The Time Structure of the Cyclic Alternating Pattern During Sleep Raffaele Ferri, MD1; Oliviero Bruni, MD2; Silvia Miano, MD1; Giuseppe Plazzi, MD3; Karen Spruyt, PsyD, PhD4; David Gozal, MD5; Mario G. Terzano, MD6 1 Sleep Research Centre, Department of Neurology I.C., Oasi Institute (IRCCS), Via Conte Ruggero 73, 94018 Troina, Italy; 2Centre for Pediatric Sleep Disorders, Department of Developmental Neurology and Psychiatry, University of Rome “La Sapienza”, Rome, Italy; 3Department of Neurological Sciences, University of Bologna, Bologna, Italy; 4Department of Cognitive and Physiological Psychology, Free University of Brussels, Brussels, Belgium; 5 Kosair Children’s Hospital Research Institute, and Division of Pediatric Sleep Medicine, Department of Pediatrics, University of Louisville, Louisville, KY; 6Sleep Disorders Centre, Department of Neurology, University of Parma, Parma, Italy
ference between stages was found in children, who showed a shift of the peak in their SWS histogram toward intervals shorter than in adults. Interval sequences were not determined by a random process in both groups. The Markovian analysis showed statistically significant lower values of entropy and higher values of time dependency, mostly in adults during SWS. Conclusions: The different CAP components of sleep occur in a nonrandom ordered fashion, and their time structure is characterized by firstorder relationships. Significance: We postulate that CAP components are the expression of a timely ordered process that exhibits specific sleep stage-related features and undergoes age-related modifications. Keywords: Sleep, cyclic alternating pattern, time structure, Markov process, entropy, EEG, human Citation: Ferri R; Bruni O; Miano S et al. The time structure of the cyclic alternating pattern during sleep. SLEEP 2006;29(5):693-699.
Study Objectives: To analyze the intervals between A phases of the cyclic alternating pattern (CAP) and to describe their time structure. This might represent an additional aspect to be studied in sleep pathologies that are accompanied by CAP changes. Methods: Sleep stages and CAP A phases were identified in polysomnographic night recordings of normal adults and children. Intervals between consecutive CAP A phases were measured, counted, and used to draw individual normalized distribution graphs. The intervals during light sleep (stages 1 and 2) were analyzed separately from those occurring during slow-wave sleep (SWS). Subsequently, we performed a Markovian analysis of intervals, in order to describe in detail their time structure. Setting: N/A. Participants: Twenty-four adults and 28 children. Measurements and Results: In adults, a preponderance of intervals shorter than 60 seconds during SWS was found; light sleep showed a higher number of intervals longer than 60 seconds. A less clear-cut dif-
this type of analysis can hardly be sufficient to describe a “functional structure” and describes only first-order aspects of the time dependencies of sleep microstructure.2,3 During non-rapid eye movement (NREM) sleep, phasic events such as K complexes, vertex waves, delta-wave bursts, and shortlasting arousals show a particular grouping, indicated as "cyclic alternating pattern" or CAP.4,5 CAP has been described as consisting of transient complexes (phase A) that periodically interrupt the tonic theta/delta activities of NREM sleep (phase B). Functionally, CAP is thought to represent a condition of sustained arousal instability oscillating between a greater arousal level (phase A) and a lesser arousal level (phase B). Because of its practical utility, the analysis of CAP has captured the attention of an increasing number of sleep researchers for its potential significance and its clinical correlations.3,6-19 The time structure of CAP has been defined, since its first description,4,5 in a simple way. CAP sequences are defined as 3 or more A phases separated from each other by at least 2 and no more than 60 seconds, with the last A phase needed to end the sequence but not included in it.20 The remaining A phases, separated by more than 60 seconds, are considered as being “isolated” and are not included into the CAP sequences and are not used in the computation of the CAP rate (percentage of NREM sleep occupied by CAP sequences). These conventional rules were established on the basis of the phenomenologic analysis of CAP, such that CAP intervals have only been described in terms of overall distribution.21 However, this approach cannot provide information on how subsequent intervals interact with each other, and such analysis requires a more quantitative approach. The aim of this study was to analyze, in detail, the intervals between subsequent CAP A phases and to describe their time structure by means of a Markovian analysis approach.
INTRODUCTION SINCE 1968, SLEEP STAGES HAVE BEEN ALMOST UNIVERSALLY EVALUATED FOLLOWING THE CRITERIA ESTABLISHED BY RECHTSCHAFFEN AND KALES1 IN their “A Manual of Standardized Terminology, Techniques and Scoring System of Sleep Stages of Human Subjects.” Under these criteria, sleep is subdivided into epochs of 30 seconds (sometimes 20 seconds) that are considered as time “units.” However, sleep is rich in shorter phasic electroencephalogram (EEG) events, such as K complexes, vertex waves, delta-wave bursts, sleep spindles, saw-tooth waves, and short-lasting arousals, all of which are indiscriminately included into the arbitrarily defined time “units” and are merely used as markers of the different sleep stages. For this reason, sleep stages are often referred to as “sleep macrostructure,” in contrast with the “sleep microstructure” dependent on the occurrence of shorter phasic events.2,3 In the vast majority of published studies, phasic events have been only evaluated in terms of number and/or rate, and, more rarely, of their intervals; Disclosure Statement This was not an industry supported study. Dr. Gozal has received research support from Astra Zeneca and is on the National Speaker Bureau for Merck Company. Drs. Ferri, Bruni, Miano, Plazzi, Spruyt, and Terzano have indicated no financial conflicts of interest. Submitted for publication August 2005 Accepted for publication December 2005 Address correspondence to: Dr. R. Ferri, Sleep Research Centre, Department of Neurology I.C., Oasi Institute (IRCCS), Via Conte Ruggero 73, 94018 Troina, Italy; Tel: 39 0935 936111; Fax: 39 0935 936694; E-mail:
[email protected] SLEEP, Vol. 29, No. 5, 2006
693
Time Structure of CAP—Ferri et al
Moreover, as CAP shows important age-related changes from childhood to adulthood,22-24 we decided to apply these new analyses to sleep recordings from 2 groups of normal subjects different for their age; the aim of this comparison was to discover eventual developmentally determined time-structure changes of CAP and to show the eventual sensitivity of these new measures that might also be used, in the future, for the analysis of sleep clinical conditions accompanied by quantitative CAP modifications.6-19
arousal strength: A1: A phase with synchronized EEG patterns (intermittent alpha rhythm in stage 1; sequences of K complexes or delta bursts in the other NREM stages), associated with mild or trivial polygraphic variations A2: A phase with desynchronized EEG patterns preceded by or mixed with slow high-voltage waves (K complexes with alpha and beta activities, k alpha, arousals with slow-wave synchronization), linked with a moderate increase of muscle tone and/or cardiorespiratory rate A3: A phase with desynchronized EEG patterns alone (transient activation phases or arousals) or exceeding two thirds of the phase A length and coupled with a clear enhancement of muscle tone and/or cardiorespiratory rate20
SUBJECTS AND METHODS Subjects Two age groups of subjects, adults and children, were used for this study; none had any serious physical, neurologic, or psychiatric disorder or history of major sleep problems, and none was taking medication at the time of the recording. The adult group was formed by 24 normal healthy subjects (15 women and 9 men, mean age 27.5 years, SD 5.37, range 19-39 years) while the children group comprised 28 subjects (15 girls and 13 boys, mean age 6.5 years, SD 1.97, range 2.9-10.3 years).
CAP sequences are defined as 3 or more A phases separated from each other by 2 to 60 seconds. The percentage of NREM occupied by CAP sequences defines the CAP rate. Figure 1 shows an example of a CAP sequence formed by CAP A1 subtypes during slow-wave sleep (SWS). All the remaining NREM sleep, not occupied by CAP sequences, is called NCAP. EEG arousals in young children often include rhythms slower than those seen in adults. In agreement with our previous study of CAP in young children,22 we included rhythmic theta (4-7 Hz) and alpha activities (8-13 Hz) or frequencies slower than 16 Hz in the rules for the identification of A2 and A3 CAP subtypes in subjects aged less than 6 years. All intervals between the onset of each subsequent CAP A phase were measured and stored for further analysis.
Polygraphic Sleep Recording Each subject underwent 1 overnight polysomnographic recording, after an adaptation night, carried out in a standard sleep laboratory. All adult subjects were recorded in the same laboratory, whereas children were recorded in 2 different laboratories; however, as already tested for previous studies, the results obtained from the 2 laboratories are not statistically different.22,24 Subjects were not allowed to have drinks containing caffeine during the afternoon preceding the recording and were allowed to sleep until their spontaneous awakening in the morning. The following parameters were included in the polysomnographic study: EEG (at least 3 channels, F3 or F4, C3 or C4, and O1 or O2, referred to the contralateral earlobe with electrodes placed following the 10-20 International System); electrooculogram (electrodes placed 1 cm above the right outer cantus and 1 cm below the left outer cantus and referred to A1), electromyogram of the submentalis muscle, and electrocardiogram (1 derivation).
Statistical Analysis of Intervals For the statistical analysis, all intervals between subsequent CAP A phases were subdivided into 25 duration classes (< 5 s, ≥ 5< 10 s, ≥ 10 < 15 s, …., ≥ 115 < 120 s, ≥ 120 < 125 s) and counted in each subject; this count was used to draw individual normalized interval-distribution graphs. The normalization was obtained by calculating the percentage of each class with respect to the total individual count. In this way, data from different individuals could be pooled together. The intervals between subsequent CAP A phases occurring during light NREM sleep (stages 1 and 2) were counted separately from those occurring during SWS (sleep stages 3 and 4). In addition, 2 different normalized distribution graphs were obtained, 1 for light sleep and 1 for SWS, in each subject group. The differences between the 2 groups and their light sleep or SWS for each interval class were evaluated by means of a between-within multivariate (repeated measures) analysis of variance. When an interaction term (stage by group) was found to be significant, a contrast analysis was used to evaluate differences between the following pairs of distributions: light sleep-children versus light sleep-adults and SWS-children versus SWS-adults. For each group, the differences between the 2 stages were tested by means of a 1-way within multivariate (repeated-measures) analysis of variance.
Sleep Scoring Sleep stages were scored following standard criteria1 on 30second epochs, which were scored as wakefulness and sleep stages 1, 2, 3, and 4 and REM sleep. Subsequently, each CAP A phase, included in a CAP sequence or isolated, was detected in each recording (on the C3/A2 or C4/A1 derivation) during NREM sleep and classified into 3 subtypes (A1, A2, and A3), according to the rules defined by Terzano et al.20 CAP was scored by 3 of the authors (RF, OB, and SM), and their interrater reliability has been validated in a study published previously.25 CAP is a periodic EEG activity of NREM sleep characterized by repeated spontaneous sequences of transient events (phase A) that clearly break away from the background rhythm of the ongoing sleep stage, with an abrupt frequency/amplitude variation, recurring at intervals up to 1 minute long. The return to background activity identifies the interval that separates subsequent A phases in a sequence (phase B).
Markovian Analysis A Markov chain is a sequence of values (states) whose probabilities at a time interval depend upon the value at the previous time. The controlling factor in a Markov chain is the
CAP A phases have been subdivided into a 3-stage hierarchy of SLEEP, Vol. 29, No. 5, 2006
694
Time Structure of CAP—Ferri et al
transition probability, which is a conditional probability for the system to go to a particular new state, given the current state of the system. From these probabilities, the entropy of the system can be computed. The basic concept of entropy in information theory deals with how much randomness is in a signal or in a sequence of events. Alternatively, the entropy gives a measure of how much information is carried by the signal. For each subject, the length of each interval (i) between subsequent CAP A phases was assigned to 4 states as follows: State 1 (i < 16 s), State 2 (16 ≤ i < 30 s), State 3 (30 ≤ i < 60 s) and State 4 (i ≥ 60 s). First, we calculated the unconditional probability of occurrence of each state from which we obtained the zero-memory Markov model entropy (H0), one for intervals in light sleep and the other for intervals in SWS; in other words, we considered each state as occurring following its own intrinsic probability, not conditioned by the previous state. Subsequently, two 4 × 4 matrixes, one for intervals in light sleep and the other for intervals in SWS, were obtained. The 16 entries in each of these matrixes were the probabilities of transition from a given state to the next state, in successive interval occurrences. For example, if the transition from State 1 to State 3 occurred 7 times in N possible transitions from State 1 to any other State (including State 1), then the transition probability in cell (1,3) of the matrix (i.e., the cell in the first row and third column of the matrix) was given the value 7/N. Of course, the sum of all probability entries in a row of the matrix (e.g., the contents of cells [2,1], [2,2], [2,3] and [2.4]) had to be 1. These transition probability (or conditional probability) matrixes (TPMs) are the state-transition probability matrixes of the Markov chain theory, and they have been used in a similar way to quantify the time structure of other sequences of events recorded during sleep.26-29 For a reliable estimation of transition probabilities, a number of transitions equal to at least 8 times the number of matrix entries is needed.30 Since the total number of CAP A phases available for each subject ranged between 155 and 499, the use of 4 × 4 matrixes (16 matrix entries) can be considered as reliable. Each matrix served for the computation of the first-order Markov model entropy (H1); if the values of H1 are lower than H0, one can suppose that first-order relationships exist between the states of the system, i.e., each state does not occur following only its own probability but is influenced by the value of the state preceding it. The degree of this influence can be described by means of the Dependency Index (H0-H1/H0), which can range from 0 (lack of first-order interdependencies) to 1 (complete firstorder dependency). Statistical Analysis of TPMs In order to test further the null hypothesis that the TPMs were generated by a random process, we shuffled the states of each sequence and recalculated H0, H1, and Dependency Index for 25 times. Random shuffling destroys all interdependencies in the sequence, and the values obtained from these shuffled state sequences can be used for the statistical validation of each single TPM. This validation was performed in 2 ways: (1) by calculating the Z-score between the value of the Dependency Index obtained from the original state sequence and the 25 values obtained from the shuffled data and (2) by comparing the original TPM with each of the 25 TPMs obtained after shuffling by means of a test based on the χ2 statistics.31
Figure 1—Example of a cyclic alternating pattern sequence during slow-wave sleep. EOG refers to electrooculogram; ECG, electrocardiogram. SLEEP, Vol. 29, No. 5, 2006
695
Time Structure of CAP—Ferri et al
Table 1—CAP Parameters Obtained in the 2 Groups of Subjects Included in this Study Adults Children p Value, (n=24) (n=28) student t test CAP Rate, total 35.9 ± 8.98 31.2 ± 10.24 NS CAP Rate, S2 22.8 ± 10.53 20.3 ± 9.59 NS CAP Rate, SWS 62.0 ± 16.95 52.4 ± 19.35 NS A1, % of total A phases 81.5 ± 7.88 76.4 ± 12.33 NS A2, % of total A phases 9.3 ± 4.26 13.5 ± 8.23 < .035 A3, % of total A phases 9.1 ± 5.07 10.2 ± 5.97 NS A1 duration, min 8.0 ± 1.93 6.7 ± 1.62 < .015 A2 duration, min 11.2 ± 1.72 9.3 ± 1.52 < .0002 A3 duration, min 15.4 ± 3.01 18.1 ± 3.83 < .008 A1 index (number/hour 39.5 ± 11.95 34.8 ± 13.91 NS of NREM sleep) A2 index (number/hour 4.1 ± 2.32 4.8 ± 2.60 NS of NREM sleep) A3 index (number/hour 3.4 ± 2.27 3.1 ± 1.94 NS of NREM sleep) B duration, min 23.0 ± 2.63 23.2 ± 3.07 NS Sequence duration, min 236.6 ± 78.17 229.2 ± 95.24 NS Number of sequences 30.3 ± 8.46 32.9 ± 7.94 NS Data are presented as mean ± SD. CAP refers to cyclic alternating pattern; NREM, non-rapid eye movement sleep
Finally, a factorial analysis of variance was performed for the 2 measures of entropy and for Dependency Index, with age group and stage as factors, followed by posthoc comparisons. The commercially available Statistica software package (StatSoft, Inc., Tulsa, OK, 2001. STATISTICA data analysis software system, version 6. www.statsoft.com) was used for this statistical analysis. RESULTS Table 1 shows the CAP scoring parameters found in the 2 groups of subjects included in this study. All these parameters (and their differences) are compatible with the values anticipated for the ages included in the 2 groups.22-24 The top panel of Figure 2 shows the comparison between the normalized CAP interval distribution graphs obtained from light sleep and SWS in children. In this comparison, a statistically significant difference can be seen for the normalized number of intervals 10 ≤ i < 25 s, higher during SWS, and for the normalized number of intervals 50 ≤ i < 60, 75 ≤ i < 90, 95 ≤ i < 100, and 115 ≤ i < 120 s, slightly higher during light sleep. The same comparison carried out in the adult group (Figure 2, bottom panel), disclosed similar differences but with a much higher number of classes for which a statistical significance was obtained. In particular, interval classes 15 ≤ i < 35 and 40 ≤ i < 45 s were higher during SWS, whereas almost all classes ≥ 60 s were higher during light sleep. Table 2 shows the results of the posthoc comparisons, performed after the repeated-measures analysis of variance. For each CAP interval class, only significant comparisons have been reported; the others were statistically not significant. The comparisons for which a statistically significant value was obtained have been indicated in Figures 2 and 3 with an asterisk. The top panel of Figure 3 shows the comparison between the normalized CAP interval distribution graphs obtained from SLEEP, Vol. 29, No. 5, 2006
Figure 2—Comparison between the normalized cyclic alternating pattern interval distribution graphs obtained from light sleep (LS; sleep stages 1 and 2) and slow-wave sleep (SWS; sleep stages 3 and 4), in our groups of children and adult subjects. Asterisks indicate interval classes significantly different in each graph (see Table 3).
children and adults during light sleep. These distributions appear to be very similar, with no statistically significant differences being apparent. The bottom panel shows the same comparison during SWS. In this case, we found statistically significant differences, with interval classes 5 ≤ i < 20 s higher in children and intervals classes 25 ≤ i
