Automatic slow eye movement (SEM) detection of sleep onset in patients
with obstructive sleep apnea syndrome (OSAS): Comparison between multiple
sleep latency test (MSLT) and maintenance of wakefulness test (MWT)
Margherita Fabbria, Fabio Pizzaa,b, Elisa Magossoc, Mauro Ursinoc, Sara Contardia,b, Fabio Cirignottaa,b,
Federica Provinia, Pasquale Montagnaa,*
aDepartment of Neurological Sciences, University of Bologna, Via Ugo Foscolo 7, 40132 Bologna, Italy
bUnit of Neurology, S.Orsola-Malpighi Hospital, University of Bologna, 40138 Bologna, Italy
cDepartment of Electronics, Computer Science and Systems, University of Bologna, Bologna, Italy
a r t i c l ei n f o
Received 23 March 2009
Received in revised form 13 May 2009
Accepted 18 May 2009
Available online 16 February 2010
Slow eye movement
Multiple sleep latency test
Maintenance of wakefulness test
Obstructive sleep apnea syndrome
a b s t r a c t
Objective: To determine whether automatic slow eye movement (SEM) analysis performs comparably to
standard sleep onset criteria at the multiple sleep latency test (MSLT) and at the maintenance of wake-
fulness test (MWT) in patients with obstructive sleep apnea syndrome (OSAS).
Methods: We compared sleep latencies obtained upon standard analysis of MSLT and MWT recordings
with automatically detected SEM latencies in a population of 20 severe OSAS patients that randomly
underwent the two tests 1 week apart.
Results: Eight of 20 OSAS patients had EDS as answered by the Epworth Sleepiness Scale (ESS). Mean SEM
latency performed comparably to standard sleep onset in both the MSLT (6.4 ± 5.5 min versus
7.4 ± 5.1 min, p = 0.25) and the MWT (25.2 ± 14.5 min versus 24.4 ± 14.0 min, p = 0.45) settings. Mean
SEM latency significantly correlated with the sleep latency at the MSLT (r = 0.52, p < 0.05) and at the
MWT (r = 0.74, p < 0.001). Finally, the Epworth Sleepiness Scale score correlated with SEM latency at
the MWT (r = ?0.62, p < 0.01), but not at the MSLT.
Conclusions: Automatic SEM detection performed comparably to standard polysomnographic assessment
of sleep onset, thus providing a simplified technical requirement for the MSLT and the MWT. Further
studies are warranted to evaluate SEM detection of sleep onset in other sleep disorders with excessive
? 2010 Elsevier B.V. All rights reserved.
Excessive daytime sleepiness (EDS) is defined as the inability to
stay awake and alert during the major waking episodes of the day
. The prevalence of EDS among adults has been reported as high
as 12% . The significant impact of EDS on car accidents, work
safety and public health has been widely reported in the literature
Objective measurements of EDS have proven problematic. Sev-
eral methods have been used in evaluating EDS, from subjective
scales such as the Stanford Sleepiness Scale  and the Epworth
Sleepiness Scale , that attempt to measure state and trait sleep-
iness, respectively, to objective laboratory based tools that evalu-
ate sleep latency in different conditions. Among the latter are the
multiple sleep latency test (MSLT)  and the maintenance of
wakefulness test (MWT) , the recommended objective tests for
the evaluation of EDS for clinical and experimental purposes .
The MSLT measures a subject’s ability to fall asleep; in each ses-
sion the subject is instructed to ‘‘please lie quietly, assume a com-
fortable position, keep your eyes closed, and try to fall asleep” [1,8].
The MWT measures a subject’s ability to stay awake in a quiet,
non-stimulating situation for a given period of time; in each ses-
sion the subject is instructed to ‘‘please sit still and remain awake
for as long as possible, look directly ahead of you, and do not look
directly at the light” [1,9]. Therefore, the MSLT assumes that sleep-
iness is equivalent to a high proneness to sleep in an appropriate
setting, whereas the MWT considers an inability to maintain wake-
fulness in a boring situation. Each MSLT and MWT session involves
the recording of EEG, EOG and EMG for the detection of sleep onset.
As recommended by the Task Force of the American Academy of
Sleep Medicine , the MSLT is not specifically aimed at the inves-
tigation of sleepiness in patients with OSAS, but represents a major
diagnostic tool for the differential diagnosis of hypersomnias of
central origin (with emphasis on sleep onset Rapid Eye Movement
1389-9457/$ - see front matter ? 2010 Elsevier B.V. All rights reserved.
* Corresponding author. Tel.: +39 051 2092950; fax: +39 051 2092963.
E-mail address: email@example.com (P. Montagna).
Sleep Medicine 11 (2010) 253–257
Contents lists available at ScienceDirect
journal homepage: www.elsevier.com/locate/sleep
[REM] period detection for the identification of narcolepsy). Con-
versely, the MWT is better adapted to evaluate individuals who
must remain awake for public safety reasons (such as truck drivers
or pilots), as well as to assess the response to treatment in patients
with EDS . MSLT and MWT are expensive, all day tests; they re-
quire a laboratory setting that vaguely resembles real life, and they
need the obligatory and dynamic presence of an expert laboratory
Slow eye movements (SEMs) are low frequency (mainly 0.2–
0.6 Hz), rolling, horizontal, bidirectional and conjugate movements
of the eyes . SEMs are a phenomenon typical of the wake–sleep
transition [10–16]. Several studies have correlated SEM appear-
ance with the EEG counterpart [15,17–19] and with the behav-
ioural responses typical of drowsiness . SEM could thus
represent a useful marker of drowsiness, sleep onset and conse-
We already developed an automatic method for the detection of
SEM [21–23] that has been validated against visual scoring on
either 8- or 24-h polysomnographic recordings. Recently, we ap-
plied our technique to MSLT recordings performed in subjects with
normal MSLT and in OSAS patients, disclosing equal performance of
SEM detection versus standard measurements of sleep onset .
The aim of the present study was to perform an automatic SEM
detection on both the MSLT and the MWT recordings of a new set
of patients with severe OSAS (i.e., having an apnea–hypopnea in-
dex [AHI] higher than 30) [25,26], associated or not with EDS, in or-
der to test our automatic method on the MSLT recordings and to
assess its relative performance versus the MWT recordings.
2. Patients and methods
Twenty men with a definite clinical diagnosis of severe OSAS
participated in the study. They were consecutive patients selected
on the basis of the following selection criteria: (1) diagnosis of se-
vere OSAS established by nocturnal portable monitoring (Emblet-
ta?–Embla Systems, Broomfield, CO) performed at the patient’s
home (showing an AHI over 30) according to current diagnostic cri-
teria [25,26]; (2) acceptance of written informed consent to partic-
ipate in the study; (3) absence of other significant medical or sleep
disorders; (4) absence of chronic use of drugs interfering with day-
time alertness; (5) absence of alcohol abuse; (6) no consumption of
more than five caffeinated beverages; and (7) smoking not more
than 10 cigarettes per day. They had a mean age (±standard devi-
ation [SD]) of 54 ± 9 years, a mean body mass index (BMI) of
31.6 ± 3.6 kg/m2and a mean AHI of 48.8 ± 13.6, with an oxygen
desaturation index (ODI) of 49.1 ± 16.6 events per hour of sleep
and a minimum O2saturation of 68 ± 12%.
Daytime sleepiness was not an inclusion criterion per se. The
mean ESS score was 9.7 ± 3.6 (range 4–16), with eight patients hav-
ing an ESS score P11 suggesting EDS (Table 1) [7,28]. MSLT and
MWT were performed according to the practice parameters for
the evaluation of objective sleepiness for research purposes, with
interruptionof single tests after the occurrence of three consecutive
epochs of stage 1 sleep or any other sleep stage (Sustained Sleep La-
tency – SusSL) to avoid interfering with the sleep homeostatic pro-
cess, or after 200(for the MSLT) and 400(for the MWT) if SusSL did
at 10:00, 12:00, 14:00 and 16:00. MSLT and MWT were performed
on two different days, 1 week apart. Patients were randomly as-
ceded by one week of sleep diary (with total sleep time and number
of awakenings per night) as previously described . A polysom-
nographic study was not performed the night before the study.
Sleep latency at MSLT and MWT was scored as the elapsed time
from lights-out to the occurrence of a single epoch of stage 1 sleep
(SL) and of three consecutive stage 1 epochs or any other sleep
stage (SusSL) for each test; individual mean sleep latency was
the average of the four SL and SusSL . Recordings were visually
examined independently by three neurologists trained in sleep
medicine (F.P., S.C. and F.C.) who had to agree on the scoring of
SL and SusSL.
2.1. EOG signal analysis procedure for the SEM
The MSLT and MWT recordings were acquired according to the
recommendations for the MSLT and MWT protocol . Unipolar
electrode montage was used for EOG acquisition as recommended
by Rechtschaffen and Kales . Therefore, the EOG active elec-
trodes were placed 1 cm below and one above the outer cantus
of the left and right eyes, respectively (E1, E2), with the reference
electrode placed on the left mastoid (A1) . The EOG traces were
subsequently exported in European Data Format (EDF?) for auto-
matic SEM detection via computerized EOG analysis according to
the wavelet-based method previously developed [21–23]. The sec-
onds within each 30-s epoch that were occupied by SEM were
automatically calculated according to previous methods [22,24].
Subsequently, in analogy with the epoch time for standard MSLT
and MWT detection of sleep onset, the epochs that were occupied
for P50% of the time by SEM were considered ‘‘SEM events” in each
recording, and a ‘‘SEM latency” was calculated as the elapsed time
from lights off to the first SEM event in each MSLT and MWT
recording. Final individual SEM latency was the average of the four
MSLT and MWT sessions.
Since MSLT and MWT should be interrupted after the occurrence
of three consecutive sleep epochs by the sleep laboratory techni-
cian, we excluded from further analysis single naps that had been
precociously interrupted and any individual mean result that was
not supported by at least 75% (3 out of 4) of correct procedures.
2.2. Statistical analysis
For each MSLT and MWT single session the following data were
considered: SL, SusSL and SEM latency. Thereafter, we calculated
mean values of SL, SusSL and SEM latency for each patient.
Mean SL and SusSL were compared with mean SEM latencies
using the nonparametric Wilcoxon Signed Rank test for paired data
(p < 0.05) separately for the MSLT and the MWT. Mean SL and Sus-
SL were then correlated with the mean SEM latency using Pearson
correlations (p < 0.05) separately for the MSLT and the MWT, and
all the latencies were correlated with the ESS score.
The Bland–Altman plot was performed between the SL and SEM
at the MSLT and MWT, and we also calculated the mean difference
(bias) and the 95% limits of agreement between SL and SEM scoring
criteria at the MSLT and MWT.
2.3. Local ethics board
The study was performed according to the standards of the local
ethics board. All patients gave written informed consent prior to
the study and were informed that all results were confidential
without any legal impact regarding their drivers licenses.
The individual data are reported in Table 1.
3.1. MSLT data
For the whole OSAS population, the mean SL was significantly
shorter than the SusSL (7.41 ± 5.11 min versus 10.43 ± 4.65 min,
M. Fabbri et al./Sleep Medicine 11 (2010) 253–257
p = 0.007). The mean SEM latency was 6.41 ± 5.50 min. SL was
68 min in 70% of the patients, SEM latency in 65% and SusSL in
25%. Mean SEM latency significantly differed from SusSL, but not
from SL at Wilcoxon Signed Rank test (p < 0.001 and p = 0.25,
respectively, for SusSL and SL), and was significantly correlated
with SusSL (r = 0.45, p = 0.044) and SL (r = 0.52, p = 0.017) using
Pearson’s correlation analysis (see Fig. 1 upper plot for correlation
between SL and SEM latency). ESS score did not correlate with SL,
SusSL, or SEM latency. In the Bland–Altman plot, the mean differ-
ence (bias) between SL and SEM scoring criteria was ?1.007 (95%
CI from ?3.432 to 1.419), the 5th limit of agreement was
?11.163, and 95th limit of agreement was 9.150.
3.2. MWT data
Two out of 20 patients were excluded from statistical analysis
because of the occurrence of a technical error in more than a single
nap. Themean SL wassignificantly
(24.42 ± 14.01 min versus 29.99 ± 11.93 min, p = 0.003). The mean
SEM latency was 25.23 ± 14.47 min. SL was 68 min in 17% of pa-
tients, SEM latency in 22% and SusSL in 6%, whereas SL was
P30 min in 39% of patients, SEM latency in 44% and SusSL in
61%. Mean SEM latency significantly differed from SusSL, but not
from SL at Wilcoxon Signed Rank test (p = 0.003 and p = 0.45,
respectively for SusSL and SL), and was significantly correlated
with SusSL (r = 0.8, p < 0.0001) and SL (r = 0.74, p < 0.0001) using
Pearson’s correlation analysis (see Fig. 1 lower plot for correlation
between SL and SEM latency). ESS score significantly correlated
with SL (r = ?0.5, p = 0.03), SusSL (r = ?0.62, p = 0.006), and SEM la-
tency (r = ?0.62, p = 0.006). In the Bland–Altman plot, the mean
difference (bias) between SL and SEM scoring criteria at the
MWT was 0.448 (95% CI from ?4.137 to 5.033), the 5th limit of
agreement was ?18.753, and 95th limit of agreement was 19.648.
Our aim was to compare the performance of automatic SEM
detection against the standard analysis of MSLT and MWT record-
ings in order to test this simplified technique for the routine
assessment of EDS [1,8,9]. Therefore, we studied 20 severe OSAS
patients with standard MSLT, MWT and subjective sleepiness mea-
sures (ESS). We used the research version of MSLT in order to have
comparable MSLT and MWT recording procedures, since each
MWT should be interrupted after the appearance of sustained
sleep and to avoid interfering with the sleep homeostatic process.
Our main findings are as follows: there was a substantial equiv-
alence of the mean SEM latency versus the mean SL, though not
versus the mean SusSL, on both MSLT and MWT recordings; there
was a clear-cut difference between SL and SusSL; and there was a
significant correlation between the ESS and the MWT but not the
We acknowledge some limitations in our study. We explored an
homogeneous population suffering from a sleep-related breathing
disorder; therefore our findings could not apply to different sleep
pathologies and central hypersomnias. Moreover, we detected
SEM latency arbitrarily assuming that 15 s of SEM within a 30 s-
epoch are equivalent to sleep onset; this arbitrary assumption
was justified by its analogy to the usual scoring criteria for SL,
but it needs to be validated by a systematic analysis of an appropri-
ate SEM threshold for sleep onset detection. Indeed, while our SEM
detection of sleep onset provided results comparable to the stan-
dard analysis of MSLT and MWT, it did suffer from a certain vari-
ability, with a bias between SL and SEM of ?1.007 min (95% CI
from ?3.432 to 1.419) at MSLT, and of 0.448 (95% CI from
?4.137 to 5.033) at MWT. That may reflect the arbitrariness of
our SEM latency determination. Such limitations notwithstanding,
we believe that our present data confirm in a new sample of OSAS
patients the results of our previous study on the MSLT recordings
 and extend the reliability of SEM analysis also to the MWT
recordings in a larger population of patients with severe OSAS.
Additionally, we highlight the closer correlation between subjec-
tive sleepiness measurements (ESS) and the MWT, whereas no cor-
relation was found with the MSLT. Such a discrepancy between
MWT and MSLT closely mirrors the different association of low
MSLT scores with subjective sleep propensity (in active and passive
situations), though not with perceived sleepiness, reported in a
large epidemiological setting .
Clinical, polysomnographic and vigilance characteristics of the OSAS population.
ID Age (years)BMI AHIODI SaO2min (%)ESS MSLT (min)MWT (min)
SL SEM LatSusSL SLSEM LatSusSL
This table shows the clinical (age and body mass index [BMI]) and polysomnographic (apnea–hypopnea index [AHI], oxygen desaturation index [ODI], minimal peripheral
oxygen saturation [SaO2min]) features of each OSAS patient, together with subjective (Epworth Sleepiness Score [ESS]) and objective (MSLT, MWT) sleepiness assessment.
[SL, mean sleep latency to the occurrence of the first epoch of sleep stage 1 NREM; SEM Lat, mean sleep latency to the occurrence of the first 30 s epoch occupied by P50% of
slow eyes movements (SEM); SusSL, sleep latency to the occurrence of three consecutive epochs of sleep stage 1 NREM or 1 epoch of another sleep stage; NA, not available].
M. Fabbri et al./Sleep Medicine 11 (2010) 253–257
The present study confirms the reliability of our automatic
method in detecting the onset of sleep in OSAS with only minimal
required equipment. As outlined by other authors [30,31], an auto-
matic, simple and reliable method to disclose EDS may be of crucial
value for routine diagnostic purposes and for the prevention of EDS
consequences. But further studies should find ways to better define
and validate the threshold for SEM sleep onset definition. More-
over, our results also warrant further investigations of the useful-
ness of automatic SEM detection in the hypersomnias of central
nervous system origin.
The authors have reported no conflicts of interest.
 Littner MR, Kushida C, Wise M, Davila GD, Morgenthaler T, Lee-Chiog T, et al.
Standards of practice committee of the American Academy of Sleep Medicine.
Practice parameters for clinical use of the multiple sleep latency test and the
maintenance of wakefulness test. Sleep 2005;28:113–21.
Fig. 1. Scatter-plot graphs of SEM latency versus sleep latency at the MSLT (upper panel) and the MWT (lower panel). This figure depicts scatter-plots graphs of SEM latency
(x-axis) versus sleep latency (y-axis) to the occurrence of the first epoch of sleep stage 1 NREM (SL) at the MSLT (upper panel), and at the MWT (lower panel), together with
the corresponding regression lines.
M. Fabbri et al./Sleep Medicine 11 (2010) 253–257
 Roth T, Roehrs TA. Etiologies and sequelae of excessive daytime sleepiness. Clin Download full-text
 Dement WC, Gelb M. Somnolence; its importance in society. Neurophysiol Clin
 Dinges DF. An overview of sleepiness and accidents. J Sleep Res 1995;4:4–14.
 Akerstedt T, Kecklund G, Gilberg M. Sleep and sleepiness in relation to stress
and displaced work hours. Physiol Behav 2007;92:250–5.
 Hoddes EDW, Zarcone V. The development and use of the Stanford Sleepiness
Scale (SSS). Psychophysiology 1972;9:150.
 Johns MW. A new method for measuring daytime sleepiness: the Epworth
Sleepiness Scale. Sleep 1991;14:540–5.
 Carskadon MA, Dement WC, Mitler MM, Roth T, Westbrook PR, Keenan S.
Multiple sleep latency test (MSLT): a standard measures of sleepiness. Sleep
 Doghramji K, Mitler MM, Sangal RB, Shapiro C, Taylor S, Walsleben J, et al. A
normativestudy of themaintenance
Electroencephalogr Clin Neurophysiol 1997;103:554–62.
 Aserinsky E, Kleitman N. Regularly occurring periods of eye motility and
concomitant phenomena during sleep. Science 1953;118:273–4.
 Liberson WT, Liberson CA. EEG records, reaction times, eyes movements,
respiration, and mental content during drowsiness. Recent Adv Biol Psychiatry
 Rechtshaffen A, Kales AA. Manual of standardized terminology techniques and
scoring system for sleep stages of human subjects. Los Angeles: Brain
Information Service, Brain Research Institute; 1968.
 Hori T. Electrodermal and electro-oculographic activity in hypnagogic state.
 Santamaria J, Chiappa KH. The EEG of drowsiness in normal adults. J Clin
 De Gennaro L, Ferrara M, Ferlazzano F, Bestini M. Slow eyes movements and
 Ogilvie RD. The process of falling asleep. Sleep Med Rev 2001;5:247–70.
 Torsvall L, Akerstedt T. Sleepiness on the job: continuously measured EEG
changes intraindrivers. Electroencephalogr
 Torsvall L, Akerstedt T. Extreme sleepiness: quantification of EOG and spectral
EEG parameters. Int J Neurosci 1988;38:435–41.
of wakefulnesstest (MWT).
electroencephalographic and electro-oculographic changes during the sleep
onset period. Conf Proc IEEE Eng Med Biol Soc 2007;2007:4006–10.
 Ogilvie RD, Mc Donagh DM, Stone SN, Wilkinson RT. Eye movements and the
detection of sleep onset. Psychophysiology 1988;25:81–91.
 Magosso E, Provini F, Montagna P, Ursino M. A wavelet method for automatic
detectionofslow eye movements:
 Magosso E, Ursino M, Zaniboni A, Provini F, Montagna P. Visual and
computerbased detection of slow eye movements in overnight and 24-h
EOG. Clin Neurophysiol 2007;118:1–11.
 Magosso E, Ursino M, Zaniboni A, Gardella E. A wavelet-based energetic
approach fortheanalysis ofbiomedical
electroencephalogram and electro-oculogram.
 Fabbri M, Provini F, Magosso E, Zaniboni A, Bisulli A, Plazzi G, et al. Detection of
sleep onset by analysis of slow eye movements: a preliminary study of MSLT
recordings. Sleep Med 2008 [Epub ahead of print].
 American Academy of Sleep Medicine Task Force. Sleep-related breathing
disorders in adults:recommendations
measurement techniques in clinical research. Sleep 1999;22:667–89.
 American Academy of Sleep Medicine. International classification of sleep
disorders. 2nd ed.: Diagnostic and Coding Manual. Westchester, Illinois USA;
 Pizza F, Contardi S, Mondini S, Trentin L, Cirignotta F. Daytime sleepiness and
driving performance in patients with obstructive sleep apnea: comparison of
the MSLT, the MWT, and a simulated driving task. Sleep 2009;32:382–91.
 Vignatelli L, Plazzi G, Barbato A, Ferini-Strambi L, Manni R, Pompei F, et al.
Italian version of the Epworth Sleepiness Scale: external validity. Neurol Sci
 Kim H, Young T. Subjective daytime sleepiness: dimensions and correlates in
the general population. Sleep 2005;28:625–34.
 Virkkala J, Hasan J, Varri A, Himanen SL, Harma M. The use of two-channel
electro-oculography in automatic detection of unintentional sleep onset. J
Neurosci Methods 2007;163:137–44.
 Wise MS. Objective measures of sleepiness and wakefulness: application to
the real world? J Clin Neurophysiol 2006;23:39–49.
E, UrsinoM, ProviniF,MontagnaP. Waveletanalysis of
apilot study. MedEngPhys
forsyndrome definition and
M. Fabbri et al./Sleep Medicine 11 (2010) 253–257