Comparison of the Indices of Oxyhemoglobin Saturation by Pulse Oximetry in Obstructive Sleep Apnea Hypopnea Syndrome

Original Research SLEEP MEDICINE

Comparison of the Indices of Oxyhemoglobin Saturation by Pulse Oximetry in Obstructive Sleep Apnea Hypopnea Syndrome*

na»

Chen-Liang Lin, Chinson Yeh, PhD; Chen-Wen Yen, PhD; Wu-Huei Hsu, MD, FCCP; and Liang-Wen Hang, MD

Objective8: To comprehensively evaluate the ability and reliability of the representative previously proposed oxyhemoglobin indexes derived automatically for predicting the severity of obstructive sleep apnea hypopnea syndrome (OSAHS). Methods: Patients with a diagnosis of OSAHS by standard polysomnography were recruited from China Medical University Hospital Centre. There were 257 patients in the learning set and 279 patients in the validation set. The presence of OSAHS was defined as apnea-hypopnea index (AHI) > 51h. Three kinds of oxyhemoglobin indexes, including the oxyhemoglobin desaturation index (001), time-domain index, and frequency-domain index, were used. Degrees of severity were AHI > 15/h and AHI > 301h, representing moderate and severe OSAHS. A total of 28 oxyhemoglobin indexes were tested in our study. Results: Among the three kinds of indexes, 001 had a better diagnostic performance than the time-domain and frequency-domain indexes, with the results coincident in the validation set and learning set. For predicting the severity of OSAHS with AHI > 15/h or > 301h, the 001 clinically had the higher correlation with AHI than time-domain and frequency-domain indexes, with sensitivity/specificity achieving 84.0%/84.3% in AHI > 15/h and 87.8%/96.6% in AHI > 301h, respectively. Conclusi01l8: Based on the smaller SEE of the AHI, the 001 had a significantly smaller SEE than the time-domain and frequency-domain indexes. The 001 index provided a high level of diagnostic sensitivity and specificity at different degrees of OSAHS severity. (CHEST 2009; 135:86-93) Key words: apnea hypopnea index; obstructive sleep apnea hypopnea syndrome; polysomnography; pulse oximetry

Abbreviations: AHI = apnea-hypopnea index; AVe = area under the receiver operating characteristic curve; .ilndex =.i index; FFT = fast-Fourier transformation; NPV = negative predictive value; 001 = oxyhemoglobin desaturation index; OSAHS = obstructive sleep apnea hypopnea syndrome; PPV = positive predictive value; ROe = receiver operating characteristic; S30-70 = ratio of the area enclosed in the periodogram within the period 30 to 70 s; Sp02 = oxyhemoglobin saturation by pulse oximetry

Obstructive sleep apnea hypopnea syndrome (OSAHS) is a sleep breathing disorder characterized by recurrent airflow obstruction caused by a total or partial collapse of the upper airway.' Cardiovascular and neuropsychologic morbidity has been demonstrated in untreated sleep apnea.2-4This morbidity, plus a tendency toward an increased risk of auto accidents and a relatively increased mortality, make treatment imperative.t At present, the "gold standard" for a definitive diagnosis of OSAHSis inlabo86

ratorypolysomnography. However, this approach has its limitations: polysomnography is expensive. time consuming, and labor intensive. Oximetry has been one of the more popular American Academy of Sleep Medicine type 4 monitoring techniques used in attempts at screening for sleep apnea in the home. Oxyhemoglobin indexes from pulse oximetry have been used to screen and predict sleep apneahypopnea severity" Not all respiratory events are accompanied by oxyhemoglobin desaturation," and Original Research

this is one of the major limitations of pulse oximetry. For its ease of application, pulse oximetry with more sensitivity indexes could be an effective tool to use clinically. The lack of airflow during apneic or hypopnea periods may lead to recurrent episodes of hypoxemia that can be detected on oxyhemoglobin as fluctuations in oxyhemoglobin saturation by pulse oximetry (Sp02).7 Three kinds of oxyhemoglobin indexes are available to measure this irregular fluctuation, including the oxyhemoglobin desaturation index (ODI), time-domain index, and frequency-domain index proposed previously.8-18Although these quantitative indexes appear to hold more promise than the visual inspection of nocturnal home pulse oximetry, there has been no systematic comparison of their relative utility in the diagnosis of OSAHS. Thus, the applicability of these indexes to the general population remains uncertain. 19 The objective of this study was to comprehensively evaluate the ability and reliability of the representative previously proposed indexes to verify whether or not pulse oximetry can reduce sleep laboratory efforts to diagnose the severity of OSAHS, using an automated digital analysis.

MATERIALS AND METHODS

Study Subjects Patients with a diagnosis of OSAHS by standard polysomnography were recruited from China Medical University Hospital Centre. The subjects in the learning set (257 patients) were enrolled from January 2004 to December 2004, and those in the validation set (279 patients) were collected from January 2005 to December 2005. Clinical data were collected retrospectively. Table 1 shows the clinical features and polysomnographic results. All subjects were assessed with the Epworth sleepiness scale before polysomnography. Apneas were classified as central when "From the Department of Mechanical and Electro-Mechanical Engineering (Drs. Lin, Yeh, and Yen), National Sun Yat-Sen University, Kaohsiung; and Sleep Medicine Center (Dr. Hang), Department of Internal Medicine, and Division of Pulmonary and Critical Care (Dr. Hsu), Department of Internal Medicine, China Medical University Hospital, Taichung, Taiwan. The study was approved by the Medical Research Ethics Committee of the China Medical University Hospital. The study number is DMR 96-IRB-17. This study was supported by a grant from China Medical University Hospital (DMR-97-029). The authors have no conflicts of interest to disclose. Manuscript received January 8, 2008; revision accepted July 14, 2008. Reproduction of this article is prohibited without written permission from the American College of Chest Physicians (www.chestjournal. orglmisc/reprints.shtml).

Correspondence to: Liang-Wen Hang, MD, Sleep Medicine Center, Department of Internal Medicine, China Medical University Hospital, No.2, Yude Bd, North District, Taichung City 404, Taiwan; e-mail: [email protected] DOl: IO.1378/chest.08-0057

www.chestjournal.org

Table l-CUnfeal Characteristic8 of the Subjects and Po~ic Re8Ult8 in the Learning and . .. Validation Set8* Clinical Characteristics Age, yr Body mass index, kglm2 Epworth sleepiness scale AHI, events/h Male/female gender Presence of central apnea event Presence of periodic limb movement disorder

Learning Set (n = 257) 42.7 ± 12.3 26.3 ± 3.9 8.9 ± 4.7 37.6 ± 23.6 209/48 37

Validation Set (n = 279) 44.6 ± 26.8 ± 9.4 ± 36.7 ±

12.4 4.4 5.0 27.0

233/46

33

p Value]

NS NS NS NS NS NS NS

*Data are presented as mean ± SD or No. NS = not Significant. t Between the learning and validation sets.

there was no airflow and no respiratory movement. Periodic limb movements were scored with the periodic, jerking leg movements by bilateral leg electromyograms. Periodic limb movement disorder was defined as periodic limb movements > 51b. Inclusion criteria were age> 16 years and a diagnosis of OSAHS with apnea-hypopnea index (AHI) > 51b. Exclusion criteria were COPD, chronic chest wall disease, and a total recording time of < 3 h. The study was approved by the Medical Research Ethics Committee of the China Medical University Hospital. The study number is DMR 96-IRB-17.

Polysomnography Polysomnography data were recorded using a computerized polysomnographic system (Alice 4; Healthdyne Technologies; Atlanta, GA). The purchase date was September 21, 2001, and the firmware version was Alice Host 1.8.03 (Healthdyne Technologies). This system included a standardized montage: twochannel EEGs (C4IA1, C3/A2), bilateral electro-oculograms, submental electromyogram, bilateral leg electromyograms, and ECG. Sp02 was recorded using a finger probe (935 Oximeter Sensor; Respironics, Murrysville, PAl. The sampling rate of the oximetry was 1 Hz. For pulse rates < 112 beats/min, the averaging calculation is based on a four-beat exponential average for SP02' For pulse rates > 112 beats/min, the averaging is doubled and then redoubled > 225 beats/min. Airflow was measured using oronasal pressure (1287 nasal flow, PTAF 2; Pro-Tech; Pittsburgh, PAl, and respiratory effort was assessed by inductance plethysmography (3240 chest effort sensor adult; Respironics). The stored data were digitized for computer analysis by data analysis software (Matlab; MathWorks Inc; Natick, MA). Artifacts were removed by eliminating all changes ofoxygen saturation between consecutive sampling intervals of> 4%/s, and any oxygen saturation < 50% by an automated algorithm." The raw data were reviewed by an experienced doctor and scored by a sleep technician certified by the Taiwan Sleep Medicine Society separately. Sleep stages were scored according to the criteria of Rechtschaffen and Kales.20 Arousals were defined as episodes lasting 2:: 3 s in which there was a return of IX activity associated with a discernible increase in electromyogram 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 -s 50% 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.20 •2 1 CHEST /135/1/ JANUARY, 2009

87

A

Methods The ODIs include three components, a certain threshold, baseline parameter, and lasting time parameter: ODI(nL(baseline parameterL(lasting time parameter). The om calculated the amount of oxyhemoglobin desaturation below a certain threshold; when n = 3 indicated a 3% decline from baseline, and n = 4 indicated a 4% decline from the baseline, The three different baseline definitions were a mean of all-night oxygen; a mean of the first 3 min of overnight oxygen recordmg'v, and a mean of the top 20% of oxyhemoglobin saturation values over the 1 min preceding the scanned oxyhemoglobin value.8.9 The definition of lasting time parameters was that the oxyhemoglobin desaturation had to continue for more than a certain criterion period. We tested three different lasting time parameters, including 1 s, 3 s, and 5 S.8.11.12 Finally, the om was the total oxyhemoglobin desaturation counts divided by the total recording time (in hours). The four time-domain indexes were minimal or mean nocturnal SP02,13--15 cumulative time spent < 90% or < 80%,5.11.13-15 threshold and fall index in Sp02 to ::; 90%,16 and the delta index (~Index). The dIndex measured the average of absolute differences of Sp02 between successive 12-s intervals.5. 13 The key idea of frequency-domain index was if a peak in the spectrum between the period boundaries 30 s and 70 s was observed, the subject was considered an OSAHS subject. However, if the peak of 30- to 70-s periods from the spectrum was absent, the subject was considered normal. The four indexes (the total area of the periodogram, the ratio of the area enclosed in the periodogram within the period 30 to 70 s [S30-70], the S30-70 with respect to the total area of the periodogram, and the peak amplitude of the periodogram in the period 30 to 70 s) were obtained as shown in Figure 1,17.18 where the periodogram is an estimate of the power spectral density of a signal. A detailed description of the frequency-domain indexes can be seen in the Appendix. In order to realize these indexes for the prediction of the severity of OSHAS, oximetry indexes first analyzed the accuracy and reliability of the detection of OSAHS. Based on the best SEE between AHI and the ODI, time-domain index, and frequencydomain index, we analyzed the sensitivity and specificity of the diagnosis of moderate (AHI 2: 15!h) and severe (AHI 2: 30!h) OSAHS patients. In presenting the results, sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) were all reported. A receiver operating characteristic (ROC) curve was constructed, and the area under the ROC curve (AVC) was calculated. ROC analysis related sensitivity and 1 - specificity. For the ROC curve, the point with the largest sum of sensitivity and specificity was chosen as a threshold. Differences in means of continuous variables were assessed with a Student t test. All data are reported as mean:!:: SD. A two-tailed value of p < 0.05 was considered Significant.

RESULTS

From polysomnography, in the learning set, AHI ;::: 15/h was confirmed in 206 of 257 patients (80,2%) and AHI > 30/h in 139 of 257 patients (54.1%). 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 AHI values ;::: 5 to < 15/h, ;::: 15 to < 30/h, and ;::: 301h, respectively. Allof the SEE, linear regression parameters, correlation. and significance level between AHI and the indexes are shown in Tables 2 and 3. 88

2.5

2.0

I

1.5

PA

<\ ••• _--

1.0

-- -

0.5 0.0 10

100

1000

B 0.00.5 0.004

r~

0,002 0.001 0.000 10

100

1000

PerIDd(IIC)

FIGURE 1. Diagram of frequency-domain indexes. Top, A: typical spectrum of an OSAHS subject. Bottom, B: typical spectrum of a normal subject. 17.18 PA = peak amplitude the periodogram in the period 30 to 70 s (sec); S = S30-70 with respect to the total area of the periodogram; STOTAL = total area of the periodogram.

or

Figure 2 is the plot between AHI and three kinds of indexes. The top three have a smaller SEE, and the bottom three have a larger SEE in this group. ODI indexes had a better diagnostic performance than time-domain and frequency-domain indexes. Figures 3 shows the Roe curves when the oxyhemoglobin index was ODI3_1M20P_3, ~Index, and S30-70. respectively, at different OSAHS thresholds. In the group with AHI ;::: 301h, the Ave of ODI3_1M20P_3, ~Index, and S30-70 were 0.95, 0.94, and 0,91, respectively. In AHI ;::: 15/h, the Ave of these three indexes were 0.90, 0.87, and 0.84. The results showed that performance was better for all three kinds of oxyhemoglobin indexes when the OSAHS threshold was AHI ;::: 301h, and also showed that the ODI indexes had a better performance than the time-domain and frequencydomain indexes. Figures 4 shows Bland-Altman plots when the oxyhemoglobin index was ODI3_ IM20P_3, ~Index, and S30-70, respectively. Original Research

Table 2-Linear Regression Parameters Between AHI and Oxyhemoglobin aDI Indices in the Learning Set Index"

SEE

Slope

Intercept

Correlation

p Value]

ODI3_A_I ODI3_A_3 ODI3 _A_5 ODI4_A_I ODI4 A 3 ODI4_A_5 ODI3_3M _I 00I3_.1M _3 ODI3 _ 3M_S ODI4_3M _I ODl4 _ 3M _:3 ODI4 _ 3M _ ,5 ODI3 _ IM20 _ I ODIL IM20P_3 O0I3 _ IM20P _ 5 00I4 _IM20P _I 00I4 _ IM20P _ 3 0014 _ IM20P _ 5

11,94 11.95 12.16 12.79 12.90 13.22 105,205 14.03 13.57 14.15 13.60 13.53 9.29 9.19 9.20 10.20 10.38 10.68

1..5.52 1.609 1.688 1.680 1.734 1.834 0.827 0.934 0.995 0.965 1.043 1.093 1.019 1.079 1.120 1.100 1.1051 1.189

16.317 17.013 17.768 20.054 20.558 21.071 17.183 16.719 17.011 20.408 20.392 20.835 1l.899 12.683 13.579 17.165 17.749 18.442

0.86 0.86 0.86 0.84 0.84 0.83 0.76 0.80 0.82 0.80 0.82 0.82 0.92 0.92 0.92 0.90 0.90 0.89

< 0.001 < 0.001 < 0.001

*Oefinition of terms: A = mean of all-night oxygen; 3M = mean of first 3 min of overnight oxygen recording; IM20 oxyhemoglobin values over the 1 min preceding the scanned oxyhemoglobin. tCorrelation and p value were between indices and AHI.

The results between the validation set and the learning set were coincident, as shown in Tables 4, 5. In the validation set, ODI indexes still had a better performance than the time-domain and frequencydomain indexes. In the group with AHI ~ 30/h, the xuc of ODI3_1M20P_3, AIndex, and S30-70 are 0.94,0.93, and 0.90, respectively. In AHI ~ 15/h, the Aues of these three indexes are 0.90, 0.89, and 0.83. The diagnostic performance was also better when the OSAHS threshold was AHI ~ 30/h than AHI ~ 15/h for all three kinds of oxyhemoglobin indexes. Tables 4, 5 show the Aue and the optimal cutoff of the Roe curve at different OSAHS thresholds in the learning set and validation set.

< 0.001

< 0.001 < 0.001 < 0.001

< 0.001 < 0.001

< 0.001 < 0.001

< 0.001

< 0.001 < 0.001 < 0.001 < 0.001 < 0.001

< 0.001

= mean

of top 20% of

DISCUSSION

Polysomnography is the "gold standard" for a definitive diagnosis of OSAHS. Pulse oximetry has also been proposed as a useful diagnostic and screening tool for OSAHS in previous studies,/i-Is We analyzed 28 oxyhemoglobin indexes for predicting the severity of OSAHS automatically, and compared the differences of the ODI, time-domain index, and frequency-domain index. The results indicated that the ODI indexes have a Significantly stronger correlation and a better diagnostic performance than others. Sp02 data were collected as part of a standard laboratory polysomnography, and this signal

Table 3-Linear Regression Parameters Between AHI and Oxyhemoglobin Time Domain and Frequency Domain Indices in the Learning Set* Index Time domain CTOO CT80 DI90 Spa2 minimum Spa2 mean A Index Frequency domain S30-70 S-total PA S

SEE

Slope

16.32 21.36 14.66 17.29 16.88 12.88

2.273 6.551 1.335 - 1.666 - 9.196 16.645

18.71 19.84 19.32 18.74

823.387 5.205E-05 0.003 1.375E-04

Correlation t

p Value]

27.189 34.174 25.430 172.573 915.557 10.465

0.72 0.43 0.78 - 0.68 - 0.79 0.84

< 0.001 < 0.001 < 0.001 < 0.001 < 0.001 < 0.001

- 158.786 - 104.518 23.663 - 52.335

0.61 0.54 0.57 0.61

< 0.001 < 0.001 < 0.001 < 0.001

Intercept

*CT90 = cumulative time spent below 90%; CT80 = cumulative time spent below 80%; 0190 Figure I legend for expansion of abbreviations. tCorrelations and p valnes were between indices and AHI. www.chestjournal.org

= threshold and

fall index in Sp02 to :s 90%. See

CHEST /135/1/ JANUARY, 2009

B9

a

b 0

0 0

0

o ooeR:>

~

~

~ "

~

0

C

y-I664Sx+IO645 00 0

~\,~09 o

0

y-823.387x.)58.786

#';,

cPo

~o

00

0

0

~

0

~ "

0

0

..

...

OM

o

0

~

0

:<

.

CTBO

...

'"~

co

ii"

o

Ut

S..."

'" f o

~ 00 sg~OO<9 OOO~c9 0 o 00 o 00 0

..

0

...........

• +-i~"""""----'-""""'--r-"""""'--r--, --~-_

s....

2. The plot between AHI and oxyhemoglobin indexes, including ODI in the learning set.

could be obtained by the pulse oximetry. Therefore, pulse oximetry could reduce sleep laboratory efforts to screen more severe OSAHS with ODI. V sing the oxyhemoglobin desaturation indexes, the baseline parameter could improve the predicted performance. The reason why ODI had a better diagnostic performance was that ODI was calculated by local relative fragmentation. Every saturation sequence was checked, and the baseline was adapted for different moments. Most authors5 •11- 16,22 have not given a baseline definition of ODI features in their studies. Gyulayet allO reported in a patient with AHI ~ 151h that ODI3 had a sensitivity of 51.0% when the baseline definition was the mean of the first 3 min of recording only. Vazquez et alB found that the top 20% of Sp02 values over the 5 min preceding the scanned value can yield a higher correlation 0.97, and Chiner et al9 showed that mean saturation in the previous 1 min can yield a higher sensitivityand specificityof 63.0% and 96.0%. Based on these data and ours, the baseline parameter defined as the upper percentage of the moving window average could improve the diagnostic performance with ODI. However, the lasting parameter, of which Vazquez et alB required at least three consecutive falls in recorded Sp02 readings, and Teramoto et al12 reported drops in Sp02 lasting at least > 5 s, did not improve the predicted effect in our study. This meant most oxyhemoglobin desaturation was at least > 5 s. In the time-domain indexes, the 4Index is more correlated with AHI than other time-domain in90

00

• +-.,.......Lr--T~--r---.--T"""""""1r--T-,

d

FIGURE

0

dexes. Almost all of the time-domain indexes were calculated by the global absolute value. The 4Index had a better diagnostic performance among timedomain indexes because this index was calculated by local difference, but the drawback was the fixed window size and nonoverlapping window way. Levy et al13 and Magalang et al5 reported that the correlation between the 4Index and AHI was 0.72 and 0.77, which is less than the 0.84 of our study. The difference may be due to the different population in terms of age and greater obesity in their study (age, 56.0 ye,ars and 48.9 years vs our 42.7 years; and body mass index, 32.0 kglm2 and 32.3 kglm2 vs our 26.3 kglm2). The frequency-domain indexes did not correlate with ARI more than ODI did. The drawback of the frequency-domain indexes was that they were only calculated by a global absolute value, so they could not present the respiratory event precisely. The higher correlation in the study by Zamarron et al17.1B was 0.74, sensitivity was 94.0%, and specificity was 65.0% on AHI ~ 101h, which was better than in our study. However, the AHI of the OSAHS and nonOSAHS groups in the study by Zamarron et aP7.HI were quite different: 40.1 ± 23.01h and 2.2 ± 2.71h, respectively. There are several limitations to this study. First, the selected OSARS threshold for the subjects may have affected the sensitivity and specificity results. As seen in Tables 4, 5, ODI indexes had a larger AVC when the OSAHS threshold was from AHI ~ 151h to ARI ~ 301h. The reason for this is that OriginalResearch

.--~-:--=oa-_. _ _.-tJI

a

•__././ .

... ,.",-~,...

..,;.

~r

OSAHS lllrel'hoid

G 0-0 AHI :-IS ih G-B-El AHI- 30 ih

cutoff value:

' 08

0

0

0

0eo

OJ

.~o

•

\

o Ql8.0

·20

0

+19650

------~-------~-------0 0 0 00 oell 0 0 18 "'4

~

I

7~O

cutoff value: 18.17

02

'Sl «!

!

AHI>-30 AUG: 0.95

02 -

J

0

20

.8

AUG: 0.90

T

0 0

i

AHI>~15

-T--

.

j'

(

08

b

a

0 0

0

0

~oP ~

0

0

Memlo

0

0

-

0 0

o oOf 0 ~ o~s:'oo 0 ~ O(<:p 0000 00 8

00

0

0

0

__(!____ !! ____ -'fJ>"--- _;I.:~5O o

~

-18 ~4

0

--r---r "T-r"'-'--1--r--'J 04

06

08

I-specificity

1

_.----..c..:.rf]

...fr-J ' . . r.-or- '•. / -

(r._ ../ ./

• -

. -

I ...OJ

i-

aSAHS dueBhold GO-OAHI,""1~41 AHI ,.co 30 Itt

G-B-E1

AHl '-15 AUe 087

cutoffvalue 1.035

Deltaindex

02

AHI >=30 AUe 094 cutoffvalue:l.~06

-

G O-OAHI ,....I~ih

G-B-E1 AHI;- 30ih

,~

'R.. 04

AHI >=15 AUC: 0.84

cutoffvalue 0 ~31

AHI>=30 AUC 091

02

cu'offvahle:0.~3~

1

40

o

o

o

+196SD -------------a---------o °lb e~75 00 0 0

08 0

~

0

000

~oe qj~

j

00 ~

o

o o

U ~

~

0019

61 o

0 0

o

_______________ 9

~

'20

o

o

It

aSAHS due-hold

&.08

10

..

~ .,.~

roo

-40

0

00

co

Mean o

o

o -196SD o _

...

r:-../, ..

/-......,.,.--"",,-

C

eo

60

~

08

......

...EJ _--------S-r . - . -

40

averlllte of AHImidYDol'.......

q

c

20

08

08

I-specificity

~

04

02

-~75

o 02

04

08

l-specificity

08

FIGURE 3. ROC curves of the oxyhemoglobin indexes for the diagnosis of 05AH5 in the learning set. Top, a: ODI3_1M20P_3. Center, b: ~lndex. Bottom, c: 530-70.

www.chestjournal.org

20

40

so

m'erageof AHI midYS30,1O

60

100

FIGURE 4. Bland-Altman plot in the learning set. Top, a: 'YOI;>1\-IM20P_3 = 1.079 X (ODI3_1M20P_3) + 12.683. Center, b: 'Y.:1Index = 16.645 X (~Index) + 10.645. Bottom, c: 'YS30-70 = 823.387 X (530-70) + 1,58.786.

CHEST/135/1/ JANUARY, 2009

91

Table 4-AUC of HOC Cu",e by Oxyhemoglobin Indices and Optimal Cutoff of the HOC Cu",e at Different OSAHS Thresholds in the Learning Set (n = 257)* AHI Cutoff of 30!h Variables AUC Sensitivity, % Specificity, % PPV, % NPV,%

I

ODI3_1M20P_3

0.95 87.8 (80.0-95.3) 96.6 (94.7-100.0) 96.8 (94.7-100.0) 87.0 (77.5-95.8)

~

AHI Cutoff of 15!h I

530-70

Index

ODI3_1M20P_3

0.91 81.3 (77.4-86.2) 91.5 (87.6-96.9) 91.9 (87.9-96.7) 80.6 (73.7-87.2)

0.94 84. 9 (76.8-91.6) 92.4 (87.5-98.5) 92.9 (87.4-98.2) 83.9 (78.3-91.3)

0.90 84.0 (77.7-92.1) 84.3 (75.2-93.1) 95.6 (93.2-98.1) 56.6 (48.2-70.3)

~

Index

0.87 70.9 (65.3-76.9) 92.2 (84.5-99.9) 97.3 (95.5-99.8) 43.9 (32.5-54.4)

530-70 0.84 63.1 (59.0-72.4) 98.0 (90.2-100.0) 99.2 (94.2-100.0) 39.7 (27.7-53.9)

*Data are presented as mean (95% confidence interval).

when AHI was lower, such as < 201h, the difference in the oximetry indexes was relatively smaller, as seen in Figure 2. Second, not all of the respiratory events were accompanied by oxyhemoglobin desaturation: 78.0% of apneas and 54.0% of hypopneas caused desaturation," which generated a false-negative error. In addition, oxyhemoglobin desaturation drops during the night were not always consistent with respiratory events, which generated a falsepositive error. The number and duration of drops were influenced by many factors, including incorrect placement of the probe, contact problems with the probe, and line trouble in the pulse oximetry and body movements, which created artifacts in the SP02' We used an artifact-free recording time to do digital processing, but we could not guarantee all of the artifacts were handled. One way to improve this problem was to increase the pulse oximetrysampling rate, such as using 10 to 25 Hz for the next steps.P Third, the oximetrysignal alone could not determine whether the subject was sleeping or not because the oxyhemoglobin indexes were obtained from the total recording time, not the total sleeping time. Hence, if the sleep efficiency of the subject was low, the error between the oxyhemoglobin indexes and AHI would increase. Therefore, if this method is to be applied to clinical practice, it would be better to measure the nocturnal Sp02 at least two times.24,25 In conclusion, we analyzed 28 representative oxyhemoglobin indexes totally to predict the pres-

ence and severity of OSAHS. This study provided a systematic comparison of the relative utility of these quantitative indexes in the diagnosis of OSAHS with large number of subjects. The results revealed that the ODI index is a more appropriate index than the time-domain and frequency-domain indexes and provides a high-level diagnostic performance. Clinically, pulse oximetry could be a helpful abbreviated testing modality for moderateto-severe OSAHS. APPENDIX: CALCULATION OF FREQUENCY-DOMAIN INDEXES The key idea of frequency-domain indexes is that if a peak in the spectrum between the period boundaries 30 s and 70 s is observed, the subject is considered as an OSAHS subject. 17· 18 However, if the peak of 30- to 70s periods from the spectrum is absent, the subject is considered as normal. The spectrum was calculated by the Fast-Fourier transformation (FFT) method. FFT of the signal inherently assumes that the data we have are a single period of a periodically repeating waveform with an infinite number of samples; however, we had a finite number of samples of a signal that was randomly cut between two points in time. This originates an artifact known as spectral leakage, which is due to the discontinuities that appear at the beginning and the end of the Signal. To avoid spectral leakage, Sp02 was windowed by initially multiplying it by a Hamming window. The Hamming window is shown as follows: w]n] = 0.54 - 0.46 cos(2'7Tll!N) for n = 1,2,3, ... , t\ - 1 Then the power spectrum was analyzed using the FFT of the

Table 5-AUC by Oxyhemoglobin Indices and Optimal Cutoff of the HOC Curve at Different OSAHS Thresholds in the Validation Set (n = 279)* AHI Cutoff of 30!h Variables AUC Sensitivity, % Specificity, % PPV,% NPV,%

I

ODI3_1M20P_3 0.94 88.2 (83.1-93.2) 89.6 (83.4-94.6) 88.8 (84.5-93.1) 89.0 (82.1-94.6)

~

AHI Cutoff of 15 !h

Index

0.93 80.7 (75.9-86.0) 92.4 (88.4-96.9) 90.8 (84.0-95.8) 83.7 (77.9-88.8)

530-70

Il

0.90 80.7 (69.0-91.2) 88.2 (83.4-93.6) 86.5 (77.2-93.8) 83.0 (77.1-91.9)

I

ODI3_1M20P_3 0.90 85.3 (75.2-93.9) 85.3 (78.5-94.2) 94.1 (89.3-97.7) 68.1 (57.6-81.1)

~

Index

0.89 79.4 (72.7-85.8) 84.0 (78.2-93.7) 93.1 (88.7-97.9) 60.0 (52.6--£8.6)

530-70 0.83 61.3 (59.5--£8.7) 98.7 (96.4-100.0) 99.2 (97.0-100.0) 48.4 (43.0-57.5)

*Data are presented as mean (95% confidence interval). 92

Original Research

Hamming-windowed signal. Power spectra show the power density or squared magnitude of the amplitude in each of the frequency components of the signal in the bandwidth defined by the interval: fmin

= M = fslN

fmax = I/fy where linin is the lower frequency boundary, f max is the upper frequency boundary, ,if is the frequency resolution, fs is the frequency of sampling, and N is the total number of points sampled. The periodogram was obtained by substituting frequency with period in the spectrum. T(s) = I/frequency (Hz)

The four indexes in the periodogram ,30 to 70 s (the total area of the periodogram, S30-70, the S30-70 with respect to the total area of the periodogram, and the peak amplitude of the periodogram in the period 30 to 70s) were obtained as shown in Figure 1, top, A.

REFERENCES Caples SM, Gami AS, Somers VK. Obstructive sleep apnea. Ann Intern Med 2005; 142:187-197 2 Lavie P, Herer P, Hoffstein V. Obstructive sleep apnea syndrome as a riskfactorfor hypertension. BMJ2000;320:479-482 3 Lanfranchi P, Somers VA. Obstructive sleep apnea and vascular disease. Respir Res 2001; 2:315-319 4 Malhotra A, White DP. Obstructive sleep apnea. Lancet; 2002; 360:237-245 .'l Magalang UJ, Dmochowski J, Veeramachaneni S, et al. Prediction of the apnea-hypopnea index from overnight pulse oximetry. Chest 2003; 124:1694-1701 6 Ayappa I, Rappaport BS, Norman RG, et aI. Immediate consequences of respiratory events in sleep disordered breathing. Sleep Med 2005; 6:123-130 7 Epstein LJ, Dorlac GR. Cost-eflectiveness analysisof nocturnal oximetry as a method of screening for sleep apneahypopnea syndrome. Chest 1998; 113:97-103 8 Vazquez J, Tsai W, Flemons W, et aI. Automated analysis of digital oximetry in the diagnosis of obstructive sleep apnea. Thorax 2000; ,')5:302-307 9 Chiner E, Signes-Costa J, Arrieiro JM, et aI. Nocturnal oximetry for the diagnosis of the sleep apnea hypopnea syndrome: a method to reduce the number of polysomnographies? Thorax 1999; ,')4:968-971 10 Gyulay S, Olson LG, Hensley MJ, et aI. A comparison of clinical assessment and home oximetry in the diagnosis of

www.chestjoumal.org

11 12 13 14

obstructive sleep apnea. Am Rev Respir Dis 1993; 147: 50-53 Golpe R, Jimenez A, Carpizo R, et al. Utility of home oximetry as a screening test for patients with moderate to severe symptoms of OSA. Sleep 1999; 22:932-937 Teramoto S, Matsuse T, Fukuchi Y. Clinical significance of nocturnal oximeter monitoring for detection of sleep apnea syndrome in the elderly. Sleep Med 2002; 3:67-71 Levy P, Pepin JL, Deschaux-Blanc C, et al. Accuracy of oximetry for detection of respiratory disturbances in sleep apnea syndrome. Chest 1996; 109:395-399 Herer B, Roche N, Carton M, et aI. Value of clinical, functional, and oximetric data for the prediction of obstructive sleep apnea in obese patients. Chest 1999; 116:15371544

15 Choi S, Bennett LS, Mullins R, et aI. Which derivative from overnight oximetry best predicts symptomatic response to nasal continuous positive airway pressure in patients with obstructive sleep apnea? Respir Med 2000: 94:89S-899 16 Urschitz MS, Wolff'], Voneinem V, et aI. Reference values for nocturnal home pulse oximetry during sleep in primary school children. Chest 2003; 123:96-101 17 Zamarron C, Romero PV, Rodriguez JR, et al. Oximetry spectral analysis in the diagnosis of obstructive sleep apnea. Clin Sci 1999; 97:467-473 18 Zamarron C, Gude F, Barcala J, et al. Utility of oxygen saturation and heart rate spectral analysisobtained from pulse oximetric recordings in the diagnosis of sleep apnea syndrome. Chest 2003; 123:1567-1576 19 Jaeschke R, Guyatt GH, Sackett DL. User's guides to the medical literature: III. How to use an article about a diagnostic test B; what are the results and will they help me in caring for my patients? The Evidence-Based Medicine Working Group. JAMA 1994; 271:703-707 20 Rechtschaffen A, Kales A. A manual of standardized terminology scoring system for sleep stages of human subjects. Washington, DC: Public Health Service, US Government Printing Office, 1968 21 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 1999; 22:667-689 22 Deegan PC, McNicholas WT. Predictive value of clinical feature for the obstructive sleep apnea syndrome. Eur Respir J 1996; 9:117-124 23 Iber C, Ancoli-Israel S, Chesson AL [r, et aI. AASM manual for scoring sleep. Westchester, IL: American Academy of Sleep Medicine, 2007 24 Douglas NJ, Thomas S, Jan MA. Clinical value of polysomnography. Lancet 1992; 339:347-350 25 Al-Jahdali HH. Obstructive sleep apnea: unjustified diagnostic challenges. Ann Saudi Med 2000; 20:24-28

CHEST1135 I 1 I JANUARY, 2009

93