityfor ApEn for time series of length N � 60 [32]. In
Improving diagnostic ability of blood oxygen saturation using CTM 17
previous studies by our own group [16,17], weshowed that ApEn was better estimated withm = 1 when applying it to SaO2 signals. Thus, inthe present study, we computed ApEn with m = 1and r equal to 0.1, 0.15 and 0.20 times the S.D. ofthe original time series, obtaining ApEn01, ApEn015and ApEn02, respectively.
3.3. Central tendency measure
Quantifying the signal variability by means of CTMstarts displaying a second-order difference plot.These kinds of scatter diagrams, given by Eq. (1),are graphs centered in the origin used to assess thedegree of chaos in a data set [19].
½xðnþ 2Þ � xðnþ 1Þ� versus ½xðnþ 1Þ � xðnÞ� (1)
The second-order difference plots are very usefulin modeling biological systems such as hemody-namics and heart rate variability. With thisapproach, rather than defining a time series aschaotic or not chaotic, the degree of variability orchaos is evaluated [19]. Such scatter plots could be auseful tool for physicians, who could make a pre-liminary diagnosis by visual inspection of the dia-grams. Fig. 2 displays the second-order differenceplots for the SaO2 signals depicted in Fig. 1. Theuncertain OSA positive subject whose SaO2 record isdisplayed in Fig. 1(c) could be initially misclassifiedas OSA negative by visual inspection. However, thecorresponding scatter plot in Fig. 2(c) shows a sig-nificantly high dispersion compared with the dia-gram for a common OSA negative subject,suggesting a different diagnosis. Thus, the sec-ond-order difference plots could help physiciansto improve preliminary diagnoses.
The CTM is used to quantify the signal variabilityfrom the second-order difference plots. The CTM is
Figure 2 Second-order difference plots (a) a commonOSA negative subject (b) an apparent OSA positive patientand (c) an uncertain OSA positive subject.
computed by selecting a circular region of radius r
around the origin and counting the number of pointsthat fall within the circle. This measure is subse-quently normalized dividing by the total number ofpoints. For a N point data series, N � 2 would be thetotal number of points in the scatter plot given byEq. (1). Then, the CTM can be computed as [35]
CTM ¼PN�2
i¼1 dðdiÞN � 2
(2)
where
dðdiÞ ¼1 if ½ðxðiþ 2Þ � xðiþ 1ÞÞ2
þðxðiþ 1Þ � xðiÞÞ2�1=2< r
0 otherwise
8><>:
(3)
The radius r is selected depending on the char-acter of the data. In the present study, we computedCTM with various radii to assess the behavior of themethod in pulse oximetry analysis. We computedCTM with three different values of r. For radii equalto 1, 3 and 6 we obtained CTM1, CTM3 and CTM6. Wecompared these parameters with the classical oxi-metric indexes and with those indexes derived fromthe ApEn analysis.
3.4. Statistical analysis
The Kolmogorov—Smirnov and Shapiro—Wilk testswere used to assess the normal distribution of thevariables involved in the study. Homoscedasticity(homogeneity of variances) was also assessed bymeans of the Levene’s test. The normal distributionand homoscedasticity could not be verified with allthe variables under study. Thus, the nonparametricMann—Whitney test was applied to look for signifi-cant differences between the OSA positive and theOSA negative groups. The Bonferroni correctionwasapplied due to the large number of variablesincluded in the study. SPSS 14 was used to performthe statistical analysis. The degree of associationbetween each index and the AHI was studied usingthe Pearson correlation test. A p-value was alsocomputed to measure the statistical significanceof the results. A ROC curve analysis was performedto assess the diagnostic capacity of each method inscreening for OSA syndrome. The following statis-tics were derived from this study: sensitivity, spe-cificity, positive predictive value (PPV), negativepredictive value (NPV), positive likelihood ratio(LR+), negative likelihood ratio (LR�), accuracyand area under the ROC curve. Correlation andROC curve analyses were used to select the bestparameter in each group of indexes. Scatter dia-grams were depicted to graphically study the rela-tion between indexes under study and the AHI. Inaddition, the linear regression (degree n = 1) and
18 D. Alvarez et al.
Table 3 Pearson correlation coefficients between oxi-metric indexes and AHI from PSG
Classical oximetric indexes
ODI2 ODI3 ODI4 CT90
r 0.608 0.617 0.603 0.280p <0.0001 <0.0001 <0.0001 <0.001
Regularity indexes from ApEn analysis
the quadratic regression (degree n = 2) curves,which best fit the plotted points in a least squaressense, were drawn in the scatter diagrams. Thepolynomials, which defined the relationshipbetween indexes and the AHI from PSG, were usedto obtain a derived AHI. Finally, Bland and Altmanplots were used to graphically measure the degreeof agreement between the derived AHIs and the AHIfrom PSG.
ApEn01 ApEn015 ApEn02
r 0.577 0.579 0.606p <0.0001 <0.0001 <0.0001
Variability indexes from CTM analysis
CTM1 CTM3 CTM6
r �0.738 �0.684 �0.561p <0.0001 <0.0001 <0.0001
r: Pearson correlation coefficient; p: statistical significance;ODI2: number of dips in SaO2 � 2%; ODI3: number of dips inSaO2 � 3%; ODI4: number of dips in SaO2 � 4%; CT90: percen-tage of time spent with SaO2 below 90%. ApEn01: approximateentropy computed with m = 1 and r = 0.1; ApEn015: approx-imate entropy computed with m = 1 and r = 0.15; ApEn02:approximate entropy computed withm = 1 and r = 0.2; CTM1:central tendencymeasure computedwith r = 1; CTM3: centraltendency measure computed with r = 3; CTM6: central ten-dency measure computed with r = 6.
4. Results
SaO2 signals from nocturnal oximetry were pro-cessed by means of ApEn and CTM. Classical oxi-metric indexes, ODI2, ODI3, ODI4 and CT90 werealso included in this study. As well as ODIs areprovided for three common desaturation thresholds(2, 3 and 4%), three different measures were com-puted for each nonlinear method varying their inputparameters: ApEn01, ApEn015 and ApEn02 werederived from the ApEn and CTM1, CTM3 and CTM6were derived from the CTM analysis.
Table 2 shows themean � S.D. values for both theOSA positive and the OSA negative groups for everyindex. Furthermore, the p-value from the Mann—Whitney test is displayed. CTM1, CTM3 and indexesfrom the ApEn analysis provide the highest signifi-cant differences between groups, improving resultsfrom the classical oximetric indexes. Table 3 showsthe correlation between all indexes and the AHI
Table 2 Average index values for the groups understudy
OSA positive OSA negative p-value
ODI2 27.18 � 24.54 6.07 � 10.03 6.9 � 10�5
ODI3 25.69 � 23.31 4.37 � 7.91 1.3 � 10�5
ODI4 24.08 � 22.46 3.12 � 6.02 1.2 � 10�6
CT90 23.94 � 29.90 4.56 � 17.05 2.3 � 10�6
ApEn01 1.16 � 0.35 0.48 � 0.27 1.1 � 10�21
ApEn015 1.14 � 0.34 0.48 � 0.27 1.0 � 10�21
ApEn02 1.09 � 0.31 0.48 � 0.26 1.4 � 10�21
CTM1 0.30 � 0.20 0.71 � 0.18 7.0 � 10�22
CTM3 0.75 � 0.23 0.98 � 0.05 8.9 � 10�22
CTM6 0.90 � 0.15 0.99 � 0.01 5.0 � 10�20
Data are presented as mean � S.D. OSA positive: patients withobstructive sleep apnea. OSA negative: patients withoutobstructive sleep apnea. ODI2: number of dips in SaO2 � 2%.2%. ODI3: number of dips in SaO2 � 3%. ODI4: number of dips inSaO2 � 4%. CT90: percentage of time spent with SaO2 below90%. ApEn01: approximate entropy computed with m = 1 andr = 0.1. ApEn015: approximate entropy computed with m = 1and r = 0.15. ApEn02: approximate entropy computed withm = 1 and r = 0.2. CTM1: central tendency measure computedwith r = 1. CTM3: central tendency measure computed withr = 3. CTM6: central tendency measure computed with r = 6.
derived from the PSG. Although CT90 was statisti-cally correlated with the AHI ( p < 0.001), itachieved the smallest Pearson correlation coeffi-cient (r = 0.280). ODIs achieved higher correlationwith AHI than CT90, slightly improving correlationvalues between ApEn and AHI. We obtained r = 0.617(p < 0.0001) with ODI3, whereas an r = 0.606(p < 0.0001) was reached with ApEn02. CTM1
Figure 3 ROC curves for the best parameter of eachindex group in terms of diagnostic accuracy and correla-tion with AHI.
Improving diagnostic ability of blood oxygen saturation using CTM 19
Table 4 Test results for the optimum decision threshold from the ROC curve analysis
Th: optimum decision threshold; S: sensitivity; E: specificity; PPV: positive predictive value; NPV: negative predictive value; LR+:positive likelihood ratio; LR�: negative likelihood ratio; A: accuracy; AROC: area under the ROC curve; ODI2: number of dips inSaO2 � 2%; ODI3: number of dips in SaO2 � 3%; ODI4: number of dips in SaO2 � 4%; CT90: percentage of time spent with SaO2 below90%. ApEn01: approximate entropy computed with m = 1 and r = 0.1; ApEn015: approximate entropy computed with m = 1 andr = 0.15; ApEn02: approximate entropy computed with m = 1 and r = 0.2; CTM1: central tendency measure computed with r = 1;CTM3: central tendency measure computed with r = 3; CTM6: central tendency measure computed with r = 6.
achieved the highest correlation with the AHI(r = �0.7382, p < 0.0001). The negative correlationindexes obtained with CTM are due to the ownnature of this method, which assigns small valuesto high variability and vice versa.
Table 4 shows the results from the ROC curveanalysis for each parameter. We can notice thatnonlinear indexes achieved better diagnostic accu-racy and area under the ROC curve (AROC) thanclassical indexes. All nonlinear measures providedsensitivity values over 83% and specificity valuesover 81%, leading to accuracy values of at least84.5% and areas under the ROC curve over 0.90.On the other hand, classical oximetric indexesshowed large differences between sensitivity andspecificity values, with very poor sensitivities.
Figure 4 Scatter plot for ODI3 and AHI. Linear regression(dashed line) and quadratic regression (solid line) curvesthat best fit the data in a least square sense.
Accuracies are around 75% and AROCs are below0.80. Diagnostic test values from ApEn analysisdo not vary very much changing the input para-meters. However, CTM diagnostic accuracy slightlydecreases when the radius is increased. The bestresults are provided by CTM1, with 90.1% sensitiv-ity, 82.9% specificity, 88.5% positive predictivevalue, 85.1% negative predictive value, 87.2% accu-racy and an area under the ROC curve of 0.924.Fig. 3 shows the ROC curves for the best parameterof each index group in terms of diagnostic accuracyand correlation with AHI: ODI3 from the classicalindex group, ApEn02 from the ApEn index group andCTM1 from the CTM index group. The ROC curvesillustrate the variation of sensitivity and specificityand hence, the diagnosis,when different thresholds
Figure 5 Scatter plot for ApEn02 and AHI. Linear regres-sion (dashed line) and quadratic regression (solid line)curves that best fit the data in a least square sense.
20 D. Alvarez et al.
Figure 6 Scatter plot for CTM1 and AHI. Linear regres-sion (dashed line) and quadratic regression (solid line)curves that best fit the data in a least square sense.
Figure 8 Bland and Altman plot for measuring agree-ment between AHI derived from the quadratic regressioncurve of ApEn02 and AHI from PSG. Solid line representsthe mean difference between both methods and dashedlines their limits of agreement.
are used. The ^ symbol represents the optimumthreshold. A threshold to the left (right) results in atest with higher specificity (sensitivity) but lowersensitivity (specificity). Nonlinear features showmore regular behavior than ODI3 when differentcut-off points are used to determine thepresence ofOSA.
Figs. 4—6 show the scatter diagram for ODI3,ApEn02 and CTM1, respectively. Regression curves,linear (dashed line) and quadratic (solid line),which best fit the data in a least square sense,are also depicted.We used the regression equationsto derive an AHI from each index. Due to theirbest fitting to the data, the quadratic curves
Figure 7 Bland and Altman plot for measuring agree-ment between AHI derived from the quadratic regressioncurve of ODI3 and AHI from PSG. Solid line represents themean difference between both methods and dashed linestheir limits of agreement.
( p2x2 + p1x + p0) were selected. Bland and Altman
plots, displayed in Figs. 7—9, were subsequentlyused to quantify the agreement between the origi-nal AHI obtained from PSG and these AHI derivedfrom ODI3, ApEn02 and CTM1, respectively. A sys-tematic bias can be shown in the Bland and Altmanplot corresponding to the AHI derived from ODI3(mean difference = 8.1), whereas the AHI from boththe ApEn02 and the CTM1 have no bias (meandifference = 0.0). The limits of agreement (�1.96S.D.) are wide (�34.8 to 51.5), indicating there is agreat lack of agreement between both techniques.The limits of agreement decreased to �37.9 whencomparing the AHI obtained from PSG with that
Figure 9 Bland and Altman plot for measuring agree-ment between AHI derived from the quadratic regressioncurve of CTM1 and AHI from PSG. Solid line represents themean difference between both methods and dashed linestheir limits of agreement.
Improving diagnostic ability of blood oxygen saturation using CTM 21
derived from ApEn02. Furthermore, limits of agree-ment in the Bland and Altman plot corresponding toCTM1 decrease by various events per hour (up to�31.5 e/h) and the number of outliers is lower.
We have also measured the performance of non-linear methods in terms of computational time.ApEn02 and CTM1 were both applied to a commonnocturnal SaO2 record (8 h long) divided in epochs of16.66 min using the same PC (AMD AthlonTM XP 3000with 1 GB RAM). While ApEn02 achieved a meancomputational time of 0.5833 s per epoch, CTM1was 1000 times faster, decreasing computationaltime up to 0.5333 ms.
5. Discussion and conclusions
Portable monitoring has been widely used as analternative technique to PSG in the diagnosis ofOSA syndrome. This study has shown that nonlinearanalysis, and particularly CTM, could enhance thediagnostic capacity of SaO2 signals recorded fromnocturnal oximetry. CTM improved the diagnostictest values of classical indexes commonly derivedfrom SaO2, e.g. ODIs and CT90. Moreover, CTMimproved the results obtained applying other non-linear measures previously studied by our own group[16—18]. We have shown that OSA patients havesignificantly lower CTM values (0.30 � 0.20,mean � S.D.) than OSA negative subjects (0.71 �0.18, mean � S.D.) according to their higher disper-sion in the second-order difference plots. Thus, wecould say that SaO2 signals from OSA patients aremore variable than those from OSA negative sub-jects.
We have divided the indexes under study in threedifferent groups: classical oximetric indexes (ODI2,ODI3, ODI4 and CT90), regularity indexes fromapproximate entropy analysis (ApEn01, ApEn015and ApEn02) and variability indexes from CTM ana-lysis (CTM1, CTM3 and CTM6). Tables 3 and 4 showthat CTM1 reached the best statistics and para-meters obtained from ROC analysis, significantlyimproving the results of classical indexes. CTM1was significantly correlated ( p < 0.0001) with ODIsand regularity indexes from ApEn analysis (r > 0.6).Furthermore, CTM1 showed the highest correlationcoefficient (r = �0.738) with AHI.
The ROC curve analysis also yielded the bestresults for nonlinear methods, and particularly forCTM1. Whereas classical indexes achieved high spe-cificities but very poor sensitivities, nonlinear meth-ods provided sensitivity and specificity values bothgreater than 80%, leading to high accuracies. Highpositive and negative predictive values, as well assmall likelihood ratios, make nonlinear methods
especially useful to help in OSA diagnosis. CTM1reached 90.1% sensitivity, 82.9% specificity, a PPVof 88.5, a NPV of 85.1, a LR+ of 5.27, a LR� of 0.12and an accuracy of 87.5%. The area under the ROCcurve was 0.924, the largest one compared withother indexes.
A significant drawback of nocturnal oximetrycommon to most sleep studies is the substantialnumber of false negative cases. Subjects involvedin these studies are typically referred to the sleepunits because they are suspected of suffering fromsleep apnea. Thus, the population under study has avery high prevalence of the disease, resulting in asmall percentage of patients who test negative witha high chance to be incorrectly classified (falsenegative result). Additionally, due to the high pre-valence, patients with a positive result are morelikely to have a true positive result than a falsepositive result [7]. This leads to a high LR+ but alsoto a high LR�. Many sleep studies used two differentthresholds to achieve both high LR+ and low LR�,with the limitation that patients presenting diag-nostic values between both cut-off points will not beclassified [7]. In the present research, we used asingle threshold. All subjects in the data set could bediagnosed, although both false positive and falsenegative cases were present. However, we achievedsignificant operating characteristics with a singlethreshold, increasing the probability that a patienttesting positive has an abnormal AHI (LR+ > 5.0) anddecreasing the probability that a patient testingnegative has an abnormal AHI (LR� < 0.2).
Previous studies based on classical oximetricindexes [14,23,36,37] provided higher sensitivitybut lower specificity, whereas others [9,38,39]achieved higher specificity but significantly lowersensitivity. On the other hand, our results demon-strate that nonlinear methods provide significantsensitivity and specificity values, leading to highaccuracies and areas under the ROC curve. Thestudy carried out by Olson et al. [40] achievedsimilar sensitivity and specificity values, althoughour results are slightly better. The study by Nuberet al. [11] reached a higher sensitivity (91.8%) andgood specificity (77.8%), but their study was basedon a small sample (40 subjects). The diagnosticaccuracy in terms of ROC analysis varies greatlyamong studies carried out by different researchers.Although these studies were probably developedunder different conditions, the major limitationwhen using ODIs is that each study uses their owndefinition of desaturation. Moreover, the thresholdused to diagnose OSA based on the AHI derived fromPSG usually varies among studies. Our studies areguided to remove these uncertainties by using uni-versally well-defined nonlinear methods. There is
22 D. Alvarez et al.
Table 5 Area under the ROC curve for each diagnosticfeature varying the AHI used to diagnose OSA syndrome
AHI: apnea hypopnea index; ODI2: number of dips inSaO2 � 2%; ODI3: number of dips in SaO2 � 3%; ODI4: numberof dips in SaO2 � 4%; CT90: percentage of time spent withSaO2 below 90%. ApEn01: approximate entropy computedwithm = 1 and r = 0.1; ApEn015: approximate entropy computedwith m = 1 and r = 0.15; ApEn02: approximate entropy com-puted withm = 1 and r = 0.2; CTM1: central tendencymeasurecomputed with r = 1; CTM3: central tendency measure com-puted with r = 3; CTM6: central tendency measure computedwith r = 6.
not a consensus neither in the definition of the AHInor in the threshold subsequently used to determinethe disease [14]. Thus, we assessed the diagnosticability of each feature involved in this study whenthe AHI threshold used to diagnose OSA changed.The AROC was computed taking into account thefollowing OSA thresholds: AHI � 5, 10, 15 and 20.Table 5 summarizes the results from this analysis,showing that CTM achieves the highest area what-ever the AHI threshold used.
Previous studies by our own group applied non-linear analysis to oximetric signals from portablemonitoring in screening for OSA syndrome. Weapplied ApEn to SaO2 signals, obtaining 88.3% sen-sitivity and 82.9% specificity [16]. Moreover, we alsoapplied ApEn to heart rate oximetric signals fromthe same population, obtaining 71.2% sensitivity and78.9% specificity [17]. LZ complexity and CTM werealso applied to SaO2 recordings. Using LZ, weobtained 86.5% sensitivity and 77.6% specificity,while with CTM the sensitivity was 90.1% and thespecificity was 82.9% [18]. In the present work, wehave extended our study applying CTM with differ-ent radii. Furthermore, we compared CTM withclassical oximetric indexes and with ApEn. CTM1significantly improves statistical and diagnostic testresults of previous studies. Moreover, the computa-tional time spent by CTM to process a commonoximetric record (approximately 8 h long) is 1000times smaller than that used by ApEn. Thus, CTM ismore suitable to be incorporated as a software toolin an oximeter. Furthermore, CTM provides a gra-phical tool, the second-order difference plots,
which could be very useful in the diagnosis of OSAsyndrome. Physicians could make a preliminarydiagnosis by visual inspection of these scatter plots.
Furthermore, we studied the ability of CTM toprovide a derived AHI. We plotted CTM1 versus theAHI from PSG and them we computed the linear andthe quadratic regression equations. The AHI derivedfrom the quadratic regression curve showed no biasand moderate agreement with the AHI from PSG,improving the results obtained with classical ODIsand with other nonlinear indexes. Hence, CTM ana-lysis of SaO2 signals from nocturnal oximetry couldprovide useful information in the development ofalternative techniques to conventional PSG.
From our study, we could derive that the recur-renceof theapneaevents typical ofOSAwere respon-sible for the high variability of SaO2 signals in the OSApositive group measured by the CTM. However, it isknown that altered respiratory patterns are notexclusive of OSA syndrome. A total of 42 subjectssuffering COPD were included in our study. We haveshown that COPD patients with OSA presented sig-nificantly higherCTMvalues thanCOPDpatientswith-out OSA. The CTM analysis provided 82.9% specificity,with 13 false positive cases. Six subjects (46%) withinthe false positive group suffered from COPD, whiletwo subjects had a BMI > 34 kg/m2. If COPD patientswere removed from the study, specificity increases to87.5%. Regarding to the OSA positive group, our CTMstudy reached 90.1% sensitivity, with 11 false nega-tive cases. Five patients (45%) within the false nega-tive group had an AHI < 15 e/h. If those patients areremoved from the study, sensitivity increases to 94%.
We should take into account some limitations ofour study. Firstly, regarding to the population understudy, the sample size could be larger. Furthermore,OSA positive patients were predominantly studied.Thus, additional work is needed to apply our meth-odology to a new and larger data set with a widespectrum of sleep-related breathing disorders, aswell as to study groups of especial interest, such ashealthy subjects, young snorers and patients withlung and/or cardiac diseases. Another limitationshould be stated in relation to the applicability ofour methodology. Oximetry signals were recordedsimultaneously with PSG, eliminating potential con-founders such as night to night variability of AHI, aswell as ensuring that oximetry data were collectedin exactly the same environment as the PSG data.However, further analyses using unattended noctur-nal oximetry in home are necessary. In addition, thedata collection process could be enhanced. Ouroximeter takes one sample every 5 s. The studyby Wiltshire et al. [41] showed that low samplingrates provide SaO2 recordings with a low number ofartifacts. However, low sampling frequencies poten-
Improving diagnostic ability of blood oxygen saturation using CTM 23
tially reduce the sensitivity and increase the speci-ficity of a diagnostic test [7]. Although the studycarried out by Warley et al. [29] showed that thissampling frequency provides reasonable resolutionin SaO2 variability, sampling at higher frequencieswe could record SaO2 signals more accurately,improving subsequent analyses. Moreover, a draw-back related to the airflow measure in the PSG studyshould bementioned. Thermistor is highly reliable indetecting static respiratory events (apneas). How-ever, it is less effective with dynamic respiratoryevents (hypopneas) [42]. The use of nasal cannulacan improve the detection of hypopneas [43].Nevertheless, one of the limitations of nasal pres-sure is false positive detection of apneas/hypopneasdue to nasal obstruction or mouth breathing, lead-ing to ambiguous results. The American Academy ofSleep Medicine (AASM) Task Force suggested thatdifferentiation of apneas from hypopneas was notnecessary in clinical practice because both eventtypes share a common pathophysiology and clinicalconsequences [27]. However, the use of nasal can-nula could lead to an overestimation of the AHI dueto the false positive events. On the other hand,thermistor may not be sensitive for detectinghypopneas, leading to an underestimation of theAHI [44]. However, thermistor is the most commonmethod for defining breathing events based on aflow measurement [7]. In the present study, wherethe thermistor is used, the underestimation of thereference standard index leads to increase thenumber of false positive diagnoses when our pro-posed nonlinear methods are used. Based on anunderestimated threshold, these subjects withoutOSA but with an AHI slightly below to this cut-offpoint will test positive, resulting in sensitivity valueshigher than the specificity ones. Table 4 shows thistrend, where sensitivity is higher than specificity inApEn01, ApEn015, ApEn02, CTM1 and CTM6. Never-theless, we obtained high accuracy and area underthe ROC curve, as well as high and low LR+ and LR�values, respectively. Moreover, we could also derivefrom Table 5 that CTM provides the highest AROCregardless of the threshold used to diagnose OSA.Finally, we would like to point out that the ability ofportable monitors in OSA diagnosis has been gener-ally assessed by comparing their results with thoseof the accepted reference standard: the sleeplaboratory-based PSG [45]. However, both PSGand portable monitors have considerable night tonight variability, which accounts for some loss ofagreement between single-night observations ofthe two tests. It is known that AHI by itself haslimited clinical significance, correlating poorly withsymptoms or with outcome of treatment [46].Respiratory events can be totally obstructive
(apneas), partially obstructive (hypopneas) or verysubtle upper airway obstructions which can lead torespiratory effort-related arousals (RERAs). It isknown that RERAs can produce fatigue and daytimesleepiness without a significant number of apneasand hypopneas [47]. Thus, an index includingapneas, hypopneas and RERAs would be a muchmore powerful reference index than AHI to diagnoseOSA and it could detect more effectively those caseswithout major physiological complications.
In summary, we have shown that a nonlinear ana-lysis by means of the CTM could enhance the diag-nostic capacity of oximetric signals recorded fromnocturnal pulse oximetry. CTM could be a usefuldiagnostic tool that improves classical oximetricindexes commonly used by physicians. Second-orderdifference plots could allow physicians to make apreliminary diagnosis by visual inspection, while CTMprovides a quantitative measure from those scatterdiagrams, making both interrelated techniques sui-table to be incorporated in the oximeters.
Acknowledgements
This work has been partially supported by a grantproject from Consejerıa de Educacion de la Junta deCastilla y Leon under project VA108A06 and SOCAL-PAR (Sociedad Castellano-Leonesa y Cantabra dePatologıa Respiratoria).
References
[1] Guilleminault C, Tilkian A, Dement WC. The sleep apneasyndromes. Ann Rev Med 1976;27:464—84.
[2] Guilleminault C, Van Den Hoed J, Mitler MM. Clinical over-view of the sleep apnea syndromes. In: Gilleminault C,Dement WC, editors. Sleep apnea syndromes. New York:Alan R. Liss; 1978.
[3] Douglas NJ. Recent advances in the obstructive sleepapnoea/hypopnoea syndrome. Ann Acad Med 2002;31:697—701.
[4] Young T, Peppard PE, Gottlieb DJ. Epidemiology of obstruc-tive sleep apnea. Am J Resp Crit Care 2002;165:1217—39.
[5] Day R, Gerhardstein R, Lumley A, Roth T, Rosenthal L. Thebehavioral morbidity of obstructive sleep apnea. Prog Car-diovasc Dis 1999;41:341—54.
[6] Agency for Health Care Policy and Research (AHCPR). Sys-tematic Review of the Literature Regarding the Diagnosis ofSleep Apnea: Summary, Evidence Report/Technology Assess-ment. Department of Health and Human Services (U.S.Public Health Service), 1999, No. 1, vol. E002.
[7] Flemons WW, Littner MR, Rowlet JA, Gay P, Anderson WM,Hudgel DW, et al. Home diagnosis of sleep apnea: a systema-tic review of the literature. Chest 2003;124:1543—79.
[8] Epstein LJ, Dorlac GR. Cost-effectiveness analysis of noc-turnal oximetry as a method of screening for sleep apnea—hypopnea syndrome. Chest 1998;113:97—103.
24 D. Alvarez et al.
[9] Sano K, Nakano H, Ohnishi Y, Ishii Y, Nakamura T, MatuzawaK, et al. Screening of sleep apnea—hypopnea syndrome byhome pulse oximetry. Nihon Kokyuki Gakkai Zasshi 1998;36:948—52.
[10] Chiner E, Signes-Costa J, Arriero JM, Marco J, Fuentes I,Sergado A. Nocturnal oximetry for the diagnosis of the sleepapnoea hypopnoea syndrome: a method to reduce the num-ber of polysomnographies? Thorax 1999;54:968—71.
[11] Nuber R, Varvrina J, Karrer W. Predictive value of nocturnalpulse oximetry in sleep apnea screening Schweiz. MedWschr(Suppl) 2000;116:120S—2S.
[12] Chaudhary B, Dasti S, Park Y, Brown T, Davis H, Akhtar B.Hour-to-hour variability of oxygen saturation in sleep apnea.Chest 1998;113:719—22.
[13] Golpe R, Jimenez A, Carpizo R, Cifrian JM. Utility of homeoximetry as a screening test for patients with moderate tosevere symptoms of obstructive sleep apnea. Sleep 1999;22:932—7.
[14] Magalang UJ, Dmochowski J, Veeramachaneni S, Draw A,Mador MJ, El-Solh A, et al. Prediction of the apnea—hypopneaindex from overnight pulse oximetry. Chest 2003;124:1694—701.
[15] Netzer N, Eliasson AH, Netzer C, Kristo DA. Overnight pulseoximetry for sleep-disordered breathing in adults. Chest2001;120:625—33.
[16] Del Campo F, Hornero R, Zamarron C, Abasolo D, Alvarez D.Oxygen saturation regularity analysis in the diagnosis ofobstructive sleep apnea. Artif Intell Med 2006;37:111—8.
[17] Zamarron C, Hornero R, del Campo F, Abasolo D, Alvarez D.Heart rate regularity analysis obtained from pulse oximetricrecordings in the diagnosis of obstructive sleep apnea. SleepBreath 2006;10:83—9.
[18] Alvarez D, Hornero R, Abasolo D, del Campo F, Zamarron C.Nonlinear characteristics of blood oxygen saturation fromnocturnal oximetry for obstructive sleep apnoea detection.Physiol Meas 2006;27:399—412.
[19] Cohen ME, Hudson DL, Deedwania PC. Applying continuouschaotic modeling to cardiac signals. IEEE Eng Med Biol1996;15:97—102.
[20] Burioka N, Cornelissen G, Halberg F, Kaplan DT, Suyama H,Sako T, et al. Approximate entropy of human respiratorymovement during eye-closed waking and different sleepstages. Chest 2003;123:80—6.
[21] Caldirola D, Bellodi L, Caumo A, Migliarese G, Perna G.Approximate entropy of respiratory patterns in panic dis-order. Am J Psychiatry 2004;161:79—87.
[22] Rechtschaffen A, Kales A. A manual of standardized termi-nology, techniques and scoring system for sleep stages ofhuman subjects. Washington, DC: US Government PrintingOffice, National Institutes of Health Publication; 1968.
[23] Zamarron C, Romero PV, Rodrıguez JR, Gude F. Oximetryspectral analysis in the diagnosis of obstructive sep apnoea.Clin Sci 1999;97:467—73.
[24] Zamarron C, Gude F, Barcala J, Rodrıguez JR, Romero PV.Utility of oxygen saturation and heart rate spectral analysisobtained from pulse oximetric recordings in the diagnosis ofsleep apnea syndrome. Chest 2003;123:1567—76.
[25] Zamarron C, Pichel F, Romero PV. Coherence between oxy-gen saturation and heart rate obtained from pulse oximetricrecordings in obstructive sleep apnoea. Physiol Meas2005;26:799—810.
[26] Meoli AL, Casey KR, Clark RW, Coleman Jr JA, Fayle RW,Troell RJ, et al., Clinical practice review committee. Hypop-nea in sleep-disordered breathing in adults. Sleep 2001;24:469—70.
[27] American Academy of Sleep Medicine Task Force (AASM).Sleep-related breathing disorders in adults: recommenda-
tions for syndrome definition and measurement techniquesin clinical research. Sleep 1999;22:667—89.
[28] Global Initiative for Chronic Obstructive Lung Disease(GOLD). Global strategy for the diagnosis managementand prevention of chronic obstructive pulmonary disease,NHI Publication 2701. Bethesda: National Health, Lung andBlood Institute (NHLBI/WHO); 2001. p. 1—100.
[29] Warley ARH, Mitchell JH, Stradling JR. Evaluation of theOhmeda 3700 pulse oximeter. Thorax 1987;42:892—6.
[30] Levy P, Pepin JL, Deschaux-Blanc C, Paramelle B, BrambillaC. Accuracy of oximetry for detection of respiratory dis-turbances in sleep apnea syndrome. Chest 1996;109:395—9.
[31] Gyulay S, Olson LG, Hensley MJ, King MT, Murree Allen K,Saunders NA. A comparison of clinical assessment and homeoximetry in the diagnosis of obstructive sleep apnea. Am RevRespir Dis 1993;147:50—3.
[32] Pincus SM. Assessing serial irregularity and its implicationsfor health. Ann NY Acad Sci 2001;954:245—67.
[34] Pincus SM. Approximate Entropy as a measure of systemcomplexity. Proc Natl Acad Sci USA 1991;88:2297—301.
[35] Jeong J, Gore JC, Peterson BS. A method for determinism inshort time series, and its applications to stationary EEG. IEEETrans Biomed Eng 2002;49:1374—9.
[36] Rodriguez JM, de Lucas P, Sanchez MJ, Izquierdo JL, Peraita R,Cubillo JM. Usefulness of the visual analysis of night oximetryasa screeningmethod inpatientswith suspectedclinical sleepapnea syndrome. Arch Bronconeumol 1996;32:437—41.
[37] Lacassagne L, Didier A, Murris-Espin M, Charlet JP, Chollet P,Leophonte-Domairon ML, et al. Role of nocturnal oximetry inscreening for sleep apnea syndrome in pulmonary medicine:study of 329 patients. Rev Mal Respir 1997;14:201—7.
[38] Ryan PJ, Hilton MF, Boldy DA, Evans A, Bradbury S, Sapiano S,et al. Validation of British Thoracic Society guidelines for thediagnosis of the sleep apnoea/hypopnea syndrome: canpolysomnography by avoided? Thorax 1995;50:972—5.
[39] Brouillette RT, Morielli A, Leimanis A, Waters KA, Luciano R,Ducharme FM. Nocturnal pulse oximetry as an abbreviatedtesting modality for pediatric obstructive sleep apnea.Pediatrics 2000;105:405—12.
[40] Olson LG, Ambrogetti A, Gyulay SG. Prediction of sleep-disordered breathing by unattended overnight oximetry. JSleep Res 1999;8:51—5.
[41] Wiltshire N, Kendrick A, Catterall J. Home oximetry studiesfor diagnosis of sleep apnea/hypopnea syndrome: limitationof memory storage capabilities. Chest 2001;120:384—9.
[42] Farre R, Montserrat JM, Rotger M, Ballester E, Navajas D.Accuracy of thermistors and thermocouples as flow-measur-ing devices for detecting hypopneas. Eur Respir J 1998;11:179—82.
[43] Montserrat JM, Farre R, Ballester E, Felez MA, Pasto M,Navajas D. Evaluation of nasal prongs for estimating nasalflow. Am J Respir Crit Care Med 1997;155:211—5.
[44] Series F, Marc I. Nasal pressure recording in the diagnosis ofsleep apnoea hypopnoea syndrome. Thorax 1999;54:506—10.
[46] Whitelaw WA, Brant RF, Flemons WW. Clinical usefulness ofhome oximetry compared with polysomnography for assess-ment of sleep apnea. Am J Respir Crit Care Med 2005;171:188—93.
[47] Masa JF, Corral J, Martın MJ, Riesco JA, Sojo A, Hernandez M,et al. Assessment of thoracoabdominal bands to detectrespiratory effort-related arousal. Eur Respir J 2003;22:661—7.
