Cross-correlation of fetal cardiac and somatic activity
as an indicator of antenatal neural development
Janet A. DiPietro, PhD,a Rafael A. Irizarry, PhD,b Melissa Hawkins, PhD,a
Kathleen A. Costigan, RN, MPH,c and Eva K. Pressman, MDc
Baltimore, Md
OBJECTIVE: In this study, we wanted to model the emergence of coupling between fetal cardiac and
somatic activity in normal and at-risk fetuses.
STUDY DESIGN: One hundred six fetuses of uncomplicated pregnancies were longitudinally monitored at
20, 24, 28, 32, 36, and 38 weeks of gestation by using a fetal actocardiograph and computerized data collection. Twenty-six fetuses of complicated pregnancies were also included. Statistical time series analysis techniques were used to examine the relation between fetal movement and fetal heart rate.
RESULTS: A linear increase was found in the magnitude of the cross-correlation function between fetal
movement and fetal heart rate as gestation advanced, with coalescence around a peak lag of 5 seconds by
32 weeks. Fetuses that delivered before term evidenced accelerated fetal movement and fetal heart rate
coupling, whereas fetuses affected by deleterious conditions showed a decline in developmental trajectory.
CONCLUSIONS: The cross-correlation between fetal cardiac and somatic activity is an indicator of neuroregulation in human fetuses. (Am J Obstet Gynecol 2001;185:1421-8.)
Key words: Fetal heart rate, fetal movement, neurologic development, preterm birth
The development of the nervous system before birth
has been of long-standing interest, but its investigation
has been hampered by the inaccessibility of the fetus. Recent renewed interest in neural development during the
prenatal period has been generated by models of fetal
programming of adult disease,1 partitioning of variance
for constitutional attributes into genetic versus that conferred by the prenatal environmental milieu,2 and recognition of the antenatal origins of many functional childhood morbidities like cerebral palsy.3
Recent advances in technology and analysis techniques
provide an opportunity to determine whether features of
fetal neurobehavior can serve as markers of neural development during pregnancy. Fetal neurobehavioral development during gestation is multifaceted and includes
changes in patterns of heart rate, motor activity, and
From the Department of Population and Health Sciences,a the Department of Biostatistics,b and the Division of Maternal-Fetal Medicine,c
Johns Hopkins University.
Supported by the National Institutes of Health (HD R01 27592)
awarded to the first author, and by a Johns Hopkins Faculty Innovation
Fund awarded to the second author.
Received for publication January 30, 2001; revised June 20, 2001; accepted July 30, 2001.
Reprint requests: Janet A. DiPietro, PhD, Department of Population and
Health Sciences, The Johns Hopkins University, 624 N Broadway, Baltimore, MD 21205. E-mail: [email protected].
Copyright © 2001 by Mosby, Inc.
0002-9378/2001 $35.00 + 0 6/1/119108
doi:10.1067/mob.2001.119108
sleep-wake cycles. At the core of these features of development lies the emergence of a temporal association between motor activity and heart rate. As gestation advances, fetal motor activity becomes increasingly
associated with transient accelerations of fetal heart rate.
Individual variation in the strength of this relationship
has been proposed as an indicator of general fetal wellbeing,4 and its developmental nature has been interpreted as a sign of the developing integration between
sympathetic and parasympathetic innervation of the autonomic nervous system.5-8
Most existing investigations of the cardiac-somatic relationship have been limited by several factors, including
lack of a longitudinal perspective and reliance on visual
inspection of tracings. More important, investigators use
different a priori criteria to define how much heart rate
must change with movement to be considered evidence
of coupling, as well as the window of time around the
movement during which a change must occur (eg,
requiring an increase in heart rate of 15 bpm for at
least 10 seconds within 15 seconds of the onset of a
movement). Such definitions can greatly influence results and are based on assumptions regarding the fetal
movement–fetal heart rate (FM-FHR) relation, not on existing empirical information regarding the actual nature
of this relation. The goal of this study is to determine how
FM and FHR change in relation to one another on a continuous, dynamic basis. We rely on a computerized system
for collecting FM and FHR data and apply time series
analysis, a statistical techique that can explicitly depict
1421
1422 DiPietro et al
the relation between FHR and FM, to examine the development of cardiac-somatic coupling during gestation.
Patients and methods
One hundred six women with uncomplicated pregnancies, and their normally developing fetuses, were
monitored at successive gestational ages of 20, 24, 28,
32, 36, and 38 weeks by using a fetal actocardiograph
(MT320, Toitu Corp, Tokyo, Japan). Eligibility was restricted to nonsmoking women aged at least 20 years,
with singleton pregnancies, of which dating was validated by early first trimester pregnancy testing, examination, or ultrasonography. As a group, the patients
tended to be well educated (mean maternal education =
16.6 years), employed (96%), and married (92%). Half
of the fetuses (51%) were male. All were born with a
normal birth weight and without significant neonatal
conditions. This research was approved by the governing institutional review board, and informed consent
was obtained from all participants after they were given
a thorough explanation of the project.
Twenty-six additional participants met eligibility criteria at the onset of the study, but either complications developed during their pregnancy or they gave birth to offspring with previously undetected conditions of fetal
origin. There were no statistically significant sociodemographic differences between the complicated and uncomplicated groups. The complications represented
were as follows:
1. Preterm (<37 weeks of gestation) delivery (14)
2. Arrested preterm labor without preterm birth (5)
3. Intrauterine growth retardation (4)
4. Other fetal conditions (3)
The fetal monitor (MT320, Toitu Corp) used in this
study simultaneously records FM and FHR through the
use of a single wide array transabdominal Doppler transducer and processes this signal through a series of filtering techniques. The actograph detects fetal movements
by preserving the remaining signal after bandpassing frequency components of the Doppler signal that are associated with FHR and maternal somatic activity. Reliability
studies that compare actograph-based versus ultrasoundvisualized fetal movements have found the performance
of this monitor to be highly accurate in detecting both
fetal motor activity and quiescence.9-11 Fetal data were
sampled at 1000 Hz by using an internal analog-to-digital
board and concurrently digitized-via-streaming software.
Subsequent processing of fetal data, including application of algorithms to interpolate segments with artifact,
was accomplished by using software developed through
our laboratory (GESTATE; James Long Company, Caroga
Lake, NY). Fig 1 presents sample-time synchronized output of digitized FM and FHR data. Fifty minutes of data
were collected for each patient at each gestational age.
The data were reduced by resampling at 1 Hz, producing
December 2001
Am J Obstet Gynecol
time series measurements (X1, Y1), (X2, Y2), ..., (XT, YT)
of FM and FHR, respectively, where T is the total number
of measurements. In this case, because there were 50
minutes of data sampled each second, T = 3000. The FM
and FHR measurements, seen in Fig 1, can be considered
outcomes of a 2-component stationary time series. A standard way to measure the association between 2 components in a time series is through the cross-correlation
function,12 which is obtained by computing the correlation coefficient between FM and FHR at various lags. The
mathematical details for this procedure are included in
the appendix. This time series analysis yields 2 descriptive
values. The first is the lag, in seconds, at which the
strongest association occurs, as measured by the crosscorrelation function. The second is the magnitude or
strength of this association.
The plots in Fig 2 show averaged sample cross-correlation functions collapsed across all subjects for each gestational age. SD for each are shown as dotted lines. Two features of this series of plots are particularly salient. First,
the magnitude of the maximum sample cross-correlation
increases linearly from a value of .12 at 20 weeks of gestation to .25 at 38 weeks. The stabilization at .24 by 36 weeks
of gestation corresponds to the standard definition of
term birth 1 week later, with little evidence of additional
development once this gestational age is attained. Note
that the cross-correlation between FM and FHR of .25 at
term is a lower value than the associations reported by
others,5-7 although the increase over gestation is the
same. This is because the manner in which the relation
was assessed in the current study is not predicated on
large movements as in earlier studies. The magnitude of
the cross-correlations we observed over gestation corresponds to 6% to 9% of shared variance between FM and
FHR when measured on a second-by-second basis.
Whereas the overall strength of this association may be
relatively low, it represents the average over a 50-minute
period that includes times of quiescence and both weak
and strong movements.
The development of the more defined peaks in subsequent plots is a result of the second feature, the consolidation of peak lag times. The peak lag, the time lapse at
which there is maximal association between FM and FHR,
appears to stabilize at a lag of 5 seconds beginning at 32
weeks of gestation and is maintained through term. A lag
close to 5 seconds is evidenced at 24 and 28 weeks of gestation (6 seconds), but before this, there is evidence of
only a diffuse relationship, which suggests that the maximum rate of development for this function occurs between weeks 20 and 28 of gestation.
The histograms presented in Fig 3 provide further information on the development of the peak cross-correlation. The range of individual peak lag times at 20 weeks of
gestation is highly dispersed but progressively coalesces
around 5 seconds as gestation advances. We suspect that
Volume 185, Number 6
Am J Obstet Gynecol
DiPietro et al 1423
Fig 1. Thirty-minute sample of digitized movement and fetal heart rate data from a fetus aged 32 weeks. The top plot provides fetal movement data in arbitrary units of 0 to 100, representing calibrated voltage outputs of the Doppler-based actograph. The second plot shows fetal heart rate. The slow change in baseline over time was present for most recordings. To
consider these measurements as outcomes of a stationary time series, the slow trend was removed using a local regression.
The estimated smooth trend is represented with a dotted line. The last plot shows only the positive part of the detrended fetal
heart rate, because we are interested in the acceleration in heart rate associated with fetal movements. This relationship becomes evident when the first (fetal movement) and last (fetal heart rate) plots are compared.
this reflects variability in maturation rates among individual fetuses, and that at 20 weeks of gestation, individuals
who do not cluster around the mean have not yet developed substantive neural integration between FM and
FHR. The robust nature of the emergent relation between FM and FHR at lag 5 seconds suggests that this
value best characterizes the coupling between FM and
FHR in normally developing fetuses; hence, we define the
magnitude of the correlation at a lag of 5 seconds as ρ(5),
ˆ
our indicator of neural integration in the fetus.
Longitudinal recordings for the subset of 26 participants
that had pregnancy complications were available until the
time of delivery and analyzed by using the same time series
techniques. Inspection of the lagged relations (not shown)
reveal curves consistent with those presented in Fig 2 for
the first group, with peak lags at 5 seconds after 28 weeks of
gestation for the group with fetal conditions, and from 4 to
5 seconds for the preterm group. Fig 4 presents average
values of the sample cross-correlation function evaluated at
lag 5 {ˆρ (5)}, stratified into 3 groups: uncomplicated, preterm birth/preterm labor, and fetal conditions (including
intrauterine growth retardation). The preterm group includes cases of preterm delivery and preterm labor. Because treatment with tocolytic agents for preterm contractions can have independent effects on the fetus, data
collection did not proceed for women in this category
once the condition was detected, thus reducing the sample
size at subsequent times. Of the participants who delivered
prematurely, one delivered at 35 weeks of gestation, all others at 36 weeks. As is typical, approximately one third of
preterm births were associated with spontaneous premature rupture of membranes and one third with spontaneous labor, and one third were delivered as a result of maternal/fetal conditions (ie, oligohydramnios, gestational
hypertension, and preeclampsia). During preliminary
analyses, no detectable differences in cross-correlation
1424 DiPietro et al
December 2001
Am J Obstet Gynecol
Fig 2. Sample cross-correlation functions of fetal movement and fetal heart rate averaged across all patients from 20 weeks
of gestation through term. Values at lags –50 seconds through +50 seconds are included. Dotted lines represent SD.
Volume 185, Number 6
Am J Obstet Gynecol
DiPietro et al 1425
Fig 3. Histogram of individual peak lags at each gestational age. For each visit of every patient, the lag at
which the maximum of the sample cross-correlation function occurred was computed. For each lag, the
height of the histogram bar represents the percentage of patients who had that peak lag.
1426 DiPietro et al
Fig 4. Development of the sample cross-correlation at lag 5 in uncomplicated, preterm, and afflicted fetuses. Values at each gestational age are plotted; numbers at each point indicate the number of patients who provided data.
functions were observed between the spontaneous and delivered preterm groups, so they were combined. Patients in
the fetal condition group include 4 with intrauterine
growth retardation (determined by a birth weight of <10th
percentile for gestational age as well as clinical ascertainment), and the following: intrauterine death at 31 weeks
(1), Down syndrome (1), and diaphragmatic hernia (1; 20week data only). The infants with intrauterine growth retardation were delivered at term; accompanying antepartum factors included fetal distress, oligohydramnios, and
gestational hypertension.
With the consideration of the developmental trajectory
of the first group of patients as the standard, the preterm
group demonstrates accelerated development beginning
at 28 weeks of gestation, whereas the affected group begins to fall off the expected course of development. If regression lines are fit to the values observed in each group,
with the use of either ordinary least squares or generalized estimating equations, the slope estimates are statistically significant as are the differences between the estimated slopes for each group.
Comment
The association between FM and FHR has been most
often attributed to centrally mediated coactivation of car-
December 2001
Am J Obstet Gynecol
diac and somatomotor processes.6-8 In animal preparations, stimulation of single central loci can rapidly increase both heart rate and blood flow to muscles,13 and
muscle paralysis does not eliminate cardiovascular responses to direct afferent stimulation.14 Cardiac-somatic
coupling is predominantly considered a function of
parasympathetic control that becomes the increasingly
prominent influence as gestation advances. Changes in
autonomic control of the heart from the medulla oblongata to higher cortical processes between 27 and 30 weeks
of gestation have been inferred from a study of fetuses
with progressive levels of neural tube defects.15 Our
demonstration of the close temporal synchrony between
FM and FHR, the orderly pattern in emergence of FMFHR sample cross-correlation functions during the latter
half of pregnancy, and the consolidation in peak lag
times that occurs between 20 and 32 weeks provide convergent support of this measure as reflective of the development of the fetal nervous system.
The etiologies of both preterm birth and growth retardation are multifaceted and not isomorphic,16 although
conditions affecting uteroplacental sufficiency and fetal
well being, including infection,17 uteroplacental dysfunction,18 and psychosocial factors,19 are among the chronic
pregnancy conditions implicated. The prospective data
generated by the subset of participants in whom pregnancy complications developed revealed similarities in
the mature lag time between the association of FM and
FHR of about 5 seconds, which suggests that this aspect of
the FM-FHR relationship may be an intrinsic and invariant aspect of neural development. However, group differences in the developmental trajectory of the magnitude
of the association were apparent. The deviation from the
developmental trajectory displayed by the afflicted fetuses was in the expected direction; that is, conditions
which jeopardized fetal growth and survival also affected
development of neural regulation, manifested by a reduction in FM-FHR integration. In contrast, development
of fetuses ultimately delivering before 37 weeks of gestation appeared to be accelerated. Although this may appear counterintuitive, there has been long-standing speculation that various sources of intrauterine stress
accelerate neurologic maturation through alterations in
the hypothalamic-pituitary-adrenal axis. The preponderance of evidence for this association is based on observation of lower-than-expected rates of respiratory morbidities in growth-retarded newborns,20 although accelerated
motor and neurologic findings have also been documented.21
In the current study, we did not find evidence of accelerated neural development in growth-retarded fetuses, but in preterm infants. Two recent lines of research provide information that may be applicable to
our findings. First, experimental manipulation of preterm contractions in sheep through pulsatile oxytocin
DiPietro et al 1427
Volume 185, Number 6
Am J Obstet Gynecol
administration has been associated with acceleration of
cardiovascular measures22 and electroencephalogram
patterns,23 which suggests functional effects on both autonomic and central maturation. Second, elevations in
corticotropin-releasing hormone have been observed as
early as 18 to 20 weeks of gestation in pregnancies that
terminate in preterm delivery.24 Thus, episodic mild
hypoxemia, as well as alterations to the HPA axis, have
been implicated as chronic and early mediators of preterm birth. On the basis of results of the current study,
we propose that conditions that generate the mild prematurity observed in the current study, without significantly afflicting the fetus, serve to accelerate neural
maturation, thus providing an adaptive mechanism in
threatened pregnancies. Conversely, conditions that are
severe enough to afflict the fetus, as manifested through
restricted growth and other morbidities, can also retard
neural development. On the basis of the data in Fig 4,
this effect becomes more pronounced the longer the
fetus remains in utero. However, conclusions regarding
the latter group are offered cautiously as the developmental pattern observed may reflect vagaries of this
small number of cases.
In summary, data generated by this investigation provide a clear illustration of the development of cardiac and
somatomotor coupling in the fetus. The time series
method allowed empirical analysis of this relation without requiring a priori constraints on the size of heart rate
or movement excursions to be evaluated. Findings are
consistent with central and autonomic neuroregulation
of these systems before birth and are among the first documentation of accelerated maturation in preterm fetuses.
Further investigative pursuit of these findings may ultimately yield clinical applications in evaluation of uncomplicated and at-risk pregnancies.
We are grateful for the insight into these processes provided by Timothy R. B. Johnson.
REFERENCES
1. Barker D. In utero programming of chronic disease. Clin Sci
1998;2:115-28.
2. Devlin B, Daniels M, Roeder K. The heritability of IQ. Nature
1997;388:468-71.
3. Nelson KB, Ellenberg JH. Antecedents of cerebral palsy: multivariate analysis of risk. N Engl J Med 1986;315:61-86.
4. Baser I, Johnson TRB, Paine LL. Coupling of fetal movement
and fetal heart rate accelerations as an indicator of fetal health.
Obstet Gynecol 1992;80:62-6.
5. DiPietro JA, Hodgson DM, Costigan KA, Hilton SC, Johnson
TRB. Development of fetal movement—fetal heart rate coupling
from 20 weeks through term. Early Hum Dev 1996;44:139-51.
6. Johnson TRB, Besinger RE, Thomas RL, Strobino DM, Niebyl
JR. Quantitative and qualitative relationships between fetal heart
rate accelerations and fetal movement. J Matern Fetal Med
1992;1:251-3.
7. Timor-Tritsch IE, Dierker LJ, Zador I, Hertz RH, Rosen MG.
Fetal movements associated with fetal heart rate accelerations
and decelerations. Am J Obstet Gynecol 1978;131:276-80.
8. Vintzileos AM, Campbell WA, Nochinson D. Relation between
fetal heart rate accelerations, fetal movements, and fetal breathing movements. Am J Perinatol 1986;3:16-8.
9. Besinger RE, Johnson TRB. Doppler recordings of fetal movement: clinical correlation with real-time ultrasound. Obstet Gynecol 1989;74:277-80.
10. DiPietro JA, Costigan KA, Pressman EK. Fetal movement detection: comparison of the Toitu actograph with ultrasound from
20 weeks gestation. J Matern Fetal Med 1999;8:237-42.
11. Maeda K, Tatsumura M, Utsu M. Analysis of fetal movements by
doppler actocardiogram and fetal B-mode imaging. Clin Perinatol 1999;26:829-51.
12. Diggle P. Time series: a biostatistical introduction. Oxford (UK):
Oxford University Press; 1990.
13. Koizumi K, Kollai M. Control of reciprocal and non-reciprocal
action of vagal and sympathetic efferents: study of centrally induced reactions. J Auton Nerv Syst 1981;3:483-501.
14. Pascoe J, Bradley D, Spyer K. Interactive resonse to stimulation
of the amygdaloid central nucleus and baroreceptor afferent activation in the rabbit. J Auton Nerv Syst 1989;26:157-67.
15. Yoshizato T, Koyanagi T, Takashima T, Satoh S, Akazawa K,
Nakano H. The relationship between age-related heart rate
changes and developing brain function: a model of anencephalic
human fetuses in utero. Early Hum Dev 1994;36:101-12.
16. Holzman C, Paneth N, Fisher R, Group TMP. Rethinking the
concept of risk factors for preterm delivery: antecendents, markers and mediators. Prenat Neonatal Med 1998;3:47-52.
17. Romero R, Gomez R, Galasso M, Munoz H, Acosta L, Yoon BH,
et al. Macrophage inflammatory protein-1 α in term and preterm parturition: effect of microbial invasion of the amniotic
cavity. Am J Reprod Immunol 1994;32:108-13.
18. Salafia C, Vogel C, Vintzileos A, Bantham K, Pezzullo J, Silberman L. Placental pathologic findings in preterm birth. Am J Obstet Gynecol 1991;165:934-8.
19. Wadhwa PD, Sandman CA, Porto M, Dunkel-Schetter C, Garite
TJ. The association between prenatal stress and infant birth
weight and gestational age at birth: a prospective investigation.
Am J Obstet Gynecol 1993;169:858-65.
20. Procianoy RS, Garcia-Prats JA, Adams JM, Silvers A, Rudolph AJ.
Hyaline membrane disease and intraventricular haemorrhage in
small for gestational age infants. Arch Dis Child 1980;55:502-5.
21. Amiel-Tison C. Possible acceleration of neurological maturation
following high-risk pregnancy. Am J Obstet Gynecol 1980;
158:303-6.
22. Shinozuka N, Yen A, Nathanielsz P. Increased myometrial contracture frequency at 96-140 days accelerates fetal cardiovascular
maturation. Am J Physiol Heart Circ Physiol 2000;278:H41-H49.
23. Sadowsky D, Martel J, Jenkins S, Poore M, Cabalum T,
Nathanielsz P. Pulsatile oxytocin administered to ewes at 120 to
140 days gestational age increase the rate of maturation of the
fetal electrocorticogram and nuchal activity. J Dev Physiol
1992;17:175-81.
24. Hobel C, Dunkel-Schetter C, Roesch S, Castro L, Arora C. Maternal plasma corticotropin-releasing hormone associated
with stress at 20 weeks’ gestation in pregnancies ending in preterm delivery. Am J Obstet Gynecol 1999;180:S257-S263.
Appendix
Under the assumption that the measurements are the
outcome of a stationary stochastic process, we define the
cross-correlation function with
E{(XtX)(Yt+h Y)}
(h) = , h = 0,1,2,K
σX σY
(1)
with µX = E(Xt) and µY = E(Yt), the expected value of X1,
K, X T, and Y1, K, YT, respectively, and σ2X and σ2Y the
variance of X1, K, XT, and Y1, K, YT, respectively. With the
assumption of stationarity, none of these quantities depend on time (t). If fetal heart rate tends to increase
1428 DiPietro et al
whenever there is fetal motor activity we expect ρ(0) (the
common definition of correlation between X1, K, XT and
Y1, K, YT is the cross-correlation at lag 0) to be relatively
high. However, if fetal heart rate tends to increase some
time (h) after fetal movement occurs, then it is ρ(h) that
we expect to be high. Notice that ρ(h) has an intuitive definition as the correlation between X1, K, XT–h and Yh+1,
K, YT. The sample cross-correlation function is:
December 2001
Am J Obstet Gynecol
_
_
(Xt X)(Yt+h Y)
ˆ(h) = ˆX ˆY
T-h
t=1
with X and Y the sample averages of X1, K, XT and Y1, K, YT,
respectively, and σ̂ 2X and σ̂ 2Y the sample variances of X1, K,
XT and Y1, K, YT, respectively. The sample auto-correlation is
the standard estimate of the cross-correlation function.
Availability of Journal back issues
As a service to our subscribers, copies of back issues of the American Journal of Obstetrics
and Gynecology for the preceding 5 years are maintained and are available for purchase
from Mosby until inventory is depleted. Please write to Mosby, Subscription Customer Service, 6277 Sea Harbor Dr, Orlando, FL 32887, or call (800)654-2452 or (407)345-4000, for information on availability of particular issues and prices. If back issues are unavailable from
the publisher, photocopies of complete issues may be purchased from Bell & Howell Information and Learning, 300 N Zeeb Rd, Ann Arbor, MI 48106-1346. Telephone (734)761-4700
or (800)521-0600.