New Doppler index for prediction of perinatal brain damage in growth-restricted and hypoxic fetuses

Abstract

Objective

To evaluate the new vascular score, hypoxia index (HI), in the prediction of sonographically detected structural brain lesions in neonates within the first week after delivery of growth-restricted fetuses.

Methods

This prospective study included 29 growth-restricted fetuses delivered between 31 and 40 gestational weeks. Doppler umbilical artery (UA) and middle cerebral artery (MCA) resistance indices (RI) were recorded at 48-h intervals for at least 2 weeks before delivery. The cerebroumbilical ratio (C/U ratio = MCA-RI/UA-RI) and the HI (the sum of the daily reductions in C/U ratio, i.e. percentage below the cut-off value of 1, over the period of observation) were calculated. After delivery, neonatal outcome was evaluated according to obstetric parameters and ultrasound examinations of the brain. Doppler indices, C/U ratio and HI, as well as neonatal clinical and biochemical parameters, were tested as potential predictors of brain lesions using the C4.5 data-mining algorithm.

Results

Neonatal brain lesions were detected in 13 growth-restricted fetuses. Of all the parameters tested by the C4.5 data-mining algorithm, only HI was identified as a predictor of neonatal brain lesions. HI also showed better correlation with neonatal biochemical parameters, such as umbilical venous partial pressure of oxygen and umbilical venous pH, compared with the C/U ratio.

Conclusions

HI, which takes into account cumulative oxygen deficit, could significantly improve the prediction of a poor neurological outcome in pregnancies complicated by growth restriction and hypoxia. Copyright © 2007 ISUOG. Published by John Wiley & Sons, Ltd.

Introduction

Placental insufficiency, which leads to a reduction in nutritive supplies and fetal oxygen partial pressure, is one of the most important factors responsible for intrauterine growth restriction (IUGR) and fetal hypoxia1. IUGR can lead to long-term metabolic consequences, such as an increased propensity for some of the most common diseases of adult life, whereas fetal hypoxia is the leading cause of perinatal morbidity and mortality2. One of the most important sequelae of fetal hypoxia is the development of perinatal brain damage, resulting in a spectrum of neurological disabilities, from minimal cerebral disorder to cerebral palsy.

Fetal hypoxia activates a range of biophysical, cardiovascular, endocrine and metabolic responses. Fetal cardiovascular responses to hypoxia, which include modification of the heart rate, an increase in arterial blood pressure and redistribution of the cardiac output towards vital organs, are probably the most important adaptive reactions responsible for maintaining fetal homeostasis3. The redistribution of blood flow towards the fetal brain is known as the ‘brain-sparing effect’. Doppler assessment of the fetal cerebral and umbilicoplacental circulations can detect fetal blood flow redistribution during hypoxia and quantify the degree of this redistribution3–5. Although the brain-sparing effect attempts to compensate for the reduced oxygen delivery to the fetal brain, it has recently become clear that this phenomenon cannot always prevent the development of brain lesions6–8.

Our recent studies have demonstrated the existence of several phases in the hemodynamic response of the fetal brain to chronic hypoxia. During the early phase of Doppler surveillance, cerebrovascular variability was still observed; this was followed by a loss of cerebrovascular variability and finally an increase in cerebrovascular resistance with a reduction in brain perfusion. Maximal redistribution of blood flow in favor of the fetal brain was reached 5–8 days prior to the onset of fetal heart rate abnormalities6, 9. On the basis of these results, a new vascular parameter, the hypoxia index (HI), which takes into account the degree and duration of cerebral blood flow redistribution, was introduced. HI was found to be a good predictor of fetal distress at delivery in pregnancies complicated by malaria or maternal hypertension and chronic fetal hypoxia10–12. However, its value in the prediction of perinatal brain tissue damage has not been evaluated.

The objectives of this study were to assess the cerebral and umbilical hemodynamic changes in growth-restricted fetuses, and to determine whether prolonged fetal brain hyperperfusion (the brain-sparing effect) is associated with a poor fetal outcome and sonographically detectable perinatal brain lesions. Another objective was to estimate the value of HI in the prediction of brain lesions caused by fetal hypoxia.

Methods

Population and protocol

Seventy-three consecutive pregnant women with clinically suspected IUGR were considered for enrolment in this prospective observational study at the University Department of Obstetrics and Gynecology, Sveti Duh General Hospital, Zagreb, during a 3-year period. The inclusion criteria were: singleton viable pregnancy of more than 28 weeks' gestation, sonographically verified IUGR, a follow-up period of at least 2 weeks between hospitalization and delivery, and an umbilical artery (UA) resistance index (RI) above the normal value for gestational age13. The exclusion criteria were: fetal malformations detected by prenatal morphological ultrasound or at delivery, and any sign of maternal or fetal intrauterine infection. Twenty-nine women from this group fulfilled all the criteria and were included in the study. The participation rate was restricted because in many cases the follow-up period was shorter than 2 weeks, some women refused to give informed consent and the others were excluded for the reasons listed above.

IUGR was defined as an estimated fetal weight (calculated on the basis of ultrasound fetal biometry) below the 10^th percentile for the local reference values for gestational age and gender, and at the same time growth rate slower than normal along a standardized growth curve. Serial measurements with an interval of at least 2 weeks showed reduced progression of growth in all 29 fetuses14, 15. In all women, gestational age was estimated by fetal measurements before the 22^nd week of gestation. Following hospital admission, Doppler assessment of the fetal middle cerebral artery (MCA) and UA was performed at 48-h intervals until delivery, over a period of at least 15 days, in order to determine the longitudinal changes in the vascular resistance of the vessels examined. Fetal well-being was assessed by at least one non-stress test daily until delivery. A relatively long period of observation was attained because the obstetricians endeavored to maintain fetuses in utero for as long as possible in order to improve fetal maturation, or because the clinical condition of the fetuses was not deemed to be critical and there was no indication for immediate delivery. The course of the pregnancy was unaltered by participation in the study. The timing of delivery was decided on the basis of the fetal heart rate pattern as well as maternal and fetal clinical data. Indications for delivery by Cesarean section included abnormal cardiotocogram (CTG) pattern (the presence of persistent poor variability or repetitive variable or late decelerations), lack of increase in estimated fetal weight with progressive oligohydramnios at two successive scans performed at least 15 days apart, pre-eclampsia, and reversed end-diastolic flow in the UA. The other Doppler data were not used for timing the delivery and clinical management. Fetuses which were near or at term, without abnormal fetal heart rate and with present end-diastolic flow in the UA during the entire observation period were delivered vaginally; there were no operative vaginal deliveries. The study was approved by the medical ethics review board and all parents gave informed consent.

Data collection

Examinations were carried out using the Aloka 2000 (Aloka, Japan) ultrasound machine, equipped with a 3.5- and a 5-MHz transabdominal probe (maximum emission energy of the device was below the limits approved for use in fetal medicine, spatial peak temporal average intensities < 100 mW/cm² in both imaging and Doppler modes). Blood flow velocity waveforms were recorded from the UA and MCA. Changes in the placental hemodynamics were quantified using the UA-RI, and cerebrovascular adaptation by the MCA-RI. The RIs were calculated on the basis of Doppler records of at least six sequential cardiac cycles. The variability of MCA-RI values was assessed in all fetuses throughout the study period. Loss of this variability was defined as a difference of ≤ 0.01 between at least three successive measurements (over at least 6 days). The blood flow redistribution between the placenta and the brain was detected and quantified by the cerebroumbilical (C/U) ratio, expressed as MCA-RI/UA-RI. Only C/U ratio values below 1 were considered abnormal3, 16. In the case of absent or reversed end-diastolic flow (AREDF) in the UA, UA RI = 1 and the C/U ratio is equal to the value of the MCA-RI. The last value of the C/U ratio recorded before delivery was used for statistical analysis.

The last CTG recorded within 24 h before delivery was also compared with the outcome parameters. The fetal heart rate traces were evaluated by obstetricians who were blinded to the other results. The fetal heart rate was considered to be pathological if it was low (< 120 bpm) or high (> 160 bpm), if it was of persistent low variability, or if it showed variable or late decelerations17.

After delivery, the HI was calculated, as suggested by Arbeille et al.12, by summing the daily reductions in C/U ratio (in percent below the cut-off value of 1) over the period of observation. Studies on animal models have shown that the decrease in C/U ratio below the cut-off value, expressed as a percentage of this limit, is proportional to the reduction in the partial pressure of oxygen (pO₂) below the lower limit of the normal range (pO₂ relative deficit)3. By summing the decreases in C/U ratio below the cut-off value over the period of observation, the cumulative relative deficit in pO₂ to which the fetus was exposed during this period can be measured. The value of HI corresponds to the area between the percentage C/U ratio curve and the time axis in days (Figure 1). Therefore, this parameter combines information on the degree of hypoxia-induced blood flow redistribution towards the brain, and the duration of the hypoxia. In our study, HI was calculated on the basis of the last eight C/U ratio values measured during the last 15 days before delivery. On non-study days, C/U ratio values were determined as a mean value of the two nearest values, measured at 48-h intervals.

Figure 1. The hypoxia index (HI) is calculated by summing the daily reductions in C/U ratio (in percent below the cut-off value of 1) over the period of observation. In this example, on 12^th day before delivery, the C/U ratio is 0.95, i.e. 5% below 1. On the 11^th day it is 10% below the cut-off value. This process is repeated over the 15 days of measurement, giving a value of HI (shaded area) of 225 (5 + 10 + 15 + 20 + 15 + 10 + 20 + 30 + 25 + 20 + 25 + 30).

Umbilical arterial and venous blood samples were collected at delivery, and pO₂, partial pressure of carbon dioxide and pH were determined. The last antenatal value of the C/U ratio, the last CTG and the HI were related to the perinatal obstetric parameters.

Ultrasound examination of the neonatal brain was performed by one experienced sonographer, who was blinded to the fetal Doppler results. Ultrasound examinations were performed through the anterior fontanelle in coronal and parasagittal projections with a 7.5-MHz probe during the first 12 h after birth, and on the 2^nd, 4^th and 7^th days. Ultrasound findings of severe periventricular echodensities or moderate to severe intracranial hemorrhage indicated the existence of brain lesions, and infants were classified into groups according to the most severe sonographic diagnosis at the end of the 1^st week. The classification of Pidcock et al.18 was used for periventricular echodensities and that of Volpe19 was used for hemorrhage. Echodensities brighter than the choroid plexus were considered severe. Moderate to severe intracranial hemorrhage was defined as any intraventricular or intraparenchymal hemorrhage.

In two neonates who died in the perinatal period, postmortem examination of the brain was performed. Brain tissue samples were processed for light microscopy examination. Assessment of brain damage was carried out on 20-µm paraffin coronal sections through the telencephalon of each neonate. Brain damage was determined by macroscopic neuropathological inspection, and routine neurohistological (Nissl staining) and immunohistochemical (antibodies to glial fibrillary acidic protein) methods.

Statistical analysis

Since the data were not normally distributed, range and median values are reported. Median values were compared by Mann–Whitney U-test. Normal or abnormal Doppler and CTG data were compared with the rates of Cesarean section and admission to the neonatal intensive care unit (NICU) using the χ² test. Correlations between the HI or C/U ratio and biochemical parameters from the umbilical blood samples were performed with Spearman's correlation test. P < 0.05 was considered statistically significant. The analysis was performed using SAS software (SAS Institute, Cary, NC, USA).

The prenatal parameters (C/U ratio, CTG, AREDF in the UA and HI), together with neonatal outcome parameters were used to generate the decision tree for prediction of perinatal brain lesions. Because the decision tree represents a relatively new method, it is described briefly below.

The decision tree is a useful predictive algorithm that has been used in many areas of bioinformatics, such as gene finding and tumor classification. An advantage of the tree-based classification algorithm is its relative ease of interpretation, which can help the user to understand and improve the classification rules. In this study, the C4.5 decision tree introduced by Quinlan20 was used. The method was performed using the WEKA machine learning package21. The C4.5 program generates a classifier in the form of a decision tree, a structure with elements that are either leaves or decision nodes20. The leaf represents a class and the decision node specifies some test to be performed on an attribute value with one branch and subtree for each possible result of the test. The starting node is usually referred to as the root node. A decision tree is used to predict a case by starting at the root of the tree and moving through it until a leaf is encountered.

The C4.5 algorithm presumes the existence of an appropriate number of learning examples described by a set of attributes and by classes representing conditions. It searches for the most informative attribute according to the gain criterion and constructs a decision tree. This search is based on Shannon's measure of information gain, which defines how well a given attribute separates the training examples according to their target classification, and selects the candidate attribute at each step of the tree22.

The sample size in our study was relatively small (n = 29), but this modern statistical method gives reliable conclusions for smaller populations23. The accuracy of the classification model was tested by performing leave-one-out cross-validation on the classifier. This estimation technique is the best method for evaluating classifier performance on small samples. For a given method and sample size (n), a classifier is generated using (n − 1) cases and tested on the single remaining case. This is repeated n times, each time designing a new classifier by leaving one out. In this way, each case in the sample is used for testing and each time all but one of the cases are used for learning (i.e. designing a classifier).

Results

The study population included 29 fetuses with sonographically verified IUGR, delivered between 31 and 40 weeks' gestation. Their birth weights ranged from 790 to 2800 (median, 2050) g. On admission, all the fetuses had an estimated gestational age-related fetal weight below the 10^th percentile on the local reference values for gender and age, and increased UA-RI.

At the initial observation, two fetuses had absent end-diastolic flow (AEDF) in the UA (UA-RI = 1), and at the last measurements in both of them, reversed end-diastolic flow was detected. They reached the highest values of HI (458 and 416) and also developed brain lesions. During the period of assessment, AEDF in the UA was noted in another seven fetuses, in six of which brain lesions were diagnosed.

Seven fetuses had a low C/U ratio from the beginning of the observation period (C/U ratio < 1), of which five later developed brain lesions and of which four also had AREDF in the UA. A decrease in the C/U ratio below 1 was later observed in another 15 infants, with brain lesions being detected in eight of these. In seven fetuses the C/U ratio did not fall below 1 throughout the observation period, and none of these developed brain lesions. According to studies on experimental models, the C/U ratio normally decreases in proportion to the decrease in fetal pO₂3, 7. Hence, during the observation period, suspicion of hypoxia was established in 22 cases. In the 13 low C/U ratio (C/U ratio < 1) fetuses with brain lesions, the median value of the lowest measured C/U ratio was 0.65 (0.55–0.79), whereas in the nine low C/U ratio fetuses without brain lesions, the median of the lowest C/U ratio was 0.92 (0.69–0.99).

The last C/U ratio to be calculated was < 1 in 16 fetuses and ≥ 1 in 13 fetuses. Perinatal outcome according to the last antenatal value of the C/U ratio is shown in Table 1. Correlation between the last C/U ratio values and umbilical venous pO₂ yielded r = 0.513 (P < 0.01) and that between the last C/U ratio values and umbilical venous pH gave r = 0.494, P < 0.01. Figure 2 shows the relationship between the last calculated value of the C/U ratio and both gestational age and neonatal neurosonography.

Figure 2. Relationship between the last value of the cerebroumbilical ratio (C/U ratio), gestational age and pathological (●) or normal (▵) neonatal neurosonography. The cut-off value between normal values and C/U ratio values < 1 is indicated (vertical line).

Table 1. Perinatal outcome in neonates with normal or low last calculated antenatal values of cerebroumbilical ratio (C/U ratio)

Characteristic	C/U ratio ≥ 1 (n = 13)		C/U ratio < 1 (n = 16)		P*
	Range	Median	Range	Median
* Mann–Whitney U-test.
Birth weight (g)	1090–2800	2340	790–2720	1805	0.039
Gestational age (days)	218–280	268	217–275	253	0.020
Apgar score at 5 min	8–10	9.5	7–10	9	0.282
Umbilical artery
pO2 (kPa)	2.1–4.1	3.2	1.4–3.5	2.8	0.008
pCO2 (kPa)	5.5–6.8	6.1	5.6–6.9	6.1	0.558
pH	7.08–7.35	7.31	7.01–7.29	7.21	0.002
Umbilical vein
pO2 (kPa)	2.4–5.4	5.1	2.3–5.5	3.3	0.009
pCO2 (kPa)	4.5–5.8	5.2	4.6–6.1	5.4	0.130
pH	7.20–7.43	7.4	7.18–7.42	7.32	0.013

At the beginning of the observation period, MCA-RI values showed variability in all fetuses, reflecting the presence of normal cerebrovascular autoregulation. Later, in five low C/U ratio fetuses (C/U ratio < 1), this variability was lost for a minimum of 6 days. In three of these, in whom the C/U ratio was 35–48% below the physiological limit, the MCA-RI finally increased, reaching its highest value. All five fetuses later developed brain lesions. The evolution of fetal hemodynamics in three fetuses is shown in Figure 3.

Figure 3. Evolution of the umbilical artery resistance index (UA-RI, ◊), middle cerebral artery resistance index (MCA-RI, ○), and cerebroumbilical ratio (C/U ratio, ▴) in three growth-restricted fetuses during the 2-week Doppler surveillance period. Although in the first case (a) the C/U ratio gradually decreased below the cut-off value of 1, the hypoxia index (HI) was 58 and after delivery neonatal neurosonography did not detect brain damage. In the second case (b), HI exceeded 74 and a brain lesion was found neonatally. However, MCA-RI variability was maintained throughout the period of monitoring. In the last case (c), a loss of MCA-RI variability appeared at a gestational age of 209 days, HI was > 74 and a brain lesion developed.

A pathological fetal heart rate pattern within 24 h before delivery occurred in 10 of the 29 fetuses. The perinatal outcome of infants with normal and those with pathological CTG is presented in Table 2.

Table 2. Perinatal outcome in neonates with normal or pathological cardiotocography (CTG)

Characteristic	Normal CTG (n = 19)		Pathological CTG (n = 10)		P*
	Range	Median	Range	Median
* Mann–Whitney U-test.
Birth weight (g)	790–2800	2300	1050–2400	1635	0.022
Gestational age (days)	217–280	266	217–271	248	0.018
Apgar score at 5 min	8–10	10	7–10	9	0.013
Umbilical artery
pO2 (kPa)	2.3–4.1	3.2	1.4–3.2	2.1	< 0.001
pCO2 (kPa)	5.5–6.8	6.1	6.1–6.9	6.6	0.001
pH	7.14–7.35	7.29	7.01–7.29	7.11	< 0.001
Umbilical vein
pO2 (kPa)	3.1–5.5	4.8	2.3–4.9	3.1	0.002
pCO2 (kPa)	4.5–5.7	5.2	5.2–6.1	5.5	0.001
pH	7.26–7.43	7.4	7.18–7.38	7.24	< 0.001

Cesarean section was performed in 21 cases, and 15 neonates required admission to the NICU after delivery. There was no statistically significant association between a last antenatal C/U ratio of < 1 and an increased incidence of Cesarean section, but a last C/U ratio of < 1 was significantly associated with more frequent admission to the NICU (P = 0.014). Similarly, a pathological CTG was associated with the need for admission to the NICU (P = 0.040), but not with an increased incidence of Cesarean section.

A normal brain ultrasound examination was recorded in 16 neonates, while perinatal brain lesions were observed in 13 cases. Moderate or severe intracranial hemorrhage was verified sonographically in six cases, and severe periventricular echodensities were recorded in another seven. Infants with brain lesions had lower birth weight and gestational age. However, birth weight and gestational age were similar regardless of the type of lesion (hemorrhage or periventricular echodensities) (Table 3). Table 4 shows the incidence of perinatal brain lesions in relation to the last calculated prenatal value of the C/U ratio, CTG and HI. Of all variables used in designing the decision tree, only HI emerged as having a prognostic value for the prediction of brain lesions. According to this classifier, neonates with an HI value ≤ 74 (n = 15) were classified into a group with normal neurosonographic results, while those with HI > 74 (n = 14) were classified into a group with sonographically detected brain injuries. With this classifier, all 13 fetuses with brain lesions were classified correctly, while one fetus with no brain lesion was classified into the group with brain lesions. The stability of the classifier was tested by 29-fold leave-one-out cross validation: overall, 93.1% of the neonates from the whole group were classified correctly using the classifier HI > 74. The sensitivity of the resulting classifier was 100% (95% CI, 80.74–100.00%) and the specificity was 93.8% (95% CI, 72.84–99.69%). χ² testing indicated a significant association between HI and brain lesions and also between CTG and neurosonography. Figure 4 presents the relationship between HI, gestational age and neonatal neurosonography.

Figure 4. Relationship between the hypoxia index (HI), gestational age and pathological (●) or normal (▵) neonatal neurosonography. The cut-off value of HI is indicated (vertical line). HI > 74 is capable of predicting neonatal brain lesions detected sonographically during the 1^st week after delivery. Note that two fetuses with normal neurosonography had the same gestational age (280 days) and HI values (HI = 0).

Table 3. Birth weight and gestational age in infants with or without sonographically detected brain lesions

Characteristic	Brain lesion detected sonographically			Type of brain lesion
	Yes (n = 13)	No (n = 16)	P*	Severe periventricular echodensities (n = 7)	Moderate/severe intracranial hemorrhage (n = 6)	P*
* Mann–Whitney U-test.
Birth weight (g, median (range))	1590 (790–2400)	2355 (1570–2800)	0.001	1870 (1050–2200)	1460 (790–2400)	0.445
Gestational age (days, median (range))	241 (217–265)	271 (246–280)	< 0.001	252 (217–262)	241 (217–265)	0.836

Table 4. Cerebroumbilical ratio (C/U ratio), cardiotocography (CTG) and hypoxia index (HI) in relation to intracranial hemorrhage and periventricular echodensities in the first 7 days after delivery

Parameter	Neurosonography		P*
	No/moderate periventricular echodensities and/or no/ subependymal intracranial hemorrhage	Severe periventricular echodensities and/or moderate/ severe intracranial hemorrhage
* χ^2 test. † The C4.5 decision tree prediction algorithm introduced by Quinlan20 indicated that of the parameters only HI with a cut-off of 74 could predict neonatal brain lesions.
C/U ratio ≥ 1 (n = 13)	10	3	0.081
C/U ratio < 1 (n = 16)	6	10
Normal CTG (n = 19)	15	4	0.002
Pathological CTG (n = 10)	1	9
HI ≤ 74 (n = 15)	15	0	< 0.001
HI > 74 (n = 14)	1	13†

The HI with this cut-off limit was then tested for its value to distinguish fetuses with normal and poor obstetric outcomes. The obstetric outcome according to HI is presented in Table 5. Although both the last calculated C/U ratio and CTG distinguished fetuses with a poor and normal obstetric outcome, the most statistically significant differences between the two outcome groups were found when the groups were allocated according to HI values. The HI showed a better correlation with the biochemical parameters umbilical venous pO₂ (r = − 0.604, P < 0.001) and umbilical venous pH (r = − 0.662, P < 0.001) than did the last calculated C/U ratio. HI > 74 was significantly associated with delivery by Cesarean section (P = 0.044) and admission to the NICU (P < 0.001).

Table 5. Perinatal outcome in neonates with low or raised hypoxia index (HI)

Characteristic	HI ≤ 74 (n = 15)		HI > 74 (n = 14)		P*
	Range	Median	Range	Median
* Mann–Whitney U-test.
Birth weight (g)	1570–2800	2370	790–2400	1635	< 0.001
Gestational age (days)	252–280	271	217–265	244	< 0.001
Apgar score at 5 min	9–10	10	7–10	9	0.002
Umbilical artery
pO2 (kPa)	2.5–4.1	3.2	1.4–2.9	2.2	< 0.001
pCO2 (kPa)	5.5–6.6	6	5.9–6.9	6.5	0.002
pH	7.26–7.35	7.3	7.01–7.26	7.13	< 0.001
Umbilical vein
pO2 (kPa)	3.2–5.5	5	2.3–3.7	3.1	< 0.001
pCO2 (kPa)	4.5–5.3	5.1	5.3–6.1	5.5	< 0.001
pH	7.33–7.43	7.4	7.18–7.33	7.28	< 0.001

The two neonates who died in the perinatal period were delivered at a gestational age of 31 weeks. Their birth weights were 790 and 1090 g. The HI in these neonates was 368 and 416, respectively. Prenatal loss of cerebrovascular variability was observed in both cases, and in one of them this was followed by an increase in the cerebral vascular resistance prior to delivery. Postmortem examination of the brain showed severe intracranial hemorrhage in both cases.

Discussion

Fetal hypoxia is the most common cause of intrauterine death and one of the major causes of cerebral palsy and neurological deficits in those that survive. Although the standard surveillance procedures for chronically hypoxic fetuses include multivascular Doppler examination in combination with non-stress testing and biophysical profile scoring, none of these tests, independently or in combination, has provided reliable guidelines for the management of such fetuses24, 25. Even if the standard tests can identify the signs of fetal distress, they cannot determine the amount and duration of hypoxia that the fetus can tolerate without developing brain lesions, indicating that our understanding of the pathophysiological events related to fetal hypoxia is still rather limited25.

Our previous studies have revealed the existence of physiological cardiovascular compensatory mechanisms that protect the fetal vital organs, particularly the fetal brain, during both acute and chronic hypoxia3, 6, 7. However, later studies on experimental models and human fetuses have indicated that hypoxic brain lesions can develop despite the presence of blood flow redistribution in favor of the fetal brain7–9. The flow redistribution between the brain and the placenta can be detected and quantified using the C/U ratio. This ratio takes into account the placental insufficiency responsible for IUGR and hypoxia, and the cerebral response to hypoxia. Values of the C/U ratio less than 1 correspond to a redistribution of the fetal blood flow towards the brain in response to a reduction in fetal pO₂3, 26. Previous experimental studies on animal models have shown that the C/U ratio decreases in proportion to fetal pO₂3, 7. The pulsatility index (PI) as compared with RI better describes the velocity waveform, especially when there is absent end-diastolic flow. However, we decided to use RI because all our previous results were obtained using this index, and the C/U-RI ratio has been validated against fetal pO₂ on animal and human fetuses while a C/U ratio calculated from PI has not3, 4, 6–12, 26. In 22 of the 29 fetuses in this study, the C/U ratio decreased to < 1 during the period of surveillance; these fetuses were considered hypoxic. Thirteen of these fetuses developed brain lesions, while none of the seven fetuses with a C/U ratio > 1 developed brain lesions. The presence of brain lesions in 13/29 neonates seems high, but their mean C/U ratio was significantly lower than that in the 16 cases without brain lesions, implying that the development of brain lesions was associated with a very low level of pO₂. Also, all fetuses included in the study had an UA-RI above the normal values for gestational age. Moreover, in eight of 13 cases with brain damage, AREDF in the UA was detected. The high sensitivity of neonatal ultrasound to detect periventricular echodensities or intracranial hemorrhage is also responsible for the relatively high number of neonates with brain lesions. ‘Brain lesion’ in our study actually refers only to a sonographically detected lesion during the 1^st week of the newborn's life, which would not necessarily lead to a neurological deficit in all cases. There was a good correlation between the last calculated value of the C/U ratio and umbilical venous pO₂. However, Scherjon et al.27 did not find an association between a normal or raised last umbilicocerebral ratio and the occurrence of perinatal brain lesions in high-risk fetuses. Our decision tree analysis also confirmed that the last calculated C/U ratio could not predict the development of brain lesions. Only repeated measurements of the fetal Doppler indices over a longer period before delivery and calculation of the cumulative relative oxygen deficit (HI) gave an accurate evaluation of the exposure to hypoxia and its consequences on the brain.

The MCA and other cerebral arteries are capable of undergoing autoregulation to preserve cerebral metabolism and function in the presence of hypoxia. This autoregulation, which permits vasoconstriction or vasodilatation to maintain constant perfusion of the cerebral tissues, is controlled by metabolic, neural and chemical mediators. The integrity of autoregulation of the cerebral blood flow in response to changes in systemic blood pressure has been demonstrated in preterm fetal lambs28. One could assume that severe and/or prolonged hypoxia would reduce the fetus's ability to compensate for changes in blood pressure, thus predisposing it to intracranial hemorrhage or periventricular leukomalacia, as has been observed in preterm neonates29. We found that, at the beginning of the observation period, MCA-RI values varied from day to day in all 22 fetuses with low C/U ratio, as observed in the seven fetuses without suspected hypoxia, indicating that cerebral hemodynamics were still under the influence of the nervous system or other regulating inputs. Later, all five of the fetuses with a low C/U ratio in whom the MCA-RI did not fluctuate from one measurement to another over a period of at least 6 days, developed brain lesions. In all cases, the loss of cerebrovascular variability started before CTG abnormalities. Quite different from the fetal heart rate variability, as measured on CTG, this variability refers to changes in brain arterial vasomotoricity and was observed in a different (much longer) time frame (as in previous studies)6, 9. Finally, in three of these five fetuses, the MCA-RI increased, reaching its highest value just before delivery. This means that the cerebral resistance, reduced in response to hypoxia, probably increased due to the vasoconstrictive process or to an increase in the intracranial pressure (perivascular edema)9. At this stage, the compensatory mechanisms were overloaded, and acidosis developed. The disappearance of blood flow redistribution and the paradoxical ‘normalization’ of the MCA-RI, associated with a very adverse perinatal outcome or fetal death, have been described by other authors, and interpreted as yet another sign of myocardial deterioration and a decrease in left ventricular output24, 30. In agreement with recent studies, our results confirmed the development of brain lesions in the presence of the brain-sparing effect, suggesting that, although this phenomenon can be considered as an adaptive process against hypoxia, it cannot guarantee that brain tissue damage will not occur after a certain period of time6, 9. Furthermore, our results indicated that, although the loss of MCA-RI fluctuation can be considered a sign of advanced brain deterioration, such a sign was not present in eight of 13 fetuses with brain damage; the damage developed in these fetuses before the loss of cerebrovascular variability.

The degree and duration of fetal hypoxia were quantified by calculating HI. The neonates with brain lesions had a significantly higher HI than had those without lesions. As it is well known that brain lesions and a poor neurological outcome may be associated with premature delivery, we also used the gestational age and birth weight as training data for the C4.5 decision tree algorithm, which is a novel, promising alternative to standard statistical methods27. However, of all the variables used in designing the decision tree, only HI was recognized as having a prognostic value for the prediction of perinatal brain lesions, with the cut-off value being calculated by the program on the basis of our results alone. Fetal heart rate abnormalities occurred after the C/U ratio reached its lowest values and the HI rose above the cut-off value of 74. The rather low cut-off value might indicate that perinatal brain lesions could develop far earlier than do signs of fetal myocardial deterioration. This finding is supported by a magnetic resonance imaging study, which has detected structural changes in the brains of neonates from pregnancies with placental insufficiency and IUGR, without detectable intrauterine hypoxia or any other pathological condition31. HI showed a better correlation with umbilical venous pO₂ and pH than did the last calculated C/U ratio. The very high values of HI were detected in the two infants who died in the perinatal period, while low values of HI (≤ 74) were not associated with any brain damage. This finding suggests that a short period of moderate hypoxia can be well-tolerated by the human fetus. The high accuracy of the resulting classifier, confirmed by the cross-validation test, indicated that HI might significantly improve the prediction of brain lesions in low C/U ratio fetuses.

It is probable that HI could also be calculated on the basis of UA- and MCA-PI values (although precise data about the correlation between C/U-PI ratio and fetal pO₂ changes are required). However, it is uncertain whether the cut-off point would be as clear and whether the MCA-PI would be sensitive enough to detect changes in cerebrovascular variability. Arbeille et al.10–12 investigated fetal hypoxia and found that an HI (calculated differently from ours, to predict the functional rather than the structural consequences of hypoxia) predicted the consequences of exposure to hypoxia (fetal heart rate abnormalities at delivery) better than did the other Doppler indices. Nevertheless, these studies evaluated only functional myocardial disorders thought to be caused by hypoxia, and the presence of brain tissue lesions was not determined.

Our results should be evaluated on a much larger number of patients, as well as on very premature fetuses with early-onset IUGR. Long-term follow-up of the infants with sonographically detected brain lesions should be performed in order to determine the neurological consequences of these findings. Furthermore, HI should be compared with other hemodynamic parameters, such as Doppler indices measured in the venous circulation, in order to determine the precise hemodynamic sequence of both central and peripheral hemodynamic changes. However, one should bear in mind that alterations in the venous circulation only reveal myocardial dysfunction, offering no information on the effects of hypoxia on fetal brain structure and function. Furthermore, this sequential deterioration can rarely be observed beyond 32–34 weeks' gestation, after which fetal growth asymmetry and brain-sparing are the only evidence of placental insufficiency and fetal hypoxia13, 26, 32.

In conclusion, the C/U ratio must be calculated continuously during the prenatal monitoring of pregnancies complicated by IUGR and hypoxia, and its variation over time should be assessed. Whether blood flow redistribution (C/U ratio < 1) can be considered as a beneficial physiological adaptation depends on both the degree of_reduction in pO₂ and the duration of exposure to hypoxia. There may be harmful effects on the brain tissue if blood flow redistribution is too prolonged or too severe. The introduction of the new vascular score, HI, which takes into account both the duration and the intensity of fetal flow redistribution, could improve considerably the prevention of hypoxic brain lesions, which are one of the most common causes of perinatal morbidity and mortality.

Acknowledgements

We would like to thanks Paško Konjevoda for invaluable advice and support in the statistical analysis. This work was supported by a grant from the Ministry of Science, Education and Sport of the Republic of Croatia in collaboration with INSERM 316, Tours, France.