www.elsevier.com/locate/sleep

Sleep Medicine 9 (2008) 511–516

Original Article

Residual subjective daytime sleepiness under CPAP treatmentin initially somnolent apnea patients: A pilot study

using data mining methods

Xuan-Lan Nguyen a, Dominique Rakotonanahary a, Joel Chaskalovic b,e,Carole Philippe c, Chantal Hausser-Hauw d, Bernard Lebeau a, Bernard Fleury a,*

a Unite de Sommeil – Service de Pneumologie, Hopital Saint-Antoine, Paris, Franceb Laboratoire de Modelisation en Mecanique, Universite Pierre et Marie Curie, Paris VI, France

c Explorations Fonctionnelles, Hopital Tenon, Paris, Franced Explorations Neurophysiologiques, Hopital Foch, Suresnes, France

e University Center of Samaria, Ariel, Israel

Received 1 January 2007; received in revised form 3 July 2007; accepted 26 July 2007Available online 24 October 2007

Abstract

Background and purpose: Despite correct treatment with positive airway pressure (PAP), obstructive sleep apnea (OSA) patientssometimes remain subjectively somnolent. The reliability of the Epworth Sleepiness Scale (ESS) has been established for healthysubjects and patients under stable conditions; the ESS may eventually vary among treated OSA patients, biasing the results of across-sectional analysis of persisting sleepiness. The objective of this study was to depict the evolution of subjective vigilance undertreatment using an index of ESS variability (DESS).Methods: In 80 OSA patients (apnea–hypopnea index [AHI] = 54 ± 26/h), initially somnolent (ESS = 15 ± 3) and treated withauto-titrating PAP (APAP) (oxyhaemoglobin desaturation index 3% [ODIAPAP] = 3.4 ± 2.2/h; daily APAP use = 5.3 ± 1.5 h) dur-ing 434 ± 73 days, ESS scores were regularly collected four times every 109 ± 36 days. DESS was calculated and data mining meth-ods (Segmentation and Decision Tree) were used to determine homogeneous groups according to the evolution of ESS scores.Results: When assessed cross-sectionally, 14–25% of the subjects were recognized as somnolent, depending on the moment whenESS was administered. Using data mining methods, three groups were clearly identifiable: two without residual somnolence – group1, n = 38 (47%), with high DESS = �2.9 ± 0.8, baseline ESS = 16.3 ± 3.3, AHI = 58.5 ± 26.1/h, mean ESSAPAP = 5.1 ± 2.4 andgroup 2, n = 31 (39%), with low DESS = �1.1 ± 0.5, baseline ESS = 13.2 ± 1.4, AHI = 53 ± 27.3/h, mean ESSAPAP = 8.8 ± 1.9;and one with persisting sleepiness; group 3, n = 11 (14%), with low DESS = �0.3 ± 0.8, baseline ESS = 16.3 ± 3,AHI = 38.7 ± 10.8/h, mean ESSAPAP = 14.1 ± 1.9. Compliance to PAP was high and comparable in the three groups. Age and bodymass index (BMI) did not differ.Conclusion: Data mining methods helped to identify 14% of subjects with persisting sleepiness. Validation needs to be done on alarger population in order to determine predictive rules.� 2008 Published by Elsevier B.V.

Keywords: Obstructive sleep apnea; Auto-titrating positive airway pressure; Residual excessive daytime sleepiness; Epworth Sleepiness Score;Epworth Sleepiness Score variation; Data mining methods

1389-9457/$ - see front matter � 2008 Published by Elsevier B.V.

doi:10.1016/j.sleep.2007.07.016

* Corresponding author. Tel.: +33 1 0149282498; fax: +33 10149282331.

E-mail address: bernard.fleury@sat.ap-hop-paris.fr (B. Fleury).

1. Introduction

Intermittent obstruction of the upper airways dur-ing sleep associated with excessive daytime sleepiness

512 X.-L. Nguyen et al. / Sleep Medicine 9 (2008) 511–516

(EDS) is a common pathology affecting approximately2% of the middle-aged population who require a treat-ment with nasal continuous positive airway pressure(CPAP) [1]. Nasal CPAP is a well established and evi-dence-based therapy for these patients, usually leadingto rapid improvement of diurnal symptoms [2,3].Therefore, there is an increasing demand for accessto effective treatment. To deal with the differencebetween demand and capacity [4], new strategies havebeen proposed for diagnosis and treatment using asimplified procedure based on an index derived fromovernight oximetry [5] and auto-CPAP [6–8].

Despite an efficient and effectively used positive air-way pressure therapy, some obstructive sleep apnea(OSA) patients, initially sleepy, remain somnolent undertreatment. EDS is a major issue which needs to be recog-nised because of its harmful consequences, in particularwith regard to risk for traffic accidents [9,10]. However,given the dramatically increasing number of CPAPprescriptions – up to 30,000 new prescriptions every yearin France – the standard practice of systematicallymaking objective measurements (multiple sleep latencytest [MSLT] or maintenance of wakefulness test[MWT]) [11] to assess changes in vigilance under treat-ment conditions has clear limitations. To screen forresidual EDS, the use of the Epworth Sleepiness Scale(ESS) [12] has been proposed. Due to its ease ofadministration, the ESS is used in the majority of sleepclinic studies [13,14] to evaluate the clinical benefit ofCPAP therapy and as a baseline in validation studiesfor molecules stimulating wakefulness [15,16]. Studiesof residual EDS with CPAP have typically used only asingle ESS value to define patients’ somnolence status[15,16], but the stability of ESS scores has only beenvalidated with healthy subjects or with subjects with asleep disorder under circumstances when daytimesleepiness is expected to remain constant. In the case ofapnea patients whose vigilance is expected to improveunder CPAP, the subjective evaluation of their state ofvigilance ‘‘in the usual way of life in recent times’’ [17]is potentially influenced by their general health status,sleep hygiene or treatment compliance during the daysand weeks preceding assessment [18]. ESS scores canconsequently vary transiently around the threshold of11, which is most commonly used to define the limit ofnormality [19]. Moreover, a significant improvement invigilance has already been described in the long-termon CPAP-treated patients [20]. Using a simple cross-sec-tional analysis could lead to a misclassification ofsomnolent patients. We, therefore, made the assumptionthat several consecutive measurements of the ESS arenecessary to define the evolution of vigilance under treat-ment and allow an optimal recognition of authenticallysomnolent patients.

The objective of the present study was to assess in theconditions of everyday clinical practice the evolution of

subjective vigilance on OSA patients from initiallysleepy to correctly treated with auto-titrating PAP(APAP). Repeated ESS scores were used to monitorsubjective vigilance over a period of time.

2. Materials and methods

2.1. Patients

The study was conducted on a prospective patientcohort of OSA patients, diagnosed during a full-nightpolysomnography in a hospital (apnea–hypopnea index[AHI] P 10/h). Patients were eligible for evaluation ofvigilance under treatment if they were initially somno-lent (ESS score > 11) and accepted APAP treatment. Aminimum follow-up of 12 months and mean dailyAPAP use above or equal to three hours per nightthrough the period of follow-up were required, as ithas been shown that cognitive performance and daytimefunction could be improved even at this low level ofCPAP use [3].

According to the recommendations of the AmericanAcademy of Sleep Medicine concerning the initiation ofAPAP treatment, patients did not have any of the follow-ing pathologies: congestive heart failure, significant lungdisease (e.g., chronic obstructive pulmonary disease),daytime hypoxemia and respiratory failure from anycause, or prominent nocturnal desaturation other thanfrom OSA (e.g., obesity hypoventilation syndrome) [21].

The study was approved by the Local Medical EthicsCommittee.

2.2. Procedures

Patients were fitted with the CPAP masks on the dayfollowing the polysomnography diagnosis; the proce-dure was identical for all patients. After a training ses-sion in the hospital, consisting of an explanation ofthe pathology and the treatment, the mask was selected,and patients were given a 30-min practice session withthe CPAP device.

The positive airway pressure device in each case wasan auto-titrating positive airway pressure device, BoraAdaptec� (Taema, Antony, France). Variations in pres-sure levels were based on the detection of apneas, hyp-opneas, airflow limitations and episodes of rebreathingconsidered to be micro-arousals. Initially, minimumand maximum pressures were empirically set at 6 and12 cm H2O, respectively.

Home treatment was carried out by a technicianspecializing in home-care medical appliances con-tracted from the service provider. Following a routineprocedure, patients were systematically visited by atechnician on days 8, 15 and 21. On each visit, CPAPmemory readings (compliance, leaks) were collected.The mask was changed in cases of leakage or discom-

Consecutive patientswith an AHI 10 / h on full-night PSG

(n = 199)

Somnolent patients ESS > 11 (n = 125)

Non-somnolent patients ESS ≤ 11 (n = 74)

Declined APAP treatment immediately

(n = 16)

Received APAP treatment (n = 109)

Withdrew Non-eligible: between 3 and 12 months mean daily APAP of follow-up (n = 10) use < 3 hours (n = 19)

80 patients with APAP treatment had at least a 12 month- follow-up

Fig. 1. Details of the enrollment precedure.

X.-L. Nguyen et al. / Sleep Medicine 9 (2008) 511–516 513

fort, and a heated humidifier was provided in cases ofnasobuccal dryness.

Overnight oximetry recording (PalmSAT� 2500,NONIN Medical, Inc. Plymouth, MN, USA) was per-formed on two consecutive nights during the first threemonths of treatment after verifying that the patientwas sufficiently comfortable. Mechanical efficacy wasdefined by an oxyhaemoglobin desaturation index(ODI) < 10/h; in case of an ODI P 10/h, the range ofpositive pressures was modified and additional visitswere made in order to control treatment efficacy [22,23].

ESS and effective APAP use, determined from thedevice memory readings, were systematically collectedduring scheduled visits: visit 1 on the third, visit 2 onthe sixth, visit 3 on the ninth, and visit 4 on the 12thmonth.

Subjective residual EDS was defined by ESS > 11.Compliance with APAP treatment was expressed by

the number of hours of mean daily effective APAP useand by the percentage of effective APAP use per week.

2.3. Data analysis

The analysis was retrospective on data collectedprospectively.

The main outcome variable was the mean ESS scorevariation (DESS) across the full study period, defined asfollows:

DESS ¼ Mean ESS variation

¼ 1

4

X4

i¼1

ðEPWÞi � ðEPWÞi�1

� �

Two groups were defined, according to the median valueof DESS: one group with ‘‘low DESS’’ (DESS 6 median(DESS)) and one group defined by complementarity, with‘‘high DESS’’ (DESS > median (DESS)).

Data mining is an activity of information extraction,whose goal is to discover hidden or a priori unknownfacts contained in databases. Using a combination ofmachine learning, statistical analysis, modeling tech-niques and database technology, data mining finds pat-terns and subtle relationships in data and infers rulesthat allow the prediction of future results. We appliedthe data mining Segmentation method that uses Deci-sion Tree learning [24] (Clementine Software, SPSSInc., USA) to our database to identify homogeneouspatterns of ESS evolution inside the two groups previ-ously cited (‘‘low DESS’’ and ‘‘high DESS’’). The mainparticular explicative variables were age, gender, bodymass index (BMI), AHI, baseline ESS, the four ESS val-ues collected during the follow-up visits and APAP usemeasured at each visit. Descriptive statistical analysisof the homogeneous groups was performed using theStudent’s t-test and the v2 test. A p value < 0.05 wasconsidered significant.

3. Results

3.1. Patient selection: the study sample

Eighty patients who met the criteria for recruiting outof a population of 199 consecutive patients diagnosed ashaving obstructive sleep apnea syndrome (OSAS) werefinally selected. Details of the enrollment procedureare shown in Fig. 1.

The study sample was comprised of 69 men and 11women, of mean age 55 ± 12 years, BMI 30 ± 5 kg/m2,AHI 54 ± 26 events per hour, and baseline ESS score15 ± 3. Under APAP treatment, mean ODI (ODIAPAP)recorded over two nights was 3.4 ± 2.2/h, mean ESSAPAP

was 8 ± 4, and mean average daily compliance over thefollow-up period was 5.3 ± 1.5 h/night.

From the third month until the end of the follow-up,four ESS measurements were collected with a meaninterval of 109 ± 36 days. Mean follow-up time was434 ± 73 days.

On visit 1, 20/80 (25%) of subjects had an ESS > 11(14 ± 2) with a mean daily CPAP use of 5.12 ± 1.44h/night.

On visit 2, 14/80 (18%) subjects had an ESS > 11(14 ± 3) with a mean daily CPAP use of 5.12 ±1.25h/night.

On visit 3, 11/80 (14%) subjects had an ESS > 11(14 ± 2) with a mean daily CPAP use of 5.37 ± 1.79h/night.

514 X.-L. Nguyen et al. / Sleep Medicine 9 (2008) 511–516

Finally, on visit 4, 15/80 (19%) subjects had anESS > 11 (15 ± 2) with a mean daily CPAP use of5.32 ± 1.77 h/night.

3.2. Applying the data mining method to the database

The Segmentation method using Decision Tree learn-ing led to the identification of three homogeneous sub-groups: two groups of ‘‘no longer somnolent’’ subjects(groups 1 and 2) and one group of subjects ‘‘with per-sisting sleepiness’’ (group 3).

Group 1 (n = 38). No residual EDS (mean ESSAPAP =5.1 ± 2.4). Subjects were characterized by a high ESSvariability (high DESS = �2.9 ± 0.8). Mean ESS at base-line was 16.3 ± 3.3, mean ESS on visit 1 was 5.9 ± 3.6,mean ESS on visit 2 was 4.9 ± 2.7, mean ESS on visit3 was 4.8 ± 2.7, and mean ESS on visit 4 was 4.7 ± 2.5.

Group 2 (n = 31). No residual EDS (mean ESSAPAP =8.8 ± 1.9). Subjects were characterized by a low DESS

(1.1 ± 0.5). Mean ESS at baseline was 13.2 ± 1.4, meanESS on visit 1 was 9.4 ± 3.1, mean ESS on visit 2 was8.2 ± 2.7, mean ESS on visit 3 was 8.9 ± 2.9, and meanESS on visit 4 was 8.8 ± 2.3.

Group 3 (n = 11). With residual EDS (mean ESSAPAP =14.1 ± 1.9). Subjects were characterized by a low DESS

(�0.3 ± 0.8). Mean ESS at baseline was 16.3 ± 3, meanESS on visit 1 was 14 ± 2, mean ESS on visit 2 was13.8 ± 1.9, mean ESS on visit 3 was 13.6 ± 2.4, andmean ESS on visit 4 was 15 ± 2.6.

3.3. Characteristics of patients in the 3 groups (see Table 1)

Subjects in the three groups did not differ in terms ofage, BMI, treatment efficacy and compliance to CPAP.In group 1, despite high baseline ESS, all subjects hadrecovered normal vigilance after three months of CPAPtreatment. Subjects in group 2 were significantly lesssomnolent at baseline; eventually one individual ESSvalue could be found higher than or equal to 11 duringfollow-up, but on the whole, mean ESS on CPAP wereclearly under 11 and showed little variation throughoutfollow-up. Despite high ESS at baseline, subjects in

Table 1Mean characteristics of subjects in the 3 groups

Age (yrs) BMI (kg/m2) AHI (per hour) Baseli

Group 1 (n = 38) 54.1 ± 12.9 30.2 ± 5.1 58.5 ± 26.1 16.3 ±Group 2 (n = 31) 55.2 ± 10.2 29.4 ± 4.2 53 ± 27.3 13.2 ±Group 3 (n = 11) 59 ± 13.5 30.1 ± 4.1 38.7 ± 10.8 16.3 ±p <0.05�

ns� <0.05� <0.05ns� ns ns* <0.05

� Group 1 compared to group 3.� Group 2 compared to group 3.* Group 1 compared to group 2.

group 3 had significantly less severe OSAS accordingto AHI; all ESS values remained higher than 11 for allsubjects throughout follow-up.

4. Discussion

In a population of 80 initially sleepy patients, cor-rectly treated with APAP with an acceptable compli-ance, data mining methods allowed us to identify14% of authentically somnolent subjects, who hadremained sleepy from the end of the third month oftreatment and throughout their follow-up. Between14% and 25% of subjects were somnolent, on the basisof cross-sectional analysis carried out at differenttimes. Thus, apart from the ‘‘always somnolent’’ indi-viduals, there were some subjects who may haveshown one ESS value above 11 once during their fol-low-up and who consequently contributed to inflatingthe number of somnolent subjects under a cross-sec-tional analysis.

We chose to assess vigilance in OSAS subjects whowere initially somnolent, as it is in patients with day-time symptoms that CPAP treatment has proven to bethe most effective [25] on daytime functioning and oncardiovascular risk [26]. In our study, sleepy patientswere recruited on one initial value of the ESS, asthe reliability of the ESS score has been confirmedin a population of sleep-disordered breathing patients[27].

We chose to evaluate somnolence after three monthsof CPAP treatment. We estimated that the efficacy, reg-ularity and stability of CPAP treatment might havebecome firmly established at the end of the first threemonths for most of the patients [20,28].

It needs to be emphasized that these results wereobtained for patients who used an auto-CPAP, withoutinitially fixing the pressure level in the laboratory. Thisunconventional treatment is nevertheless largely usedin common practice [6]. The effectiveness of auto-CPAPas an alternative method of CPAP titration or for con-tinual routine use has already been reported in theresearch literature [7,29–31]. In a recent study by Masa

ne ESS DESS ODI APAP(per hour)

Daily APAPuse (h/night)

WeeklyAPAP use (%)

3.3 �2.9 ± 0.8 2.9 ± 1.7 5.6 ± 1.6 82.8 ± 24.51.4 �1.1 ± 0.5 3.1 ± 2.1 5.5 ± 1.5 84.5 ± 22.53 �0.3 ± 0.8 2.8 ± 1.6 5.4 ± 1.9 81.4 ± 19.4

*

� ns ns ns

X.-L. Nguyen et al. / Sleep Medicine 9 (2008) 511–516 515

et al. [8], no difference was found in the mechanical effec-tiveness of nasal CPAP, the objective compliance totreatment or the improvement of vigilance as measuredby the ESS, when comparing fixed pressure to autoad-justed CPAP. In our patients, the effectiveness of APAPtreatment was clearly confirmed by the data from repet-itive nocturnal oximetry recordings under ventilation,with an average ODI = 3.4 ± 2.2/h. Oximetry hasindeed proven to be a convenient and accurate meansof detecting clinically relevant upper airway obstruc-tions [22,23,32,33]. One could reasonably think that thisODIAPAP implies a satisfying correction of the nocturnalabnormal oxyhaemoglobin desaturations thought to beresponsible for cardiovascular risk. We are neverthelessconscious that some inspiratory flow limitations result-ing in micro-arousals may have been underestimated[34], which could lead to persisting sleepiness and justifya full-night polysomnography reevaluation.

Because the results of a cross-sectional analysis,between 14% and 25% of somnolent subjects, could seemsimplistic, we found it more interesting to use data min-ing methods for a retrospective assessment. An impor-tant use of the data mining method is to be able toplace categorical variables into classes, and eventuallyto predict the category of categorical data by buildinga model based on some predictive variables. Thanks todata mining methods, three different kinds of patternsin the evolution of vigilance could be described. Subjectswho would remain somnolent (group 3) were typicallyhighly somnolent at baseline and had less severe OSAS,and their ESS under treatment did not vary much; theyremained somnolent despite compliance to CPAP whichwas quite compelling (5.4 ± 1.9 hours per night,81.4 ± 19.4% of days per week). On the contrary, therewas another group of subjects (group 1) also highly som-nolent at baseline but with severe OSAS, whose ESSdecreased rapidly under treatment; this group could beconsidered to be no longer somnolent after three monthsof CPAP. Finally, a third subgroup was identified (group2), comprised of subjects with severe OSAS, moderatelysomnolent at baseline, whose ESS decreased also but inless significantly. Compliance with CPAP was equallysatisfactory in the three groups (indeed mean CPAPuse exceeded five hours per night in groups 1, 2 and 3[18] and could not, therefore, be an explanation for suchdifferences in the evolution of subjective vigilance).

The data mining approach seemed to us to be moreinformative than a single cross-sectional study of vigi-lance at a particular moment of the follow-up; we arenevertheless conscious that the number of subjects is rel-atively small and that studies on larger populations areneeded in order to establish predictive rules. Knowingthat a patient will remain somnolent can help to antici-pate and can lead to a complete reevaluation, lookingfor an associated and before unrecognised diagnosis; apolysomnography looking for respiratory effort-related

arousals and eventually objective vigilance tests shouldbe programmed without delay. Inversely, any furtherexam should be considered dispensable in the case of arapid normalization of ESS scores. Patients of group 2(tending less rapidly towards improvement) may becompatible with those subjects who need time toimprove [20] and need not be urgently reevaluated.

In conclusion, a merely cross-sectional analysis of theESS in a population of treated patients without takinginto account the evolution of ESS would only imper-fectly reflect the impact of treatment on subjective day-time vigilance. Several values of the ESS, over anextended interval of time associated altogether with ameasurement of the trend (DESS) could give a betterinsight into the patients’ results, and certainly predictthe long-term evolution of vigilance. This method needsnevertheless to be validated prospectively on a largernumber of patients, in order to determine a predictiveequation, based on the values of baseline and thirdmonth ESS, the strongest predictive variables, and theinitial severity of the sleep apnea syndrome having beentaken into account.

References

[1] Stradling JR, Davies RJO. Obstructive sleep apnea/hypopneasyndrome: definitions, epidemiology, and natural history. Thorax2004;59:73–8.

[2] Lamphere J, Roerhs T, Wittig R, Zorick F, Conway W, Roth T.Recovery of alertness after CPAP in apnea. Chest1989;96:1364–73.

[3] Engleman HM, Martin SE, Deary IJ, Douglas NJ. Effect ofcontinuous positive airway pressure treatment on daytime func-tion in sleep apnoea/hypopnoea syndrome. Lancet1994;343(8897):572–5.

[4] Flemons WW, Douglas NJ, Kuna ST, Rodenstein DO, WheatleyJ. Access to diagnosis and treatment of patients with suspectedsleep apnea. Am J Respir crit Care Med 2004;169(6):668–72.

[5] Mulgrew AT, Fox N, Ayas NT, Ryan CF. Diagnosis and initialmanagement of obstructive sleep apnea without polysomnogra-phy. Ann Intern Med 2007;146:157–66.

[6] Berry RB, Parish JM, Hartse KM. The use of auto-titratingcontinuous positive airway pressure for treatment of adultobstructive sleep apnea. An American Academy of Sleep Med-icine review. Sleep 2002;25(2):148–73.

[7] Hukins C. Comparative study of autotitrating and fixed-pressureCPAP in the home: a randomized, single-blind crossover trial.Sleep 2004;27(8):1512–7.

[8] Masa JF, Jimenez A, Duran J, Capote F, Monasterio C, MayosM, et al. Alternative methods of titrating continuous positiveairway pressure. A large multicenter study. Am J Respir Crit CareMed 2004;170:1218–24.

[9] Teran-Santos J, Jimenez-Gomez A, Cordero-Guevara J. Theassociation between sleep apnea and the risk of traffic accidents.Cooperative Group Burgos-Santander. N Engl J Med1999;340(11):847–51.

[10] George CF. Reduction in motor vehicle collisions followingtreatment of sleep apnea with nasal CPAP. Thorax2001;56:508–12.

[11] Littner MR, Kushida C, Wise M, Davila DG, Morgenthaler T,Lee-Chiong T, et al. Standards of Practice Committee of the

516 X.-L. Nguyen et al. / Sleep Medicine 9 (2008) 511–516

American Academy of Sleep Medicine. Practice parameters forclinical use of the multiple sleep latency test and the maintenanceof wakefulness test. Sleep 2005;28(1):14–5.

[12] Douglas NJ. Why am I sleepy? Am J Respir Crit Care Med2001;163(6):1310–3.

[13] Patel SR, White DP, Mahotra A, Stanchina ML, Ayas NT.Continuous positive airway pressure therapy for treating sleepi-ness in a diverse population with obstructive sleep apnea, Resultsof a meta-analysis. Arch Intern Med 2003;163:565–71.

[14] Kapur VK, Baldwin CM, Resnick HE, Gottlieb DJ, Nieto FJ.Sleepiness in patients with moderate to severe sleep-disorderedbreathing. Sleep 2005;28:472–7.

[15] Kingshott RN, Vennelle M, Coleman EL, Engleman HM,Mackay TW, Douglas NJ. Randomized, double-blind, placebo-controlled crossover trial of modafinil in the treatment of residualexcessive daytime sleepiness in the sleep apnea/hypopnea syn-drome. Am J Respir Crit Care Med 2001;163(4):918–23.

[16] Schwartz JR et alfor the United States Modafinil in OSA StudyGroup. Modafinil as adjunct therapy for daytime sleepiness inobstructive sleep apnea: a 12-week, open-label study. Chest2003;124:2192–9.

[17] Johns MW. A new method for measuring daytime sleepiness: theEpworth Sleepiness Scale. Sleep 1991;14:540–5.

[18] Weaver TE, Maislin G, Dinges DF, Bloxham T, George CFP,Greenberg H, et al. Relationship between hours of CPAP use andachieving normal levels of sleepiness and daily functioning. Sleep2007;30(6):711–9.

[19] Punjabi NM, Bandeen-Roche K, Young T. Predictors of objectivesleep tendency in the general population. Sleep 2003;26(6):678–83.

[20] Meurice JC, Paquereau J, Neau JP, Caron F, Dore P, Ingrand P,et al. Long-term evolution of daytime somnolence in patientswith sleep apnea/hypopnea syndrome treated by continuouspositive airway pressure. Sleep 1997;20:1162–6.

[21] Littner M, Hirshkowitz M, Davila D, Anderson WM, KushidaCA, Woodson BT, et al. Practice parameters for the use of auto-titrating continuous positive airway pressure devices for titratingpressures and treating adult patients with obstructive sleep apneasyndrome. An American Academy of Sleep Medicine report.Sleep 2002;25(2):143–7.

[22] Douglas NJ, Thomas S, Jan MA. Clinical value of polysomnog-raphy. Lancet 1992;339(8801):1113.

[23] Whitelaw WA, Brant RF, Flemons WW. Clinical usefulness ofhome oximetry compared with polysomnography for assessmentof sleep apnea. Am J Respir Crit Care Med 2005;171(2):188–93.

[24] Witten IH, Frank E. Data mining: practical machine learningtools and techniques with Java implementations. San Fran-cisco: Morgan Kaufmann Publishers; 2000.

[25] Barbe F, Mayoralas LR, Duran J, Masa JF, Maimo A,Montserrat JM, et al. Treatment with continuous positive airwaypressure is not effective in patients with sleep apnea but nodaytime sleepiness. a randomized, controlled trial. Ann InternMed 2001;134(11):1015–23.

[26] Pepperell JC, Ramdassingh-Dow S, Crosthwaite N, Mullins R,Jenkinson C, Stradling JR, et al. Ambulatory blood pressureafter therapeutic and subtherapeutic nasal continuous positiveairway pressure for obstructive sleep apnoea: a randomisedparallel trial. Lancet 2002;359(9302):204–10.

[27] Johns MW. Reliability and factor analysis of the EpworthSleepiness Scale. Sleep 1992;15(4):376–81.

[28] McArdle N, Devereux G, Heidarnejad H, Engleman HE, MackayTW, Douglas NJ. Long-term use of CPAP therapy for sleepapnea/hypopnea syndrome. Am J Respir Crit Care Med1999;159:1108–14.

[29] Hudgel DW, Fung C. Sleep a long-term randomized, cross-overcomparison of auto-titrating and standard nasal continuousairway pressure. Sleep 2000;23(5):645–8.

[30] Massie CA, Mc Ardle N, Hart RW, Schmidt-Nowara WW,Lankford A, Hudgel DW, et al. Comparison between automaticand fixed positive airway pressure therapy in the home. Am JRespir Crit Care Med 2003;167(1):20–3.

[31] West SD, Jones DR, Stradling JR. Comparison of three ways todetermine and deliver pressure during nasal CPAP therapy forobstructive sleep apnoea. Thorax 2006;61(3):226–31.

[32] Levy P, Pepin JL, Deschaux-Blanc C, Paramelle B, Brambilla C.Accuracy of oximetry for detection of respiratory disturbances insleep apnea syndrome. Chest 1996;109:395–9.

[33] Netzer N, Eliasson AH, Netzer C, Kristo DA. Overnight pulseoximetry for sleep-disordered breathing in adults: a review. Chest2001;120(2):625–33.

[34] Guilleminault C, Stoohs R, Clerk A, Cetel M, Maistros P. Acause of excessive daytime sleepiness. The upper airway resistancesyndrome. Chest 1993;104(3):781–7.