Comparisons of EEG Sleep State-Specific Spectral. Values Between Healthy Full-Term and Preterm. Infants at Comparable Postconceptional Ages. *tMark S.
Sleep, /7(1):47-5/ © 1994 American Sleep Disorders Association and Sleep Research Society
Comparisons of EEG Sleep State-Specific Spectral Values Between Healthy Full-Term and Preterm Infants at Comparable Postconceptional Ages *tMark S. Scher, :j:§Mingui Sun, §Doris A. Steppe, ~David L. Banks, *:j:Robert D. Guthrie and :j:§Robert J. Sclabassi *Department of Pediatrics, tDepartment of Neurology and tDepartment of Neurosurgery, Children's Hospital of Pittsburgh and Magee- Womens Hospital §Department of Electrical Engineering, University of Pittsburgh, School of Medicine flDepartment of Statistics, Carnegie-Mellon University, Pittsburgh, Pennsylvania, U.S.A.
Summary: Differences in state-specific electroencephalographic (EEG) spectral values are described between groups of preterm and full-term neonates at comparable postconceptional term ages. Eighteen healthy preterm neonates of ,,; 32 weeks gestation were selected from an inborn population of a neonatal intensive care unit. Twenty-fourchannel recordings were obtained at a full-term age and compared with studies of 22 healthy full-term neonates. The initial three hours of each 12-hour study were recorded on paper from which EEG sleep state scores per minute were visually assessed. Six mean spectral values (i.e. total EEG, electromyogram, delta, theta, alpha and beta energies) were calculated from each corresponding minute of digitized data, which was also assigned one of six EEG sleep states. Each neonatal group displayed statistically significant differences among sleep-state segments for all spectral values. The alpha- and beta-range spectral values of the preterm group, compared to the full-term control group, were lower during all sleep state segments. Spectral values for the theta band were lower during both quiet sleep segments only, whereas spectral values for delta were lower during all sleep stages, except trace-altemant quiet sleep. Significant differences in EEG spectral values were noted among states of sleep for both pre term and fullterm infants of similar postconceptional term ages. These data also suggest differences in central nervous system maturation between neonatal populations. These findings strengthen our previously stated contention that there is a functional alteration in brain development of the preterm infant as reflected in sleep organization that results from a prolonged extrauterine experience and/or prematurity. Key Words: EEG sleep- Neonate-Preterm-Spectral values.
The sleep cycle ofa full-term neonate comprises two active and two quiet sleep segments. A coalescence of electrographic (EEG) and polygraphic measures define the predominant physiologic state (1,2). Pattern recognition of each sleep-state segment is based in part on the electrographic expression of specific combinations of EEG frequencies over multiple channels, organized as either continuous or discontinuous electrographic patterns. In general, EEG activities reflect summated oscillatory potentials emanating from difPresented in part at the Association of Professional Sleep Societies Meeting, Toronto, Canada, June 1991. Accepted for publication August 1993. Address correspondence and reprint requests to Mark S. Scher, M.D., Director, Developmental Neurophysiology Laboratory, Magee-Womens Hospital, 300 Halket Street, Pittsburgh, PA, 152133180,412-641-4144. t'
47
ferent neuronal aggregates within the brain. Synchronized sleep in the newborn (i.e. quiet sleep) may represent projected thalamo-cortical activity, whereas de synchronized sleep (i.e. active sleep) represents predominantly neocortical activity (3). Spectral values of EEG frequencies are quantitative measures of EEG activities. Comparisons of digitized neurophysiologic data with visually scored neonatal EEG sleep patterns can provide additional information concerning maturation of brain function (4) and differences in sleep organization between preterm and full-term neonatal groups at similar postconceptional term ages, as previously described (5,6). The purpose of this study was to compare the spectral values within visually identified minutes of EEG sleep states in preterm and full-term infants who were at comparable postconceptional term ages. The pro-
48
M. S. SCHER ET AL.
longed extrauterine experience or shortened intrauterine development of the preterm infant may influence brain maturation, as expressed by EEG sleep organization when compared with the full-term infant who experienced a complete intrauterine gestation.
TABLE 1. Clinical and demographic data for preterm and full-term neonates
Patient selection
Length
40.9 37-43
3,580 2,400-4,082
36.5 33.5-38.2
47.0 30-52.5
40.4
3,554
34.9
51.9
Mean Range Full-termb Mean Range
Preterm
METHODS
Birthweight
Occipital frontal circumference
Postconceptional age a
The clinical and demographic data for these forty 2,700-4,670 32-37.5 47-57.5 37-42 neonates are listed in Table 1. Eighteen preterm infants a Preterm group: n = 18; 12 White, 6 Black; 9 female, 9 male. h Full-term group: n = 22; 15 White, 7 Black; 12 female, 10 male. of ::;32 weeks estimated gestational age (EGA) were recruited from a neonatal population admitted to the Neonatal Intensive Care Unit (NICU) ofMagee-WomDigitized neurophysiological data for each minute ens Hospital. Selection was based on review of mater- of sleep during the first 3 hours were compared with nal and neonatal medi~al records, combined with con- the contemporaneous minute of sleep, which was visultation with the attending neonatologist. All infants sually assigned one of four sleep states according to were clinically asymptomatic. None were treated for a traditional criteria (1,2) (two active sleep and two quiet major organ system illness (e.g. hyaline membrane dis- sleep segments). EEG sleep records in the preterm inease, sepsis, etc.) fant at postconceptional full-term ages were used for Twenty-two, appropriate-for-gestational-age, full- the present study. Each full-term subject received a term infants were selected from eight well-child nurs- single study between 2 and 4 days of life before diseries. Careful review of the medical records as well as charge. physical examinations were carried out to verify the A neonatal research nurse provided clinical care for healthy status of these full-term infants. each infant during the recording session. Sleep, feeding All infants received a neurodevelopmental assess- behavior, diaper changes, medication administration ment on the day of each recording (i.e. Dubowitz's and technical comments (i.e. equipment malfunctions, examination) (7), and all information regarding feeding environmental measures, etc.) were documented on schedule, sleep behavior and medical care were re- our computer data base. No infants received medicacorded. tions during the studies. No male children were cirThirty-four of the 40 subjects were followed for at cumcised prior to the study. least 18 months, and some were followed through 3 years of life. Neuro-developmental status of all chil- Recording parameters dren was assessed to be normal. Psychometric testing Digitized data were collected from II cerebral elecincluded the Bayley Motor and Mental Performance Scales, Carey Temperament Scales and Vineland So- trodes placed according to the modified international cial Maturity Scales. The other six subjects were nor- 10/20 system for neonates with longitudinal and tanmal by parental reports. gential electrode arrays simultaneously utilized, yielding 14 channels ofEEG data. Ten noncerebral channels and an event marker recorded viscerosomatic funcEEG sleep recordings tion. These noncerebral measures, however, were not Electroencephalogram studies were carried out in an analyzed in the present study. A time constant of 0.3 environmentally controlled setting in which sound, seconds and a low-pass filter of70 Hz were maintained. light, humidity and tactile stimulation were monitored. Data were recorded onto the hard disc of the computer All infants were studied while sleeping in a prone po- workstation and transferred onto magnetic tape for sition in an open bed, which was their usual sleeping storage (Danford, Inc., San Pedro, CA) prior to data position in the nursery. Continuous recordings began processing. Minute-by-minute computations of all after a feeding and diaper change between 2000 and spectral values were entered into a relational data base 2100 hours and ended between 0800 and 0900 hours with the corresponding minute of visually scored sleep. on the following day. The entire 12-hour study was digitized on an Apollo computer workstation (Hewlett- Data processing Packard, Inc., Palo Alto, CA). The initial 3 hours were simultaneously recorded on paper using a 21-channel Data were sampled at 64 samples per second and EEG machine (Nihon Kohden model 4221). stored in a block of 1,024 digital data points. These Sleep, Vol. 17, No.1, 1994
49
EEG STATE-SPECIFIC SPECTRAL VALUES IN NEONATES TABLE 2. Mean spectral values during specific eeg sleep state segments for preterm and full-term neonates
Spectral values
Mixed-frequency active sleepa FT
PT
High-voltage slow quiet sleep FT
Total EEGb 46,455 51,001 35,030 EMGb 1,447 1,568 1,114 0.42 0.36 0.53 Delta' 0.033 0.028 0.047 Theta' 0.011 0.007 0.012 Alpha' 0.009 0.005 0.008 Beta' a FT = full-term, PT = preterm. 2 h Measured in volts • 'Ratio of band energy to total EEG energy.
Trace-altemant quiet sleep
PT 47,085 1,487 0.46 0.0035 0.008 0.005
samples were appended with zeros to make a time series of 4,096 samples. After multiplication by a Kaiser window function, the time series was transformed into the frequency domain using an efficient fast fourier transform algorithm (8). This is the standard method for creating a time series for spectral analyses. Mean values for six spectral measures were then computed for each minute of EEG data (total EEG 0.5-22 Hz; electromyogram (EMG) 0.5-32 Hz; delta band 0.5-4 Hz; theta band 4-8 Hz; alpha band 8-13 Hz and beta band 13-22 Hz) with a frequency of 0.0 156 Hz/sample. Because the sampling rate was not greater than or equal to twice the maximum frequency defined by the low-pass filter setting, an error caused by aliasing was considered. A quantitative analysis of this error caused by aliasing was conducted on the data (9) and only 5.3% of error caused by aliasing was present when averaged over all 14 EEG channels.
Statistical analyses Minutes were eliminated from the analysis when the sleep state was indeterminate, when it was during a feeding and when it was within 5 minutes of a diaper change or an electrode adjustment. Also, filter settings eliminated sweat artifact (i.e. activity < 0.3 seconds). Analyses were performed using statistical packages (SPSS, Inc., Chicago, IL, and SAS Inc., Cary, NC) with data extracted from our Interbase relational data base (Borland Inc., Scotts Valley, CA). Exploratory calculations included tabulations of mean values, standard deviations and standard errors for all variables. Scatter plots and histograms helped define patterns of distribution as well as outliers. A normalizing transformation was applied to five of the six variables. Total EEG, total EMG, alpha and beta were skewed to the right, therefore, a logarithmic transformation was used. Delta was skewed to the left, so all delta values were squared. No transformation was used on theta values. Multivariate analysis of variance (MANOV A) and Scheffe comparisons were employed to detect whether
Low-voltage irregular active sleep
FT
PT
FT
PT
34,420 1,041 0.57 0.048 0.012 0.007
29,223 1,398 0.53 0.041 0.009 0.004
57,415 1,332 0.35 0.029 0.010 0.008
45,751 1,431 0.30 0.024 0.007 0.004
differences were present among sleep-state segments as well as between study groups. Visual inspection of graphs showed evidence that the covariance matrices of full-term and preterm infants were unequal. This did not appear to affect the conclusions of the analysis, except for the delta and theta variables in Table 4, where MANOV A found univariate differences but not global differences.
RESULTS Table 2 lists the mean spectral values for aU six EEG variables by state and between study groups. Using MANOV A, significant differences were present among states of sleep (p < 0.00 I), as well as between preterm and full-term groups overall (p < 0.005). Some interaction was also noted between the neonatal group (preterm and full-term groups) and sleep state, although this was only of borderline significance (p = 0.093). The state differences were further investigated using Wilks' Lambda and ANOVA. Results are summarized in Table 3. Each neonatal group (preterm and fullterm) displayed statistically significant differences among the four states of sleep for all six mean spectral values. Among full-term neonates, five of the six spectral variables showed significant differences overall among the four sleep states, and the sixth variable, alpha, was borderline (p = 0.0880). Four of the six spectral variables (total EEG, total EMG, delta and theta) discriminated between either active sleep state (mixed frequency or low-voltage irregular) and either quiet sleep state (trace altemant or high-voltage slow). Delta and theta energies discriminated between any two states, but theta energies did not discriminate between high-voltage slow and trace altemant. Alpha discriminated between low-voltage irregular and each of the other states. Beta discriminated between trace-altemant and mixed-frequency or high-voltage slow as well as between mixed-frequency and low-voltage irregular active sleep states. Among preterm neonates (Table 3), all six of the variables were significantly different overall among the Sleep, Vol. 17, No.1, 1994
M. S. SCHER ET AL.
50
TABLE 3. Comparisons (p- values) of spectral values between pairs of EEG sleep segments in full-term and preterm groups Overall
M/HVS
M/TA
LVIIHVS
LVIITA
M/LVI
HVS/TA
Active/Quiet
Full-term EEG EMG Delta Theta Alpha Beta
0.001* 0.002* 0.001* 0.001* 0.088** 0.001*
0.011* 0.003* 0.001* 0.001* 0.244** 0.953**
0.001* 0.002* 0.001* 0.001* 0.492** 0.001*
0.001* 0.010* 0.001* 0.001* 0.019* 0.080**
0.001* 0.004* 0.001* 0.001* 0.030* 0.342**
0.002* 0.426** 0.001* 0.001* 0.037* 0.008*
0.876** 0.069** 0.038* 0.681** 0.388** 0.007*
0.001* 0.002* 0.001* 0.001* 0.046* 0.585**
Preterm EEG EMG Delta Theta Alpha Beta
0.001* 0.010* 0.001* 0.001* 0.024* 0.008*
0.124** 0.039** 0.001* 0.003* 0.159** 0.119**
0.001* 0.005** 0.001* 0.001* 0.092** 0.015*
0.272** 0.726** 0.001* 0.003* 0.004* 0.173**
0.005* 0.154** 0.001* 0.001* 0.005* 0.827**
0.514** 0.012* 0.001* 0.005* 0.010* 0.001*
0.001* 0.094** 0.001 * 0.023* 0.416** 0.156**
0.001* 0.060** 0.001* 0.001* 0.008* 0.330**
M = mixed-frequency active sleep, HVS = high-voltage slow quiet sleep, T A = trace-alternant quiet sleep, LVI = low-voltage irregular active sleep, Active = M + LVI, Quiet = HVS + TA. * p < 0.06, ** p ~ 0.06.
four sleep states. Delta and theta discriminated any two states. Total EEG only discriminated between tracealtemant and any active sleep state. Total EMG only discriminated between low-voltage irregular and mixedfrequency active sleep; alpha discriminated between low-voltage irregular and either quiet sleep state, as well as between low-voltage irregular and mixed-frequency, and beta only discriminated between mixedfrequency and tracc-altemant or low-voltage irregular sleep states. Table 4 summarizes MANOV A and Scheffe comparisons between the neonatal groups. For all significant differences, the lower value was in the preterm group. Significant differences were found between preterm and full-term groups overall (p = 0.005) and for three of the six variables: total EEG (p = 0.005), alpha (p = 0.01) and beta (p = 0.002). Delta energies were lower for all sleep segments except during trace alternant. Theta energies were lower during both quiet sleep states (i.e. trace altemant and high-voltage slow). Alpha and beta energies were lower throughout all segments of the sleep cycle.
infant differed from the full-term infant during specific segments of the neonatal EEG sleep cycle. Mean total EEG and alpha/beta frequency energies were lower in the preterm neonate overall. However, spectral values in the delta frequency ranges were lower in all segments except during the trace-altemant quiet sleep. Spectral theta frequencies were lower during both quiet sleep segments, Alpha and beta spectral values remained consistently lower in the preterm group throughout the sleep cycle. These findings support our previously published data emphasizing differences between preterm and full-term neonates at matched postconceptional ages (6). We reported a longer sleep cycle length (i,e. 70 versus 50 minutes), more abundant trace-altemant quiet sleep (i.e. 32% versus 28%) and a less abundant low-voltage irregular active sleep (i.e. 12% versus 17%). Fewer arousals and body movements during quiet sleep and REM sleep were also noted. We also reported that relative spectral values (i.e. the ratio of specific EEG frequency bandwidth com-
DISCUSSION
TABLE 4. Significance levels (p-values) for comparison of state-specific spectral measures between preterm andfull-term groups
Significant differences in EEG spectral values were noted among states of sleep for both preterm and fullterm neonates. All spectral values except beta-range frequencies for both groups and EMG frequencies in preterm neonates could distinguish between active sleep (i.e. mixed-frequency and low-voltage irregular segments) and quiet sleep (i.e. high-voltage slow and tracealtemant sleep segments). Each neonatal group differed with respect to how each spectral value discriminated between each of six combinations of sleep-state segment pairs. Mean EEG spectral values per minute in the preterm Sleep, Vol. 17, No.1, 1994
Spectral valuesa
LowMixed High- Trace voltage All fre- voltage alter- irreg- Active Quiet ular sleep sleep states quency slow nant
Total EEG 0.005 NS NS NS EMG NS NS Delta NS 0.01 0.02 Theta NS NS 0.01 Alpha 0.01 0.001 0.001 Beta 0.002 0.001 0.001 Mean preterm value is lower than ferences. a Absolute spectral power.
NS NS NS NS NS NS NS NS NS 0.05 0.02 0.05 NS NS 0.01 0.05 0.01 0.001 0.001 0.001 0.001 0.001 0.001 0.001 full-term for all significant dif-
EEG STATE-SPECIFIC SPECTRAL VALUES IN NEONATES pared to total EEG frequency) for an entire 12-hour recording were lower in the preterm neonate for theta, alpha and beta ranges, whereas delta spectral values remained unchanged (5). Calculations of delta energies per minute detected reduced spectral values for delta frequencies during specific sleep-state segments that were not detected on spectral calculations of the entire 12-hour recording period without sleep-state differentiation. We have reaffirmed work of previous investigators who have indicated that most spectral energy in the neonatal EEG is concentrated in the delta frequency range (10-1 7). Ou~ finding of higher delta frequency during the trace-alternant segment is in contrast with the lack of difference noted between active and quiet sleep segments by Schulte and Bell (12). Other authors (14), however, have also described higher values in the delta range during quiet sleep. Without referring to specific EEG sleep segments, Duffy et al. (17) also reported lower amplitudes for EEG spectral measures in preterm compared with fUll-term groups. These differences in spectral values for the preterm infant suggest that fewer functional neuronal aggregates generate less abundant spontaneous oscillatory potentials. Delta range activities, on the other hand, remain equivalent to the full-term infant, but only during the trace-alternant quiet sleep segment. Neuronal aggregates responsible for these oscillatory potentials alter their behavior during active or quiet sleep segments in all neonates. However, differences in brain maturation in the pre term neonate lead to modifications in these state-specific oscillatory potential values. Despite an increased percentage of tracealternant quiet sleep segments in the sleep cycle noted in our previous study, spectral values in the theta through beta band range frequencies remained lower for the preterm infant when compared with the fullterm neonate. One limitation of the present study is our arbitrary definition of sleep states based on visual inspection of EEG patterns. Changes in body motility, arousals and autonomic behavior must coalesce with specific EEG patterns before a sleep state is achieved. Subtle changes in cardiorespiratory patterning or EEG reactivity during transitions between sleep-state segments (i.e. indeterminate sleep) would better define active and quiet sleep. Future studies need to incorporate both cerebral and noncerebral criteria into an automated classification of active, quiet and indeterminate sleep states. Also, we recognize that differences in spectral energies between neonatal groups may reflect, in part, time of night or adaptation effects. Spectral EEG energies throughout the remainder of the 12-hour recording need to be investigated to determine if differences persist over the nighttime between sleep states
51
and neonatal groups. Adaptation effects, however, can only be assessed by multiple night studies for each patients, and these have not yet been performed. Acknowledgements:
This research was supported in part
by grants NSOlllO and NS26793 to M.S.S., the Scaife Family Foundation, the Twenty-Five Club of Magee-Womens Hospital, the Cradle Roll Auxiliary and the Magee-Womens Hospital Research Fund. Ms. Margie Phillips provided secretarial assistance.
REFERENCES I. Anders T, Emde R, Parmelee A, eds. A manual of standardized terminology and criteria for scoring of states of sleep and wake· fulness in newborns. Los Angeles: UCLA Brain Information Service, 1971. 2. Pope JE, Werner SS, Birkford RG. Atlas of neonatal electroencephalography, 2nd ed. New York: Raven Press, 1992. 3. Striade M, Gloor P, Llimls RR, Lopes de Silva FH, Mesulam MM. Basic mechanisms of cerebral rhythmic activities. Elec· troencephalogr Clin NeurophysioI1990;76:481-508. 4. Scher MS, Sun M, Hatzilabrou GM, Greenberg N, Cebulka G, Sc1abassi R. Computer analyses of EEG sleep in the neonate: methodological considerations. J Clin NeurophysioI1990;7(3): 417-41. 5. Scher MS, Sun M, Steppe DA, Guthrie RD, Sc1abassi R. Extrauterine influence on EEG-sleep in the healthy preterm neonate at term. Ann Neurol 1991 ;30:488. 6. Scher MS, Steppe DA, Dahl RE, Asthana S, Guthrie RD. Comparison ofEEG-sleep measures in healthy full-term and preterm infants at matched conceptional ages. Sleep 1992; 15(5):442-8. 7. Dubowitz LMS, Dubowitz V, Goldberg C. Clinical assessments of gestational age in the newborn infant. J Pediatr 1970;77: 110. 8. Sun M, Li CC, Sekhan LN, Sc1abassi RJ. Efficient computation of discrete pseudo Wigner distribution. I. EEG transactions acoustics speech, signal processings. ASSP 1989;37: 1135-42. 9. Sun M, Scher MS, Dahl DE, Ryan N, Sc1abassi RJ. Analysis of aliasing problems in EEG data acquisition. IEEE 12th Southern Biomedical Engineering Conference, 1993 (in press). 10. Nolte R, Schulte FJ, Michaels R, Jurgen V. Power spectral analysis of the EEG of newborn twins in active and quiet sleep. In: Kellaway P, Petersen I, eds. Clinical electroencephalography of children. Stockholm: Alinquist and Wiksell, 1968:89-96. II. Parmelee AH. EEG power spectral analysis of newborn infants' sleep states. Electroencephalogr Clin Neurophysiol 1969;27: 690-1. 12. Schulte FJ, Bell FF. Bioelectric brain development. An atlas of EEG power spectrum in infants and young children. Neurope· diatrics 1973;4:30-45. 13. Havlicek V, Childigeva R, Chernick V. EEG frequency spectrum characteristics of sleep states in fullterm and preterm infants. Neuropaediatrie 1975;6:24-40. 14. Willekens H, Dumermuth G, Due G, Mieth D. EEG spectral power and coherence analysis in healthy fullterm neonates. Neu· ropediatrics 1984; 15: 180-90. 15. Connell JA, Oozeen R, Dubowitz V. Continuous four-channel EEG monitoring: a guide to interpretation with the normal values in preterm infants. Neuropediatrics 1987; 18: 138-45. 16. Joffe S, Chernick V. Development of the EEG between 30 and 40 weeks gestation in normal and alcohol-exposed infants. Dev Med Child Neurol 1988;30:797-807. 17. Duffy FH, Als H, McAnulty GB. Behavioral and electrophysiological evidence for gestational age effects in healthy preterm and fullterm infants studied two weeks after expected due date. Child Develop 1990;61: 1271-86. Sleep, Val. 17, No.1, 1994
