=Paper=
{{Paper
|id=Vol-1419/paper0080
|storemode=property
|title=Is Statistical Learning Affected by Sleep Apnoea?
|pdfUrl=https://ceur-ws.org/Vol-1419/paper0080.pdf
|volume=Vol-1419
|dblpUrl=https://dblp.org/rec/conf/eapcogsci/ArciuliVDOSMWRG15
}}
==Is Statistical Learning Affected by Sleep Apnoea?==
https://ceur-ws.org/Vol-1419/paper0080.pdf
Is Statistical Learning Affected by Sleep Apnea?
Joanne Arciuli
Faculty of Health Sciences, The University of Sydney, Australia
Corresponding author: joanne.arciuli@sydney.edu.au
Andrew Vakulin
Adelaide Institute for Sleep Health, Repatriation General Hospital, School of Medicine, Flinders University, Australia
Woolcock Institute of Medical Research, Sleep and Circadian Research Group, Central Clinical School, The University of Sydney
Angela D’Rozario
Woolcock Institute of Medical Research, Sleep and Circadian Research Group, Central Clinical School, The University of Sydney
Hannah Openshaw
Woolcock Institute of Medical Research, Sleep and Circadian Research Group, Central Clinical School, The University of Sydney
David Stevens
Woolcock Institute of Medical Research, Sleep and Circadian Research Group, Central Clinical School, The University of Sydney
Doug McEvoy
Adelaide Institute for Sleep Health, Repatriation General Hospital, School of Medicine, Flinders University, Australia
Keith Wong
Woolcock Institute of Medical Research, Sleep and Circadian Research Group, Central Clinical School, The University of Sydney
Caroline Rae
Neuroscience Research Australia, University of New South Wales, Australia
Ron Grunstein
Woolcock Institute of Medical Research, Sleep and Circadian Research Group, Central Clinical School, The University of Sydney
Abstract
Introduction
It is thought that sleep is important for learning yet we know
strikingly little about the relationship between sleep and a The brain’s ability to detect statistical regularities in the
form of implicit learning known as statistical learning (SL). environment is known as statistical learning (SL). SL is
SL can be assessed using a distinct familiarisation phase in implicit in that it occurs without any intention to learn,
which participants are exposed to a stream of stimuli that without any form of reinforcement, and without conscious
contains statistical regularities but are not given any awareness (Perruchet & Pacton, 2006). SL has been shown
instruction to learn or any form of reinforcement. This is
followed by a surprise test phase some time later where to operate across a range of modalities and different types of
implicit learning of those statistical regularities is assessed. In stimuli including speech, musical tones, geometric shapes,
the current study we investigated the relationship between scenes, and cartoon figures. With regard to sequentially
sleep and SL across a period that included night time sleep presented stimuli it has been shown to operate across both
with polysomnography (PSG) monitoring. Participants were adjacent and non-adjacent dependencies. SL is present from
47 adults (mean age = 48.79 years, sd = 8.76 years) known to infancy. This powerful learning mechanism is thought to
have impaired sleep due to obstructive sleep apnea. They
underpin a wide variety of critical mental tasks. For a
were exposed to familiarisation in the evening prior to going
to sleep and undertook the surprise test phase the following review of SL research in recent decades see Arciuli and von
evening, approximately 24 hours later (mean overnight sleep Koss Torkildsen (2012).
time = 399.60 mins). Results revealed that, as a group,
participants showed statistically significant SL despite the Lasting learning is desirable in a mechanism underpinning
substantial delay between familiarisation and test phases. mental activity. However, there has been relatively little
Although we found no relationship between individual research investigating the lasting effects of SL. In fact there
differences in SL and clinical measures of sleep apnea
severity, our findings revealed that SL is positively correlated
is only a handful of previous studies that have examined SL
with non rapid eye movement sleep (NREM). over time. Some of these previous studies have been
conducted with infants, while others have examined adults.
Here we restrict our discussion to previous studies of adults.
Keywords: statistical learning; VSL; sleep; sleep apnea;
sleep apnoea; sleep-dependent learning
Before discussing these previous studies we review the
embedded triplet paradigm which is often used to assess SL.
494
on participants’ sleep activity. Direct measurement of sleep
Although SL can be assessed using a number of different activity is required for a better understanding of the link
methods, the embedded triplet task is frequently used to test between sleep and SL.
sensitivity to adjacent dependencies in sequentially
presented stimuli (e.g. Arciuli & Simpson, 2011, 2012; A previous study of healthy adults examined the
Aslin et al., 1991; Brady & Oliva, 2008; Evans et al., 2009). relationship between sleep and SL using polysomnography
This task includes a familiarisation phase where participants (PSG). Durrant et al. (2011) examined SL using sequentially
are exposed to a stream of individually presented stimuli presented auditory stimuli. Their SL task, which was not
containing statistical regularities (in this case, embedded based on the embedded triplet paradigm, was comprised of a
triplets). After this initial phase has been completed distinct familiarisation phase containing statistical
participants undertake a forced-choice test phase where they regularities, followed by an immediate test phase and, later,
are asked to identify regularities that occurred in the a delayed test phase. Experiment 2 included a wake group
familiarisation phase across a number of trials: on each trial and a day time sleep group. The sleep group underwent PSG
they make a decision between an embedded triplet versus a monitoring in a lab – with approximately 4 hours delay
foil triplet and the number of correct responses is tallied as a (mean sleep time = 83.08 mins) between immediate and
percentage of the total number of trials. Although delayed testing. The data revealed some marginal results
participants usually report no conscious sense of familiarity which made it difficult to draw firm conclusions from group
during the test phase, average group performance comparisons in Experiment 2. However, there was a
demonstrates that healthy participants identify embedded significant correlation between the duration of slow wave
triplets at a rate greater than chance (using a one-sample t- sleep (a type of non rapid eye movement sleep, NREM,
test to compare the group average against the chance level known as stage 3) and improvement between immediate and
of 50%). delayed testing within the sleep group. Durrant et al. (2011)
interpreted their findings in terms an active role for sleep in
A study by Kim et al. (2009) investigated SL in healthy the consolidation of learning.
adults using an embedded triplet task with visual stimuli.
When the test phase was administered 24 hours after the A recent study by Nemeth et al. (2012) examined implicit
familiarisation phase participants showed significant levels sequence learning using the alternating serial reaction time
of SL.1 In their study of healthy adults using an embedded task (ASRT) with visual stimuli in 20 adults with
triplet task, with different visual stimuli, Arciuli and obstructive sleep apnea and 20 controls. They found intact
Simpson (2012) manipulated the time between and equivalent learning in both groups. However, that study
familiarisation and a surprise test phase. Specifically, they did not examine learning over time. Csabi et al. (2014) used
incorporated 5 time intervals: 30 mins, 1 hour, 2 hours, and the same ASRT task to assess learning over 10–12 hours
24 hours. Their between-participants design revealed (including overnight sleep) in 17 adults with sleep apnea
statistically significant SL at each time interval and no and 17 controls. Group comparisons indicated that sleep
difference in the magnitude of learning at each interval. disturbance did not affect learning. It is important to note
Like Kim et al. (2009), Arciuli and Simpson (2012) found that neither of these previous studies explored the
that SL can be observed 24 hours after exposure to relationship between individual differences in learning and
statistical regularities. However, Arciuli and Simpson’s individual differences in sleep apnea severity.
(2012) finding regarding equivalent levels of SL at each of
five different time intervals raises questions about whether In summary, previous research examining the link between
sleep plays a role in consolidating SL. sleep and SL has been limited. A study of healthy adults
used PSG but the sleep period was a four hour day time nap.
If sleep plays an important role in consolidating SL we Two other studies have looked at learning in adults with
might have expected to see more learning in the participants sleep apnea. One of these examined immediate learning
who were in the 24 hour delay group in the study reported while the other investigated learning over 10–12 hours of
by Arciuli and Simpson (2012). In previous research on night time sleep. In the current study we sought to
sleep and learning (not specifically SL), it has been investigate the link between sleep and SL over a 24 hour
hypothesised that sleep may play an active role in period in sleep apnea – with participants undergoing PSG
consolidating learning (especially during NREM sleep: e.g., for a full period of night time sleep. In view of previous
Diekelmann & Born, 2010). By contrast, some have research, we expected that we would likely see intact SL
suggested that sleep may play a more passive role by based on group performance, as well as a link between SL
offering an absence of stimulation (e.g., Ellenbogen, Payne and NREM sleep. We were unsure as to whether we would
& Stickgold, 2006). Also, it is noteworthy that neither Kim see a link between SL and clinical measures of sleep apnea
et al. (2009) nor Arciuli and Simpson (2012) collected data severity because, as far as we are aware, our study is the
first to look at these kinds of individual differences in this
1
Kim et al. used a rapid serial visual presentation (RSVP) reaction population.
time test during their test phase rather than a forced-choice test of
familiarity.
495
Method four groups of three (four embedded triplets): ABC, DEF,
GHI and JKL Each triplet appeared in the familiarsation
Participants stream 24 times (i.e., a total of 96 triplets). In 6/24
instances, one character was presented twice in a row (i.e.,
Participants were 47 adults (3 females) who had previously repeated) in order to provide a cover task: participants were
received a clinical diagnosis of sleep apnea (mean age = required to press the space bar of the computer every time
48.79 years, sd = 8.76 years). they saw the same alien appear twice in a row. These
repetitions were counterbalanced among the three characters
Study Protocol within each triplet to ensure that repetitions did not draw
Participants arrived for a two night weekend stay at the participants’ attention to the existence of triplet boundaries.
sleep laboratory at the Woolcock Institute for Medical The order of triplets presented during the familiarisation
Research at 7.00 pm. Following brief patient interview phase was randomised although it was ensured that the same
participants underwent the familiarisation phase of the SL triplet would never appear twice in a row.
task. Following polysomnography (PSG) set up patients
went to bed at an average time of 9:52 pm and were The surprise test phase included 64 trials. For each forced-
awakened at an average time of 5:58 am. The surprise test choice trial participants were presented with two triplets,
phase of the SL task took place approximately 24 hours one following the other: an embedded triplet which had
after familiarisation. occurred during familiarisation and a foil triplet. Each of the
four foil triplets contained one character from three different
Overnight Polysomnography (PSG). Participants had embedded triplets. These foil triplets never occurred in the
overnight PSG with the following recordings: familiarisation phase (AEI, DHL, GKC, and JBF). Thus,
Electroencephalography (EEG: F3, C3, O1, Fz, Cz, Pz, Oz, there was a statistical contrast between embedded versus
F4, C4, O2 lead placements), left and right electro- foil triplets: embedded triplets had high internal transitional
oculograms, submental electromyogram, nasal cannula to probabilities (i.e., B followed A and C followed B with
measure nasal pressure, leg movement sensors, chest and almost perfect certainty), whereas the internal transitional
abdominal motion bands, lead II electrocardiography and probabilities of the foil triplets were 0.2 The presentation
arterial oxygen saturation (finger pulse oximetry). Signals order of the embedded versus foil triplets was
were digitized and stored using Embla Titanium hardware counterbalanced. During test trials each embedded triplet
and REMLogic software sleep system. Sleep and EEG and each foil triplet was seen an equal number of times (16
arousals were scored using standardized criteria (Iber et al., times), and each individual character was seen 32 times.
2007). Apneas were defined as cessations of nasal flow Order of test trials was randomised for each participant.
lasting ≥10 s and hypopneas as a >50% decrease in nasal
flow (or in both thoracic and abdominal excursions) and For each participant the number of correct responses was
associated ≥3% oxygen desaturation or an EEG arousal. divided by the total number of test trials and reported as a
Two clinical indexes of severity were obtained: an apnea percentage.
hypopnea index (AHI); ≥30/hour is considered to be severe
(Ruehland et al., 2009, based on AASM criteria), and an
oxygen desaturation index (ODI). REM and NREM sleep Results
(including a breakdown across stage 1, stage 2, and stage 3)
was calculated as a percentage of total sleep time.
Clinical Measures of Sleep Apnea (AHI and ODI)
Statistical Learning Task. We used the same visual SL On average, the sample had moderate to severe sleep apnea
task originally reported by Arciuli and Simpson (2011), (AHI: 35.35 events/hour, SD = 22.61) with frequent dips in
presented on a laptop via E-Prime software (Schneider, oxygen saturation (ODI: 26.88 events/hour, SD = 21.84).
Eschman & Zuccolotto, 2002). Stimuli were eighteen Participants slept for an average of 399.60 mins (SD =
cartoon-like characters that could not be easily verbalized 45.13) and the average amount of total sleep time spent in
(i.e. did not resemble known animals, people, or popular NREM sleep was 80.68% (remainder was REM sleep).
cartoon characters). Of these, six were used during
instruction and practice leaving 12 for exclusive use during Descriptive Statistics from PSG
the familiarisation phase and subsequent surprise test phase Descriptive statistics for the entire sample of 47 participants
(see Appendix of Arciuli & Simpson, 2011, for stimuli). relating to total sleep time (TST: in mins), percentage of
These 12 characters were divided into four groups of three total sleep time spent in REM, percentage of total sleep time
(i.e. four triplets). spent in NREM and percentage of total sleep time spent in
stages 1/2/3 are presented in Table 1.
The familiarisation phase consisted of a continuous stream
of individually presented characters which appeared on the 2
Each character in each embedded triplet was occasionally
screen for 400 ms each. The characters were divided into repeated in the familiarisation phrase, in order to provide a cover
task. Thus, internal TPs of the embedded triplets was 0.92.
496
sleep apnea that were assessed in the current study showed
Table 1: Descriptive statistics. statistically significant SL approximately 24 hours after
exposure to regularities (a period which included a mean
Mean SD overnight sleep duration of 399.60 mins). This finding is in
TST (mins) 399.60 45.13 line with recent studies using the ASRT task that found
% REM 19.32 6.13 intact implicit sequence learning in participants with sleep
% NREM 80.68 6.13 apnea (Nemeth et al., 2012, who looked at immediate
% stage 1 4.26 2.29 learning; Csabi et al., 2014, who looked at learning over 10–
% stage 2 60.94 10.10 12 hours of night time sleep).
% stage 3 15.48 8.32
Our results revealed no relationship between SL and two
Statistical Learning commonly used clinical measures of sleep apnea severity
(AHI and ODI). Furthermore, there was no relationship
The average SL of the group was significantly greater than
between SL and total sleep time. However, as we had
chance (mean = 55.39, sd = 15.07, t(46) = 2.45, p = 0.018,
hypothesized, we did discover a significant relationship
Cohen’s d = 0.722). There was variability in performance
between SL and percentage of total sleep time spent in
across participants and no evidence of a floor effect.
NREM sleep. Our data revealed that greater SL was
associated with a larger percentage of NREM sleep. There
Relationship between SL and Sleep Variables were no significant relationships between SL and any of the
Bivariate correlational analyses using Pearsons r were
stages within NREM sleep.
conducted in order to examine the relationship between SL
and AHI and ODI, as well as total sleep time (TST: in
More research is needed to investigate why NREM sleep is
mins), percentage of total sleep time spent in REM,
associated with SL in both healthy and sleep impaired adult
percentage of total sleep time spent in NREM, and
populations. In line with Durrant et al. (2011) we expect that
percentage of time spent in stages 1/2/3.
this relationship may point to an active role for sleep in the
consolidation/facilitation of learning (e.g. Diekelmann &
There was no significant relationship between SL and either
Born, 2010). However, we note that while Durrant et al.
of our two measures of sleep apnea severity. For SL and
(2011) observed a relationship between SL and stage 3
AHI we observed r = -0.064, p = 0.668. For SL and ODI we
(slow wave) sleep in their study of healthy adults (which
observed r = -0.066, p = 0.659.
included brief day time sleep), we did not observe a
relationship between SL and stage 3 sleep in the current
There was a positive correlation between SL and NREM, r
study of overnight sleep in participants with sleep apnea.
= 0.297, p = 0.042 (and, due to dependency in the data, an
equivalent negative correlation between SL and REM).
One possible link between SL and NREM sleep is via the
There was no correlation between SL and total sleep time (r
production of sleep spindles. Sleep spindles (waxing and
= -0.019, p = 0.897). There was no significant correlation
waning oscillations of 0.5–3 seconds, usually ranging 11–16
between SL and stage 1 (r = 0.033, p = 0.827), stage 2 (p =
Hz) are thought to play an important role in the maintenance
0.103, p = 0.491), or stage 3 (r = 0.085, p = 0.569).
of NREM sleep continuity (McGinty & Szymusiak, 2011).
Another potential importance of sleep spindles is for new
learning and memory consolidation (Fogel & Smith, 2011).
Discussion Slowing of spindles in sleep apnea patients compared with
healthy controls, predominantly in frontal brain regions, has
Our participants undertook an SL task which used been observed in two studies (Himanen et al., 2003,
sequentially presented visual stimuli. They were exposed to Schonwald et al., 2012). However, sleep spindles are
a familiarisation phase containing statistical regularities generally associated with a particular type of NREM sleep
before undergoing a full night of sleep monitoring via (stage 2 sleep) – our analyses did not reveal a significant
polysomnography (PSG). Approximately 24 hours later they correlation between SL and stage 2 sleep.
were given a surprise test phase to assess implicit learning
of embedded regularities in the familiarisation stream. In conclusion, our results suggest that, on average,
participants with sleep apnea demonstrate intact SL after 24
Our results revealed that statistical learning is remarkably hours and that current clinical measures of sleep apnea
long lasting, even in participants with sleep apnea. Previous severity are not related to individual differences in SL.
studies of healthy adults by Kim et al. (2009) and by Arciuli Individual differences in SL are linked with NREM sleep
and Simpson (2012) observed significant SL some 24 hours and it would be worthwhile exploring this relationship
after participants had been exposed to a familiarisation further using additional techniques such as direct
phase containing embedded statistical regularities. investigation of production of sleep spindles.
Likewise, based on a group average, the participants with
497
Acknowledgments Evans, J., Saffran, J., & Robe-Torres, K. (2009). Statistical
learning in children with Specific Language Impairment.
This research was conceptualised by Arciuli (who presented
Journal of Speech Language and Hearing Research, 52,
her idea for collaborative research at an invited presentation
321–335.
at the Woolcock Institute for Medical Research in 2012).
Arciuli undertook primary responsibility for the data
analysis and writing. Vakulin and D’Rozario supervised Fogel, S. M., & Smith, C. T. (2011). The function of the
data collection and contributed to some sections of the sleep spindle: a physiological index of intelligence and a
writing (listed in order of their contributions). Openshaw mechanism for sleep-dependent memory consolidation.
and Stevens were research assistants who collected data Neuroscience & Biobehavioral Reviews, 35, 1154–1165.
(listed in alphabetical order). The project was supported by
a National Health and Medical Research Council (NHMRC) Himanen, S., Virkkala, J., Huupponen, E., & Hasan, J.
grant awarded to the final four authors (CIA: Grunstein, (2003). Spindle frequency remains slow in sleep apnea
CIB: Rae, CIC: Wong, CID: McEvoy). Arciuli was patients throughout the night. Sleep Medicine, 4, 361–366.
supported by a mid-career fellowship granted by the
Australian Research Council (ARC).
Iber C, Ancoli-Israel S, Chesson A, & Quan S. (2007) for
References the American Academy of Sleep Medicine. 1st ed. The
AASM manual for the scoring of sleep and associated
Arciuli, J., & Simpson, I. (2011). Statistical learning in events: rules, terminology and technical specification
typically developing children: the role of age and speed of Westchester: IL: American Academy of Sleep Medicine.
stimulus presentation. Developmental Science, 14, 464–473.
Kim, R., Seitz, A., Feenstra, H., & Shams, L. (2009).
Arciuli, J., & Simpson, I. (2012). Statistical learning is Testing assumptions of statistical learning: is it long-term
lasting and consistent over time. Neuroscience Letters, 517, and implicit? Neuroscience Letters, 461, 145–149.
113–135.
McGinty, D., & Szymusiak, R. (2011). Neural Control of
Arciuli, J., & Von Koss Torkildsen, J. (2012). Advancing Sleep in Mammals. In Principles and Practice of Sleep
our understanding of the link between statistical learning Medicine. M. H. Kryger, Roth, T. and Dement, W. C. (Eds)
and language acquisition: the need for longitudinal data. Missouri: Elsevier. 76–91.
Frontiers in Psychology – Language Sciences, 3, Article
324.
Nemeth, D., Csabi, E., Janacsek, K., Varszegi, M., & Mari,
Aslin, R., Saffran, J., & Newport, E. (1998). Computation of Z. (2012). Intact implicit probabilistic sequence learning in
conditional probability statistics by 8-month old infants. obstructive sleep apnea. Journal of Sleep Research, 21,
Psychological Science, 9, 321–324. 396–401.
Brady, T., & Oliva, A. (2008). Statistical learning using Perruchet, P., & Pacton, S. (2006). Implicit learning and
real-world scenes. Psychological Science, 19, 678–685. statistical learning: one phenomenon two approaches.
Trends in Cognitive Sciences, 10, 233–238.
Csabi, E., Varszegi-Schulz, M., Janacsek, K., Malecek, N.,
& Nemeth, D. (2014). The consolidation of implicit Ruehland, W., Rochford, P., O'Donoghue, F., Pierce, R.,
sequence memory in obstructive sleep apnea. PLOS One, 9, Singh, P., & Thornton, A. T. (2009). The new AASM
e109010. criteria for scoring hypopneas: impact on the apnea
hypopnea index. Sleep, 32(2), 150–7.
Diekelmann, S., & Born, J. (2010). The memory function of
Schneider, W., Eschman, A., & Zuccolotto, A. (2002). E-
sleep. Nature Reviews: Neuroscience, 11, 114–126.
Prime Reference Guide. Pittsburgh: Psychology Software
Tools Inc.
Durrant, S., Taylor, C., Cairney, S., & Lewis, P. (2011).
Sleep-dependent consolidation of statistical learning.
Neuropsychologia, 49, 1322–1331. Schonwald, S., Carvalho, D., de Santa-Helena, E., Lemke,
N., & Gerhardt, G. (2012). Topography-specific spindle
frequency changes in obstructive sleep apnea. BMC
Ellenbogen, J. M., Payne, J. D., & Stickgold, R. (2006). The Neuroscience, 13, 89.
role of sleep in declarative memory consolidation: Passive,
permissive, active or none? Current Opinion in
Neurobiology, 16, 716–722.
498