Rhythmic Organization of Neonatal Heart Rate and its Relation to Atypical Fetal Growth PHILIP SANFORD ZESKIND Virginia Polytechnic Institute and State University Blacksburg, Virginia
DENNIS M. GOFF Randolph-Macon Woman’s College Lynchburg, Virginia
TIMOTHY RAY MARSHALL University of Maryland College Park, Maryland
The rhythmic organization underlying long-term heart rate variability was examined in 36 newborn infants. Heart rate was registered every 30 s for 2 continuous hr while infants rested in a temperaturecontrolled isolette. Spectrum analysis of the time-series of the 240 observations detected rhythmically organized changes in the heart rates of 33 of the 36 infants. Thirty of the 33 infants showed a basic rhythm at 1.5 2 .5 cycles per hr (one cycle every 30 to 60 min). While 9 infants showed this single cycle in behavioral activity, 24 infants showed additional cycles at a wide range of faster frequencies. Infants with signs of atypical fetal growth less often showed evidence of these multiple cycles, had reliably fewer cycles in heart rate, and had a marginally lower power in their basic cycle than infants with typical patterns of fetal growth. Infants with multiple cycles in the power spectra, independent of fetal growth group, were more often observed in Alert and Active Alert behavioral states and less often in Active Sleep than comparison infants. Results indicate that 1) heart rates of newborn infants show evidence of the 30- to 60-min cycle characteristic of the Basic Rest-Activity Cycle found in other behaviors, and 2) the complexity of behavioral rhythms may be affected by prenatal malnutrition. Viewed within a dynamical systems approach to development, results suggest that the complexity of rhythms in behavior may reflect the complexity of behavioral organization.
Behavioral organization has long been implicated in the processes of learning, emotion, motivation, and perception (Hebb, 1949) and is fundamental to some Reprint requests should be sent to Sandy Zeskind, Department of Psychology, Virginia Polytechnic Institute and State University, Blacksburg, VA 24061, U.S.A. Received for publication 19 August 1990 Revised for publication 8 March 1991; 20 June 1991 Accepted at Wiley 16 August 1991 Developmental Psychobiology 24(6):413-429 (1991) 0 1991 by John Wiley & Sons, Inc.
CCC 00 12-16301911060413- 17$04.OO
414
ZESKIND ET AL.
definitions of behavioral development (Gottlieb, 1991; Schnierla, 1957; Thelen, 1989). We can conceptualize one form of behavioral organization as existing in the temporal domain. That is, the occurrence or strength of a given behavior may rise and fall with a cyclic o r rhythmic pattern of energy. Rhythmicities in several behavioral domains of the newborn infant have long been observed (Wolff, 1967). Hidden among the newborn’s seemingly random motor behavior, for instance, are rhythmically organized patterns of motility (Robertson, 1982) and coordinated leg movement (Thelen, 1979, 198.5). Newborn and young infants also show 40- to 60min cycles in vocalizations and REM activity, 3- to 4-hr cycles in sleep-wakefulness, and 12-hr cycles in skin conductance (Anders, Keener, Bowe, & Shoaff, 1983; Ernde, Swedberg, & Suzuki, 197.5; Hellbrugge, 1974; Kleitman, 1963; Roffwarg, Muzio, & Dement, 1966; Stratton, 1982). Rather than reflecting the effects of single behavioral oscillators, many of these behavioral cycles result from the combined effects of multiple oscillators that interact within a hierarchically organized set of behavioral and physiological systems (Berg & Berg, 1987; Meier-Koll, 1979; Moore-Ede, Sulzman, & Fuller, 1982). Importantly, the complexity of this rhythmic activity has been hypothesized to reflect both the maturity and integrity of the infant’s developing nervous system (Field, 1981; Stratton, 1982; Thelen, 1979, 1985; Thoman, 1986; Zeskind & Marshall, in press). Examining cyclicity in heart rate may be a particularly useful method by which the complexity of behavioral organization can be studied. Changes in heart rate reflect the summed effects of a wide range of internally and externally derived sensory conditions including respiration, therrnoregulation, blood pressure, behavioral state, and vasomotor activity. As such, heart rate may be conceptualized as a collective vuriuble, one that reflects the compressed activity of multiple subsystems (see Thelen, 1989). Behavioral state may also be an example of a collective variable that reveals the summed effects of multiple systems (see Berg & Berg, 1987; Moore-Ede, et al., 1982; Prechtl & O’Brien, 1982; Wolff, 1987). However, unlike behavioral state which is measured on a nominal scale, heart rate is a continuous variable, the range of which can be subjected to parametric forms of spectrum analysis for tests of rhythmicity. The power spectrum produced by analysis of heart rate may detect the rhythmic activity of the multiple subsystems compressed into changes in heart rate. For example, peaks in the power spectrum of heart rate at .3.5 Hz, . I Hz, and .025 Hz reveal effects of rhythmic changes in respiration, vasomotor activity, and body temperature, respectively (Sayers, 1973). Similarly, by examining shorf-term beat-to-beat variability in the frequency band of .3-1.3 Hz, spectrum analysis has been used to estimate the tone of the vague nerve (Porges, 1986; Porges, McCabe, & Yongue, 1982) and assess individual differences in behavioral organization and development (DiPietro, Larson, & Porges, 1987; Fox & Porges, 1985). In contrast to examining specific neural function, spectrum analysis of longterm heart rate variability has mostly been utilized to detect the emergence of timing mechanisms typical of development. For example, cycles in heart rate with periods of 2-3 and 3-4 hr have been detected by the third trimester of pregnancy (Hoppenbrouwers, Combs, Ugartechea, Hodgman, Sterman, & Harper, I98 l ) , and a diurnal cycle in heart rate appears at 7 postnatal weeks as multiple systems are coordinated (Hellbruge, 1974). Interestingly, a long-term cycle common to several behaviors is one with a period that ranges from 30-60 min (see Berg &
RHYTHMIC ORGANIZATION O F HEART RATE
415
Berg, 1987; Stratton, 1982). Behavioral rhythms with this period length have been found as early as 26 weeks of gestation in 30- to 50-min cycles in fetal motor activity (Sterman & Hoppenbrouwers, 1971) and are often attributed to the effects of the Basic Rest-Activity Cycle (BRAC) described by Kleitman (1963). While a cycle in heart rate at the frequency of the BRAC has been detected in the fetus (Hoppenbrouwers et al., 1981) and adult (Sayers, 1973), some suggest that newborn infants lack sufficient autonomic stability following birth for this basic cycle to appear (Kripke, 1982). One purpose of the present study was to determine whether the newborn’s long-term heart rate variability is sufficiently organized to show rhythmic activity at a frequency approximating the BRAC. Perhaps because studies have mostly focused on behavioral rhythmicities typical of development, little is known about individual differences in long-term heart rate variability and the previous experiences that contribute to their development. A population that may be particularly suited to address these issues is found among full-term infants who show atypical patterns of fetal growth, as assessed by the ponderal index (PI). The PI is a weight-length ratio used to assess fetal malnutrition (Miller & Hassanein, 1971) and is associated with behaviors that have been proposed to reflect poorer behavioral organization (Lester, 1984; Lester & Zeskind, 1982; Zeskind, 1985). Infants with either atypically low or high PIS show similar behaviors, as measured by the Neonatal Behavioral Assessment Scale (NBAS) and differential cry features, that distinguish them from infants with average PIS (Zeskind, 1981, 1983; Zeskind & Lester, 1981). Other work shows that infants with these differential behaviors have heart rate patterns that are more erratic (Zeskind & Field, 1982) and that infants with atypical PIS are less eficient in their energy utilization (Woodson, Field, & Greenberg, 1983). To the extent that the spectrum analysis of heart rate describes the disposition of the infant’s behavioral organization, differences may also be detected among infants who are hypothesized to vary in behavioral and autonomic organization. Recent work, emphasizing a dynamical systems approach to development, provides the conceptual basis for hypotheses regarding how these individual differences may be manifested in the temporal domain (Zeskind & Marshall, 1991). Our view, like the view of others (e.g., Gottlieb, 1991; Thelen, 1989), is that development is characterized by increasing behavioral complexity as multiple systems become increasingly coordinated over time. The coordination of systems, and thus apparent complexity of behavior, may be manifested in the complexity of rhythmic activity evident in the power spectrum of heart rate. In essence, spectrum analysis can determine whether a series of sinusoidal waves accounts for significant portions of behavioral variability. Interestingly, sinusoidal, or “pendulum-mode” , oscillations have been hypothesized to result from energy-efficient timing mechanisms on behavior (Stratton, 1982). Thus, simply stated, a less-well organized, less energy-efficient behavioral system may show fewer sinusoidal oscillations in the temporal domain of its behavior than a well-organized, energyefficient system. As such, we proposed that infants with atypical PIS would show fewer cycles in long-term heart rate variability than infants with an average PI. A second hypothesis regards the power of the basic rhythm in the power spectrum. Reviews of the chronobiology literature suggest that poorer coordination among secondary timing systems, or subsystems that comprise a collective uariable, may result in a lower power of the overt rhythm (Moore-Ede et al., 1982; Winfree,
416
ZESKIND ET AL.
Table 1 Comparisons of Birth Characteristics of Average- and Nonaverage-PI Infants. Average PI
-
-
SD
M
33.6 39.7
20.5 1.3
36.4 39.1
3291 50.8 99 8.1 8.9 21.5 25.9
348 1.8 17.7 1.4 1.2 4.0 14.2
11.2
1.8
M
Age (hrs) Gestational Age (wks) Birthweight (gm) Birthlength (cm)
ocs
Apgar ( I min) Apgar ( 5 min) Maternal Age (yr) Maternal Weight Gain (Ibs) Maternal Education (yr)
Nonaverage PI
SD
P < -
25.8 1.6
1(34) .37 1.32
3117 50.4 91 8.0 9.2 20.8 24.9
559 2.5 17.2 1.3 .5 5. I 10.6
1.16 .20 1.4 .21 .87 .48 .22
.25 .53 .I7 .84 .39 .63 .82
10.6
2.3
.86
.39
-
.72 .I9
1987). To the extent that the behavioral systems contributing to their heart rate variability are less-well coordinated, we hypothesized that infants with an atypical PI would show a lower power in the peak of their basic rhythm than infants with an average PI.
Method Subjects Subjects were thirty-six 1- to 4-day-old full-term (237 weeks gestation) infants who resided in the normal newborn nursery of an urban hospital. All infants were from low socioeconomic status (SES) families as defined by their participation in a prenatal care program for the indigent. Half of the subjects were male; half were female. Twenty-two infants were of Caucasian descent and 14 were of AfricanAmerican background. Routine physical and neurological examinations showed no central nervous system or physical anomalies. The PI was calculated as the infant’s birthweight (gm) x 100, divided by the cube of the birth length (cm) (Miller & Hassanein, 1971). Birth length was determined by eliciting a tonic-neck reflex and measuring the infant’s length from crown to heel. Measurements of length were repeated until reliability was achieved. The random selection of 36 infants from low-SES families at the urban hospital resulted in the recruitment of 21 infants who had an average PI (average weight-for-length) and 15 infants who had a nonaverage PI (atypical weight-for-length). Ten infants with a nonaverage PI were underweight-for-length (PI 5 2.20, 5 10th percentile); 5 were overnight-for-length (PI 2 2.82, 5 90th percentile). As shown in Table 1, average- and nonaverage-PI infants did not reliably differ in their age at testing, gestational age, birth weight, birth length, Obstetric Complications Scale (OCS) score, Apgar scores at both 1 and 5 min, or the mothers’ age, weight gain during pregnancy, or years of education. Chi square analyses showed that average- and non-average-PI infants also
RHYTHMIC ORGANIZATION OF HEART RATE
417
did not differ in their distributions of ethnic background, gender, circumcision, or breast-feeding (all p ’ s > .35). Comparisons of the two groups within the nonaverage-PI group (low- and highPI infants) on the obstetric and parturition variables described above also showed no reliable differences in postnatal age ( p < 361, gestational age ( p < .39), OCS ( p < .21), Apgar at 1 min ( p < .42), Apgar at 5 min ( p < 1.0), maternal age ( p < .54), maternal weight gain during pregnancy ( p < .99), and years of maternal education ( p < 32). Fisher’s Exact tests showed no reliable differences in the distributions of ethnic background ( p < 1.01, sex ( p < .33), circumcision ( p < l.O), and breast-feeding ( p < S1). As may be expected from the variables used to calculate the PI, low-PI infants had lower birth weights (M = 2915.2, SD = 436) than high-PI infants (M = 3519.6, SD = 600), t(13) = 2.24,p < .04, and marginally longer birth lengths (M = 51.25, SD = 2.16) than high-PI infants (M = 48.7, SD = 2.4), t(13) = 2.1, p < .06). One infant in each of the 3 PI groups was born to a mother who had diabetes during pregnancy.
Procedure Infants were studied midway between feedings as they rested in a soundattenuated and temperature-controlled (32°C) isolette. Electrodes were securely attached to the infant’s pectoral and abdominal regions and a Corometric 515a neo-trak monitor provided a digital display of the infant’s heart rate per min averaged with a moving window approximately 10 s in duration. Behavioral observations began 10 min after electrode placement. Heart rate was recorded every 30 s for 2 continuous hr, thus providing 240 time-series data points for each infant. Figure I shows an example of the points of heart rate for one infant. Concurrent with measurements of heart rate, the infant’s behavioral state was determined every 30 s from the consensus of three observers. State was rated along a 6-point nominal scale (for descriptions & criteria, see Brazelton, 1984) including the states of (1) Quiet Sleep, (2) Active Sleep, (3) Drowse, (4) Alert, (5) Active Alert, and (6) Crying. The three observers were blind to the PI classification of the infant and the hypotheses of the study. After training to criterion, interobserver reliability was assessed on independent observations for three infants interspersed through the study and showed 98% agreement. As would be expected in observations of newborns, infants were observed while either in the Sleep or Drowse states during 91% of the 30-s epochs. Rarely were infants observed in the Alert (3%), Active Alert (4%) or Crying (2%) behavioral states. The heart rate data were analyzed according to the procedures detailed by Gottman (1981). Linear trends were removed before the heart rate spectra were computed in order to improve the stationarity of the time series. The linear trends accounted for an average of 10% of the variance (SD = .12) and were reliable for 27 infants (at p < .05). The residual variance of each time series was then subjected to a Blackman-Tukey discrete Fourier analysis to partition the total variance into independent components in the frequency domain. The spectral density function for each infant was derived by smoothing the periodogram with a Hanning window (lag = 80). As suggested by Jenkins and Watts (1968), spectral peaks were judged to be significant when their 95% confidence interval was above the theoretical cumulative distribution of white noise as determined by a Kolmogorov-Smirnov
418
ZESKIND ET AL.
t 1
15
30
45
60 MINUTES
75
90
105
120
Fig. 1. This figure shows 2 hours of heart rate assessed every 30 s. The X axis represents the time at which the behavior was sampled; the Y axis represents the value of the heart rate at that point. A rise and fall in heart rate at a 40- to 60-min rhythm is visible, as is the increase in arousal over time that was removed by the linear regression.
test. This analysis allowed for the detection of rhythms ranging from 1-60 cph with a resolution of .5 cph.
Results
Evidence of the BRAC in Neonatal Heart Rate The spectrum analyses of the long-term heart rate variability showed that the cumulative variance distribution departed significantly from that of white noise for all but 3 of the 36 infants (all p s < .01). That is, all but 3 of the 36 infants showed significant rhythmic cycles in heart rate. These results indicate that at least some of the long-term variability in newborn heart rate during the 2-hr observation period was rhythmically organized and not random. The first peak was evident at 1.5 +- .5 cph (45 +- 15 min) for 30 of the 33 infants who showed reliable cyclic activity in heart rate: 19 infants at 1 cph (one cycle every 60 rnin), 7 infants at 1.5 cph (one cycle every 40 min), and 4 infants at 2 cph (one cycle every 30 min). Two infants showed their first and highest peak at 2.5 cph (one cycle every 25 min) and 1 infant showed its first and highest peak at 19.50 cph (one cycle every 3 min). Clearly differentiating the infants in this sample was the number of reliable peaks evident in the power spectra. While 9 infants showed a single cycle in heart
RHYTHMIC ORGANIZATION OF HEART RATE
419
Table 2 Descriptive Variables of Heart Rate Variability in Total Sample. Variables Mean Heart Rate Variance of Heart Rate Variance of Time Series R’ of Regression Number of Reliable Peaks in Power Spectra Slowest Cycle (cph) Fastest Cycle (cph) Power of Basic Cycle
M
SD
Minimum
Maximum
126 174 157
11.6 166.5 151
103 27 26
150 129 676
0 0
61 6
1 2 14
44
.10 1.9 1.9 8.8 220
.12 1.13 3.2 10 391
20 2301
rate, 24 infants showed additional peaks at higher frequencies in the spectrum. Infants showed as many as six reliable peaks with a mode of two: 16 infants showed two reliable peaks, 7 infants showed three, and 1 infant showed six. The additional peaks occurred at a wide range of frequencies. The second reliable peak occurred at periods ranging from 2.0-25.5 cph; the third reliable peak occurred with periods ranging from 4.5-31. The 6 reliable peaks in the power spectrum of the 1 infant were at 3.0,5.5, 8.5, 12.5, 17.5, and 22.5 cph. In 10 of the 24 spectra with multiple cycles, the peaks appeared to be multiples of the basic cycle (within a .5 cph resolution of the analysis), thus suggesting that they possibly represent harmonics in the power spectra. No detectable pattern was found in the 14 other cases. Table 2 shows the means, standard deviations, and ranges of the measures of heart rate for the total sample.
Atypical Patterns of Fetal Growth and Cycles in Heart Rate The number of peaks in the power spectrum was associated with the infant’s pattern of fetal growth. Whereas 7 of the 15 infants with a nonaverage PI showed multiple peaks in the power spectrum, 17 of the 21 infants with an average PI showed this evidence of multiple cycles in heart rate variability, a distribution that departed from that expected by chance, X2(1) = 4.63, p < .03. Two of the 3 infants with no reliable cycles in heart rate had a nonaverage PI. As predicted, infants with a nonaverage PI showed reliably fewer cycles in heart rate than infants with an average PI (average PI: M = 2.2; SD = 1.1; nonaverage PI: M = 1.5, SD = .9), one-tailed t(34) = 1.77, p < .04. Infants with a nonaverage PI (M = 11.5; SD = 124) also showed a marginally lower power in their basic heart rate cycle than ~ .06. (Different average-PI infants (M = 295; SD = 493), one-tailed t(24) = 1 . 6 0 < df reflect the use of separate variance estimates due to heterogeneity of variance). Figure 2 shows a three-dimensional landscape of the 36 power spectra. As seen in this figure, the peaks in the power spectrum flatten as values proceed from the center of the Z coordinate (average PI) to its edges in either direction (towards increasingly lower or higher PI values). The flatter landscape connotes fewer reliable peaks at lower powers. Importantly, the number of cycles in the power spectrum was not directly related to the possible confounding effects of other measures of heart rate or
420
ZESKIND ET AL.
e.0
rso
Fig. 2. This figure shows a three-dimensional landscape of an interpolation of the power spectra of 36 infants who varied in the ratio of their birthweight for birth length. The power spectra are lined up according to the value of each infant’s ponderal index, from lower-PI values to higher-PI values on the Z axis. The further an infant is from an average ponderal index (middle of the Z axis), fewer reliable peaks are observed in the power spectra of their heart rates ( X axis) at lower amplitudes ( Y axis). The lines along the Z axis represent interpolations of the 36 power spectra or what the power spectrum would look like at each given value of the PI. Interpolations are based on the real data points in the power spectra adjacent values.
differences in several standard pediatric assessments. A loglotransformation was performed on the variance scores to normalize their distribution (Porges, Arnold, & Forbes, 1973). Comparisons with t tests showed that infants with typical and atypical patterns of fetal growth did not reliably differ in heart rate variance before ( p < .20) or after ( p < .24) the regressions were conducted. Pearson productmoment correlations showed that the number of cycles was not reliably related to the infants’ mean heart rate ( r = .03), the variance removed by the linear regression
RHYTHMIC ORGANIZATION OF HEART RATE
421
Table 3 Comparisons of Subsample of Average- and Nonaverage-PI Infants in Sleep States. ~
Average PI
M Number of Cycles Power of Basic Cycle Total Variance (log 10) Variance of Time Series (log 10) Quiet Sleep (Proportion) Active Sleep (Proportion) Number of State Transitions
2.3 95.9
Nonaverage PI
SD
.49 103
M 1.4 66
SD
__
((13) -
P
DENNIS M. GOFF Randolph-Macon Woman’s College Lynchburg, Virginia
TIMOTHY RAY MARSHALL University of Maryland College Park, Maryland
The rhythmic organization underlying long-term heart rate variability was examined in 36 newborn infants. Heart rate was registered every 30 s for 2 continuous hr while infants rested in a temperaturecontrolled isolette. Spectrum analysis of the time-series of the 240 observations detected rhythmically organized changes in the heart rates of 33 of the 36 infants. Thirty of the 33 infants showed a basic rhythm at 1.5 2 .5 cycles per hr (one cycle every 30 to 60 min). While 9 infants showed this single cycle in behavioral activity, 24 infants showed additional cycles at a wide range of faster frequencies. Infants with signs of atypical fetal growth less often showed evidence of these multiple cycles, had reliably fewer cycles in heart rate, and had a marginally lower power in their basic cycle than infants with typical patterns of fetal growth. Infants with multiple cycles in the power spectra, independent of fetal growth group, were more often observed in Alert and Active Alert behavioral states and less often in Active Sleep than comparison infants. Results indicate that 1) heart rates of newborn infants show evidence of the 30- to 60-min cycle characteristic of the Basic Rest-Activity Cycle found in other behaviors, and 2) the complexity of behavioral rhythms may be affected by prenatal malnutrition. Viewed within a dynamical systems approach to development, results suggest that the complexity of rhythms in behavior may reflect the complexity of behavioral organization.
Behavioral organization has long been implicated in the processes of learning, emotion, motivation, and perception (Hebb, 1949) and is fundamental to some Reprint requests should be sent to Sandy Zeskind, Department of Psychology, Virginia Polytechnic Institute and State University, Blacksburg, VA 24061, U.S.A. Received for publication 19 August 1990 Revised for publication 8 March 1991; 20 June 1991 Accepted at Wiley 16 August 1991 Developmental Psychobiology 24(6):413-429 (1991) 0 1991 by John Wiley & Sons, Inc.
CCC 00 12-16301911060413- 17$04.OO
414
ZESKIND ET AL.
definitions of behavioral development (Gottlieb, 1991; Schnierla, 1957; Thelen, 1989). We can conceptualize one form of behavioral organization as existing in the temporal domain. That is, the occurrence or strength of a given behavior may rise and fall with a cyclic o r rhythmic pattern of energy. Rhythmicities in several behavioral domains of the newborn infant have long been observed (Wolff, 1967). Hidden among the newborn’s seemingly random motor behavior, for instance, are rhythmically organized patterns of motility (Robertson, 1982) and coordinated leg movement (Thelen, 1979, 198.5). Newborn and young infants also show 40- to 60min cycles in vocalizations and REM activity, 3- to 4-hr cycles in sleep-wakefulness, and 12-hr cycles in skin conductance (Anders, Keener, Bowe, & Shoaff, 1983; Ernde, Swedberg, & Suzuki, 197.5; Hellbrugge, 1974; Kleitman, 1963; Roffwarg, Muzio, & Dement, 1966; Stratton, 1982). Rather than reflecting the effects of single behavioral oscillators, many of these behavioral cycles result from the combined effects of multiple oscillators that interact within a hierarchically organized set of behavioral and physiological systems (Berg & Berg, 1987; Meier-Koll, 1979; Moore-Ede, Sulzman, & Fuller, 1982). Importantly, the complexity of this rhythmic activity has been hypothesized to reflect both the maturity and integrity of the infant’s developing nervous system (Field, 1981; Stratton, 1982; Thelen, 1979, 1985; Thoman, 1986; Zeskind & Marshall, in press). Examining cyclicity in heart rate may be a particularly useful method by which the complexity of behavioral organization can be studied. Changes in heart rate reflect the summed effects of a wide range of internally and externally derived sensory conditions including respiration, therrnoregulation, blood pressure, behavioral state, and vasomotor activity. As such, heart rate may be conceptualized as a collective vuriuble, one that reflects the compressed activity of multiple subsystems (see Thelen, 1989). Behavioral state may also be an example of a collective variable that reveals the summed effects of multiple systems (see Berg & Berg, 1987; Moore-Ede, et al., 1982; Prechtl & O’Brien, 1982; Wolff, 1987). However, unlike behavioral state which is measured on a nominal scale, heart rate is a continuous variable, the range of which can be subjected to parametric forms of spectrum analysis for tests of rhythmicity. The power spectrum produced by analysis of heart rate may detect the rhythmic activity of the multiple subsystems compressed into changes in heart rate. For example, peaks in the power spectrum of heart rate at .3.5 Hz, . I Hz, and .025 Hz reveal effects of rhythmic changes in respiration, vasomotor activity, and body temperature, respectively (Sayers, 1973). Similarly, by examining shorf-term beat-to-beat variability in the frequency band of .3-1.3 Hz, spectrum analysis has been used to estimate the tone of the vague nerve (Porges, 1986; Porges, McCabe, & Yongue, 1982) and assess individual differences in behavioral organization and development (DiPietro, Larson, & Porges, 1987; Fox & Porges, 1985). In contrast to examining specific neural function, spectrum analysis of longterm heart rate variability has mostly been utilized to detect the emergence of timing mechanisms typical of development. For example, cycles in heart rate with periods of 2-3 and 3-4 hr have been detected by the third trimester of pregnancy (Hoppenbrouwers, Combs, Ugartechea, Hodgman, Sterman, & Harper, I98 l ) , and a diurnal cycle in heart rate appears at 7 postnatal weeks as multiple systems are coordinated (Hellbruge, 1974). Interestingly, a long-term cycle common to several behaviors is one with a period that ranges from 30-60 min (see Berg &
RHYTHMIC ORGANIZATION O F HEART RATE
415
Berg, 1987; Stratton, 1982). Behavioral rhythms with this period length have been found as early as 26 weeks of gestation in 30- to 50-min cycles in fetal motor activity (Sterman & Hoppenbrouwers, 1971) and are often attributed to the effects of the Basic Rest-Activity Cycle (BRAC) described by Kleitman (1963). While a cycle in heart rate at the frequency of the BRAC has been detected in the fetus (Hoppenbrouwers et al., 1981) and adult (Sayers, 1973), some suggest that newborn infants lack sufficient autonomic stability following birth for this basic cycle to appear (Kripke, 1982). One purpose of the present study was to determine whether the newborn’s long-term heart rate variability is sufficiently organized to show rhythmic activity at a frequency approximating the BRAC. Perhaps because studies have mostly focused on behavioral rhythmicities typical of development, little is known about individual differences in long-term heart rate variability and the previous experiences that contribute to their development. A population that may be particularly suited to address these issues is found among full-term infants who show atypical patterns of fetal growth, as assessed by the ponderal index (PI). The PI is a weight-length ratio used to assess fetal malnutrition (Miller & Hassanein, 1971) and is associated with behaviors that have been proposed to reflect poorer behavioral organization (Lester, 1984; Lester & Zeskind, 1982; Zeskind, 1985). Infants with either atypically low or high PIS show similar behaviors, as measured by the Neonatal Behavioral Assessment Scale (NBAS) and differential cry features, that distinguish them from infants with average PIS (Zeskind, 1981, 1983; Zeskind & Lester, 1981). Other work shows that infants with these differential behaviors have heart rate patterns that are more erratic (Zeskind & Field, 1982) and that infants with atypical PIS are less eficient in their energy utilization (Woodson, Field, & Greenberg, 1983). To the extent that the spectrum analysis of heart rate describes the disposition of the infant’s behavioral organization, differences may also be detected among infants who are hypothesized to vary in behavioral and autonomic organization. Recent work, emphasizing a dynamical systems approach to development, provides the conceptual basis for hypotheses regarding how these individual differences may be manifested in the temporal domain (Zeskind & Marshall, 1991). Our view, like the view of others (e.g., Gottlieb, 1991; Thelen, 1989), is that development is characterized by increasing behavioral complexity as multiple systems become increasingly coordinated over time. The coordination of systems, and thus apparent complexity of behavior, may be manifested in the complexity of rhythmic activity evident in the power spectrum of heart rate. In essence, spectrum analysis can determine whether a series of sinusoidal waves accounts for significant portions of behavioral variability. Interestingly, sinusoidal, or “pendulum-mode” , oscillations have been hypothesized to result from energy-efficient timing mechanisms on behavior (Stratton, 1982). Thus, simply stated, a less-well organized, less energy-efficient behavioral system may show fewer sinusoidal oscillations in the temporal domain of its behavior than a well-organized, energyefficient system. As such, we proposed that infants with atypical PIS would show fewer cycles in long-term heart rate variability than infants with an average PI. A second hypothesis regards the power of the basic rhythm in the power spectrum. Reviews of the chronobiology literature suggest that poorer coordination among secondary timing systems, or subsystems that comprise a collective uariable, may result in a lower power of the overt rhythm (Moore-Ede et al., 1982; Winfree,
416
ZESKIND ET AL.
Table 1 Comparisons of Birth Characteristics of Average- and Nonaverage-PI Infants. Average PI
-
-
SD
M
33.6 39.7
20.5 1.3
36.4 39.1
3291 50.8 99 8.1 8.9 21.5 25.9
348 1.8 17.7 1.4 1.2 4.0 14.2
11.2
1.8
M
Age (hrs) Gestational Age (wks) Birthweight (gm) Birthlength (cm)
ocs
Apgar ( I min) Apgar ( 5 min) Maternal Age (yr) Maternal Weight Gain (Ibs) Maternal Education (yr)
Nonaverage PI
SD
P < -
25.8 1.6
1(34) .37 1.32
3117 50.4 91 8.0 9.2 20.8 24.9
559 2.5 17.2 1.3 .5 5. I 10.6
1.16 .20 1.4 .21 .87 .48 .22
.25 .53 .I7 .84 .39 .63 .82
10.6
2.3
.86
.39
-
.72 .I9
1987). To the extent that the behavioral systems contributing to their heart rate variability are less-well coordinated, we hypothesized that infants with an atypical PI would show a lower power in the peak of their basic rhythm than infants with an average PI.
Method Subjects Subjects were thirty-six 1- to 4-day-old full-term (237 weeks gestation) infants who resided in the normal newborn nursery of an urban hospital. All infants were from low socioeconomic status (SES) families as defined by their participation in a prenatal care program for the indigent. Half of the subjects were male; half were female. Twenty-two infants were of Caucasian descent and 14 were of AfricanAmerican background. Routine physical and neurological examinations showed no central nervous system or physical anomalies. The PI was calculated as the infant’s birthweight (gm) x 100, divided by the cube of the birth length (cm) (Miller & Hassanein, 1971). Birth length was determined by eliciting a tonic-neck reflex and measuring the infant’s length from crown to heel. Measurements of length were repeated until reliability was achieved. The random selection of 36 infants from low-SES families at the urban hospital resulted in the recruitment of 21 infants who had an average PI (average weight-for-length) and 15 infants who had a nonaverage PI (atypical weight-for-length). Ten infants with a nonaverage PI were underweight-for-length (PI 5 2.20, 5 10th percentile); 5 were overnight-for-length (PI 2 2.82, 5 90th percentile). As shown in Table 1, average- and nonaverage-PI infants did not reliably differ in their age at testing, gestational age, birth weight, birth length, Obstetric Complications Scale (OCS) score, Apgar scores at both 1 and 5 min, or the mothers’ age, weight gain during pregnancy, or years of education. Chi square analyses showed that average- and non-average-PI infants also
RHYTHMIC ORGANIZATION OF HEART RATE
417
did not differ in their distributions of ethnic background, gender, circumcision, or breast-feeding (all p ’ s > .35). Comparisons of the two groups within the nonaverage-PI group (low- and highPI infants) on the obstetric and parturition variables described above also showed no reliable differences in postnatal age ( p < 361, gestational age ( p < .39), OCS ( p < .21), Apgar at 1 min ( p < .42), Apgar at 5 min ( p < 1.0), maternal age ( p < .54), maternal weight gain during pregnancy ( p < .99), and years of maternal education ( p < 32). Fisher’s Exact tests showed no reliable differences in the distributions of ethnic background ( p < 1.01, sex ( p < .33), circumcision ( p < l.O), and breast-feeding ( p < S1). As may be expected from the variables used to calculate the PI, low-PI infants had lower birth weights (M = 2915.2, SD = 436) than high-PI infants (M = 3519.6, SD = 600), t(13) = 2.24,p < .04, and marginally longer birth lengths (M = 51.25, SD = 2.16) than high-PI infants (M = 48.7, SD = 2.4), t(13) = 2.1, p < .06). One infant in each of the 3 PI groups was born to a mother who had diabetes during pregnancy.
Procedure Infants were studied midway between feedings as they rested in a soundattenuated and temperature-controlled (32°C) isolette. Electrodes were securely attached to the infant’s pectoral and abdominal regions and a Corometric 515a neo-trak monitor provided a digital display of the infant’s heart rate per min averaged with a moving window approximately 10 s in duration. Behavioral observations began 10 min after electrode placement. Heart rate was recorded every 30 s for 2 continuous hr, thus providing 240 time-series data points for each infant. Figure I shows an example of the points of heart rate for one infant. Concurrent with measurements of heart rate, the infant’s behavioral state was determined every 30 s from the consensus of three observers. State was rated along a 6-point nominal scale (for descriptions & criteria, see Brazelton, 1984) including the states of (1) Quiet Sleep, (2) Active Sleep, (3) Drowse, (4) Alert, (5) Active Alert, and (6) Crying. The three observers were blind to the PI classification of the infant and the hypotheses of the study. After training to criterion, interobserver reliability was assessed on independent observations for three infants interspersed through the study and showed 98% agreement. As would be expected in observations of newborns, infants were observed while either in the Sleep or Drowse states during 91% of the 30-s epochs. Rarely were infants observed in the Alert (3%), Active Alert (4%) or Crying (2%) behavioral states. The heart rate data were analyzed according to the procedures detailed by Gottman (1981). Linear trends were removed before the heart rate spectra were computed in order to improve the stationarity of the time series. The linear trends accounted for an average of 10% of the variance (SD = .12) and were reliable for 27 infants (at p < .05). The residual variance of each time series was then subjected to a Blackman-Tukey discrete Fourier analysis to partition the total variance into independent components in the frequency domain. The spectral density function for each infant was derived by smoothing the periodogram with a Hanning window (lag = 80). As suggested by Jenkins and Watts (1968), spectral peaks were judged to be significant when their 95% confidence interval was above the theoretical cumulative distribution of white noise as determined by a Kolmogorov-Smirnov
418
ZESKIND ET AL.
t 1
15
30
45
60 MINUTES
75
90
105
120
Fig. 1. This figure shows 2 hours of heart rate assessed every 30 s. The X axis represents the time at which the behavior was sampled; the Y axis represents the value of the heart rate at that point. A rise and fall in heart rate at a 40- to 60-min rhythm is visible, as is the increase in arousal over time that was removed by the linear regression.
test. This analysis allowed for the detection of rhythms ranging from 1-60 cph with a resolution of .5 cph.
Results
Evidence of the BRAC in Neonatal Heart Rate The spectrum analyses of the long-term heart rate variability showed that the cumulative variance distribution departed significantly from that of white noise for all but 3 of the 36 infants (all p s < .01). That is, all but 3 of the 36 infants showed significant rhythmic cycles in heart rate. These results indicate that at least some of the long-term variability in newborn heart rate during the 2-hr observation period was rhythmically organized and not random. The first peak was evident at 1.5 +- .5 cph (45 +- 15 min) for 30 of the 33 infants who showed reliable cyclic activity in heart rate: 19 infants at 1 cph (one cycle every 60 rnin), 7 infants at 1.5 cph (one cycle every 40 min), and 4 infants at 2 cph (one cycle every 30 min). Two infants showed their first and highest peak at 2.5 cph (one cycle every 25 min) and 1 infant showed its first and highest peak at 19.50 cph (one cycle every 3 min). Clearly differentiating the infants in this sample was the number of reliable peaks evident in the power spectra. While 9 infants showed a single cycle in heart
RHYTHMIC ORGANIZATION OF HEART RATE
419
Table 2 Descriptive Variables of Heart Rate Variability in Total Sample. Variables Mean Heart Rate Variance of Heart Rate Variance of Time Series R’ of Regression Number of Reliable Peaks in Power Spectra Slowest Cycle (cph) Fastest Cycle (cph) Power of Basic Cycle
M
SD
Minimum
Maximum
126 174 157
11.6 166.5 151
103 27 26
150 129 676
0 0
61 6
1 2 14
44
.10 1.9 1.9 8.8 220
.12 1.13 3.2 10 391
20 2301
rate, 24 infants showed additional peaks at higher frequencies in the spectrum. Infants showed as many as six reliable peaks with a mode of two: 16 infants showed two reliable peaks, 7 infants showed three, and 1 infant showed six. The additional peaks occurred at a wide range of frequencies. The second reliable peak occurred at periods ranging from 2.0-25.5 cph; the third reliable peak occurred with periods ranging from 4.5-31. The 6 reliable peaks in the power spectrum of the 1 infant were at 3.0,5.5, 8.5, 12.5, 17.5, and 22.5 cph. In 10 of the 24 spectra with multiple cycles, the peaks appeared to be multiples of the basic cycle (within a .5 cph resolution of the analysis), thus suggesting that they possibly represent harmonics in the power spectra. No detectable pattern was found in the 14 other cases. Table 2 shows the means, standard deviations, and ranges of the measures of heart rate for the total sample.
Atypical Patterns of Fetal Growth and Cycles in Heart Rate The number of peaks in the power spectrum was associated with the infant’s pattern of fetal growth. Whereas 7 of the 15 infants with a nonaverage PI showed multiple peaks in the power spectrum, 17 of the 21 infants with an average PI showed this evidence of multiple cycles in heart rate variability, a distribution that departed from that expected by chance, X2(1) = 4.63, p < .03. Two of the 3 infants with no reliable cycles in heart rate had a nonaverage PI. As predicted, infants with a nonaverage PI showed reliably fewer cycles in heart rate than infants with an average PI (average PI: M = 2.2; SD = 1.1; nonaverage PI: M = 1.5, SD = .9), one-tailed t(34) = 1.77, p < .04. Infants with a nonaverage PI (M = 11.5; SD = 124) also showed a marginally lower power in their basic heart rate cycle than ~ .06. (Different average-PI infants (M = 295; SD = 493), one-tailed t(24) = 1 . 6 0 < df reflect the use of separate variance estimates due to heterogeneity of variance). Figure 2 shows a three-dimensional landscape of the 36 power spectra. As seen in this figure, the peaks in the power spectrum flatten as values proceed from the center of the Z coordinate (average PI) to its edges in either direction (towards increasingly lower or higher PI values). The flatter landscape connotes fewer reliable peaks at lower powers. Importantly, the number of cycles in the power spectrum was not directly related to the possible confounding effects of other measures of heart rate or
420
ZESKIND ET AL.
e.0
rso
Fig. 2. This figure shows a three-dimensional landscape of an interpolation of the power spectra of 36 infants who varied in the ratio of their birthweight for birth length. The power spectra are lined up according to the value of each infant’s ponderal index, from lower-PI values to higher-PI values on the Z axis. The further an infant is from an average ponderal index (middle of the Z axis), fewer reliable peaks are observed in the power spectra of their heart rates ( X axis) at lower amplitudes ( Y axis). The lines along the Z axis represent interpolations of the 36 power spectra or what the power spectrum would look like at each given value of the PI. Interpolations are based on the real data points in the power spectra adjacent values.
differences in several standard pediatric assessments. A loglotransformation was performed on the variance scores to normalize their distribution (Porges, Arnold, & Forbes, 1973). Comparisons with t tests showed that infants with typical and atypical patterns of fetal growth did not reliably differ in heart rate variance before ( p < .20) or after ( p < .24) the regressions were conducted. Pearson productmoment correlations showed that the number of cycles was not reliably related to the infants’ mean heart rate ( r = .03), the variance removed by the linear regression
RHYTHMIC ORGANIZATION OF HEART RATE
421
Table 3 Comparisons of Subsample of Average- and Nonaverage-PI Infants in Sleep States. ~
Average PI
M Number of Cycles Power of Basic Cycle Total Variance (log 10) Variance of Time Series (log 10) Quiet Sleep (Proportion) Active Sleep (Proportion) Number of State Transitions
2.3 95.9
Nonaverage PI
SD
.49 103
M 1.4 66
SD
__
((13) -
P
