An Investigation Of Fetal Growth In Relation To Pregnancy Characteristics

AN INVESTIGATION OF FETAL GROWTH IN RELATION TO PREGNANCY CHARACTERISTICS by

Joe Max Mongelli MB BS, B Sc (Sydney) MRCOG Thesis submitted to the University of Nottingham for the degree of Doctor of Medicine, November 1994

CONTENTS Title page Contents Abstract Acknowledgements Abbreviations

1 2 3 4 5

Part I - Literature Review Chapter 1 Chapter 2 Chapter 3 Chapter 4 Chapter 5

The Determinants of Birth Weight Adjustable Birth Weight Standards Ultrasonic Methods of Fetal Weight Estimation Models of Fetal Growth Screening Strategies for Abnormal Fetal Growth

6 15 22 29 37

Part II- Development of Research Techniques Chapter 6 Chapter 7 Chapter 8 Chapter 9 Chapter 10

Principles of the Customised Growth Chart Methods of Gestational Age Estimation Forward Projection of Fetal Weight Estimate Selection of Ultrasonic Weight Formula Ultrasonic Study of Fetal Growth: Patients and Methods.

49 57 64 68 76

Part III - Clinical Findings Chapter 11 Chapter 12 Chapter 13 Chapter 14 Chapter 15

An Ultrasound Standard for Fetal Weight Gain Symphysis-fundus Height in Relation to Gestational Age and Fetal Weight Fetal Growth Kinetics in Relation to Pregnancy Characteristics Customised Growth Charts in Relation to Neonatal Outcome The Prediction of Birth Weight

83 95 102 115 126

Part IV - General Discussion Chapter 16 References

Comments and Conclusions

143 156

2

3

4

5

1. THE DETERMINANTS OF BIRTH WEIGHT 1.1 Introduction Birth weight is one of the most important measures we have of the health status of a population, being a strong predictor of both mortality and morbidity, and reflecting nutritional status and growth rates. Yet the estimation of the normal growth potential -and hence the definition of growth retardation - for a given individual has remained an elusive objective. Neonatal size can be influenced by a large number of variables. Kramer (1987), in a lengthy review on low birth weight, listed 43 potential causes, subdivided into 7 groups, while admitting that his literature search may not have been complete. For the purposes of our discussion, we will attempt to classify them as pathological or physiological, depending whether or not they are associated with adverse perinatal outcome. This classification will be arbitrary for many of these factors, because of our limited knowledge in this field.

1.2 Pathological Factors

A large number of pregnancy complications are associated with reduced birth

weight. Classically, growth retardation has been

classified as either symmetric or asymmetric (Pearce & Campbell, 1985), depending on whether the fetal body dimensions are proportionately reduced, or whether some degree of ‘head sparing’ has occurred. In practice, asymmetric IUGR is associated with either preeclamptic toxaemia or recurrent abruption, while most other causes lead to symmetric IUGR.

6

Hypertensive Disease. Essential, uncomplicated hypertension poses little or no risk to the fetus. In early onset or severe pre-eclampsia birth weight may be reduced by 300-500g and birth length by 1-3 cm, whereas late onset or mild

pre-eclampsia

has

no

such

association

(Fedrick

&

Adelstein,1978; Long et al, 1980). Reduced utero-placental blood flow is considered to be responsible for reduced growth.

Chronic Maternal Illness Maternal cyanotic heart disease is associated with fetal growth retardation in up to 52% of pregnancies, as opposed to 9% in the acyanotic group (Shime et al, 1987). Diabetes has complex effects on fetal growth, with a tendency towards larger babies unless associated with vascular disease and advanced maternal age, when growth retardation is more likely. It is the main pathological cause of fetal macrosomia. Severe chronic respiratory diseases such as poorly controlled asthma (Greenberger and Patterson, 1983), cystic fibrosis (Palmer et al, 1983) and bronchiectasis (Thaler et al, 1986) may lead to reduced fetal growth. In the case of

anaemia, it is difficult to separate the effects of

anaemia per se from the underlying nutritional problems. However, women with sickle cell disease, sickle-thalassaemia and sickle haemoglobin-C disease have an increased incidence of growth retarded infants (Powers et al, 1986).This is most likely due to placental micro-infarcts resulting from episodes of sickling, leading to placental insufficiency. Chronic renal disease of moderate severity is associated with IUGR in up to 24% of cases (Katz et al, 1989), although some of this effect may be related to hypertension.

7

Systemic lupus erythematosus has been implicated in fetal growth retardation (Carlson,1988), though this may be a result of the underlying renal disease or drug therapy.

Maternal Addictions Smoking mothers have babies that are 150 - 200g lighter than those of non-smokers (Wilcox et al,1993a); this appears to be caused by smoking in itself rather than other factors associated with the smoker (Doughterty,1982). Excessive alcohol intake may result in small babies with shortened palpebral fissures and a small head - the fetal alcohol syndrome (Lemoine P et al, 1968). Fetal weights are reduced by 165-200 grams among mothers who drink the equivalent of more than 50 ml of absolute alcohol per day (Ouellette et al, 1977). Heroin addiction has also been associated with reduced fetal growth (Naeye et al, 1973).

High Altitude Babies born at high altitude are lighter than those born at ground level (Lubchenko, 1963); this appears to be a 'dose-response' effect related to chronic hypoxia, with smaller babies being born at higher altitude (Yip, 1987). Interestingly, both of these studies showed higher rates of preterm delivery.

Malnutrition Maternal malnutrition, when severe, may result in adverse neonatal outcome. Stein and

Susser (1975) analysed the birth statistics in

Holland during the famine of 1944-1945; they found that birth weights declined by about 10% only when under-nutrition occurred in the third trimester with caloric intake below 1500 g. This is to be expected, since the period of greatest absolute growth is in the last 10 weeks of

8

gestation, when the average fetus gains about 2000g during this interval (Hadlock et al, 1991).

Placental Disorders. Recurrent antepartum haemorrhages in the first and second trimesters is strongly related to reduced fetal growth, possibly due to impaired development of the utero-placental circulation (Fedrick & Adelstein, 1978). Other placental anomalies associated with SGA infants include circumvallate placenta, and large chorioangiomas.

Infection Viral infections, particularly rubella and cytomegalovirus, may reduce fetal weight and length to 80-85% of normal values ( Miller, 1981; Naeye, 1967). In fatal cases of rubella, the growth restriction is associated with markedly reduced cell numbers in the fetal organs (Naeye, 1965). Listeriosis is sometimes associated with IUGR. In global terms, malaria is probably the most important infectious agent associated with growth restriction on a world-wide basis. This is related to haemolytic anaemia and placental insufficiency related to placental infestation. The pathological changes observed in the placenta include perivillous fibrinoid deposits, syncytiotrophoblast necrosis and partial loss of microvilli. A brownish discoloration may be observed (plasmodial placental pigmentation), and these cases are associated with significantly lower birth weights (Garin et al, 1985; Walter et al,1982).

Chromosomal Defects Most chromosomal and genetic disorders are associated with impaired fetal growth of varying severity. In Downs’ syndrome the birthweight is 80-90% of normal, while for trisomy 13 and Turner’s syndrome it is

9

80% and 84% respectively (Polani, 1974). More severe growth restriction is observed in foetuses with trisomy 18, with an average birth weight of only 62% of normal. Some genetic disorders such as the Seckel and Russell-Silver syndromes are associated with severe dwarfing apparent at birth. On the opposite side of the spectrum, we find a group of genetic disorders that are associated with fetal growth acceleration. These include the Beckwith-Wiedemann syndrome, in which trisomy for the IGF-II gene has been implicated, and ‘stood’ conditions associated with a dramatic increase of fibrous tissue (Elejalde et al, 1977). In Sotos’ syndrome, characterised by cerebral gigantism , the birth weight is not significantly increased but the birth length is increased to a mean of 55.2 cm.

1.3 Physiological Factors

Duration of pregnancy. The length of gestation is the most important determinant of birth weight (Wilcox et al,1993b), and also of perinatal mortality and morbidity in the pre-term period (Allen et al, 1993). The 'terminal flattening’ seen in birth weight standards based on menstrual data is artefactual; it is much less marked in standards derived from ultrasound-dated populations (Wilcox et al, 1993a).

Parental size. The relationships between birth weight and parental size have been studied extensively, both in humans and in animals. The classic studies by Walton and Hammond (1938) on crosses between the Shire horse and the Shetland pony showed that the birth weights of foal born to Shetland dams of Shire sires were close to those of pure Shetlands; conversely, foals of shire dams by Shetland sires were close to those

10

of the pure breed. Evidence in humans on the preponderance of the maternal effects on fetal growth has been presented by Cawley (1954) and Ounsted (1966). Both maternal height and weight have positive correlations with birth weight, the latter being the stronger factor. Low maternal pre-pregnancy weight is significantly correlated with both preterm delivery and low birth weight (Garn 1990). Obesity, as measured by the body mass index, is only weakly correlated (Abrams, 1986; Wilcox et al, 1993b). The parents' own birthweight is significantly correlated with that of their offspring (Alberman et al, 1992).

Parity. The positive effect of parity on birth weight has been documented in most races and many mammalian species (Ounsted, 1973; Bantje, 1985), suggesting that its mechanism may have an evolutionary advantage. Garn (1990) has argued, on epidemiological grounds, that the effect of parity is a result of the increase in the maternal prepregnancy weight seen in developed countries, rather than an independent factor. This does not agree with multiple regression analysis of birth weight, which shows parity to be a factor independent of mid-pregnancy weight (Wilcox et al, 1993b); this may be because the mid-pregnancy weight is a compound variable, dependent on both the pre-pregnancy weight and the maternal weight gain. There is some evidence that the effect is partner-specific, i.e. a change in partner may be associated with a reduction in the birth weight of the first born of the new relationship (Warburton & Naylor, 1971). It may be argued that because of the significant increase in the perinatal mortality of first-born infants and those of mothers of high parity, this factor should be classified as a pathological variable. The odds ratios, however are only slightly elevated (Kirkup & Welch,1990), and do not justify this re-classification.

11

Race. Meredith (1970) published an extensive description of the variations in birth weight among different ethnic groups. Of the 78 groups considered, the largest newborns were found in Anguilla and Nevis, weighing a mean of 3.88 Kg. The smallest babies were those of the Lumi tribe in the Toricelli mountains in New Guinea, with a mean birth weight of 2.4 Kg. These ethnic differences clearly persist in mixed populations from the same location (Cheng et al, 1972; Wilcox et al, 1993b). Birth weight variations do not always correlate with trends in perinatal mortality. In Singapore, Malay babies have a much higher perinatal mortality than Indian babies even though their percentage of

low-birthweight (<2500) is significantly smaller

(Hughes, 1984), both groups living in

similar socio-economic

conditions with total health care coverage. Similarly, Californian black babies under 3001g have much lower mortality rates than whites, even though their birthweights are lower (Williams et al, 1982).

Sex The female newborn weighs on the average 118 g less than the male (Wilcox et al, 1993b), and this has been observed in most ethnic groups studied (Meredith, 1970).Animal studies suggest that the XY embryo has a growth advantage at the earliest stages of organogenesis (Snow, 1989); hormonal differences are not responsible, since the sex differences are noted even in anencephalic foetuses. It is of interest that in spite of this, females born preterm have lower mortality and morbidity than males (Allen et al, 1993). This may be due to female infants having relatively more energy stores in adipose tissue than males (Oakley et al, 1977).

12

Others Women with a history of SGA infants are more likely to give birth to small babies. It is not clear whether this is due to genetic factors, or recurrence of genetic/pathological factors. Work performed by Ounsted (1965,1966) has shown that mothers who have borne SGA infants had themselves lower than average birth weights, although their adult height did not differ significantly from that of women who had given birth to babies of normal weight. Consanguinity in parents has been shown to cause a significant reduction in birth weight in Pakistan (Shami et al, 1991), Japan (Morton, 1958) and Norway ( Magnus et al, 1985). A seasonal trend has been observed in birth weights, with significantly lower values in summer and a peak in winter/early spring (Matsuda, 1992). These fluctuations are rather minor, occurring within a 100g band.

1.4 Discussion The distinction between 'pathological ' and 'physiological' factors is to some extent arbitrary, with a significant 'grey zone' of uncertainty. In terms of defining normal growth potential, the genetic components of the physiological factors are probably more important, and this was stressed by Lazar and colleagues (1975). This distinction is, however, an important exercise in order to develop valid adjustable growth standards. It is likely that fetal development is under the control of inhibitory and stimulatory growth factors, and that some physiological and pathological factors may well act through common pathways. Animal studies have shed some light on the relative importance of fetal genome and maternal effect; these have been reviewed by Snow (1989). There is good evidence to suggest that maternal effects operate

13

late in gestation, whereas early fetal growth is controlled by fetal genetic mechanisms. Winick (1971) studied the cellular basis of fetal growth using animal models. Three phases of growth were identified: cellular hyperplasia, followed by both hyperplasia and hypertrophy, and then predominantly hypertrophy. Depending on the timing and duration of experimental insults, different forms of growth restriction were observed. Disturbances in early pregnancy restricted the total cell number, so that no 'catch-up' growth was possible, whereas later in pregnancy

cell size was predominantly affected with minimal

reduction in cell numbers , and post-natal recovery was possible with adequate nutrition. This points to the heterogeneous nature of growth disturbances and to the need for using appropriate standards. It is difficult to determine how much of the differences observed among different ethnic groups are due to genetic factors, as opposed to environmental

factors such

as nutrition

and

socio-economic

conditions. Hence customising for ethnicity can only be justified when the adjustment factors are derived from sub-populations in the same location and ideally living under similar socio-economic conditions. The observed differences in neonatal morbidity between different sub-populations do not always agree with the birth weight differences. This lends weight to the argument that, for optimal performance, fetal weight for gestation as an index of morbidity needs to be evaluated in relation to other pregnancy characteristics.

14

2. ADJUSTABLE BIRTH WEIGHT STANDARDS

2.1 Introduction Birth weight on its own is only a crude indicator of neonatal welfare, being a retrospective measure from which

it is difficult to make

accurate inferences about prenatal growth kinetics. The definition of 'low birth weight infant' as being a newborn weighing below 2500g was used as an index of prematurity until the 60's, when it was realised that a considerable proportion of these cases were in fact growth restricted (Ounsted, 1970). It has been nevertheless a convenient tool for epidemiologists, since reliable

data on the

duration of gestation is often difficult to obtain, particularly in developing countries. This, however, fails to make the important distinction between infants who are small because they were born preterm and those term babies who are small because of constitutional or pathological factors.

2.2 Pathological Implication of Abnormal Birth Weight This has prompted the search for birth weight for gestation standards, so that given the appropriate variables this distinction can be made. When birth weight is analysed as a function of gestation, some important relationships with poor perinatal outcome emerge. In a large study by Patterson et al (1986) of a database of 44 811 cases, a U-shaped relationship was found between birth weight centile and the incidence of morbidity , with minimal morbidity near the middle ranks of birth weight ; the percentage of the total poor perinatal outcome occurring below the tenth or above the 90th centiles increased linearly from 16% at 28-29 weeks to 57% at 40-41 weeks. There is mounting evidence that being small for gestational age has pathological implications extending into childhood and adulthood. Hill et al (1984) , in a small study, related the outcome of term infants

15

with nutritional parameters. Malnutrition in the newborn was defined in terms of subcutaneous tissue thickness. About 45% of the 33 malnourished infants had birth weights below the 10th centiles; in this group, poor outcome included reduced educational achievement up to the age of 14. Rantakallio(1985) studied a cohort of 12,000 children in Finland followed up to 14 years; it was found that the incidence of neuro-behavioural disturbances was significantly higher in weight percentile classes below the median. Most of these and similar studies are flawed by the use of menstrual dates in determining gestation . As the error tends to be towards overestimation (Gardosi & Mongelli, 1993), more babies would be classified incorrectly as below the 10th centile than those assigned above the 90th centile.

2.3 Assignment of Gestational Age With very few exceptions, gestation is usually estimated from menstrual dates, rounded down to 'completed weeks'. Typically, when birth weight is plotted against gestation, the resulting standard curves show considerable flattening near term, and this has been attributed either to placental ageing/insufficiency, or to physical restriction of growth. More recent standards in which gestational age has been calculated on the basis of early ultrasound measurements show a much more linear relationship between duration of pregnancy and birth weight (Lindgren, 1988; Wilcox et al,1993a). This inaccuracy in the estimation of gestational age is also likely to lead to a greater apparent variance in the birthweight distribution for any given week of gestation.

2.4 Preterm Delivery and Birth Weight Standards Birth weight standards have in the past been referred to as 'fetal growth curves'. Apart from the fact that these are cross-sectional studies, values derived from preterm deliveries cannot be regarded as

16

representative of normal growth. Furthermore, unless the population sample is very large, the number of babies born preterm is relatively small, leading to a greater incidence of sampling errors. Preterm delivery may be associated with growth restriction , and birth weight norms at these gestations may be well below those derived from serial ultrasound weight estimations (Ott, 1993). An experimental model of growth retardation supporting this epidemiological data was described by Alexander (1964). He, and subsequent workers, found that the excision of endometrial caruncles in the sheep (before pregnancy) resulted not only in an increased rate of growth retardation, but also in increased preterm labour rates and intrauterine death. Another indicator of pathology in the preterm period is the statistical distribution of birth weights. Whereas the distribution of birth weights at term shows a significant positive skewness, in the preterm period this becomes negative (Wilcox et al, 1993a), most likely because of the greater number of growth-retarded babies born at these gestations. It has also been shown that preterm babies delivered electively are significantly lighter than those born spontaneously, yet even when the former are excluded from the analysis the negative skewness is reduced but not entirely eliminated (Yudkin, 1987).

2.5 Effect of Environment Reference standards may also be strongly affected by the environment and population characteristics. For example, Lubchenko's (1962) birth weight chart has been used widely in the United States and elsewhere. This was derived from a population in Denver, Colorado, at an altitude of about 10000 ft. Not only did this high altitude result in lower birth weights for all gestations than any other published standard, but also the percentage of births occurring preterm was much higher than expected. There is also some evidence that birth weight standards may

17

change historically as living standards and social characteristics vary (Alberman 1991; Ulizzi & Terrenato, 1992).

2.6 Adjustable Standards The realisation that these genetic or physiological effects are in operation led to attempts to introduce adjustment factors to allow for maternal characteristics. Thomson and colleagues (1968) published birth weight for gestation tables allowing for sex, parity and inclusive of correction factors for maternal weight . Altman and Coles (1980) produced nomograms for the calculation of birth weight centiles based on this data that included correction factors for parity, maternal height and weight, and fetal sex. Lazar and colleagues (1975) used multiple regression analysis to derive correction factors for both maternal and paternal weight and height, claiming that paternal weight is as important as maternal weight; they believed that the effect of these variables is largely of genetic origin, and in order to improve their predictive power they estimated what the parental values of height and weight would be at the age of 20 before entering them in their regression model. Parity and ethnic group were not considered in their analysis, and their model was not tested prospectively. Voigt (1989) and Mamelle(1989) published elaborate tables to allow adjustment for these variables, but the fact that they are not in general use attests to their complexity. Some interesting similarities among different birth weight standards were described by Dunn (1989). When the centile cut-off points were expressed as percentages above or below the population median and plotted against gestation, virtually identical values were obtained for all the standards. This remarkable correspondence led to the construction of the Bristol Perinatal Growth Chart, a method that would allow the production of antenatal and post-natal

growth

18

standard for population sub-groups. The method assumes that the latter are normally distributed.

2.7 The Birth Weight Ratio An alternative variable to describe size for gestation , the 'birth weight ratio' , was described by Brooke and colleagues (1989) in order to analyse factors affecting birth weight. This is simply the observed weight divided by the weight expected for a given gestation. Morley and colleagues (1990) described a relationship between birthweight ratio and the need for mechanical ventilation and post-neonatal mortality in preterm infants; this, however, was not observed in the study of Brownlee et al (1991). More recently, Wilcox and colleagues (1993b) analysed a large database of 31 561 computerised records of term deliveries in order to develop a multiple regression model to predict birth weight. The variables included gestation, sex, maternal height, weight, parity and ethnic group. The ratio of the observed birth weight to predicted weight ('individualised birth weight ratio', IBR) can then be calculated by a computer program and expressed as a centile value. This method has been reported to identify a higher proportion of truly growth retarded infants, as defined by neonatal ponderal index and skinfold thickness measurements (Sanderson, 1994). The drawback of Wilcox's program is that in its present form it is only applicable to babies born at term, and cannot be used for screening purposes in the antenatal period. It can be shown that when birth weight ratios are transformed into centile values, these are very similar to the corresponding birth weight centiles, provided the reference standards used to obtain the mean and standard deviation are similar (chapter 16). 2.8 Discussion The large number of birth weight standards in existence is a reflection of the importance given to this parameter, as well as the need to relate

19

birth weight to local conditions. In the English literature alone, Goldenberg and colleagues (1989) were able to review 13 such standards published since 1963. Large differences were noted in the 10th centile cut-off point, greater than 500g for some gestational ages. These discrepancies are partly due to inconsistencies in methodology. For example, McKeown and Gibson (1951) included both live and still births in their analysis of the Birmingham data, whereas most investigators have restricted their samples to live born infants. The treatment of outliers in the data can vary between studies. A variety of corrections for bimodal or skew birthweight distributions have also been adopted (Gruenwald, 1966; Milner & Richards, 1974). A number of studies are also limited by small sample sizes, making it difficult to estimate centile distributions of birthweight with any degree of accuracy. Another major source of error is gestational age assignment. Assessment of fetal well-being, by whatever means,

requires an

accurate estimate of gestational age. The introduction of routine early ultrasound scanning in the United Kingdom has eliminated large errors, but the use of '10-day' or '7-day' rules whereby menstrual dates are used in preference to ultrasound determined dates if in agreement, may lead to some loss of accuracy (see Chapter 7). Those women who book late tend to have poorer outcomes, and ultrasonography may be of special benefit in this group. Although algorithms have been developed for the accurate determination of gestational age up until 32 week's gestation (Sabbagha et al, 1978), these have not gained widespread acceptance. All of the adjustable standards of fetal growth are limited by their use of cross-sectional birth weight data. While they may be valid for the assessment of relative size, they are not suitable for assessing serial weight estimates, i.e. growth (Altman, 1994).

20

With the exception of the IBR, the correction factors are usually presented in tabular form, and do not take into account gestationdependent variations. Thomson and colleagues (1968) stated that these adjustments should not be made for gestations under 37 weeks, since their numbers was limited in that range. Nevertheless, their parity differences were statistically different even at 32 weeks. Even if there were sufficient numbers, it is doubtful that these adjustment factors would be applicable to intrauterine fetal weight estimates. In practice most clinicians do not adjust beyond sex and parity, probably because of the inconvenience in using complex tables or graphs. In the standard published by Yudkin and colleagues (1987) - widely used in paediatric units in the UK-, no adjustment is made apart from fetal sex. Although the importance of accurate and valid fetal growth standards has long been acknowledged, the validity of specific growth standards when applied to a particular population or study sample is seldom tested. As a result, the assessment of growth retardation and evaluation of screening procedures may be inaccurate and biased.

21

3. ULTRASONIC METHODS OF FETAL WEIGHT ESTIMATION

3.1 Introduction Fetal weight estimation plays an important part in clinical obstetrics decision-making, often used in screening for IUGR, the management of diabetes in pregnancy and pregnancy complicated by breech presentation. In current practice, ultrasonography remains the most accurate method for determining the estimated fetal weight (EFW). Although newer imaging techniques such as computerised tomography and nuclear magnetic resonance are likely to be much more accurate (Baker et al, 1994), their cost will prevent widespread use; their main role in the immediate future will remain as research tools. Threedimensional ultrasound equipment, on the other hand, is now affordable, and the better, more reliable definition of anatomical planes (Kuo, 1991) should lead to reduced operator error and hopefully to better performance of the existing formulae.

3.2 Fetal Weight Estimation Formulae The equations for fetal weight estimation in terms of given ultrasound parameters are usually derived by applying a model of fetal weight composition to a source population examined shortly before delivery. The measured ultrasound parameters and the birth weights, are entered and the relevant coefficients are estimated by multiple regression analysis. The performance of the formula in term of its prediction errors is then tested on a separate sample, and 'target' population. One of the first such formulae to be developed was that of Campbell & Wilkin (1975 ); this was based on the fetal abdominal circumference (AC), and it is still in common use in the UK A considerable number of other formulae that usually employ more than one ultrasound parameter have since been published. A sample of these are listed in

22

table 3.1. They can be broadly classified as exponential or nonexponential, depending on the type of mathematical expression. Exponential formulae take the form of:

EFW = exp[F(P1,..Pn)] where F(p1,..pn) is a polynomial function of the ultrasound parameters P1 to Pn. Other approaches to weight prediction have been explored. A computer neural network program has been developed specifically to estimate weights in foetuses at risk of macrosomia (Farmer et al, 1992);

ultrasound

parameters

were

combined

with

clinical

measurements such as fundal height, with a reported accuracy of around 5%. Birnholz (1986) published an algorithmic method for weight estimation, whereby one of two formulae are chosen by a computer program depending on the body proportions of the individual. About 90% of cases had an error less than 80 g/Kg . This method requires regression analysis of the fetal ultrasound parameters in the population under study.

3.3 Clinical Performance of Weight Estimation Formulae This is usually assessed by the statistical analysis of the errors. They may be expressed as signed or absolute percentage errors, absolute error in grams, errors in grams per Kg of fetal weight, and percent of errors beyond a given threshold. A typical weight formula employing more than one parameter will estimate 75% of cases within 15% of the actual weight (Thompson et al, 1990).The most common practice is to report the mean error and its standard deviation (SD); the former gives a measure of the tendency to under- or over-estimate (the systematic error), whereas the latter indicates the spread of the errors. It has recently been suggested that the standard deviation should be replaced

23

by the 95% confidence limit of the errors (Bland & Altman, 1986); this is certainly preferable in situations when the distribution is not Gaussian. There is general agreement that equations employing two or more ultrasound parameters are more reliable than those using only one (Hadlock et al, 1985; Guidetti et al, 1990). Formulae developed within an institution tend to perform better than those from other centres (Thompson et al, 1990), probably because of significant interobserver variability (Chang et al, 1993) and differences in equipment and populations. For example, formulae derived from Chinese populations perform better on Chinese patients than those developed from European populations (Chang et al, 1991). It has been reported that continual review of the results obtained by the methods used by an obstetric ultrasound department may further enhance its performance (Thompson et al, 1990). In Hadlocks' studies (1985), the prediction errors of equations employing three or four ultrasound parameters (BPD, HC, FL and AC) had a SD of around 8%. Slightly better values were reported by Issel and colleagues (1991), a SD of about 7% by measuring up to 7 ultrasound parameters. In clinical practice these errors tend to be somewhat higher (Miller et al, 1988). The problem of systematic over- or under-estimation of fetal weight is frequently reported when such formulae are used by centres other than the one where the formula originated (Robson et al, 1993). Some of this error may be due to the variations in the lag times between ultrasound examination and delivery, which is not usually allowed for by the authors of the formulae; this means that a fetus examined some days before delivery will be slightly lighter than at birth, and when the birth weight is entered into the regression analysis without due modification, a small but appreciable over-estimation will take place. This problem was appreciated by Spinnato and colleagues (1993), who introduced a time component into the established formulae, valid up to 35 days before delivery. A more serious and common problem is the

24

existence of

trends in the errors. There may be significant and

negative correlation with the size of the fetus (Robson et al, 1993; Miller et al, 1988); hence small babies tend to be overestimated while large ones are under-estimated. This can lead to serious data distortion when producing normal values in growth curves for fetal weight, and may affect the performance of screening programs for small- and large- for gestational age infants.

3.4 Discussion Early retrospective studies on the detection of growth retarded foetuses by measuring the biparietal diameter suggested that this parameter could be helpful in defining groups of cases with higher perinatal mortality and preterm delivery rates (Persson et al, 1978). The technique, however, was subsequently found by Campbell & Dewhurst (1971) to have a false positive rate for SGA of 25%. This is not surprising, since the correlation coefficient of BPD with birth weight is not as high as other ultrasound parameters such as the abdominal circumference and femur length (Favre et al, 1993). Several studies have been published on the performance of different ultrasound parameters in the detection of the SGA fetus. Neilson and colleagues (1984) measured a series of fetal parameters in the third trimester; they found that the product of trunk area and crown-rump length (as an index of fetal weight) was superior to the trunk diameter alone. Dudley and colleagues (1990) showed that EFW was the best of four ultrasound parameters in identifying the small-for-dates infant. Similarly, Chang and colleagues (1993) reported that a single EFW estimate based on multiple ultrasound parameters was superior to abdominal circumference in predicting 2 out of 3 indices of neonatal nutritional deprivation. The efficacy of growth screening programs continues to be limited by the error of ultrasonic EFW and by the lack of a uniform standard for fetal growth and growth velocity. The

25

normal standards of EFW published to date show considerable disagreement, and at least some of the differences may be attributable to the choice of weight estimation formula. This is further discussed in chapter 9. Volumetric formulae for fetal weight estimation have been proposed by several workers, including Combs (1993 ) and Birnholz (1986), on the grounds that fetal volume is proportional to weight when the specific gravity is constant. There are at least two theoretical objections to this argument. Firstly, the gestational age-dependent changes in specific gravity have not been described, and the magnitude of error from this factor is unknown. Secondly, growth retarded babies would have considerably less fat stores, increasing their specific gravity and thus leading to underestimation of weight. That this may be the case is suggested by the fact that Birnholz noted systematic underestimation of fetal weight in the under-1000g group, for whom he had to apply a recursive correction formula. In any case, the claimed improvements in accuracy of their methods have not been confirmed by independent workers. Birnholz (1986) has suggested that , on the grounds of information theory, averaging serial fetal weight estimates would improve the final estimate, with the expected improvement being related to the square root of the number of measurements. This makes the assumption that, for a given individual, the error in fetal weight estimation on each occasion is random, i.e. the signed error values are not correlated with each other. This particular issue has not been reported on in the literature. All of the commonly used formulae place an emphasis on bony landmarks and do not use any other soft tissue measurements apart from the AC. While bony landmarks are accurate for the purposes of estimating gestational age, the emphasis on these parameters could explain their relative inaccuracy in estimating weight.

26

New weight estimation formulae should be explored that include additional measures of soft tissue parameters. These were explored by Favre and colleagues (1993), who reported better performance in the small for dates group using the thigh circumference and femur length. The standard deviation of the error was nevertheless fairly high at 15.9%, and they did not compare their formulae with older established equations. Other measures that should improve not only accuracy of fetal weight estimation but also performance of other tasks include continual audit and quality control, to ensure consistent techniques and peak performance of equipment. Plastic ‘phantoms’ have been designed in order check the technique and accuracy of the measurements performed by ultrasonographers, but these are not in common use in the UK. The scope for making major errors in estimating growth velocity from ultrasound fetal weight estimation has been pointed out in correspondence by Gardosi (1994b). Substantial gains in accuracy will be needed before abnormalities in growth velocity can be reliably detected.

27

Table 3.1 Ultrasound fetal weight estimation formulae.

Authors Campbell & Wilkin

Exponential formulae wt=1000*exp(-4.564 +0.282*fac -0.00331*fac*fac)

(1975) Hadlock et al (1985)

wt=exp(2.695 +0.253*fac -0.00275*fac*fac)

Hadlock et al (1985)

log10(wt)=1.3598 +0.051*fac +0.1844*fl -0.0037*fac*fl

Hadlock et al (1985)

log10(wt)= 1.4787 -0.003343*fac*fl +0.001837*bpd*bpd +0.0458*fac +0.158*fac

Hadlock et al (1985)

log10(wt)=1.3596 -0.00386*fac*fl +0.0064*hc +0.00061*bpd*fac +0.0424*fac +0.174*fl

Shepard et al (1982)

wt=1000*exp(-1.7492 +0.166*bpd +0.046*fac 0.002646*fac*bpd)*(ln(10))

Warsof et al (1977)

wt=1000*exp(2.302585*(-1.599 +0.144*bpd +0.032*fac -0.000111*bpd*bpd*fac))

Persson et al (1986)

wt=exp(ln(10)*(0.972*ln(bpd)/ln(10) +1.743*ln(ad)/ln(10) +0.367*ln(fl)/ln(10) -2.646))

Balouet et al (1992)

wt=0.1135exp(0.739*ln(fac) +1.179*ln(ethc) -0.041*ln(ithc)) Non-exponential formulae

Combs et al (1993)

wt=0.23718*fac*fac*fl +0.03312*hc*hc*hc

Dudley et al (1990)

wt=4.1*fl*apa +0.86*fl*hpa

Shinozuka (1987)

wt=0.23966*fac*fac*fl +1.6230*bpd*bpd*bpd

Birnholz (1986)

wt=(3.42928*bpd*ad*ad/1000) +41.218)

Birnholz (1986)

wt=1.0206*{1.88496*[0.01*fl*ad+0.01667*bpd*ad + 0.01*bpd*bpd]*[(((-0.0069558*fl) +1.7394)*fl/10) -3.3626]} -61.537

28

4. MODELS OF FETAL GROWTH

4.1 Introduction

Awareness of the limitations of birth weight standards as indicators of fetal growth have led to the pursuit of ultrasound-defined intrauterine growth standards. But while it is relatively easy to obtain birthweight data from large populations, to derive ultrasound-defined standards requires considerable effort in manpower, logistics and equipment. Usually this data originates from ultrasound departments, and often does not contain the clinical details of individual cases. As a result, all of the fetal weight standards published to date have been derived from relatively small samples. The norms for commonly measured ultrasound parameters such as the AC, FL and BPD are well established, yet relatively few studies specifically address the issue of intrauterine weight gain. This is partly because for the purposes of growth monitoring,

most

ultrasound departments plot the individual measurements rather than weight estimates. This in spite of several studies suggesting that the EFW is at least as good as the AC for the detection of IUGR (Chang et al, 1993; Hedriana & Moore,1994). Here we review the literature on intrauterine weight curves, and describe an 'average' growth curve, based on published data.

4.2 Comparative Analysis of Ultrasound-Derived Growth Curves

A total of seven studies describing intra-uterine weight gain were retrieved .Studies on the growth of linear ultrasound parameters without weight estimations were excluded, since derived fetal weight curves can differ markedly depending on the weight equation being used (see chapter 8 ). The characteristics of these studies are

29

summarised in table 4.1; these include population samples, weight estimation formulae, mean birth weight and the methods of data analysis. The fetal growth kinetics described by each study are displayed in table 4.2. In order to describe the shape of the growth curves independently of the predicted term weight, growth can be expressed as a percentage of the predicted 280-day fetal weight, and plotted as fractional growth curves. This allows comparisons in terms of arbitrary descriptive landmarks such as gestation at which 50% of the term weight is reached (G50), and the percentage of term weight that is expected at 28, 37 and 42 weeks (P28, P37, P42). Figure 4.1 shows the medians of the ultrasound EFW curves plotted to 42 weeks and also the corresponding birth weight data derived from the East Midlands Obstetric Database (Wilcox et al, 1993a). Figure 4.2 shows the derived fractional growth curves, as a percentage of term weight. The equation for average fractional curve was obtained by taking the arithmetic average of the coefficients of the derived growth functions listed in table 4.1. This is plotted in figure 4.3; only a minimal degree of deceleration is noted at term.

4.3. Alternative models of fetal growth

Rossavik and Deter (1986) proposed a sigmoid function to describe fetal growth of any parameter, including weight. This function is of the form: P= c(t) k+st where P is the ultrasound parameter, t is the duration of growth, k a fixed coefficient determined by the anatomical characteristics, c and s constants related to growth regulatory processes. This function allows the prediction of individual 'normal' growth channels based on two

30

separate ultrasonic examinations before 27 weeks. This model was applied prospectively by Simon et al (1989) to a number of parameters including fetal weight. They found a small but significant systematic error of overestimation for most of the parameters and fetal weight; the standard deviation of the errors for fetal weight ranged from 6.7 % to 9.4% , depending on gestation. This is well within the range of the published errors of weight estimation formulae. The advantage of this model is that reference charts are no longer needed; instead, growth disturbances may be detected as deviations from the individually projected standard. The main drawbacks are the need for two ultrasound examinations before 27 weeks' gestation, spaced at least 5 weeks apart, and the need for appropriate computer equipment and software to carry out complex calculations.

4.4 Discussion

Most of the differences between the published ultrasound growth curves become apparent during the term period. They agree within a 100g band up to about 36 weeks gestation (figure 4.1). Beyond this point, Jeanty's curve shows marked deceleration, whereas Deter's and Otts display moderate acceleration. The remaining four curves continue a linear trend evident from about 28 weeks. Jeanty's abdominal circumference values are markedly below other standards in late pregnancy, and this probably accounts for the deceleration in his weight curve. The overall fractional average curve (figure 4.3) shows only minimal deceleration at term, and this is in contrast to birth weight standards based on menstrual data. Another approach to deriving this curve would have been to use a weighted average; the problem here is that two of the studies are cross-sectional. In any case, the largest studies are included in the middle five curves, and thus it is

31

doubtful that a weighing procedure would change the shape of the curve significantly. On the basis of their published birth weight data, Ott's and Deter's weight formulae overestimate fetal weight at term by 255g and 471 g respectively, and this may account for the steeper slopes of their curves. Larsen's study is the only one to use birth weight in order to select the optimal growth model; yet even here there is a systematic overestimation at term of about 170g. This contrasts with Hadlock's study, of similar (cross-sectional) design, where an overestimation of only about 20g was observed, probably because of better performance of the weight estimation formula. Of the five studies whose weight formula was not developed locally, none compares more than 3 different weight formulas in order to select the best. As will be discussed in chapter 8, the type of weight equation selected may result in differences of more than 300g at term. Hence, when producing standards of intrauterine weight gain, measures should be taken to correct any systematic error due to the weight estimation formula, since other centres will not necessarily employ the same formula. It is unlikely that the method of gestational dating makes a significant contribution to the observed differences among these studies. This is because, with the exception of Ott's study, menstrual dates were used only if in close agreement with early ultrasound measurements; if they did not agree, gestation age was estimated from the early ultrasound measurements of the biparietal diameter . The study by Larsen and colleagues is the only one to produce separate standards for males and females; they describe a mean weight difference between the sexes of 3.8%, but do not elaborate on whether this holds true for all gestations or only for part of pregnancy. It is apparent from figure 4.1 that all of the ultrasound derived medians are higher than the birth weight data, by an average of about 100g.

32

Systematic weight estimation errors may account for some of this difference, but other factors may be at play , since the differences persist in the studies where this error was minimal, such as Hadlock's or even negative, as in Jeanty's. The most likely explanation is that infants born preterm are more likely to be growth retarded, and preterm delivery in these cases is an escape mechanism from an adverse intra-uterine environment. Complex mathematical models such as Rossavik's, irrespective of their validity, are unlikely to improve birth weight prediction in view of the magnitude of ultrasound error. A recent study by Shields and colleagues (1993) has shown that serial plotting of fetal measurements on normal curves is as accurate in this respect as complicated mathematical modelling. This would also be expected on the basis of information theory (Birnholz, 1986).

33

Figure 4.1. Ultrasound-derived fetal growth standards compared with Nottingham’s birth weight standard. The continuous curves represent the ultrasound-derived standards published by Hadlock et al, Gallivan et al, Ott, Persson & Weldner, Deter et al, Larsen et al and Jeanty (using Shepard’s weight formula). The values for the birth weight standard by Wilcox et al are displayed as triangles. The middle four curves (Hadlock, Gallivan, Larsen and Persson) are closely related.

34

Figure 4.2. Proportional growth curves for ultrasound-derived fetal growth standards. The ultrasound-derived standards published by Hadlock et al, Gallivan et al, Ott, Persson & Weldner, Deter et al, Larsen et al and Jeanty (using Shepard’s weight formula) have been transformed into ‘proportional’ growth curves, whereby the values for each gestation represent the percentage of the predicted 280 day fetal weight.

35

Figure 4.3. Average proportional fetal growth curve. The transformed proportional fetal growth curves of the ultrasound-derived standards published by Hadlock et al, Gallivan et al, Ott, Persson & Weldner, Deter et al, Larsen et al and Jeanty (using Shepard’s weight formula) have been averaged arithmetically to yield an average curve .

36

5. SCREENING STRATEGIES FOR ABNORMAL FETAL GROWTH

5.1 Introduction

In spite of the widespread introduction of obstetric ultrasound, our clinical ability to detect the small-for-dates fetus remains poor, with only about 30% -50% of cases detected in the ante-natal period (Jones, 1986; Hepburn & Rosenberg, 1986) . We are also rather ineffective in detecting fetal macrosomia, with sensitivities of about 50% (Sandmire, 1993; Pollack et al, 1992). The question has been raised on several occasions of whether antenatal detection of growth disturbances is going to significantly affect neonatal prognosis . While there is no long-term follow-up data on this issue, there is some evidence that those SGA infants that are detected tend to have a better short-term outcome than the undetected cases (DeCourcy-Wheeler, personal communication). Hence we should persist in our efforts to improve the antenatal detection of the potentially compromised fetus. There are two basic methods in practice for the detection of fetal growth anomalies: obstetric ultrasound and assessment of the fundal height.

5.2 Fetal Ultrasonography

Obstetric ultrasound has been investigated as a screening technique since the early 70's (Campbell, 1971).While it has proved successful in the detection of congenital abnormalities (Chitty et al, 1991) and in establishing gestational age, the detection of growth restriction and growth acceleration have remained much more elusive goals. Fetal macrosomia is a common cause of concern for obstetricians, and it is

37

common practice to refer cases to the ultrasound department for fetal weight estimation. There is, however, mounting evidence that so far the performance of ultrasound in this weight group is poor compared with other categories (Sandmire,1993). Two ultrasonic methods are in common use for the detection and assessment of fetal growth abnormalities: serial and static measurements of fetal anatomy . At least five randomised trials have been performed since 1984 to assess the efficacy of antenatal ultrasound as a screening tool for growth restriction or developmental anomalies. These were published by the following authors:

1. Bakketeig LS and collegues (1984) 2. Waldenstrom U and collegues (1988) 3. Ewigman B and collegues (1990) 4. Ewigman B and collegues (1993) 5. Newnham JP and collegues (1993)

The results have been somewhat contradictory and inconclusive. A significant reduction in the number of SGA infants, a modest increase in the mean birth weight and a significant reduction in the induction rate was demonstrated by Waldenstrom and colleagues, following routine scanning at 12 weeks. This was attributed to a reduction in smoking due to visualisation the baby. On the other hand, in a group undergoing both Doppler and ultrasound imaging on up to 5 occasions, Newnham and colleagues noted a slight but significant increase in the SGA frequency in the screened group. In the largest randomised study to date (4) involving 15151 low-risk pregnancies , no significant differences in outcome were noted between the screened and the routine management group. The latter, however, did undergo ultrasound examination when clinically indicated, thus limiting the scope of the inferences that can be made from this study.

38

Optimal strategy for the detection of growth restricted fetus continues to be a controversial issue (Daniellan and colleagues,1994). The study by Chang and colleagues (1993) suggests that the use of either growth velocity or single fetal weight estimates is rather limited at detecting the truly growth restricted fetus as defined by neonatal morphometric indices; at a false positive rate of 10% only 20-40% of the growth retarded fetuses were detected. Similar figures were reported by Daniellan and colleagues (1993). Both of these studies used morphometric indices as the definitive criteria for IUGR; the limitations of these indices are discussed in chapter 9. The data presented by Hedriana and Moore (1994) suggests that a single ultrasound examination is nearly as good as multiple examinations in predicting the birth weight, but they did not test the hypothesis that growth velocity as assessed by multiple measurements is a better predictor of poor outcome than fetal weight estimated from a single measurement.

5.3 Fundal height assessment

5.3.1 Introduction The earliest report on measuring the symphysial-fundal height (SFH) was published in the German literature by Spiegelberg in 1891. Rumboltz and

McGoogan (1953) were the first to describe a

relationship between reduced growth of the uterine fundus and 'placental insufficiency'. Since then, many conflicting reports have been published on screening for growth disturbances by the clinical measurement of the symphysial-fundal height (SFH).

5.3.2 Precision and accuracy of SFH measurements. The

estimation

of

symphysis-fundus distance

is

subject

to

considerable error. Bagger and colleagues (1985) reported an average

39

intra-observer variation of 1.5-2 cm and an inter-observer variation of 4 cm; these were not correlated with the actual SFH measurements. Some observers were found to consistently overestimate or underestimate .The accuracy of SFH measurements was checked by comparing clinical measurements with those obtained by ultrasound guided measurement; the differences between the former and the latter ranged from -0.2 to +2.7 cm. Calvert (1982) found intra-observer and inter-observer coefficients of variation of 4.6% and 6.4%, slightly lower values than those published by Bagger's group. The limits of agreement of the inter-observer variation were estimated in a study by Bailey and colleagues (1989) to be -5.0 to +1.6 cm, corresponding to a coefficient of variation of 4%. This study highlights what is probably the major shortcoming of SFH measurents: that the error due to interobserver variation, even between experienced practitioners, is too wide in relation to the standard deviations in the published reference charts.

5.3.3 Fetal weight estimation by fundal height measurement. Estimation of fetal weight by unaided clinical palpation was reported by Loeffler (1967) to be accurate within 450g of the birth weight in 80% of cases; it is of interest that in this study the accuracy of the individual observers improved with experience. The first attempt to estimate fetal weight by measuring the fundal height was reported by Johnson and colleagues (1954). This method included correction factors for engagement of the fetal head and obesity.The standard deviation of the reported errors was 353g, which is slightly greater than ultrasound estimation using Campbell's formula for abdominal circumference(Campbell & Wilkin,1975). More recently, a Belgian study of an African population showed that SFH was more closely related to fetal weight than gestation (De Muylder and colleagues,1988).

40

5.3.4 Derivation of SFH standards. In all the studies, women without 'sure' dates or an early dating ultrasound scan were excluded. In six of the nine papers reviewed the data was filtered by removing those cases whose birth weights were outside arbitrary limits; depending on the degree of restriction, the standard deviation of the SFH values thus obtained would be narrower. The inclusion criteria are summarised in table 3. In all of the above studies there appears to be some flattening of the SFH curve near term. As pointed out by Westin, miscalculation of gestational age may lead to serious error. In all the standards so far published, gestation was reckoned on basis of menstrual dates, ultrasound dating being used routinely to 'confirm' dates (Pearce),or reserved for those cases whose last menstrual period was unknown (Calvert, Quaranta) or otherwise excluding those cases without a known LMP (others). Geirsson (1991) has convincingly argued that even when certain, LMP dates are less reliable than those derived from ultrasound, with an overall tendency to overestimate gestation. He pointed out that birth weight standards in populations whose gestations are derived from ultrasound dating show a much less marked 'terminal flattening' of the reference curves at term. This has also been our experience (Wilcox et al, 1993a). It is likely that SFH reference standards are also subject to the same effect.

5.3.5 Ethnic variations in SFH standards. Table 5.1 shows the differences in SFH standards by ethnic group. For comparative purposes, the 40-week median value is given for each group; the SD deviation is omitted because of the widely different methodologies used. It thus appears that for European populations there are only small differences among the published standards. Indian

41

populations, however, have lower term values of around 33 cm, compared with 36 cm for Europeans. Grover and colleagues (1991) published a reference standard derived from 200 low-risk Indian women with birth weights within +/- 1 SD of the local standard. Compared with European curves, their fundal height increments were similar from 20 weeks until 32 weeks (1 cm/week); some slowing was noted thereafter, resulting in term values that were 3 -4 cm lower. Similar values were published by Mathai and colleagues (1987) in South India. However, Depares and colleagues (1989),on comparing European and Pakistani SFH values in Bradford (UK), could not detect significant differences.This may be because of the small samples in their study.Oguranti studied SFH in 581 unselected Nigerian women; their values were also lower than European standards pre-term, but reached similar values at term. These differences in SFH standards among ethnic groups may arise from the well-known differences in birth weights, but could also be due to other factors such as maternal body build and prevalence of fetal pathology.

5.3.6 Clinical performance of SFH measurements. The definitions of 'positive for SGA' by SFH measurements differ in the literature. The populations tested also differ, some being high-risk, hence artificially increasing the detection rate. In most cases at least two or three consecutive readings have to be below the 10th centile. Theoretically, increasing the number of abnormal measurements in order to diagnose SGA should reduce the false positive rate. In a large uncontrolled study of low risk, uncomplicated pregnancies Westin (1977) in Sweden showed that SFH measurements were superior to maternal weight gain, maternal girth measurements, and biochemical analytes (uE3, HPL) for the detection of the SGA infant.The routine introduction of reference SFH charts in the case notes of all their patients was associated with a significantly steeper fall in the local

42

perinatal mortality rate compared with the overall Swedish statistics. Pearce and Campbell (1987) compared serial SFH mesurements with a single fetal abdominal circumference (FAC) obtained by ultrasound as screening tests for SGA. No significant differences were noted between the two, when specificities were set equal

at 79%.

Interestingly, they found a peak sensitivity at 34 weeks, similar to Quaranta's peak at 32 weeks. The only randomized controlled trial on the clinical performance of SFH measurement versus clinical palpation was reported by Lindhard and colleagues (1990), in a population of 1639 women. No significant differences were found between the two methods in terms of the detection rate of SGA, number of interventions, additional diagnostic procedures or the condition of the newborn. Table 5.3 summarises the clinical performance of SFH measurements in detecting SGA infants. Because fundal height standards and the definitions of an abnormal SFH test vary, it is not possible to pool results in order to arrive at average values.Persson and colleagues (1986) summarise positive predictive values for various studies including their own, which is the largest. They range from 13% to 79%; there is a tendency for larger studies to show lower PPV's. This is consistent with the hypothesis that the larger the number of observers, the greater is the effect of inter-observer variability and hence the poorer the tests' performance.

5.4 Discussion If , as one would expect, antenatal detection of IUGR improves neonatal outcome, then an effective screening strategy for growth disturbances is a major target in perinatal medicine. That randomised studies have not been able to document a definite improvement in outcome following routine ultrasound examinations may be due to a number of factors. To some extent this is likely to reflect the limited

43

accuracy of ultrasound in estimating fetal size; in none of the studies was fetal weight rather than the individual biometric parameters plotted. Another factor is the threshold for clinical intervention. In Newnham's study, the induction rates of the screened women did not differ significantly from the regular group even though the former were significantly more likely to be given the diagnosis of IUGR. This suggests some reluctance on the part of the clinicians to act on the basis of the ultrasound findings. Not to be discounted is the lack of an appropriate standard for detecting deviations in growth velocity. The concept that fetal growth could be monitored by such a simple and inexpensive tool as a tape measure has generated wide interest.The lack of agreement in the literature on the efficacy of SFH measurements is not surprising, given the wide differences in definitions and population sampling. The fact that the median 40-week values for different ethnic groups reflect their differences in mean birth weights provides additional support to the notion

that SFH

measurements are an indicator of fetal size. At least three studies compared traditional clinical palpation with SFH measurements for the detection of SGA fetuses. Secher and colleagues (1990) found no significant differences betweeen these two methods. Similar results were obtained by Pschera and colleagues (1984), and by Lindhard and colleagues (1990). This may be due to the clinicians’ longer experience with clinical palpation as opposed to the newer SFH measurement, and hence the results may have been biased by this factor. There is no good evidence that introduction of routine SFH measurements leads to a reduction in perinatal mortality rates. The improved figures reported by Westin may well have been a chance result, since this study was not properly controlled. In view of the magnitude of the error due to inter-observer variability, it is likely that SFH measurements are clinically more useful when

44

performed serially and frequently by the same observer using a consistent technique. There is some evidence to suggest that test performance for SFH may be optimal around 32-34 weeks, and it is of interest that this coincides with the period of peak performance for ultrasonic fetal weight estimation of 32 to 36 weeks (Hedriana & Moore, 1994). If ultrasound growth screening is to be performed as a one-stage routine, then this gestational age interval offers the best hope for success. It could also be possible to improve the accuracy of ultrasonic fetal weight estimate by combining it with the SFH; this approach was described by Farmer and colleagues (1992), who, in addition toultrasound data and the SFH also included maternal characteristics such as height and parity. They developed a trained neural network which, in the case of suspected macrosomia, was significantly more accurate in estimating fetal weight than either Hadlock’s or Shepard’s formula; its mean percentage error was 4.7% with a standard deviation of 3.9%. Accurate fetal weight estimation is the key to an effective screening program for growth disturbances. This is an area that continues to evolve, and improvements may be brought about by advanced information

processing

techniques,

using

current

clinical

measurements.

45

Table 5.1 Criteria for Population Selection of SFH Standards

Study

Population selection criteria for derivation of standard.

Quaranta

Birth weight between 25th and 90th centiles

Belizan

Birth weight between 10th and 90th centiles

Westin

Mean birth weight +/- 1 SD

Calvert

Birth weight between 10th and 90th centiles

Pearce

Birth weight between 10th and 90th centiles

Grover

Birth weight within mean +/- 1sd

Mathai

Term delivery of live infant

Rosenberg

Birth weight between 25th and 90th centiles

Ogunranti

All patients sure of their dates.

Persson

Infant weight/length ratio between 10th and 90th centile

46

Table 5.2 Ethnic variation in SFH standards.

Study

Ethnic Group

Sample size

40-week value

Quaranta

European

138

36.5

Belizan

Latin American

139

34.5

Westin

European

428

36

Calvert

European

381

36

Pearce

European

699

37

Persson

European

1350

36

Ogunranti

Afro-caribbean

581

39.4

Grover

Indian

200

33

Mathai

Indian

250

33.8

47

Table 5.3 Clinical performance of SFH measurements.

Study

Definition of abnormal result

FP

sens

spec

Quaranta

2 cons or 3 isolated vals <10th cent

21

73

80

Belizan

1 single val <10th

21

86

90

Westin

1 single val<2cm below median or 54

75

64

41

28

97

80

76

60

3 cons static or decreasing vals

Lindhard

As above

Calvert

1 single val<2cm below median or 3 cons static or decreasing vals

Pearce

1 single val below 10th centile

64

76

79

Grover

1 single val < 1sd below mean

16

81

94

Mathai

1 single val < 1sd below mean

23

78

88

Rosenberg

20% of measurements below 10th centile

21

62

85

Cnattingius

'catch-up and low' SFH growth

NA

79

92

Persson

outside 2sd's

NA

27

88

48

6. PRINCIPLES OF THE CUSTOMISED GROWTH CHART.

6.1 Introduction

To produce an adjustable standard that takes into account the physiological factors influencing fetal weight is a computational task that is not easily performed by using tables and graphs. The logical solution to this problem is computer software. The principles of a computer-generated growth standard which carries out this task were first published in the Lancet in 1992 by J.Gardosi, Professor A.Chang and colleagues.Unlike previous attempts that relied on fixed correction factors from tables applied to birth weight standards, the calculations were performed by computer software and growth charts could be displayed on screen and printed. Ultrasoundderived fetal growth standards rather than birthweight standards were used for generating growth curves, and corrections factors that included maternal weight at booking, maternal height, parity, ethnic group and fetal sex were scaled up or down depending on the gestational age.

6.2 The prediction of normal growth potential

The initial obstetric database consisted of 4179 pregnancies with ultrasound-confirmed dates. Multiple regression analysis showed that in addition to gestation and sex, maternal weight at booking, height , ethnic group and parity were factors that significantly affected birth weight. This was confirmed by analysis of variance.The multiple regression analysis was repeated by Mr Mark Wilcox on a much larger sample of

38114 cases, smoking being entered as an independent

variable.Continuous variables such as gestation, height and weight were centered around their means so as to minimize computational

49

problems. The details of the analysis have been published elsewhere (Gardosi et al, 1994), and the coefficients are given in table 3.1. This regression model allows us to estimate the fetal weight at 40 weeks for any combination of maternal characteristics. In the prediction of normal birthweight, the confounding effect of smoking is dealt with by entering the non-smoking coefficient for all cases.The model can only explain 31% of the variability of birth weight in the database, but this is likely to be an underestimate, since up to 55% of

records in

obstetric databases may have at least one error (Dombrowski, 1994); the East Midlands Obstetric Database is not subject to rigorous quality controls.

6.3 The generation of normal fetal growth curves The method of generating growth curves relies on the working hypothesis that, for normally grown fetuses, the morphology of the growth curves is approximately the same irrespective of birth weight. This means that if the mean curves from population subgroups are described in terms of a polynomial function of gestation, division of the polynomial coefficients by the 40-week weight will yield a new function whose coefficients will not vary appreciably between subgroups. Some indirect support for this postulate comes from the work by Dunn (1989) and Thomson (1968). In chapter 4 we reviewed the literature on ultrasound-derived fetal weight growth curves and for each mean curve we derived a 'proportional' curve, by dividing the coefficients of the original by its predicted 40 week weight . An average growth curve was produced by taking the arithmetic mean of the respective coefficients; this function will estimate the percentage of term weight for any gestation. Multiplying this function by the predicted 40-week weight obtained from the regression model will yield an individual 'ideal' antenatal growth curve. The 10th and 90th centile reference curves are derived

50

from the standard error of the regression analysis, and are adjusted at each gestation so that the ratio of the standard error to fetal weight (coefficient of variation) remains constant at 11%. Some paired examples of the charts are shown in figures 6.1 to 6.4. Mrs Average (fig 6.1) is a European woman of average height and weight (163 cm and 63.4 Kg) who has had a previous delivery of a male infant weighing 2700 g at 37 weeks. The value within the square (9) is the centile value of this weight for 37 weeks. The expected birth weight at 40 weeks is just over 3800 grams. Figure 6.2 shows a woman of the same parity and size but of Indo-Pakistani origins; the term birth weight expectation is reduced to 3600 grams, but the previous delivery is not classified as SGA (centile 19). The chart of a large European lady with the same obstetric history is shown in figure 6.3; the previous birth weight is given a centile value of 4. In contrast, a short and light lady with the same history would be given a centile value of 24 (figure 6.4).

6.4 Other functions The early versions of the customised growth charts also allowed the entry of fundal height measurements by including a fundal-height yaxis on the right side. This was calibrated to approximate the standard published by Pearce and Campbell (1987). An axis for the fetal abdominal circumference was also displayed, based on the standard of Deter and colleagues (1982). Previous deliveries and their birth weight centiles may be entered and displayed on the same chart. The x-axis displays gestation as exact weeks and also the calculated corresponding dates. The expected date of delivery, maternal height and weight, parity and ethnic origin are displayed on the top left hand corner of the chart. In the latest version of the chart, the maternal body mass index is

51

displayed if this falls below the 10th centile for our pregnant population at booking, as this indicates the possibility of malnutrition in the periconceptional period.

6.5 Clinical performance The initial sample of 4179 deliveries contained 385 cases with a birth weight below the 10th centile by unadjusted criteria (SGA). Of these, only 278 were still below the 10th centile following adjustment for maternal characteristics. Hence 107 (28%) would have been given a false positive diagnosis of SGA by conventional standard. Adjustment 90 cases that would have been missed by conventional assessment. Babies that by the conventional standard only were deemed SGA had significantly fewer instances of low Apgar scores.

6.6 Discussion

It is apparent from the foregoing that this method of producing an adjustable growth standard relies on many hypotheses on the physiology of fetal growth. Yet these are necessary if a model is to be developed. These may be summarised as follows:

1. The physiological variables affecting fetal weight at term are also effective in the antenatal period in proportion to the fetal weight.

2. The intrinsic shape of the normal fetal growth curve is the same for all subgroups, differing only by a scalar, or 'magnification' factor proportional

to the predicted term weight. This postulate will be

referred to as the ‘proportionality ‘ principle.

3. The average fetal growth curve entered in the program is close to the true population average.

52

4. The distribution of fetal weights is approximately normal for all gestations.

5. The variance of fetal weights is a constant fraction of the gestational median , ie a constant coefficient of variation of 11% (derived from term birth weights).

6. The selected variables of maternal weight, height, parity and ethnic group have mainly a

physiological rather than pathological

significance.

A major problem is to separate physiological from pathological effects e.g. to what extent is a low maternal weight at booking due to nutritional factors as opposed to constitutional factors. In view of the known adverse effects of malnutrition in early pregnancy on fetal growth, measures are needed to prevent the application of unduly small adjustments for maternal weight in cases where the low values are due to undernutrition at booking. While this is an infrequent problem in western populations, this is not the case in the developing world. To deal with this issue, the current version of the customised growth chart calculates the body mass index (BMI) at booking; if this is below the 10th centile the maternal weight at booking is corrected so that the BMI is at the 10th centile. The birthweight expectation is thus the one for a normally nourished individual at the lower end of the normal range. A similar algorithm is applied at the 90th centile of the BMI. The rationale for making adjustments on the basis of parity is an issue open to debate. The primigravid state, although 'natural', would be a relatively infrequent finding in a female population of reproductive age unaffected by contraceptive practices, as the statistics from a

53

century ago show. Goldenberg and colleagues (1989) stated that the genetic potential for fetal growth in primigravidas and multigravidas is identical and that the differences noted between these two groups are attributable to growth-restricting factors operating in primigravidas. Henced they advocated a parity-independent standard derived from a population of mixed

parity. On the other hand, Thomson and

colleagues (1968) argued in favour of adjusting for parity, believing that the effect of parity is a physiological factor present in the fetal environment. The effects of smoking on birthweight are relatively large. An alternative method to obtaining regression coefficients applicable to the whole population would be to include non-smokers only. But this process could theoretically lead to the

selection of a genetically

'supra-normal' population, and thus adjustment factors that may not be universally applicable. More robust estimates of the coefficients for the different ethnic groups would have required a much larger sample . We did not have sufficient numbers of Far Eastern women to separate them from the heterogenous 'others' grouping, and hence we do not have a reliable coefficient for these. In view of the known pathological effects of malnutrition on fetal growth, measures are needed to prevent the application of unduly small adjustments for maternal weight in cases where the low values are due to undernutrition at booking. While this is an infrequent problem in western populations, this is not the case in the developing world. In the current version of the customised growth program, a lower limit is entered for the weight adjustment. If the maternal booking weight is below this limit, the adjustment does not decrease; the birthweight expectation is thus the one for a normally nourished individual at the lower end of the normal range.

54

Large databases containing maternal characteristics, accurate antenatal fetal weight estimates and birth weights would be needed to address these issues.

55

Table 6.1 Multiple regression coefficients for the prediction of normal growth potential . CONSTANT STANDARD ERROR ADJUSTED R SQUARE NUMBER OF CASES

3409.853 389.032 0.31824 38114 Coefficient

GESTATION (from 280 days) Gest 1 Gest 2 Gest 3

20.667 -0.21289 -0.000167

SEX Male Female

- 116.871 -233.742

MATERNAL HEIGHT (from 162 cm) BOOKING WEIGHT (from 64.3 kg) Weight 1 Weight 2 Weight 3

7.764

8.676 -0.11740 0.000716

ETHNIC GROUP European Indian Sub-cont. Afro-Caribbean Other

31.670 -154.263 -95.789 -33.446

PARITY Para 0 Para 1 Para 2 Para 3 Para 4

4.898 112.904 153.458 154.767 154.690

SMOKING Non-smoker Smokes 1-10 11-20 > 20

31.9160 -120.602 -182.568 -214.112

56

7. METHODS OF GESTATIONAL AGE ESTIMATION

7.1 Introduction An accurate assessment of gestation is crucial in the development of fetal growth standards. It is also of importance in evaluating most of the variables in current use for fetal monitoring and in the management of post-term pregnancy. Gestational age may be estimated ultrasonically by measuring fetal parameters such as the crown-rump length (Robinson & Fleming, 1975) until 12 weeks and the biparietal diameter (Campbell & Newman, 1971) until about 20 weeks gestation. Measurement of the BPD in the mid-trimester has been shown to be 5% more accurate in predicting the date of delivery than impeccable dates (Pearce and Campbell,1983). In practice, most ultrasound departments follow either the '10-day rule' or the '7-day rule', whereby preference is given to menstrual dates if these are within 7 or 10 days of the ultrasound estimate by the BPD. Here we study the reliability of these methods by analysing the East Midlands Obstetric Database.

7.2 Materials and Methods The computerised obstetric records of three major maternity units in the East Midlands (City and University Hospitals in Nottingham and Derby City Hospital) date from 1986. So far more than 60000 cases are available for analysis. A significant drawback is that the database does not allow the reliable exclusion of induced labour. Multiple pregnancies, stillbirths, congenitally abnormal babies, late bookers (over 24 weeks), preterm deliveries (<37 weeks) and those with unknown menstrual dates were excluded. A total of 31747 cases with both menstrual and ultrasound data were retrieved for analysis. Gestational age at delivery was calculated in days from 1. The

57

biparietal diameter according to the dating charts by Campbell (1971) if over 13 weeks at booking, otherwise the crown-rump length was used (Robinson & Fleming,1975) and 2. The last menstrual period. The error in predicting the EDD was calculated in days for three different dating methods: ultrasound data only, the 7-day rule and the 10-day rule as follows: Error (days) = estimated gestational age at delivery - 280 These were expressed as signed and as absolute values. Statistical analyses were performed with the 'SPSS for Windows' statistical package.

7.3 Results Table 7.1 shows the means, standard deviations and skewness of the errors for each method. The distribution of the error is weakly but significantly skewed to the left. Analysis of the signed errors suggests that ultrasound on its own and the 7-day rule tend to underestimate the EDD, whereas the 10-day rule overestimates it. Because of the significant skewness of the data, the differences between the methods were evaluated non-parametrically using Wilcoxon Matched -PairsSigned-Ranks Test; the results are shown in table 7.2 . The mean and the standard deviation of the absolute error using ultrasound alone is slightly smaller than the other methods. Dating by ultrasound only is significantly more accurate in predicting the EDD than either the 7day or 10-day rules.

7.4 Discussion The length of gestation has traditionally been calculated from the first day of the last menstrual period using Naegele's rule -i.e., by adding 7 days and 9 months to the date of the last menstrual period. Although this formula has been attributed to Franz Karl Naegele (1778-1851), it was first proposed by professor Herman Boerhaave (1709) at the

58

University of Leyden (Speert, 1958). The formula implies a mean duration of pregnancy of 274 days from the LMP, which is at variance with the observed value of 280 days (Doring,1962). If the dates were not 'reliable', gestation was estimated by clinical palpation. There is mounting evidence that even among women with 'reliable' menstrual dates, considerable error may arise in the calculation of gestation. This is because the onset of ovulation within the menstrual cycle is relatively erratic, particularly in younger women (Geirsson,1991), and may also vary from cycle to cycle. We have studied the error of menstrual dates using the BPD-derived gestation as the reference standard, in a database of 31561 cases (Gardosi & Mongelli, 1993). For 21.5% of pregnancies ultrasound scan dates were outside plus or minus 7 days from the dates based on menstrual dates. Furthermore, the distribution of the error associated with menstrual dates was significantly skewed, such that the 95% confidence interval for gestational age derived from menstrual history was -27 to +9 days. In our unit the ultrasound department calculates the EDD at booking by using the 10-day rule if reliable menstrual dates are available, otherwise the BPD or the FL is employed. About 6% of all cases are induced for post-maturity on this basis, but these cannot be identified reliably from the computerised records. Therefore one would expect that this iatrogenic interference on the normal duration of pregnancy would bias the statistics in favour of the 10-day rule. However, in spite of this bias, we found that dating by ultrasound alone appears the best method for predicting the EDD; bearing in mind the direction of the bias, it is likely that the advantage of using ultrasound only may be greater than what our figures indicate. It has been suggested that the use of the BPD alone in estimating gestational age is flawed, in that the larger babies would be assigned a longer gestation than the smaller babies (Henriksen & Wilcox, 1994). Persson and colleagues (1978) did find a relationship between birth

59

weight centiles and the size of the BPD in early pregnancy, but the difference between the small (<10th centile) and large (>90th centile) babies amounted to only about 1.5 mm at 18 weeks. This difference is equivalent to less than one day's variation in gestational age by Campbell's dating standard. We recently investigated this issue using a database of 19 singleton pregnancies precisely dated from in-vitro fertilisation or artificial insemination. No significant correlation was found between the BPD centiles at the booking ultrasound examination and the birthweight centiles (Gardosi et al, 1994). In this study we assumed that the modal length of normal pregnancy of 280 days is equally applicable for all population subgroups. Only a limited number of studies have been published on this issue, all of them using menstrual dates. They report trivial differences in duration of pregnancy between social classes or ethnic groups (Butler & Bonham, 1963; Henderson, 1967). There is also some evidence that BPD standards do not vary appreciably between ethnic groups (Vialet et al, 1988; Simmons et al, 1985), and thus this should not be an important source of bias. Our findings do not support the use of the 7-day or 10-day rule in the assignment of gestational age when valid ultrasound measurements are available. Although the error associated with their use may not matter in clinical practice (Mongelli & Gardosi, 1994), these methods are best avoided in defining normal standards in pregnancy.

60

Table 7.1 Descriptive statistics for the errors resulting from each dating method.Values expressed in days (N = 31747).

Variable

Mean

SD

Skew

S.E.

Min

Max

Skew Signed Errors

Ultrasound

-0.72

8.50

-0.12

0.01

-20

28

-0.20

8.65

-0.15

0.01

-27

28

0.18

8.79

-0.16

0.01

-30

30

only 7-day rule 10-day rule

Absolute Errors

Ultrasound

6.91

5.00

0.69

0.01

0

28

7-day rule

7.02

5.06

0.70

0.01

0

28

10-day rule

7.15

5.12

0.73

0.01

0

30

only

61

Table 7.2

Significance of differences in absolute errors between

dating methods. Wilcoxon Matched -Pairs Signed Ranks Test.

Pairs: A Vs B

Mean

A>B

A
Rank Ultrasound only 8443.96 Vs 7-day rule

Ties

Z-value

2-tailed P

A=B

value

8251

8989

14507

-7.1

<0.00005

9231

10438

12078

-11.4

<0.00005

1449

980

29318

-10.4

<0.00005

8782.91

Ultrasound only 9493.64 Vs 10-day rule

10137.31

10-day rule

1265.5

Vs 7-day rule

1140.34

62

8. FORWARD PROJECTION OF FETAL WEIGHT ESTIMATE The proportionality principle applied to dynamic fetal weight estimation.

8.1 Introduction. Fetal weight estimation from ultrasound parameters or other methods is almost invariably made from static measurements. In practice, delivery often does not occur until days or weeks after the ultrasound examination, and the weight estimation is not corrected for this time interval. This problem was appreciated by Spinnato and colleagues (1988) who published a series of dynamic weight estimation formulae to allow forward projection of the fetal weight estimate to the expected date of delivery. Here we present an alternative method that, in addition to forward weight projection,

should also allow for backward fetal weight

interpolation from the birth weight.

8.2 Subjects and Methods. In order to compare our method with Spinnato’s method, we selected only those cases who delivered within 35 days of the last ultrasound examination. Two hundred forty two cases were available for analysis; these were liveborn infants, inclusive of adverse outcomes and congenital malformations. The forward projection equation was derived from the principle that the shape of the fetal growth curve is similar for all cases irrespective of the birth weight. It is in fact a form of proportional extrapolation, which we will refer to as the ‘prop-ex’ method. Hadlock’s growth formula was selected because: 1. The weight estimation formulae considered were also developed by Hadlock and colleagues and 2. This growth formula is close to the average of previously published growth formulae (Gardosi et al, 1994). If Hadlock is the growth function, EFW the ultrasound fetal

63

weight estimation, u is the gestational age at ultrasound examination and d is the gestational age at delivery, the following relationship applies:

EFWu : EFWd = Hadlock(u) : Hadlock(d)

Therefore: Projected fetal weight at delivery: EFWd = EFWu * Hadlock(d)/ Hadlock(u)

Likewise, Spinnato’s equivalent dynamic weight equation for the Hadlock weight formulae is:

Log (EFWd) = 1.0009 * log (EFWu ) +0.0043(d -u)

Both techniques were applied to our sample. Prediction errors were calculated both as signed and as absolute percentage errors of the birth weight (BWT) as follows:

Percentage error = 100* (EFWd -BWT)/BWT

Trends in the error in relationship to birth weight were examined using Spearman’s rank correlation coefficient. Differences between the two methods were tested by Wilcoxon’s matched pairs signed-ranks test.

8.3 Results The signed and absolute percentage errors for the two methods are shown in Table 8.1. Table 8.2 displays the statistical significance of the differences between these two methods. The errors from using Spinnato’s method were not significantly correlated with true weight ( R = -0.0880, P>0.05), whereas with our

64

method there was a weak but statistically significant inverse correlation with the true weight (R= -0.23, P<0.0005). This correlation was not significant if birth weights below 3200g were excluded. For both techniques, there was no significant correlation between either the signed or absolute errors and the lag-time interval.

8.4 Discussion These statistics show that the proportional extrapolation method we describe for birth weight prediction from remote ultrasonographic examination is significantly better than Spinnato’s technique in our population. This may be because the latter uses the lag-time difference between ultrasound examination and delivery, without considering the actual values of the gestational ages of these events. The random errors described in Spinnato’s original paper are slightly lower than when his method was applied to our population (11 vs 12 %), with a tendency towards underestimation as opposed to overestimation in our sample. Two other advantages of the prop-ex technique are that: 1. It is applicable to any fetal weight estimation technique, whereas Spinnato’s method requires different equations for different weight estimating formulae. 2. It can be modified to allow for retrospective fetal weight estimation from the birth weight by interpolation as follows:

IFWu = BWT* Hadlock(u)/ Hadlock(d)

where IFWu is the interpolated fetal weight. This formula, however, cannot be validated without an independent, highly accurate method of fetal weight estimation such as NMR. The results of these studies validate the concept of incorporating a lapse-time factor in equations to predict the birth weight from remote

65

ultrasonographic data. As the errors are not correlated with the lagtime interval, both methods are accurate throughout the range of time intervals measured. It is likely that accuracy will be lost in cases affected by growth disturbances, and this is reflected in the negative correlation between the signed errors and the birth weight, which is lost for birth weights over 3200 grams. These findings also support the ‘proportionality’principle - a key aspect of the customised growth chart program- which is the assumption that fetal growth in normal populations is essentially the same, once growth is expressed in terms independent of the actual birth weight.

66

Table 8.1.Signed and absolute percentage errors for projected fetal weight estimation. Formula Prop-ex Spinnato Prop-ex Spinnato

Systematic Error (%) 6.01 8.56 Mean Absolute Error (%) 9.9 11.8

Random Error (%) 11.0 12.0 Standard Deviation (%) 7.7 8.9

Table 8.2. Non-parametric comparison of the errors generated using the two methods on the same population. Wilcoxon Matched-Pairs Signed-Ranks Test.

Differences in Ranks

Mean Rank

- Ranks (Spinnato < Prop-ex) + Ranks (Spinnato > Prop-ex) Ties (Spinnato = Prop-ex)

172.70 130.75

No Cases 51 225 1 Total = 277

Z = -7.7646

2-Tailed P-value <0 .00005

67

9. SELECTION OF ULTRASONIC WEIGHT FORMULA

9.1 Introduction The performance of ultrasound fetal weight estimation formulae may have

important implications for growth standards and antenatal

screening. In choosing the best formula, two main selection criteria need to be satisfied: (a) Minimal overall error. (b) Minimal bias in the error in relation to fetal weight. The first condition is self-evident. The significance of minimising error trends in relation to fetal size has been somewhat underestimated in the literature. If a particular formula has a strong tendency to overestimate the small and underestimate the large fetus, both the weight standard and screening performance may be adversely affected. One may argue that fetal growth standards should be weight formulaspecific.If this was the case, then centres where a particular formula is either not in use or is not suitable will not be able to use the standard. In this study we examine the effect of different weight estimation formulae on apparent growth kinetics, and their clinical performance in our population is evaluated .

9.2 Patients and Methods Persson and colleagues (1986) have shown that the average fetal weight curve derived from the means of the ultrasound parameters for each gestation is very similar to that derived from the means of the individual weights. Hence, to illustrate the effect of different weight formulae on growth curves, we studied the means published by Chitty and Altman (1993) for individual ultrasound parameters. These were input for the following weight estimation formulae:

- Hadlock (BPD, AC, FL)

68

- Hadlock (AC, FL) - Warsof (BPD, AC) - Shepard (BPD, AC) - Campbell (AC) - Modified Persson (BPD, AC, FL)

Persson's original formula used the abdominal diameter (AD); this was modified by converting the AD into AC as follows: log10 EFW (grams) = 0.972* log10 BPD + log10 (AC/3.1416) +0.367*log10 FL -2.646 The calculations were performed by computer software, and the different growth curves for each formula displayed graphically using Cricket Graph software. In order to determine the best formula for the population in our study, we selected those cases that delivered within 14 days of an ultrasound examination. A sample of 129 cases was retrieved, from a total of 171 cases that had been seen one year before the completion of the growth study. A correction for the interval growth between time of ultrasound examination and delivery was made by projecting forward the estimated fetal weight using Hadlock's fetal growth formula as follows:

Predicted weight = EFW * H(gestation at delivery)/H(gestation at scan)

where EFW is the ultrasound-estimated fetal weight and H is Hadlock's fetal growth function as described in chapters 4 and 8. Using the weight formulae listed above, the signed percentage error for each case was computed thus:

Percentage error = (Predicted weight - Birth weight)/ Birth weight

69

A positive value indicates overestimation whereas a negative value indicates underestimation of true fetal weight. The weight formulae were then corrected for systematic error by multiplying them by a correction factor as follows: Correction factor = 1/(1 + systematic error) where the systematic error is expressed in fractional terms. Statistical analyses were performed with the SPSS for Windows statistical package. The means, standard deviations, 95% confidence limits and the distributions of the errors were computed. The trends between error and true weight were expressed in terms of Pearson's moment correlation coefficients.

9.3 Results Figure 9.1 shows the fetal growth curves obtained for Chitty and Altman's data using different fetal weight estimation formulae . For our population, the mean and standard deviations of the errors for each formula and the correlation of the error with the observed weight are displayed in table 9.1. Table 9.2 shows the errors at the extremes of birth weight. All the formulae tested on our population tend to overestimate true weight, and this applies also to the extremes of fetal weight

(except

for

Campbell's

and

Persson's

formulae

for

weights>4000g). The errors were recalculated after applying the correction factor for systematic error and the results are shown in table 9.3.

9.4 Discussion For any given population, it is apparent from Figure 9.1 that from about 36 weeks onwards the type of fetal weight formula can have a marked effect on the apparent fetal growth kinetics, particularly at term. Hence it is important to select the weight formula that is most

70

suitable for the operator and the population under study. This could explain some of the variability in the fetal growth kinetics exhibited by previously published standards, which also appears greatest after 36 weeks. In our sample all of the formulae studied tended to overestimate the true fetal weight. The error associated with the use of Hadlock's formula for BPD, AC and FL was not correlated with the true weight, whereas all of the other formulae produced errors that were significantly correlated. Robson and colleagues (1993) also found a trend to overestimate true fetal weight. They noted a significant correlation of the error with birth weight for all the formulae they tested, but these did not include the Hadlock formula included in our study. This systematic overestimation may arise from differences in ethnicity between North American and British populations, or in techniques and equipment. After applying the correction factor for overestimation, Hadlock's formula for the BPD, AC and FL showed the smallest SD of the error, and this was the formula selected for our study. If the BPD could not be measured

because of suboptimal

visualisation, the formula used was Hadlock's for the AC and FL. The modified weight formulae were then applied retrospectively to this sample and prospectively to the remaining half of the population. Our approach to selecting weight formulae and correcting for them makes the assumption that the errors observed at term are similar to those obtained in the preterm period. To verify this would require a substantial number of infants born preterm who had had an ultrasound examination. Another solution is to use echo-planar NMR for accurate weight estimation in the preterm period and to compare this with the ultrasonic weight estimates; but this is not as yet an economic proposition.

71

Table 9.1 Errors of ultrasonic weight formulae in relation to birth weight for cases delivering within 14 days of examination (N =129).

Formula

Systematic

SD

Error (%)

of Error

Correlation Skew

(R)

with P-value

birth weight

of R

Persson *

0.43

10.79

0.11

-0.2145

0.015

Campbell

0.89

11.85

0.06

-0.4226

0.000

3.23

11.52

0.2

-0.1246

0.159

3.78

11.76

0.08

-0.2299

0.009

5.32

1.21

.17

-0.1230

0.165

9.24

12.31

0.09

-0.2442

0.005

Hadlock (AC, FL) Warsof (BPD,AC) Hadlock (BPD,AC,FL) Shepard (BPD,AC)

* original formula using the AD modified to include the AC as follows: log EFW = 0.972* log BPD + log (AC/3.1416) +0.367*log FL 2.646

72

Table 9.2 Error of ultrasonic weight formulae at the extremes of birth weight (all cases).

Formula

Birthweight

Birthweight

> 4000g (N = 37)

< 2500g (N = 17)

Mean Error (SD)

Mean Error (SD)

Persson

-1.3 (10.36)

7.2 (12.5)

Campbell

-3.2 (10.86)

14.0 (16.3)

(AC, FL)

0.9 (10.8)

6.4 (13.8)

Warsof

2.7 (11.8)

11.2 (12.2)

(BPD,AC,FL)

3.4 (10.4)

8.3 (13.2)

Shepard

7.9 (12.3)

17.4 (12.8)

Hadlock

Hadlock

73

Table 9.3 Error of ultrasonic weight formulae for cases delivering within 14 days of examination after correction for systematic error (N =129).

Formula

Correction

Mean

SD

factor

Error (%) of Error

Skew

Persson

0.9957

0.00

10.7

0.11

Campbell

0.9911

-0.01

11.7

0.06

(AC, FL)

0.9687

0.00

11.2

0.2

Warsof

0.9636

0.01

11.3

0.08

(BPD,AC,FL)

0.9495

0.00

10.6

0.17

Shepard

0.9154

0.01

11.3

0.09

Hadlock

Hadlock

74

Figure 9.1. Fetal weight curves derived from data by Chitty & Altman. This is an illustration of how different weight estimation formulae can affect the kinetics of the resultant fetal growth curve. The curves were obtained from the data by Chitty and Altman (1994) on the BPD, AC and FL, transformed into fetal weight according to the weight estimation formulae by Hadlock, Persson, Shepard, Warsof and Campbell.

75

10. ULTRASONIC STUDY OF FETAL GROWTH: PATIENTS AND METHODS.

10.1 Introduction It is apparent from reviewing the literature that considerable uncertainty exists on fairly basic aspects of fetal growth. Published ultrasound-based fetal growth standards vary widely, and little is known on ultrasound-defined fetal growth patterns in human subpopulations. The analysis of intrauterine fetal weight gain using ultrasound requires fairly complex techniques, and these are to some extent arbitrary. Much of our work attempts to address these issues, as these are at least as important as establishing the clinical validity of customised growth charts.

10.2 Study Design Ethics Committee approval was obtained prior to commencing the study. We aimed to obtain a population sample suitable for the development of normal standards and to study fetal growth in relation to maternal characteristics. Inclusion criteria for the study were: - Singleton pregnancy. - Maternal age no greater than 35 years. - Gestational age at booking no greater than 22 weeks. - 'Low risk' pregnancy at booking. Smokers and cases who developed pregnancy complications were not excluded, as long as neonatal outcome was normal on clinical grounds (see 10.6). Suitable cases were recruited in the antenatal clinic following their booking ultrasound examination. Patients were given information sheets and informed consent was obtained in writing. Women were examined at intervals of 2-3 weeks commencing at 24 to 32 weeks, for a maximum of 4 examinations (excluding the booking examination).

76

This schedule allowed us to obtain ultrasound data close to delivery and in the post-term period. A total of 352 women were recruited. One patient moved out of the district and delivered elsewhere; delivery details could not be obtained. Twenty-one cases (6%) did not attend any planned visits following recruitment; these were excluded from the analysis. The remaining 325 cases underwent at least one ultrasound examination. The distribution of the number of examinations (in addition to the booking ultrasound) is shown in Fig 10.1; 46% attended 4 examinations, 29% attended 3, and the remainder one or two examinations.

In addition to the booking

ultrasound scan, a total of 1021 ultrasound examinations were performed. Patients whose fetal growth curves or symphysis-fundus heights were a cause for concern were referred to their Consultants, but not excluded unless the neonatal outcome was abnormal.

10.3 Population characteristics The maternal and neonatal characteristics are summarised in tables 10.1 and 10.2 respectively. As one would expect from the inclusion criteria, there is a higher proportion of primiparae and a lower percentage of smokers than in the general population; there are slightly higher proportions of Europeans and Indo-Pakistanis. The mean birth weight is about 100g higher than the population average, and this is explained by the lower preterm delivery rate (5.8% Vs 7.2%) and lower percentage of smokers.

10.4 Equipment and Methods Ultrasound examinations were performed using either a Kontron Sigma 1AC or a Corometrics Aloka 500 with a curvilinear array. Calculations of gestational length, estimated fetal weight and printing

77

of growth charts were carried out on IBM-compatible personal computers, using especially designed computer programs. Measurements of the biparietal diameter, femur length, and abdominal circumference were taken using standard techniques (Chudleigh & Pearce, 1986) by one of two experienced operators (JM and AD). The fundal height was measured in cm with a flexible tape, from the top of the fundus to the upper border of the symphysis pubis along the longitudinal axis of the uterus; care was taken to ensure that the bladder was emptied. Nearly all of these measurements were taken by one observer (JM). Maternal height was measured in cm with stadiometers, weights at booking were measured in Kg using standard scales .Gestation was calculated in days from the fetal biparietal diameter according to Campbell's dating chart, using specially-designed computer software.

10.5 Inter-observer and intra-observer variability. Possible bias in the ultrasound measurements arising from interobserver variability was studied in a subset of 12 cases. Measurements of the BPD and FL were in agreement within +/- 1 mm and thus were not examined further. The AC is closely related to EFW, and this was studied in more detail. Twelve randomly selected cases were measured by both observers. Differences between the paired readings were tested using the Wilcoxon Matched-Pairs Signed-Ranks Test. No significant difference between the observers could be detected (Z = 0.6276, 2-Tailed P = 0.5303). The intra-observer variability was estimated by measuring the AC of the same baby twice, for 10 cases. In the case of observer AD, the differences between the two readings ranged from -1.1 mm to 1.9 mm, the mean being 0.52 (SEM 0.383) and SD of 1.21, with a normal distribution.

78

10.6 Measures of adverse neonatal outcome Normal neonatal outcome was defined by the absence of all of the following: (a) Congenital abnormalities. (b) Admission to the Neonatal Intensive Care Unit. (c) Umbilical cord pH < 7.20. (d) Umbilical cord base excess < -8. (e) Apgar score at 5 minutes < 7. (f) Preterm delivery (<259 days).

Two hundred and eighty-three neonates satisfied these conditions. We did not use neonatal anthropometric measurements because of their questionable value, as will be discussed below.

10.7 Calculation of customised and unadjusted fetal and birth weight centiles. These were calculated in batches using computer software compiled in Turbo-Pascal. Birth weight and fetal weights were entered in grams; gestation was reckoned in days according to early ultrasound measurement of the biparietal diameter. For the calculation of unadjusted z-scores and centiles, the average ‘proportionality curve’ described in chapter 4 was fitted through the Nottingham birthweight mean of 3443.5g at 280 days and the standard deviation for each gestation was calculated as 11% of the median weight for that gestation. The methods outlined in chapter 6 were used to calculate the customised centiles and z-scores.

10.7 Discussion This was essentially an observational study analysed retrospectively. Its aim was to investigate some basic aspects of fetal growth in the

79

second half of pregnancy and to validate at least some of the principles on which the customised antenatal growth chart is based. There is far too much uncertainty on defining normal growth in order to attempt to draw a valid protocol suitable for a controlled field trial . This could be one of the reasons why no such trial has to date shown that ultrasonic screening for fetal growth anomalies improves perinatal outcome. We did not exclude cases affected by pregnancy complications such as pre-eclampsia as long as the condition of the neonate at birth was normal. Hence mild cases of IUGR and macrosomia are probably included in the sample, since we believe that there are no reliable and sensitive techniques to identify them. Morphometric indices such as the neonatal ponderal index, mid-arm to head circumference ratio and skin-fold thickness have not been used in this study.Although they have been employed extensively in diagnosing IUGR, there is little evidence that they are superior to birth weight (Chard et al 1992, 1993). On the contrary, the work of Roemer and colleagues (1991) on over 5000 neonates showed that birth weight centiles are more closely correlated with acid-base status at birth than either the ponderal index of Rohrer or the birthweight to length ratio. Likewise, Wolfe and colleagues (1990) found that the ponderal index or the weight/length ratio can explain only 52% of the variance in estimated neonatal body fat; multiple regression analysis of their sample of 119 neonates showed that

birth weight centile and

weight/length ratio were equally good predictors of skin-fold thickness. Another weakness of such indices is that they have been derived from populations whose gestation has been estimated from menstrual data (Oakley et al ,1977; Georgieff et al, 1988 ). Furthermore, the measurement of neonatal length is subject to considerable error; the inter-observer variability being greater than 1 cm in 40% of cases, and

80

the intra-observer variability greater than 1 cm in 15% of cases (Rosenberg et al, 1992). This error will be cubed in the calculation of the ponderal index. Changes in growth velocity as assessed by serial ultrasound have also been used as indicators of growth disturbances (Chang et al,1993), but these are poorly related to outcome and in any case have not been standardised. Deter and colleagues (1990) proposed a neonatal growth assessment score derived from multiple neonatal measurements including weight, crown-heel length, head, chest , abdominal and thigh circumferences, and related to the ultrasonic Rossavik growth coefficients for these parameters. However, their sample was small (37 infants), and the method has not found wide acceptance. A case could be made for excluding smokers, as Ott (1988 ) did, but this may introduce further bias in term of the socio-economic composition of the population; about 37% already belonged to Class I or Class II, and by excluding smokers this proportion would increase significantly, and also reduce the total sample size. Although a special effort was made to recruit women from nonEuropean ethnic groups, the final numbers were fairly close to the overall population norms. These women were less likely to agree to participate in the study, often because they had large families or could not afford the extra time for the required additional clinic visits.

81

Table 10.1 Maternal characteristics.

Age (years), mean ( SD) Height (cm), mean ( SD) Weight (Kg), mean ( SD) Ethnic group (%): European Indo-Pakistani Afro-Caribbean Other Socio-economic groups (%): Class I Class II Class III (M +N) Class IV Class V Unclassified Parity: Percent of primiparae Smoking: Percent of smokers at booking

Study Group N = 325 26.7 ( 4.8) 163.6 (6.4) 66.5 (11.8) 93.8 4.3 1.2 0.6

General Population* 26.5 (5.3) 162.3 (6.4) 65.7 (12.5) 92.8 3.8 2.5 0.8

22 (6.8) 100 (30.8) 119 (36.6) 64 (19.7) 16 ( 4.9) 4 (1.2)

N/A

49.5

43.8

16.6

27.0

* Data from Wilcox M, et al (1993b).

Table 10.2. Newborn characteristics (N =325). Birth weight (g) mean (SD) range Sex: percent of males Preterm delivery (%) Developmental abnormalities (%) Admission to neonatal ICU (%) Acidosis at birth

3406 (553) 1000 - 4900 50.2 5.8 1.5 3.4 3.4

82

11. AN ULTRASOUND STANDARD FOR FETAL WEIGHT GAIN

11.1 Introduction The considerable variation in the published literature on normal intrauterine fetal weight gain as assessed by ultrasound examinations has prompted us to derive a standard applicable to our local population. Only one such standard has been published so far for the United Kingdom (Gallivan et al, 1993), and this was derived from a sample of 67 cases. Our collection of ultrasound data is larger than any of the previous studies, enabling us to exclude cases with abnormal outcome. The purpose of this study is to derive a standard for fetal growth based on serial ultrasound observations of a normal population, and to compare this with other norms.

11.2 Materials and Methods

Subjects The analysis included women who were smokers and pregnancies which developed complications at a later stage, but excluded those pregnancies which had an abnormal neonatal outcome, as defined in chapter 9. A total of 283 of the 352 pregnancies had a normal outcome by these criteria. After excluding cases who delivered elsewhere and those who did not attend for at least two ultrasound scans (in addition to the booking scan) , 267 pregnancies were suitable for analysis.

Estimation of fetal weight Ultrasound equipment consisted of either a Kontron Sigma 1AC or a Corometrics Aloka 500 with curvilinear array transducers. Ultrasound fetal weight estimation was based on a modified Hadlock's formula

83

for fetal abdominal circumference, femur length and the biparietal diameter as described in chapter 8.

Modelling of fetal growth A minimum of 4 points were used to calculate the fetal growth curve for each individual pregnancy. This included an 18 week fetal weight value of 223g (Hadlock et al, 1991) which we used as an invariant point for all cases; at least two EFWs during the third trimester; and the birth weight. Fetal weight curves were individually fitted using the least-squares method and a computer program was written to allow graphical display. Three growth models were considered: (a) Simple 2nd or 3rd degree polynomial of gestation. (b) Rossavik growth model. (c) Logarithmic transformation of fetal weight expressed as a 2nd or 3rd degree polynomial of gestation. Although the Rossavik model was slightly better at predicting the birth weight than the others, the log polynomial model produced a visually more accurate interpolation. By either method , in nearly all cases the R-square value was greater than 0.99, as one would expect by fitting a relatively small number of points. The log-polynomial model was then applied to all cases thus: ln(EFW) = a0 +a1 * GA + a2 * GA2 + a3 * GA3 where GA is the gestational age in days according to the booking ultrasound. The average curve was obtained by determining the median weight for each day of gestation, excluding those cases that had delivered, i.e. there was no extrapolation beyond the observed data. The standard deviation and the skewness were also calculated for each day. Functions describing the median curve and the SD were obtained

84

by weighted multiple regression, the number of undelivered cases for each day acting as the weight. The growth velocities for our standard and for older standards were derived by taking the first derivative of the original growth functions describing the mean curves.

11.3 Results The medians, standard deviation and skewness for each week are displayed in table 11.1. Figure 11.1 shows the average growth curve and the 10th and 90th centiles. The function describing this curve is as follows: FWT = 3411.9469-337.82996*GA+9.44545*GA2-0.000000369939*GA6

where FWT is the weight in grams and GA is the gestational age in exact weeks (e.g. 30.35 weeks). The distribution of fetal weights was checked for each week of gestation; although a trend towards positive skewness was noted, it was not significantly different from normal as assessed by the Kolgorov-Smirnov test ( p>0.2). The mean birth weight for those cases delivering between day 273 and day 287 was 3462 g, which is in close agreement with the predicted 40-week value of the standard curve (3496g). The Nottingham mean curve is compared with previously published standards in figure 11.2. The growth velocity curve is shown in figure11.3, and compared with velocities of published ultrasound standards in figure11.4. The kinetics in terms of fractional growth may be derived from figure 11.5, and they are as follows: G50 = 31 weeks 2 days; P28 = 34%; P37 = 83%; P42=110%;

85

11.4 Discussion In deriving this standard we, like all previous workers, made the tacit assumption that fetal growth is monotonic. Growth spurts have been documented in preterm infants treated in neonatal intensive care units (Gairdner & Pearson, 1971), and it is quite possible that they may be occurring in the intrauterine environment. To be able to detect these would require a much greater number of observations, and also considerably more accurate means of estimating both fetal weight and gestational age than what is available from current technology. Ours is the only study to our knowledge that combines birth weights and ultrasound-estimated fetal weights in the derivation of a fetal growth standard. This is a valid procedure, provided that any systematic error in the weight estimation formula is corrected in order to avoid artefactual accelerations or decelerations in the growth curve. The main advantage of this method is to significantly increase the range and accuracy of data points into the term and post-term period. Many of the previously published standards do not give values beyond 40 weeks' gestation, because it is uncommon to perform ultrasound examinations close to the day of delivery and also because of the small number of cases recruited. Since the measurements at the booking ultrasound examination did not allow fetal weight estimation and are used for precise dating of the pregnancy, we added an invariant 18week fetal weight, based on the study by Hadlock and collegues (1991). This is justified by the fact that the individual variation in size at such early gestations is consistently small in absolute terms, with standard deviations ranging from 18.5g (Ott,1988) to 33.5g (Persson & Weldner,1986 ). The true variation may be even less if imprecision due to gestational dating is also considered. Early differences in fetal size, due to physiological or pathological reasons, will exist but unless extreme, they are not likely to be detected by current techniques of ultrasound assessment. The inclusion of a fixed 18 week weight point

86

stabilises the growth curve and also allows for accurate interpolation of fetal weight at the earlier gestations. There has been some discussion in the recent literature on what should be the best method of deriving standards of fetal size. Altman (1994) believes that the method should reflect the purpose of the chart; longitudinal data should be used for assessment of growth, whereas cross-sectional data is most suitable for assessment of size. He argues that individual curve-fitting applied to longitudinal studies would lead to unduly narrow variances, by 'smoothing ' fluctuations due to measurement error or growth processes. As Piwoz and colleagues (1992) pointed out, the resulting centile reference grid would thus be narrower, and could have potential for misclassification of cases. Analysis of the published ultrasound-derived fetal weight curves suggests that this is not likely to lead to major differences. The two cross-sectional standards of Ott and Hadlock had coefficients of variation at term of 6.2% and 12.7% respectively, whereas the longitudinal standards ranged from 9.3% to 19.4% (table 4.1). The degree of observed variation in the published standards resulting from population and methodological differences is likely to exceed the differences that may result from choice of sampling method. Another methodological issue is whether abnormal cases should be included. Altman (1994), in describing a cross-sectional study, believed they should be. We did not include them because their possibly anomalous growth patterns may distort our standard which is based on serial data. On inspection of the graphs, the morphology of our fetal growth curve appears similar to that of Hadlock and Gallivan, and also to the average of previously published ultrasound growth curves, in that there is only minimal deceleration ('flattening') at term. This is supported by the indices of fractional growth (P50, G28, G37, G42), which are very similar to these two standards. Our growth model is

87

identical to that used by Gallivan's study, but differs considerably from Hadlock's cross-sectional study. The coefficient of variation (SD/EFW) at 40 weeks is 11.6%, very close to Gallivan's 11.5%, but slightly lower than Hadlock's 12.7% (see table 4.1) . The median fetal weight values tend to be lower than other standards, and this may be due to our use of a correction factor to allow for ultrasound overestimation of fetal weight. It differs considerably from birth weight standards based on menstrual dates, where the apparent deceleration at term is more marked. In our local birth weight standard based on ultrasound-dated pregnancies (Wilcox et al, 1993a), there is a reversal in the direction of skewness from positive at term to negative in the preterm period. In contrast, our ultrasound derived standard shows positive skewness at all gestations except 42 weeks; the negative value here is probably spurious, since the sample consists of only 7 cases. The skewness is minimal at around 40 weeks, possibly because of the stabilizing effect of the birth weight data; some of the skewness may be due to ultrasound error. These differences between the ultrasound-derived and the cross-sectional birthweight standards lend further weight to the theory that a substantial number of preterm deliveries are associated with intrauterine growth retardation.

88

Table 11.1 Nottingham ultrasound growth standard.

Sample

Median

Week *

size

(grams)

24

267

674

25

267

26

SD

10th

90th

Skewness

centile

centile

109

534

813

0.656

779

109

639

919

0.645

267

899

115

752

1046

0.614

27

267

1033

124

874

1192

0.575

28

267

1180

138

1003

1356

0.539

29

267

1338

155

1140

1537

0.520

30

267

1508

176

1283

1733

0.525

31

267

1688

199

1434

1942

0.547

32

267

1876

224

1590

2163

0.572

33

267

2072

250

1752

2392

0.590

34

267

2273

277

1918

2628

0.592

35

267

2479

304

2089

2868

0.574

36

265

2686

330

2263

3109

0.525

37

262

2894

355

2440

3348

0.444

38

245

3100

376

2618

3582

0.355

39

208

3301

394

2797

3806

0.253

40

135

3496

407

2975

4017

0.159

41

54

3681

414

3152

4211

0.308

42

7

3854

413

3326

4382

-0.692

* Gestation in exact weeks dated by ultrasound.

89

Figure 11.1 Nottingham ultrasound-derived fetal growth standard. The median, 10th and 90th centile curves for fetal weight are shown from 24 weeks, derived from 267 women who underwent serial ultrasound examination. The original individual curves have been forced through a fixed 18-week point and the birth weight. Gestational age on the x-axis is in exact weeks, calculated on the basis of the BPD measurement at booking.

90

Figure 11.2 Nottingham fetal growth standard compared with others. The Nottingham standard is shown as a thick line. Note the similarity with the four middle standards (Hadlock, Persson, Gallivan and Larsen). Our values tend to be systematically lower than these, possibly because of the use of a corrected weight estimation formula.

91

Figure 11.3 Nottingham fetal growth velocity. The growth velocity in grams per week was obtained by plotting the first derivative of the median for the growth standard. Peak velocity is reached at 36 weeks (210g/week), followed by a rapid decline thereafter

92

Figure 11.4 Nottingham fetal growth velocity compared with others.

The Nottingham growth velocity in grams per week is compared with the previously published standards of Hadlock et al, Persson & Weldner, Ott, Larsen et al, Jeanty et al (using Shepard’s formula), Deter et al and Gallivan et al

93

Figure 11.5 Nottingham proportional fetal growth curve.

The Nottingham median fetal growth curve has been transformed into a ‘proportional’ function, whereby for each week of gestational age the weight is given as a percentage of the predicted 280 day value. This allows comparison of

growth

dynamics

independent

of

the

absolute

weight.

94

12. SYMPHYSIS-FUNDUS HEIGHT

IN RELATION TO

GESTATION AND FETAL WEIGHT.

12.1 Introduction Clinical estimation of fetal size, and sometimes gestation, is often performed by measuring the symphysis-fundus height. This may be plotted against gestation on a reference chart or, more commonly, McDonald's rule of equating one centimetre of fundal height for each week of gestation is used. An alternative clinical method is to estimate fetal weight by simple palpation. In the early version of the customised growth chart, a fundal height axis was included on the right side.This was based on the standard published by Pearce & Campbell (1987), to allow the evaluation of fetal size by this parameter. The aim of this study was to evaluate the relationships between SFH, gestation and EFW, and also to derive a new scale for the fundal height axis of the growth chart based on local data. We also examined the relationships between fundal height growth velocity and birthweight centiles.

12.2 Patients and methods For the purpose of obtaining a fetal weight estimation formula based on SFH, two populations were studied: the derivation set and the validation set. The derivation set comprised 284 prospectively recruited low-risk singleton pregnancies , examined on 3 to 5 occasions prior to delivery.The exclusion criteria for abnormal neonatal outcome described in Chapter 9 were applied. The final study group included 267 singleton pregnancies. In order to study the correlations between SFH and fetal size and SFH with gestation, only one set of measurements was selected randomly from each patient.Ultrasound

95

and fundal height measurements were performed as described in Chapter 9. The validation set consisted of a separate population of 130 unselected patients who were examined just prior to elective caesarian section or induction of labour. This was done in order to eliminate error arising from the lag time between measurement and delivery. The fundal height, engagement of the presenting part, booking height and weight, and parity were recorded. Birth weight was entered in grams. Weight estimation errors were expressed as percentages of the true weight as follows:

Percentage Error = 100*(Predicted weight - observed weight)/(observed birth weight)

To examine the relationship between mean SFH velocity and birthweight-for-gestation z-scores, the whole study group was examined, including cases with abnormal outcome. Customised and uncustomised birth weight centiles and z-scores were calculated using the software described in chapter 9. The mean SFH growth velocity for cases with at least three SFH measurements was calculated by computer software, fitting a straight line using the least squares method. Relationships between pairs were expressed in terms of Pearson's moment correlation coefficients. Interrelated variables were examined by stepwise multiple regression analysis, with P<0.05 as the inclusion criterion. Statistical analyses were performed using SPSS for Windows (SPSS Inc.,Chicago]. Graphs were plotted using either the SPSS graphics facility or Cricket Graph for Windows (Computer Associates, San Diego, California).

96

12.3 Results Figures 12.1 and 12.2 are scatterplots of SFH versus EFW and SFH versus gestation respectively. The correlation between SFH and EFW is stronger than that between gestation and SFH ( R-square values of 0.74 versus 0.69 ).We found that there was no significant improvement in weight prediction by allowing for engagement or including girth (results not shown). A very slight improvement could be attained by allowing for maternal body mass index . The regression line for EFW in terms of SFH is :

EFW (grams) = 225.99 * SFH - 5012.29

When this formula was applied to the validation set, the mean error was found to be -3.7%, with a standard deviation of 18.1%. The distribution of the errors was not significantly different from normal. In the 244 cases with more than 2 SFH measurements, there was a statistically significant positive correlation between SFH velocity and birthweight z-scores. For customised birthweight z-scores, the Pearson correlation coefficient was 0.3201 (P<0.0001), which was better than that between SFH velocity and unadjusted birthweight centiles (R = 0.2921, P<0.0001).

12.4 Discussion In his review of the study by Lindhard and colleagues (1990) published in the Cochrane Database, Neilson (1993) cautiously concluded that ‘it would seem unwise to abandon the use of SFH measurements unless a much larger trial likewise suggests that it is unhelpful’. This is a fair representation of the uncertainty in the literature on the value of this obstetric parameter.

97

Some of the variability in the results obtained by previous workers is likely to be due to the different approaches in deriving local reference charts. Unlike most of the previous standards for SFH that show some deceleration of the curve at term, our study suggested a linear relationship between gestation and SFH throughout pregnancy. This is most likely due to the increased precision in gestational age estimation resulting from the use of early ultrasound measurements, without using menstrual dates. If we assume that SFH gradients reflect fetal growth velocity, then the finding of a better correlation of SFH velocity with customised birth weight z-scores than with unadjusted z-scores supports the view that adjusting for pregnancy characteristics gives a better indication of fetal growth dynamics than otherwise. However, the low values of the correlation coefficients for these observations implies that mean SFH velocity on its own is likely to be of limited clinical utility in the detection of the SGA infant. Before the advent of obstetric ultrasound, SFH measurement or clinical palpation were in common use as a way of estimating fetal size. There is considerable evidence to show that, for fetal weight estimation, SFH measurement is at least as good as clinical palpation ( Secher,1990; Pschera et al, 1984; Lindhard et al, 1990), and the latter is

accurate to within 450g of the birth weight in 80% of cases

(Loeffler, 1967). However, it is quite possible that clinicians making fetal weight estimates on the basis of ‘clinical palpation’ may subconsciously also be using other information to them available, such as gestational age and maternal size; in Loeffler’s study the patients were in labour, and the clinicians had full access to all the clinical data. If this is confirmed in a well designed study, it would mean that as an objective measure clinical palpation may be inferior to SFH measurement.

98

Our finding of a better correlation of fundal height with fetal weight than with gestation is in agreement with the work of De Muylder and colleagues (1988), and lends some support to using this technique for growth screening. Furthermore, by using ultrasound fetal weight estimates, we were able to show that SFH measurements can be used to estimate fetal weight in the pre-term period. No data has been published on the accuracy of simple clinical palpation for fetuses delivering at these gestations. Since clinicians have ‘calibrated’ their fetal weight estimation skill by clinical palpation on infants delivering mostly at term, it may well be that in the pre-term period clinical palpation is not as accurate. However, our figures suggest that the accuracy of isolated measurements is , not clinically useful unless the values are extreme. This is not surprising , since fundal height is a combined measure of maternal and fetal tissues, and is in any case more likely to reflect fetal length than weight. It may have greater power if performed serially and frequently , thus allowing trend analysis, and by the same observer, to reduce error from inter-observer variability.

99

100

101

13.

FETAL

GROWTH

KINETICS

IN

RELATION

TO

PREGNANCY CHARACTERISTICS.

13.1 Introduction Although the relationships betwen pregnancy characteristics and birth weight have been extensively documented, little is known on the relationships between these characteristics and ultrasound-estimated fetal weight in the antenatal period. One of the assumptions of the customised growth chart is that the factors influencing birth weight are also operative earlier in the third trimester. In this study, we attempt to explore this issue using three techniques: (1) graphic display of mean growth curves derived from different subgroups, (2) non-parametric assessment of differences between groups, and (3) multiple regression analysis. We also examine fetal growth among babies born pre-term.

13.2 Patients and methods Two hundred and sixty seven cases had clinically normal outcome as described in chapter 5 and also sufficient data points to generate individual growth curves. Ultrasound-estimated fetal weights were calculated from the individual growth curves at 26, 28, 30, 32, 34 and 36 to 40 exact weeks, without extrapolation. Hence the samples became progressively smaller from 37 weeks. For the purpose of comparing mean growth curves, the population was subdivided into the following groups: primiparas and multiparas, male and female fetuses, European and Indo-Pakistani, smokers and nonsmokers, tall vs medium vs short stature, heavy vs medium vs light maternal weight (at booking). For height and weight, the population was divided in three equal groups, corresponding approximately to the 33rd and 67th centiles of these variables. Mean growth curves for each group were derived using the methods described in chapter 10.

102

Descriptive statistics for the individual growth coefficients were calculated. Differences in birthweight and EFW's at each gestation were evaluated using non-parametric tests. For two independent samples, the Mann-Whitney U test was used, whereas for continuous variables such as maternal height and weight Spearman's rank correlation coefficient was employed. Stepwise multiple regression analysis (selection criteria : prob. in = 0.05; prob. out =0.10) was used to identify factors that were significantly related to fetal weight and also to birth weight. Categorical variables were recoded as dummy variables ( 1 or 0), while continuous variables such as height and weight were entered without transformation. Gestation was not entered, since the EFW’s were analysed at fixed gestational points (26 to 40 weeks). In order to compare the growth patterns of the cases born preterm with the normal cases born at term a different statistical procedure was adopted, because of the 17 preterm cases only 11 had sufficient datapoints to generate individual growth curves. An average growth curve for these cases was plotted by the method described above. All the ultrasound-estimated fetal weights (up to 37 weeks) for the normal cases born at term (the control group) were transformed into zscores according to the longitudinal fetal growth standard described in chapter 10, and pooled together to yield a dataset of 736 points. For the preterm deliveries, both the birth weights and the EFW’s were transformed into standard deviates by the same growth standard, leading to a dataset of 55 points. The differences between the means of these z-scores could then be analysed using Student’s T-test. Statistical analyses were carried out using SPSS for Windows (ver 6.0). Graphs were plotted using Cricket-Graph graphic software.

103

13.3 Results A description of the population characteristics is shown in table 13.1; these are similar to the overall population figures shown in tables 9.1 and 9.2. The distribution of fetal weights at each week did not differ significantly from normal, as assessed by the Kolgorov-Smirnov test at the 0.05 level (table 10.1). The fetal growth curves in different subgroups are shown in Figures 13.1 to 13.7. Ethnic groups other than Indo-Pakistani were not plotted because of the small numbers in these categories. The basic statistics for the individual growth coefficients of the log polynomial equations are shown in table 13.2. These are significantly skewed and with unequal variances; furthermore they are closely correlated to each other (R>0.99 for all pair-wise combinations). The results of the non-parametric tests are shown in tables 13.3 and 13.4. The multiple regression constants, coefficients and R-square values are shown for each week of gestation in table 13.5; the coefficients are entered as zero if they are not statistically significant. The differences between the z-scores of EFW’s of pre-term and term deliveries are shown in table 13.6. Infants delivering pre-term had fetal and birthweights that were on the average 0.30 standard deviates (or 10 centiles ) below those of babies delivering at term, a difference that is statistically significant. This analysis was repeated by excluding birthweights from the preterm group; the latter’s values were still significantly lower by the Wald-Wolfowitz Runs Test (z =-1.6537, P= 0.0491).The mean growth curve for the preterm group is displayed graphically in figure 13.7.

13.4 Discussion Non-parametric and multiple regression analysis of EFW at different gestational ages confirms that at least some of the pregnancy characteristics that are affecting birth weight are also operative in the

104

antenatal period. The differences in growth patterns evident in the graphs for different subgroups are interesting, but are rather refractory to statistical analysis. Multivariate tests such as Hotelling T2 in relation to the individual growth coefficients could be applied, but this would not be a valid exercise because the coefficients have significantly unequal variances

and also a markedly skewed

distribution (table 13.2). These variables cannot be transformed without changing the interpretation of the results. The most marked differences in the growth curves are seen in comparing groups of maternal size .Maternal weight is highly correlated with EFW from 26 weeks onwards, whereas the effect of height does not become significant until 32 weeks. This is supported by the multiple regression analysis, which identifies maternal weight and its powers as significant throughout the gestational interval examined. The growth kinetics for primiparae and multiparae appear similar up until 36 weeks; thereafter primiparae show some relative slowing in fetal growth, whereas in multiparae it continues almost linearly. The differences do not become statistically significant by the nonparametric test until 39 weeks, although by multiple regression analysis parity is an independent significant variable except from 32 to 36 weeks. This suggests that most of the known variation in birthweight due to parity develops late in the third trimester, and may be related to differences in the intrauterine environment rather than fetal genetic factors. Differences in growth between Europeans and Indo-Pakistanis become significant by the Mann-Whitney U-test in our study from 34 weeks. It is of interest that although the kinetics of their growth curves are dissimilar, these differences are not borne out by

the multiple

regression analysis. This may be due to differences in maternal size being responsible for most of the differences, and also the small

105

sample size. The very few cases belonging to the Afro-Caribbean group and ‘Others’ are nevertheless identified as significant independent variables intermittently from 26 weeks. Chang and colleagues (1992) reported a longitudinal ultrasound study on fetal growth in a sample of 20 Bangladeshi and 67 European women; statistically significant differences between the mean estimated fetal weights of the two groups were noted from 28 weeks onwards, but their analysis did not allow for the confounding effects of maternal size and parity. Two articles have been published on ultrasonic fetal growth parameters in different ethnic groups. Vialet and colleagues (1988) studied two groups from the same location by ultrasound: 201 African women and 201 European women. These were matched by socioeconomic status and parity. The BPD, FL and AD growth curves for the two populations were obtained using the Rossavik model. They found that while the AD growth curves were similar, African fetuses tended to have significantly smaller BPD's and longer FL's in the second half of pregnancy. Their birth weights were on average 200g lighter than Europeans, with no significant differences in the duration of pregnancy. Although weight gain curves were not produced, these observed differences suggest that the weight estimation formulae in current use may yield biased results in this population, since they were derived from largely European populations.Simmons and colleagues (1985) were able to study a group of Bengali patients longitudinally; they measured the BPD and the abdominal area from 14 weeks gestation, and plotted their results against Campbell and Newman's standard for Europeans.The mean BPD measurements were below the median from about 18 weeks, and the abdominal areas were also lower from about 30 weeks, in both cases never below the 5th centiles. That their mean birthweight was 300g lower than the European mean is to be expected, because of the strong correlation of abdominal area with fetal weight. Both of these studies suggest that fetal growth does not

106

vary significantly within ethnic sub-groups up to about 20 weeks, maternal influences becoming effective thereafter. This is in agreement with the animal work discussed in chapter 1 (Snow, 1989), showing that maternal effects tend to operate late in pregnancy. Male fetuses were heavier than females from 26 weeks, but these differences did not reach statistical significance until 36 weeks in our study. Sex-related differences in individual ultrasound parameters (BPD, HC and AC) have been reported from as early as 24 weeks (Parker et al, 1984) . Cigarette smoking, as reported in mid-trimester, results in lower fetal weights in the third trimester which in our sample shows statistical significance from 36 weeks. This is consistent with the effect of cigarette smoking on birthweight which appears independent of other characteristics such as maternal height and booking weight (Wilcox et al, 1993a) . The lower R-square values in the multiple regression analysis for the earlier gestations imply that less of the variability in EFW can be explained by pregnancy characteristics , about 10% as opposed to 22% at 40 weeks. It is likely that this is due to ultrasound error and interpolation error in estimating fetal weight for the required gestations from the growth curves. Apart from the study by Persson and colleagues (1978) on growth of the biparietal diameter, previous reports of increased prevalence of retarded growth among cases born pre-term were based on comparison of birthweight datasets with ultrasound derived standards (Ott,1993; Secher et al,1987; Persson, 1989). We were able to show that the zscores of

EFWs and birthweights of babies born preterm were

significantly lower than the EFWs measured before 37 weeks by the same standard in babies delivering at term. Since only 30.9% (17/55) of the data points from the preterm group were birthweights, it is unlikely that the inclusion of this data could be a source of bias; the

107

Hadlock ultrasound weight estimation formula had been corrected for systematic error, and also the error from this formula is not correlated with fetal weight. Exclusion of preterm birthweights from the analysis still shows lower values for this group, but the statistical significance is weakened due to the smaller sample. Since our pooled values are gestation-independent, and also because of the small numbers, we cannot determine at what gestation these differences become significant. Persson and colleagues (1978) were able to detect significantly smaller biparietal diameters in the babies destined to be born prematurely from 26 weeks onwards, and since this parameter tends to be relatively ‘spared’ in IUGR, it is possible that fetal weight differences may exist even before this gestation. It also suggests that the growth retardation pattern in this group of babies is of the symmetric type. Our data show that factors which are known in the first half of pregnancy - such as maternal height and booking weight, parity and ethnic group - and which have an effect on birth weight, are also associated with variation in fetal weight in the third trimester of pregnancy. These findings suggest that no single standard can accomodate for the variation of fetal growth, which needs to be assessed in the context of individual pregnancy characteristics.

108

Table 13.1 Demographic characteristics of normal population (N = 267).

Ethnic Group: European

No. (%) 251 (94)

Indo-Pakistani

13 (4.9)

Afro-Caribbean

2 (0.7)

Other

1 (0.4)

Smokers:

40 (15.0)

Fetal sex: Males

135 (50.6)

Primiparas

127 (47.6)

Parity:

109

Figure 13.1 Maternal weight at booking and fetal growth. Fetal growth is plotted for two groups: those whose mothers had booking weights above the 67th percentile and those below the 33rd percentile. Fetuses of the heavier mothers display accelerated growth from the 26th week of gestational age.

110

Figure 13.2 Maternal height and fetal growth. Fetal growth is plotted for two groups: those whose mothers had heightsabove the 67th percentile and those below the 33rd percentile. Fetuses of the taller mothers display

accelerated

growth

from

the

32nd

week

of

gestation.

111

Figure 13.3. Sex and fetal growth. Fetal growth is plotted for male and female fetuses. Both sexes have a similar growth pattern, but males are significantly heavier

from the 36th week of gestational

age.

112

Figure 13.4 Parity and fetal growth. Fetal growth is plotted for primiparae and multiparae. Growth follows a similar pattern until 36 weeks, therafter the primiparae show a slightly decelerative course. Statistically

significant

differences

are

noted

from

39

weeks.

113

Figure 13.5 Ethnicity and fetal growth. European vs Indo-Pakistani. Fetal growth is plotted for Europeans and Indo Pakistani. Growth is slower in the Indo-Pakistani from about 30 weeks. Statistically significant differences are noted from 34 weeks, but these could not be shown to be independent of maternal size.

114

Figure 13.6 Effect of smoking on fetal growth. Fetal growth is plotted for smokers and non-smokers. Fetuses of smoking mothers are lighter throughout the gestational interval studied. Statistically significant differences are noted from 36 weeks.

115

Figure 13.7 Preterm delivery and fetal growth. Fetal growth is plotted for fetuses delivering preterm (<259 days) and those delivering at term. Fetuses delivering preterm are significantly lighter than those proceeding to deliver at term.

116

14. CUSTOMISED GROWTH CHARTS IN

RELATION TO

NEONATAL OUTCOME.

14.1 Introduction In order to assess two different methods for the detection of a condition a third independent method, or 'gold standard', would be required for an unbiased comparison. Regrettably, in the case of IUGR there is no generally accepted neonatal test that is accurate and reproducible. To overcome this difficulty, this study was limited to comparing the performance of adjusted and unadjusted growth charts in a population with clinically normal neonatal outcome. This would allow us to determine which method is best at defining normality, by giving us estimates of the true negative and false positive rates.

14.2 Materials and methods Only those cases with normal neonatal outcome, as defined by the criteria listed in chapter 9, were selected. A total of 267 pregnancies with a number of ultrasound examinations sufficient to plot individual growth curves were included in this analysis. In comparing the customised and the uncustomised growth charts, care was taken that the standard deviations entered for the two methods had the same coefficients of variation. Curve-fitting was carried out according to the method described in chapter 7. A computer program was developed in order to check whether the growth curve for each case was wholly within the 10th and 90th centile boundaries or crossed either from 27 weeks until delivery. Both customised and uncustomised reference grids were used. The uncustomised boundaries were defined by using the 'average' proportionality growth curve described in chapter 6, forced through the Nottingham population mean for 40 exact weeks of

117

3447g. Thus each case could be classified by both customised and uncustomised criteria in one of four groups as follows:

1. Crosses the 10th centile. 2. Within the 10th and 90th centiles. 3. Crosses the 90th centile. 4. Crosses the 10th and 90th centiles.

Differences between the two methods could then be tested using nonparametric tests for related samples. Group 4 cases were excluded from analysis, since these are more likely to contain large errors in ultrasound fetal weight estimation . We also investigated the relationship between the standard deviation entered in the customised growth chart program and the percentage of cases that would cross the 10th centile boundary. The coefficient of variation (SD/median) in the computer program described above was increased in steps of 1 per cent, and the resulting proportion of cases crossing the 10th centile was plotted.

14.3 Results Table 14.1 shows the percentage of cases within each group according to the type of reference boundary. The differences due the classification method are statistically significant Wilcoxon

Matched-Pairs

Signed-Ranks

test

according to the (table

14.2);

customisation of the reference range results in significantly fewer normal cases crossing below the 10th centile, but more cases cross the 90th centile. If the population is re-grouped in two categories, as either crossing or not crossing the 10th centile, the differences due to the classification method remain highly significant (table 14.3), with fewer of the normal cases crossing below the adjusted 10th centile boundary. Fewer cases cross both the 90th and the 10th centile

118

boundaries using the customised compared with the uncustomised chart (6 Vs 9), but this is not significant. A more detailed breakdown of the categorical shifts in terms of cases crossing the 10th centile is given in table 14.4. Many more of the cases labelled as SGA by the unadjusted method are reclassified as not SGA by the customised method than the reverse. In order to study the systematic trends of the two methods, the EFW's were calculated for all cases from their growth curves at 28 weeks, and then transformed into customised and uncustomised centiles. The mean customised centile was 33, whereas the mean uncustomised centile was 43 (P<0.00001, Wilcoxon Matched-Pairs Signed-Ranks Test). Figure 14.1 shows the relationship between the coefficient of variation of the standard deviation entered in the customised chart program and the percentage of cases crossing the 10th centile reference line; it is likely that most of these would be false positives for IUGR.

14.4 Discussion Statistical analysis of these results suggest that when serial ultrasound examinations are performed, fewer of the cases with normal outcome will be labelled as SGA using customised growth charts than using a fixed reference standard. McNemar's Test is the best method to analyse this data, because it is a non-parametric test that can be used to test whether dichotomous variables generated by one method differ significantly from dichotomous variables generated by another method applied to the same sample. A 2 x 2 table is constructed, and the significance level is determined by either the Chi-square test or by the binomial distribution if fewer than 25 cases are re-categorised by the second variable. Our selection criteria for this study - being independent of birthweight and morphometry - will not exclude the milder, asymptomatic cases of

119

IUGR. Conversely, not all of the neonates with abnormal outcome have been affected by growth disturbances. These considerations will prevent us from making an objective assessment of true negative and false positive rates for IUGR, but should not affect our conclusions when we are comparing the two methods applied to the same essentially normal population. The rather high percentage of cases crossing the 10th and 90th centile boundaries is likely to be due to fluctuations in the growth trajectory resulting from ultrasound error, and also the relatively small number of ultrasound examinations carried out per patient. As a result the, the 'SGA' rates of 33.2% by uncustomised and 25.5% by customised growth charts are too high by both methods. This suggests that the standard deviation generated by the program is too narrow in view of the ultrasound error. As figure 14.1 shows, the coefficient of variation should be at least 16% in order to reduce the percentage of cases labelled as SGA in this normal population to well under 10%. This approaches the figure of 19.4% in Jeanty's reference standard (1984). If this growth chart were to be used on its own as is, the action thresholds would need to be reduced to a degree commensurate with the local ultrasound error rates in order to avoid excessive referrals and unnecessary parental anxiety. Allowing for maternal characteristics in defining the 10th centile cut off is likely to result in a significant reduction in the false positive rate for IUGR. Of the cases classified as SGA by the uncustomised method, 27.5% (25/91) are reclassified as normal by the customised chart. Conversely, some 2.3% (4/174) of the cases classified as normal by the unadjusted chart are reclassified as abnormal by the customised chart. Although a higher proportion of cases may be crossing the 90th centile using the customised method, this boundary is less related to neonatal morbidity than the 10th centile (Patterson,1986 ), and hence less clinically important. Work in progress in our unit on a sample of

120

more than 1000 neonates shows a progressive reduction in neonatal morbidity with increasing birthweight rank (Dr T. Muls, personal communication), rather than the U-shaped distribution described by Patterson between 38 and 41 weeks' gestation . Our finding of fewer cases crossing the 10th centile boundary is against an overall trend for customised centiles to be lower than uncustomised centiles (chapter 14), which is evident for both birth weights (chapter 14) and EFW. This suggests that the customisation method is operating more selectively than uncustomised charts. Further support is the fact that fewer cases cross both centile boundaries using the customised charts than with the uncustomised charts. The design of this study does not allow us to make any statements about the positive predictive value of customised charts, since as discussed above we do not have clear criteria to define IUGR. This is a problem shared with other tests used in perinatal medicine, such as the biophysical profile, in that they have high negative predictive values but limited positive predictive values. It is likely that to some degree this may be due to our limited diagnostic arsenal in identifying and classifying perinatal pathology.

121

Table 14.1 Percentage of fetal growth curves crossing 10th and 90th uncustomised and customised centiles for 274 cases with neonatal normal outcome.

Category

Group

Uncustomised

Customised

No.

Number

Number

%

of cases Cross the 10th centile Within

10th

and

1 90th 2

%

of cases

91

33.2

70

25.5

133

48.5

136

49.6

41

15.0

62

22.6

9

3.3

6

2.2

274

100.

274

100.0

centiles Cross the 90th centile

3

Cross both 10th and 90th 4 centiles Total

0

122

Table 14.2 Customised versus uncustomised centile boundaries. Differences

in

group

ranks

according

to

the

classification

method.(Group 4 excluded). Rank 1: crosses 10th centile. Rank 2: wholly within 10th and 90th centile boundaries. Rank 3: crosses 90th centile . Wilcoxon Matched-Pairs Signed-Ranks Test:

Mean rank

Cases

Direction

24.03

38

- Ranks (UNCUST < CUST)

21.00

8

+ Ranks (UNCUST > CUST)

218

Ties (UNCUST = CUST)

264

Z = -4.0697

Total

2-Tailed P <0 .00005

123

Table 14.3 Customised versus uncustomised centile boundaries. Population reclassified as either Group (1): crosses the 10th centile; or Group (2): does not cross the 10th centile ( valid N= 265). McNemar Test.

Cust. Gp 2

Cust. Gp 1

Totals

Uncustomised Gp1

25

66

91

Uncustomised Gp2

170

4

174

195

70

Totals

Cases = 265; Chi-square = 13.8; P = 0.0002;

Table 14.4 Shifts in categories according to classification method. (Group 4 excluded.)

From:

To:

Frequency

SGA by

not

customised

uncustomised

SGA by

not

uncustomised

customised

unchanged

unchanged

236

89.1

Total:

265

100.0

SGA

SGA

Percent

by 4

1.5

by 25

9.4

124

Figure 14.1 Relationship between coefficient of variation and false positive rate for SGA. The graph shows the percentage of individual growth curves crossing the 10th percentile as a function of the coefficient of variation (SD/EFW), as the latter is artificially widened or narrowed. The standard deviation needs to be approximately 15.5% of the median weight in order to allow 10% of the population to cross the 10th percentile cut-off.

125

15. THE PREDICTION OF BIRTH WEIGHT

15.1 Introduction As demonstrated in the preceding chapters, the combination of ultrasound weight estimation error and error in gestational age can result in substantial fluctuations in the apparent fetal growth curves, to the extent that some 25% of cases will cross the 10th centile reference line using customised growth charts. Birth weight, although it is not an accurate indicator of individual growth kinetics, remains one of the most reliable measurements in this study. Both the customised growth chart and the IBR programs base their calculations on a predicted birth weight at a given gestation. It is important therefore to compare the predictive ability of the adjusted versus the unadjusted standards.We also analysed the birthweights in our normal sample using three standardisation methods: the customised growth chart, the unadjusted standard and the IBR .

15.2 Patients and Methods We analysed both the population with normal outcome (N = 282) and also those with abnormal outcome ( N = 42), to see the extent the weight prediction error would be increased by including pathological cases. The unadjusted growth standard was that described in chapter 13, whereas the customised program was the same as described in chapter 6. Software for the calculation of the IBR and IBR centiles was compiled with the assistance of Mr Mark Wilcox. The predictive ability of the customised and uncustomised standards was assessed by calculating the signed and the absolute prediction errors for each case, both in grams and as a percentage of the true weight. The percentage error was calculated thus:

126

Percent Error = 100 * (Predicted - Observed)/Observed

This was done by modifying the computer programs described in chapter 13 ; the predicted weight by the customised program was calculated using the multiple regression model, while by the unadjusted standard the corresponding value was the median for gestation. The cumulative frequency distribution of the errors by both methods was then plotted. The significance level of the differences between the two methods was calculated by applying Wilcoxon matched-pairs signed-ranks test. The frequency distribution of the number of cases in each decile category according to the different birthweight standards were plotted as histograms, for the overall population and also for non-smokers, since smoking is probably the most common cause of mild IUGR. In order to examine time trends in relation to customised and IBR centiles, the centile difference between the two was calculated for each case and plotted as a function of gestation.

The strength of this

relationship was quantified by calculating Spearman's rank correlation coefficient. The relationships between customised, uncustomised and IBR centiles were also studied using Spearman's rank correlation coefficients, and displayed graphically. The percentile values obtained by the three methods were categorised in ten groups, 1 to 10, according to the corresponding decile values. Differences in categorisation between two standards were tested using the Wilcoxon matched pairs signed-ranks test, for the three possible combinations: customised vs uncustomised, customised vs IBR and uncustomised vs IBR .

127

15.3 Results The descriptive statistics (including the Kolgomorov-Smirnov test for normality) for the term birthweights in the population with normal outcome are shown in table 15.1. The distribution did not differ significantly from normal, with only a very slight degree of positive skewness. The customised growth chart program was significantly better at predicting birthweights than the unadjusted chart, with a mean absolute error of 303.5 g versus 342.6 g respectively (z = -4.11,P < 0.00001). The cumulative frequency distribution of the absolute error for the two methods is shown in figure 15.1. The error was less than 250g in 50% of the population using multiple regression, whereas using the unadjusted, gestation-specific standard this figure was reduced to 46% The Spearman correlation coefficient between the predicted weight by multiple regression and the birth weight was 0.9642. The standard deviation of the signed prediction errors for the normal population was 5.35% of the true weights for the customised program and 5.63% using the unadjusted method. The corresponding figures in the neonates with adverse outcomes were 7.65% and 8.02% (table 15.2 ). The frequency distributions of deciles according to the three different methods tested

are shown as histograms in figures 15.2 to 15.4.

These are significantly different from each other, as detailed in table 15.2. Frequency distributions were also drawn for the non-smokers and these are shown in figures 15.5 to 15.7. For the unadjusted standard, the exclusion of smokers results in a reduction of cases below the 10th centile by 1.7% while the corresponding figures for the IBR and customised chart are 1.1% and 2.8%. Spearman's correlation coefficient for customised and IBR centiles was 0.9939; the scatterplot for these values is shown in figure 15.8. A highly significant negative

128

correlation was noted between gestation and the difference between customised and IBR centiles (R = -0.607, P< 0.0001 ).This is shown in figure 15.9.

15.4 Discussion The multiple regression model in use by the adjustable standards performs significantly better than the unadjusted standard in predicting birth weights. What is more remarkable is that the overall predictive performance of the models using gestation alone or gestation in combination with maternal characteristics is superior to that of the ultrasound weight-estimation formula used in our study (SD of ultrasound error: 10.2%, table 9.3). Even in the abnormal population, where one would expect an increased prevalence of growth disorders, the SD of the error using multiple regression (7.65%) matches the best reported figures in the ultrasound literature (Hadlock et al, 1985). In the multiple regression analysis of the East Midlands Obstetric Database, the standard error of the regression model, which was centred on 40 weeks, was 389 grams , which is about 11% of the overall mean birthweight at 40 weeks (3443g). The reason for the improved performance of the model in our population is probably due to the presence of frequent inaccurate values in the obstetric database, as opposed to our carefully checked entries in our study group. That multiple regression using pregnancy characteristics only may be as good as ultrasound was also suggested by Rogers and colleagues (1993) in Hong Kong. They developed a multiple regression model for the prediction of birthweight (inclusive of the same variables as our model) from half of a Chinese database of 23750 singleton deliveries, and analysed its performance by testing it against the other half. The standard deviation of their absolute errors was 130g/Kg, i.e.13% of actual birthweight. The corresponding value in our sample was at worst 4.32% (table 15.2).

129

Likewise

the

correlation

coefficient

between

predicted

and

birthweights in their sample was 0.53, as opposed to ours of 0.96. Their poorer performance is almost certainly due to the fact that their population was not routinely dated by ultrasound, and not selected on the basis of pregnancy outcome. These findings suggest that accurate knowledge of only two factors is sufficient in order to make reasonable fetal weight estimates: a population-specific average growth curve and gestational age. It also emphasises the effect of gestational age on fetal weight. The use of the two adjusted standards results in a much smoother distribution of deciles than the unadjusted standard. This means that the random fluctuations in the centile values of the unadjusted standard due to the small sample size tend to even out when processed by the customised standard together with the variation in pregnancy characteristics. It may well be that this feature of the adjustable standards makes them more suitable for analysing small data sets. There is a relatively high number of cases below the 20th centile using the unadjusted and the customised standards (26.5% and 22.6% respectively, versus the expected 20%) , while the corresponding figure for the IBR is 19.7%. The customised charts results in values that are usually lower than the IBR centiles; this is probably because the customised growth chart is predicting birthweights for a non-smoking population, whereas in the IBR program the smoking factor is ignored. This difference in the way the two programs handle the influence of smoking is also reflected in the changes in the percentage of cases below the 10th centile when the smokers are excluded from the term population. This change is greatest for the customised growth chart, which resulted in 2.8% fewer cases being labelled as ‘SGA’ when the smokers were excluded.

130

The highly significant negative correlation between gestational age and the difference between customised and IBR centiles is likely to be due to the different growth functions; the customised growth chart uses an average function which is almost a linear relationship between fetal weight and gestation, whereas the IBR uses the birthweight curve derived from the East Midlands Obstetric Database, which shows some degree of 'flattening' at term. Hence with increasing gestation, the difference in birthweight expectation between the customised and the IBR programs widens, resulting in an increasing difference between the centiles calculated by the two methods. This implies that, of the infants born post-term, more would be labelled as ‘SGA’ by the customised standard than using the IBR program, suggesting greater sensitivity of the customised standard at this critical gestational age. The very high correlation between the IBR and customised centiles is indirect evidence in favour of the clinical efficacy of the customised chart program, since the IBR program has been shown by Sanderson and colleagues (1993) to be better able at detecting IUGR than the unadjusted birthweight standard.

131

Table 15.1 Descriptive statistics for term birth weights for infants with normal outcome. Mean (SD) S.E. of mean Skewness (SE) Kolmogorov-Smirnov test for normal distribution Range Valid No.

3474.8 (490.8) g 29.2 0.086 (0.145) z = 0.6248 P = 0.8298 2890 - 4900 g 283

Table 15.2 Analysis of birthweight prediction errors in the populations with normal and abnormal neonatal outcomes. Sample

Method

Systematic Error (%)

Normal (N=283)

Unadjusted (Gestation only) Multiple Regression Model Unadjusted (Gestation only) Multiple Regression Model

Abnormal (N= 42 )

0.67

Standard Deviation (%) 5.63

Mean Absolute Standard Error (%) Deviation (%) 4.93 2.79

0.18

5.39

4.57

2.81

0.65

8.02

6.71

4.32

-0.94

7.65

6.39

4.19

132

Table 15.3 Differences in classification by standardisation method. Wilcoxon matched-pairs signed-ranks test for the population of infants born at term. Pair comparison: A with B

AB A = B (ties)

Number

Mean rank

P-value

Customised with IBR

Customised < IBR Customised > IBR Customised = IBR

161 0 121

81.0 0.0

< 0.00001

Uncustomised with customised

Uncust < cust Uncust > cust Uncust = cust

113 54 115

91.32 68.68

< 0.00001

Uncustomised with IBR

Uncust < IBR Uncust > IBR Uncust = IBR

171 18 94

99.47 52.50

< 0.00001

133

Figure 15.1 Birth weight prediction errors: customised vs uncustomised charts. The graph shows the cumulative percentage of absolute birth weight prediction errors using either the pregnancy characteristics-adjusted (‘customised’) or unadjusted (‘uncustomised’) charts when applied to the population with normal neonatal outcome. Customised charts give significantly better predictions than the unadjusted charts.

134

135

136

137

138

139

140

141

142

16. COMMENTS AND CONCLUSIONS

Adjusting for pregnancy characteristics. The concept that growth standards should be adjustable according to pregnancy characteristics has been rather controversial. It was postulated by Thomson and colleagues (1968), and was based on the known relationships between these characteristics and birth weight. Their birth weight standard was parity and sex-specific; the adjustment coefficients for maternal height and weight were fixed, because not enough data was available to analyse gestational-age dependent changes. They did not provide any clinical evidence, however, that adjusting for maternal characteristics leads to superior performance than unadjusted charts, and their arguments were based on rather intuitive grounds. These issues were debated in a discussion chaired by Professor Whittle (1989), and attended by Nicolaides, Alberman, Wigglesworth, Steer, Campbell and others. It was pointed out that although differences in mean birthweights between subgroups may appear small, shifts of such magnitude will affect the tails of the distributions significantly and these can actually be associated with important changes in perinatal mortality.We also have found that small shifts in the median values will lead to major changes in the number of cases that are reclassified as SGA (Gardosi et al, 1994). It was agreed at this discussion that it is legitimate to correct for sex and plurality, because in these instances the mortality of the smaller group is lower; but no agreement could be reached on parity and maternal weight. It was felt that because maternal malnutrition was a common cause of IUGR in the developing world it is not legitimate to adjust for maternal weight. Steer believed the effect of parity to be mediated through maternal pre-conceptional weight, and therefore should not be considered as an independent adjustment factor. He referred to work by Van Der Spuy and collegues (1983), purporting to show that

143

women who were underweight at the time of conception (as defined by the body-mass index) had double the risk of preterm delivery and a three-fold increase in the incidence of SGA.This was even greater if ovulation was induced. While this may be true for malnourished women, it disregards the relationship between maternal weight and birthweight for women within the normal range of body mass index. Another limitation of this study is that maternal weights were recorded at booking (<16 weeks), rather than pre-pregnancy. This introduces the confounding factor of weight gain, which may be considerable by 16 weeks, and is significantly correlated with birth weight. In our multiple regression analysis of the East Midlands Obstetric Database we found parity to be a significant and independent factor influencing birth weight (Wilcox et al, 1993a). This has also been the experience of other workers (Thomson et al, 1968; Voigt et al, 1989). Our data shows that in terms of predicting birth weight, adjusting for pregnancy characteristics is significantly more accurate than using gestational age alone. We were able to show that in our sample the factors influencing birth weight are also operative in the antenatal period from as early as 26 weeks. There seems a rough inverse relationship between the importance of the maternal factor and the apparent gestational age of onset of the factor, with the stronger factors being operative from an earlier stage than the weaker factors. Parity, maternal weight, and ethnic group appear to be major independent significant variables, although less of the variability in EFW can be explained by these factors , probably because of ultrasound error. This supports the principle of

adjusting the standard for pregnancy characteristics

throughout the third trimester. The origin of the observed differences in fetal size within maternal subgroups remain obscure. The studies using animal models suggest that genetic influences do not become important until the second half

144

of pregnancy, and this is supported by the human studies showing minimal inter-ethnic differences in the ultrasound dating parameters. It is not possible at the moment to determine whether the differences in birth weights between two groups are due to a preponderance of growth stimulatory effects in one or growth retarding effects in the other. It may be that effects of physiological origin are mediated by growth stimulatory factors, whereas differences due to pathological factors are mediated by growth retarding factors.

Design of the customised growth chart. We found the prediction of birth weight using the multiple regression model to be unexpectedly good. Given gestation and pregnancy characteristics, the accuracy of the model was in fact better than the ultrasound fetal weight estimation formula we used, and matches the published figures for such formulae. This degree of accuracy is most likely to be due to the use of early ultrasound in the estimation of gestational age, and it supports the validity of the ‘proportional’ fetal growth curve. This method is unlikely to be accurate in cases where some degree of growth disturbances is suspected, since the multiple regression model is designed to predict median values. Although it is effective in predicting weight, we do not know whether the assigned percentile values are indeed better related to perinatal morbidity and mortality. Some of the other principles on which the customised growth chart is based upon remain open to discussion. The method of deriving the adjustment coefficients for pregnancy characteristics by multiple regression analysis was a compromise between selecting a 'supranormal', non-smoking population and an unselected population without making any allowance for smoking. Other methods of obtaining these coefficients should be explored, as they could possibly lead to improved accuracy.As maternal weight is one of the most

145

important adjustment parameters, the customised growth chart program has a 'range-checking' mechanism to prevent making inappropriate adjustments for

weight when there is evidence of

malnutrition or gross obesity. This is based on the normal values of body mass index in mid-pregnancy. An alternative technique to using multiple regression for the prediction of term birth weight is a computer neural network. This is a method used in artificial intelligence whereby observational data related to a particular outcome is fed repeatedly to a multi-layered network of inter-related 'neurones', which will then form weighted connections. The network will then be able to make predictions on outcome when faced with a new set of data. This has been applied to the East Midlands Obstetric Database, and interestingly the predictive power of the network was not superior to the multiple regression model (Mr Mark Wilcox, personal communication). The use of a single fractional fetal growth curve derived from an average of previous studies may be questioned. Prior to our work, there was no ultrasound data on fetal weight to suggest major morphological differences between subgroups, and hence there was no alternative to using a single type of growth curve. Apart from the evidence favouring EFW over the individual ultrasound measurements as a screening tool for growth disturbances (Chang et al,1993; Hedriana & Moore,1994), the main reason we chose this parameter is that the adjustment coefficients used in the customised growth chart have been derived from birthweight data. Another reason is that very little has been published on the relationship between pregnancy characteristics and

individual

ultrasound parameters.

146

Fetal Size Versus Growth Velocity There has been considerable debate in the recent literature over the relative merits of fetal size and growth velocity in the prediction of fetal compromise (Gardosi, 1994; Chang, 1993; Danielian, 1993). While it is indisputable that fetal size at any one time is the result of average growth velocity since conception, it is also likely that disturbances in growth velocity immediately before that time may be of pathological significance. For instance, at a given gestation a fetus may be of above average size because of a high average velocity, but growth retardation may have been occurring for the previous few weeks. Some evidence to support growth velocity as a predictor of poor neonatal outcome was provided by the work of Chang and colleagues (1993). In this study, changes in the standard deviation scores of EFW and AC between the first and the last ultrasound examination were compared with the final values of the AC, EFW and Doppler indices. Poor neonatal outcome was defined in terms of morphometric indices of IUGR. It was found that serial assessment of EFW and AC was an overall better predictor of IUGR than the other parameters. One of the difficulties of this study was that the changes in the standard deviate scores were not expressed per unit time, and hence mathematically these are not accurate measures of velocity. Another issue is the validity of the morphometric indices; if the analysis is restricted to subscapular skinfold thickness - which is perhaps the most logical index of IUGR- there appear to be only minor differences in the areas under the ROC curves for all the parameters measured. One of the practical problems with using growth velocity is the additive effect of the ultrasound errors of two or more measurements; in order to compensate for this error, frequent serial measurements would have to be taken to observe significant trends. Another problem is the timing interval. The importance of this factor has been discussed

147

in correspondence by Gardosi (1994). The shorter the timing interval, the earlier the detection of anomalies but greater the relative error in estimating velocity; on the other hand, lengthening the interval will make timely intervention difficult. An additional difficulty is the lack of any practical published standard for growth velocity. It is probable that both fetal size and growth velocity need to evaluated in the optimal assessment of fetal well-being. The often-made statement that 'size does not matter as long as the baby is growing well' is based on very little empirical evidence, and may well be misleading in a clinical context.

Defining IUGR in the neonate. The ideal method to compare the performance of different tests on the same sample is by analysing receiver-operating characteristic curves (Zweig & Campbell, 1993) . This requires a clear division between affected and unaffected populations. In our sample this could not be made because of the lack of a 'gold standard' in defining either IUGR or macrosomia. Hence our finding of significantly lower rate of SGA in a normal population using the adjustable standard does not imply a better positive predictive value or detection rate for IUGR, since this could not be tested against a sample of 'truly' growth retarded fetuses. Nevertheless, indirect evidence that this may be the case comes from the work of Sanderson and colleagues (1994 ), who found that the individualised birth weight ratio is more closely related to morphometric indices of IUGR, including skinfold thickness. We have found a very high correlation (r = 0.99) between customised centiles and IBR centiles ; this is because the method of adjusting for maternal characteristics is the same, differences being due to the growth curve selected and in determining the standard deviation for a given gestational age.

148

Our choice of the 10th centile as the definition of SGA was arbitrary, based on conventional clinical practice, and we are certainly not advocating this as a definition of IUGR. Chard and colleagues (1993) have argued that fetal size for gestation at birth, as opposed to maturity, is irrelevant to outcome and have questioned the existence of fetal growth retardation at term. They believe that most small term infants are not at risk, and that a considerable number of babies thought to be small for dates are only so because of inaccurate gestational age estimation. This may well be so because of the imprecision in estimating weight -for-gestation rank resulting from error in gestational age assignment and the use of inappropriate, unadjusted birthweight standards (eg Lubchenko's standard applied to a sea-level population ). No epidemiological studies have yet been published on abnormal neonatal outcome or neurodevelopmental

disability

in

pregnancies

dated

by

ultrasonography. The study on neurodevelopmental handicap by Taylor and Howie (1989) showed that affected infants

were

significantly lighter at birth than controls, with lower birthweight centiles, but only when complications of pregnancy were present. In their sample only 16% of the population was pre-term, but at the time these children were born, gestational dating by ultrasound was still in its infancy, and hence this figure is suspect. Work in progress in our unit on ultrasound-dated populations has shown a clear, inverse relationship between birthweight centiles and neonatal morbidity, which is even more marked for customised birthweight centiles (Dr Muls, personal communications). Hence we believe that there is a role for estimating fetal size in antenatal risk assessment.

Defining normal fetal growth. Our work was essentially a longitudinal, observational study designed to investigate

fetal growth in the third trimester and to explore the

149

clinical viability of customised growth charts. For this purpose we recruited a 'low-risk' population in order to focus on normal fetal development. The choice of ultrasound weight-estimating formula was an important aspect of the project. We studied this problem (chapter 9) in a subset of the population that delivered near term. We had to make the assumption that the findings applicable to this group and the correction factors for the modified weight-estimation formulas would also be valid in the pre-term period, as we did not have a sufficiently large group of infants born below 37 weeks’ gestation. In choosing an appropriate formula, accuracy is a major consideration, but another important issue which is often disregarded is the correlation of the error with the size of the fetus. Such a correlation, if significant , will lead to data distortion at the extremes of the measured range. We found that the Hadlock formula for BPD, AC and FL was not only the most accurate, but also was free of any significant trend in the error. We decided to develop a longitudinally-derived growth standard , as opposed to a

cross-sectional one, because in clinical practice the

EFW's of high-risk fetuses are usually plotted serially. As pointed out by Altmann (1994), longitudinal data is most suitable for defining growth process. The variance would thus be somewhat smaller than that derived from a cross-sectional standard , depending on the degree of ultrasound error, but the median curve should remain unchanged. As a consequence, the 10th centile may be higher than if the standard was obtained by cross-sectional analysis. In clinical terms, this would lead to an increased test sensitivity but a reduction in specificity, which can be easily corrected by shifting the 'action threshold' downwards, say the 5th centile instead of 10th. Using the standard error of the multiple regression analysis in order to derive the standard deviation for the growth charts may be criticised in that it leads to reference ranges that are too narrow in view of the ultrasound error. We feel that it is not practical to widen the standard deviation to

150

accomodate ultrasound error in EFW because the latter varies considerably according to formula and the local conditions. The decision of which action threshold to use thus needs to be dealt with at a local level in view the performance in estimating fetal weight. Our finding of a virtually linear relationship between fetal weight and gestation using serial ultrasound data is at variance with older standards derived from samples whose gestational age is estimated from menstrual data. These are characterised by marked deceleration of growth at term. To a large extent our linear relationship is due to the improved accuracy in gestational age estimation using early BPD measurements; another factor is our population selection criteria which excludes neonates with abnormal outcome and likely anomalous growth patterns. It is unlikely to be an artefact due to the fetal weight estimation formula, since the error associated with the Hadlock formula we used was not correlated with fetal size and also because of the stabilising effect of including birthweights in the individual growth curves.

Preterm Delivery and Growth Retardation. In a substantial proportion of preterm deliveries the cause is unknown, or undetected. We were able to confirm previous reports (Ott,1993; Secher,1987; Persson,1989) of impaired fetal growth in infants born pre-term, using an ultrasound-derived fetal growth standard applied to both the EFWs and birth weights of infants born preterm. In clinical practice these cases are usually missed, either because of the use of cross -sectional, birthweight-derived reference standards or from the inaccuracy of ultrasound measurements. We also found that the distribution of our ultrasonic fetal weight estimates was positively skewed, but not to a statistically significant degree. This is in contrast to the significant negative skewness of the

151

preterm birthweight data (Wilcox, 1993) and corroborates the concept that a significant proportion of the infants born preterm may be growth retarded. In view of the poor detection rate of growth retardation using current methods, the practice of long-term tocolysis in pregnancies with threatened preterm labour should be viewed with caution. Preterm delivery may well be an escape mechanism for those babies whose nutritional needs are not being met by the utero-placental unit, and its pharmacological suppression could lead to deterioration in fetal condition.

Recommendations for further research. It will be difficult to improve the predictive value of any method for the detection of growth disturbances without first developing an accurate, quantitative standard for the diagnosis of IUGR in the newborn. The problems arising from the lack of a gold standard in the definition of deficient growth were discussed by Keirse (1984), and it is of interest that little progress has been made since then. Ideally this standard should give an indication of the energy stores of the neonate, since these are depleted in conditions of sub-acute or chronic hypoxia. This is a field of research which is being actively pursued in our department. An alternative to using indices of IUGR is to use actual clinical outcomes, such as acid-base status, admission to neonatal intensive care, etc..This allows the population to be divided two or more groups, according to neonatal outcome. The problem with this approach is that the pathological group will be heterogeneous, with only some of the morbidity being due to growth retardation. Hence the positive predictive value of the test cannot be clearly defined, unless it is restricted to clear-cut clinical conditions within a sufficiently large sample.

152

It may be possible that a single method of adjusting for maternal characteristics may not work optimally for all conditions, so that for instance a system that works optimally for IUGR may not be as effective in screening for macrosomia. Alternative

methods of

adjusting for maternal smoking need to be evaluated. In order to develop optimal customisation methods, large databases of cases with a variety of abnormal outcomes would be needed. The current program uses a single fractional growth curve for all maternal subgroups. Our work suggests that maternal subgroups may differ in the shape of their fractional curves, as is the case for nulliparous and multiparous populations. An issue to explore is whether better performance from the customised growth chart may be obtained by using more than one type of fractional growth curve. In other words, the customisation process could perhaps be optimal if it includes not only the prediction of birth weight at term, but also the likely shape of the growth curve. The distribution of our ultrasound fetal weight estimates in the normal sample showed a non-significant trend towards positive skewness at all gestations, which agrees with the observed significant positive skewness in term birthweights extracted from the East Midlands Obstetric Database. The customised growth chart does not incorporate this skewness, making the assumption of normality at all gestations. It would be a relatively simple procedure to reproduce this skewness in the computer program, and it would be interesting to compare the efficacy of this version of the program. The multiple regression coefficients allowing the prediction of birth weight at term were derived from a database that has been accumulating over many years. It is possible that, if the population characteristics change significantly with time, the coefficients may also change. The significance of this issue needs to be investigated by extracting the coefficients from the East Midlands Obstetric Database

153

for a series of time intervals. If this is the case, then the performance of the customised growth chart could improved by using 'up-dated' coefficients. Irrespective of whether adjusted or unadjusted charts are used, the error in fetal size estimation is a major limiting factor to the test performance. It is a possible explanation for the relative lack of success of screening programs using ultrasonic fetal growth parameters. Two approaches are possible in order to tackle this problem. Firstly, better formulas may be developed that, in addition to the AC, include other measures of soft tissue, such as limb circumferences (Balouet et al, 1992). Secondly, more advanced equipment is likely to lead to better results. Three-dimensional ultrasound machines, by improving the definition of tissue planes, should reduce the error in measuring ultrasound parameters, and will also allow more reproducible measuring of limb circumferences. The use of echo-planar magnetic resonance imaging (Baker et al, 1994) is a very promising technique for weight estimation, but will not be in common use until its prohibitive cost is reduced drastically. This should be case once low-cost high temperature super-conductors are available. Another significant source of error is the inter-observer variability, which for ultrasonic weight estimation by two observers ranges from -187g to 140g (Chang et al, 1993). Increasing the number of observers will result in greater variability. The ideal of having only one observer performing all the serial examinations for the same patient is probably unattainable in a busy obstetric ultrasound department, but should be within reach of community midwifery care, using symphysis-fundus height measurements. A community-based project is now underway in order to assess the value of serial SFH measurements performed by the same observer and plotted on the customised charts in terms of detecting growth disturbances. Cases that are screen-positive on the basis of abnormal serial or single SFH

154

measurements are referred for ultrasound examination and biophysical profiles including Doppler when indicated.

155

REFERENCES Alberman E. Are our babies becoming bigger? J Roy Soc Med (1991), 84:257-260. Alberman E, Emanuel I, Filakti H et al. The contrasting effects of parental birthweight and gestational age on the birthweight of offspring.Pediatric and Perinatal Epidemiology (1992), 6:134-144. Alexander G. Studies on the placenta of the sheep (Ovis aries L.).J Reprod Fertil (1964), 7:307-322. Allen MC, Donohue PK, Dusman AE. The limit of viability-neonatal outcome of infants born at 22 to 25 weeks' gestation. NEJM (1993), 329(22):1597-1601. Altman DG, Chitty LS. Charts of fetal size: 1. Methodology.Br J Obstet Gynaecol (1994), 101(1): 29-34. Altman DG, Coles EC. Assessing birth weight-for-dates on a continuous scale. Annals of Human Biology (1980),7(1):35-44. Altman DG, Coles EC. Nomograms for the precise determination of birth weight for dates. Br J Obstet Gynaecol (1980), 87(2):81-86. Bagger PV, Eriksen PS, Secher NJ, Thisted J, Westgaard L..The precision and accuracy of symphysis-fundus distance measurements during pregnancy. Acta Obstet Gynecol Scand (1985), 64:371-374. Bailey SM, Sarmandal P, Grant JM. A comparison of three methods of assessing interobserver variation applied to measurement of the symphysis-fundal height. Br J Obst Gynaecol (1989), 96:1266-1271. Baker PN, Johnson IR, Gowland PA et al. Fetal weight estimation by echo-planar magnetic resonance imaging. Lancet (1994), 343:644-645. Bakketeig LS, Jacobsen G, Brodtkorb CJ, Eriksen BC, Eik-Nes SH, Ulstein MK, Balstad P, Jorgensen NP.Randomised controlled trial of ultrasonographic screening in pregnancy. Lancet (1984), ii: 207-211. Balouet P, Speckel D, Herlicoviez M. Estimation echographique du poids foetal. Interet de la mesure de la graisse des membres. J Gynecol Obstet Biol Reprod (1992), 21: 795-802. Bantje H. A multiple regression analysis of variables influencing birth weight. Tropical and Geographical Medicine (1986), 38:123-130. Barth JA. Birthweight dependency on parental length and weight with special reference to gestational age.Arztliche Jugendkunde(1990), Band 81:3. Batra A, Chellani HK, Mahajan J et al.Ultrasonic variables in the diagnosis of intrauterine growth retardation.Indian J Med Res (1990), 92B:399-403. Beazley JM, Underhill R. Fallacy of the fundal height.BMJ (1970), 4:404-406. Belizan JM, Villar J, Nardin JC, Malamud J, Sainz de Vicuna L. Diagnosis of intrauterine growth retardation by a simple clinical method: measurement of uterine height. Am J Obstet Gynecol (1978), 131:643-646.

156

Benacerraf BR, Gelman R, Frigoletto FD. Sonographically estimated fetal weights:accuracy and limitations. Am J Obstet Gynecol (1988),159(5):1118-1121. Bergsjo P, Denman III DW, Hoffmann HJ et al. Duration of human singleton pregnancy. Acta Obstet Gynecol Scand (1990),69:197-207. Bhargava S.K., Sachdev HPS,Iyer PU et al. Current status of infant growth measurements in the perinatal period in India.Acta Paediatr Scand (1985), Suppl.319:103-110. Birnholz JC. An algorithmic approach to accurate ultrasonic fetal weight estimation. Investigative Radiology (1986), 21(7):571-576. Bower S, Campbell S. Doppler in obstetrics.In:Ultrasound in obstetrics and gynaecology. Eds. Dewbury K,Meire H,Cosgrove D. London: Churchill Livingstone (1993), pp251-270. Boyce A, Mayaux MJ, Schwartz D. Classical and true gestational postmaturity. Am J Obstet Gynecol (1976),125(7):911-914. Brandt, I. Growth dynamics of low-birth-weight infants.Acta Paediatr Scand (1985), Suppl.319:38-47. Brooke OG. Energy expenditure in the fetus and neonate: sources of variability. Acta Paed Scand (1985), Suppl 319:128-134. Brooke OG, Anderson HR, Bland JM. Birth weight,stress and behaviour; principal results from a prospective population study. BMJ (1989), 298:795-801. Butler NR, Bonham DG. Perinatal mortality: the first report of the 1958 British Perinatal Mortality Survey under the auspices of the National Birthday Trust Fund. E. and S. Livingstone, Edinburgh 1963. Calvert JP, Crean EE, Newcombe RG et al. Antenatal screening by measurement of symphysis-fundus height. BMJ (1982),285:846-849. Campbell S, Dewhurst CJ. Diagnosis of the small for dates fetus by ultrasound cephalometry. Lancet (1971),2:1002. Campbell S, Newman GB. Growth of the fetal biparietal diameter during normal pregnancy.J Obstet Gynaecol Brit C'wlth ( 1971),78: 513-519 Campbell S, Wilkin D.Ultrasonic measurement of fetal abdomen circumference in the estimation of fetal weight.Br J Obstet Gynaecol (1975), 82(9):689-697. Carlson DE. Maternal diseases associated with intrauterine growth retardation. Semin Perinatol (1988), 12:17-22. Cawley RH, McKeown T, Record RG. Parental stature and birthweight. Am J Human Genet (1954),6:448-456. Chamberlain P. Composite sonographic assessment of fetal health. Current Opinion Obstet Gynaecol (1992), 4:256-263. Chang TC. Is obstetric and neonatal outcome worse in fetuses who fail to reach their own growth potential? (Correspondence).

157

Br J Obstet Gynaecol (1993). Chang TC, Robson SC, Spencer JAD, Gallivan S. Fetal growth: A comparison between Bangladeshi and Caucasian women. Br J Obstet Gynaecol (1992), 99(12):1027-1028. Chang TC, Robson SC, Spencer JAD, Gallivan S. Ultrasonic fetal weight estimation: analysis of inter- and intra-observer variability. J Clin Ultrasound (1993), 21:515 -519. Chang TC, Robson SC, Spencer JAD, Gallivan S. Identification of fetal growth retardation: comparison of Doppler waveform indices and serial ultrasound measurements of abdominal circumference and fetal weight. Obstet Gynecol (1993), 82(2):230-236. Chard T, Costleloe K, Leaf A. Evidence of growth retardation in neonates of apparently normal weight. Eur J Obstet Gynaecol Reprod Biology (1992), 45:59-62. Chard T, Yoong A, Macintosh M. The myth of fetal growth retardation at term. Br J Obstet Gynaecol (1993),100:1076-1081. Chauhan SP, Lutton PM, Bailey KJ, Guerrieri JP, Morrison JC. Intrapartum clinical, sonographic and parous pateints' estimates of newborn birth weight. Obstet Gynecol (1992), 79(6):956-958. Chudleigh P, Pearce JM. Obstetric Ultrasound. London: Churchill Livingstone 1986. pp 51-58, 135-137. Chitty LS, Altman DG.Charts of fetal size.In: Ultrasound in Obstetrics and Gynaecology. Eds. Dewbury K, Meire H, Cosgrove D .London:Churchill Livingstone 1993; 513-595. Chitty LS, Hunt GH, Moore J, Lobb MO. Effectiveness of routine ultrasonography in detecting fetal structural abnormalities in a low risk population. BMJ(1991); 303:1165-9. Cheng MCE,Chew PCT,Ratnam SS. Birthweight distribution of Singapore Chinese, Malay and Indian infants from 34 weeks to 42 weeks gestation. J Obst Gyn Br Cmwth (1972), 79:149-153. Cnattingius S, Axelsson O, Lindmark G.. Symphysis-fundus measurements and intrauterine growth retardation. Acta Obstet Gynecol Scand (1984), 63:335-340. Combs CA, Jaekle RK, Rosenn B, Pope M, Miodovnik M, Siddiqi T. Sonographic estimation of fetal weight based on a model of fetal volume. Obstet Gynaecol (1993), 82(3):365-370. Colley NV, Tremble JM, Henson GL. Head circumference/abdominal circumference ratio, ponderal index and fetal malnutrition.Should head circumference/abdominal circumference ratio be abandoned? Br J Obstet Gynaecol (1991), 98:524-527. Danielian PJ, Allman ACJ, Steer PJ. Is obstetric and neonatal outcome worse in fetuses who fail to reach their own growth potential? (Correspondence).Authors' reply.

158

Daniellan PJ, Allman ACJ, Steer PJ. Is obstetric and neonatal outcome worse in fetuses who fail to reach their growth potential? Br J Obstet Gynaecol (1993), 99:452-454. Danielian PJ, Allman ACJ, Steer PJ. Is obstetric and neonatal outcome worse in fetuses who fail to reach their growth potential? (Letter) Br J Obstet Gynaecol (1994),100:87. Davies PSW, Lucas A. Quetelet's index as a measure of body fatness in young infants. Early Human Development (1989), 20:135-141. De Muylder X, De Somer ML, Bouckaert A. Early labour symphysis-fundus height as a predictor of birthweight. Ann Soc Belge Med Trop (1988), 68:61-65. Depares JC, Thornton JG, Clayden AD. Symphysis-fundus measurements in Asian and Caucasian women in Bradford. Eur J Obstet Gynecol Reprod Biol (1989),31:201-206. Deter RL, Harrist RB,Hadlock FP, Poindexter AN. Longitudinal studies of fetal growth with the use of dynamic image ultrasonography. Am J Obst Gyn (1982), 143(5):545-554. Deter RL, Harrist RB,Hill RM. Neonatal growth assessment score: A new approach to the detection of intrauterine growth retardation in the newborn. Am J Obstet Gynecol (1990), 162:1030-1036. Deter RL, Hill RB, Tennyson RN: Predicting the birth characteristics of normal fetuses 14 weeks before delivery. J Clin Ultrasound (1989),17:89-93. Deter RL, Rossavik IK, Carpenter RJ. Development of individual growth standards for estimated fetal weight: II.Weight prediction during the third trimester and at birth. J Clin Ultrasound (1989),17:83-88. Deter RL, Rossavik IK, Harrist RB, Hadlock FP. Mathematic modelling of fetal growth: Development of individual growth curve standards. Obstet Gynecol (1986), 68(2):156-161. Deter RL, Rossavik IK, Harrist RB. Development of individual growth curve standards for estimated fetal weight:I. Weight estimation procedure. J Clin Ultrasound (1988),16:215-225 Deter RL, Rossavik IK. A simplified method for determining individual growth curve standards. Obstet Gynecol (1987), 70(5):801-805. Dombrowski MP, Tomlinson MW, Bottoms SF, Sokol RJ. Computerised admission forms: a better idea. Abstracts: Fourth World Symposium 'Computers in Obstetrics, Neonatology and Gynaecology' Munich 1994. Doring GK. Uber die Tragzeit post ovulationem. Geburtshilfe Frauenheilkunde (1962),22:1191-1195.

159

Doubilet PM, Benson CB. Fetal growth disturbances. Seminars in Roentgenology (1990), 25(4):309-316. Dougherty CRS, Jones AD. The determinants of birth weight. Am J Obstet Gynaecol (1982), 144:190. Dubowitz LMS, Goldberg C. Assessment of gestation by ultrasound in various stages of pregnancy in infants differing in size and ethnic origin. Br J Obstet Gynaecol (1981),88:255-259. Dudley JN, Lamb MP, Hatfield JA et al. Estimated fetal weight in the detection of the small-for-menstrual-age fetus. J Clin Ultrasound (1990),18:387-393. Dunn PM. The creation of a perinatal growth chart for international reference. In :Fetal growth retardation: Diagnosis and treatment. Eds Beazley JM, Kurjak A. CRC Press, Boca Raton, Florida,1989. Elejalde BR, Giraldo C, Jimenez R, Gilbert EF. Acrocephalopolydactylous dysplasia. Birth Defects, Orig Art Series (1977),13(3B):53-67. Evans T, Farrant P, Gowland M et al. Report of the Fetal Measurements Working Party of the BMUS (1990). Ewigman BG, Crane JP, Frigoletto FD, LeFevre ML, Bain RP, McNellis D. Effect of prenatal ultrasound screening on perinatal outcome. NEJM (1993), 329 (12):821-827. Ewigman B, LeFevre M, Hesser J. A randomised trial of routine prenatal ultrasound. Obstetrics Gynecol (1990),76(2):189-194. Falkner F.Some introductory concepts of human growth: an overview. Acta Paediatr Scand (1985), Suppl. 319:17-20. Farmer RM, Medearis AL, Hirata GI, Platt LD. The use of a neural network for the ultrasonographic estimation of fetal weight in the macrosomic fetus. Am J Obstet Gynecol (1992), 166(5):1467-1472. Favre R, Nisand G, Bettahar K, Grange G, Nisand I.Measurement of limb circumferences with three-dimensional ultrasound for fetal weight estimation. Ultrasound Obstet Gynecol (1993), 3:176-179. Fay RA, Dey PL, Saadie CMJ et al. Ponderal index: A better definition of the 'at risk' group with intrauterine growth problem than birth-weight for gestational age in term infants. Aust NZ J Obstet Gynaecol (1991),31(1):17-19. Fedrick J, Adelstein P. Factors associated with low birth weight of infants delivered at term. Br J Obstet Gynaecol (1978), 85:1-7. Fescina RH, Martell M, Martinez G, Lastra L, Schwarcz R.Small for dates: evaluation of different diagnostic methods. Acta Obstet Gynecol Scand (1987),66:221-226.

160

Forbes JF, Smalls MJ. A comparative analysis of birthweight for gestational age standards. Br J Obstet Gynaecol (1983), 99: 297-303. Gairdner D, Pearson J. A growth chart for premature and other infants. Arch Dis Child (1971), 46:783-787. Gallivan S, Robson SC, Chang TC, Vaughan J, Spencer JAD.An investigation of fetal growth using serial ultrasound data. Ultrasound Obstet Gynecol (1993), 3:109-114. Gann P, Nghiem L, Warner S. Pregnancy characteristics and outcomes of Cambodian refugees. Am J Public Health (1989),79:1251-1257. Gardosi J. Is obstetric and neonatal outcome worse in fetuses who fail to reach their own growth potential? (Correspondence) Br J Obstet Gynaecol (1993), 100:295. Gardosi J. Is obstetric and neonatal outcome worse in fetuses who fail to reach their own growth potential? (Correspondence) Br J Obstet Gynaecol (1994), 100:295. Gardosi J, Chang A, Kalyan B et al. Customised antenatal growth charts. Lancet (1992), 339:283-287. Gardosi J, Mongelli M. Risk assessment adjusted for gestational age in maternal serum screening for Down's syndrome. BMJ (1993), 306:1509. Gardosi J, Mongelli M, Chang A. Screening and assessment of fetal growth. in : European Community Concerted Action Project: 'New methods for perinatal surveillance', Eds. HP Van Geijn & Copray. Elsevier 1994. Chapter 11,437-450. Gardosi J, Mul T, Mongelli M, Wilcox M. Ultrasound dating and birth weight at term. BMJ (1994), 308:1635. Gardosi J, Mul T, Mongelli M. Ultrasound dating and birth weight at term. BMJ (1994), 308:1635. Garin YJ, Blot P, Walter P, Pinon JM, Vernes A. Malarial infection of the placenta. Parasitologic, clinical and immunological aspects. Arch Fr Pediatr (1985), 42 (Suppl 2):917-920. Garn S. Pre-pregnancy weight. In: Maternal Nutrition and Pregnancy Outcomes. Eds: Krasovec K, Anderson MA. Pan American Health Organization, Scientific Publication No. 529 (1990), 69-85. Geirsson RT. Ultrasound instead of last menstrual period as the basis of gestational age assignment.Ultrasound Obstet Gynecol (1991),1:212-219. Geirsson RT, Persson P. Diagnosis of intrauterine growth retardation using ultrasound. Clinics in Obstetrics and Gynaecology (1984), 11(2):457-479.

161

Georgieff MK, Sasanow SR, Chockalingam UM et al. A comparison of the mid-arm circumference/head circumference for the evaluation of newborn infants after abnormal intrauterine growth. Acta Paediatr Scand (1988),77: 214-219. Gerhard I, Vollmar B, Runnebaum et al: Weight percentile at birth. II Prediction by endocrinological and sonographic measurements. Eur J Obstet Gynecol Reprod Biol (1987) , 26:313-328. Goldenberg RL, Cutter GR, Hoffman HJ et al. Intrauterine growth retardation: standards for diagnosis. Am J Obstet Gynecol (1989),161:271-277. Goldenberg RL, Davis RO, Cutter GR, Hoffman HJ, Brumfield CG, Foster JM. Prematurity, postdates and growth retardation: The influence of use of ultrasonography on reported gestational age. Am J Obstet Gynecol (1989), 160(2):462-470. Goldkrand JW, Fox E, Foggo BM. Coronal biparietal diameter. A reliable alternative to the traditional transverse biparietal diameter. J Ultrasound in Med (1990), 9(10): 555-557. Greenberger PA, Patterson R. Beclomethasone dipropionate for severe asthma during pregnancy. Ann Intern Med (1983), 98:478-480. Grover V, Usha R, Kalra S, et al. Altered fetal growth: antenatal diagnosis by symphysis-fundal height in India and comparison with western charts. Int J Gynecol Obstet (1991), 35:231-234. Gruenwald P. Growth of the human fetus. I. Normal growth and its variations. Am J Obstet Gynecol (1966), 94:1112-1119. Guidetti DA, Divon MY, Braverman JJ et al. Sonographic estimates of fetal weight in the intrauterine growth-retarded population. Am J Perinatol (1990), 7(1):5-7. Hadlock FP, Harrist RB, Carpenter RJ. Sonographic estimation of fetal weight. Radiology (1984),150:535-540. Hadlock FP. Harrist RB, Shah YP et al. Sonographic growth standards - are current data applicable to a racially mixed population? J Ultrasound in Medicine (1990), 9:157-160. Hadlock FP, Harrist RB, Martinez-Poyer J. In-utero analysis of fetal growth: a sonographic weight standard. Radiology (1991),181:129-133. Hadlock FP, Harrist RB, Sharman RS, Deter RL, Park SK.Estimation of fetal weight with the use of head, body and femur measurements - a prospective study. Am J Obstet Gynecol (1985), 151(3):333-337. Harvey D, Prince J, Bunton J, Parkinson C, Campbell S.Abilities of children who were small for gestational age babies. Pediatrics (1982),69: 296-300.

162

Hakala TH, Ylikorkala O. Effective prenatal care decreases the incidence of low birthweight. Am J Perinatol (1989),6(2): 222-225. Hedriana HL, Moore TR. A comparison of single versus multiple growth ultrasonographic examinations in predicting birth weight. Am J Obstet Gynecol (1994), 170(6):1600-1605. Henriksen TB, Wilcox A. Ultrasound dating subject to bias. BMJ (1994), 308:201. Henderson, M. Differences in the duration of pregnancy. Negro and white women of low socioeconomic class. Arch Environ Hlth (1967),14:904. Hendricks CH. Delivery patterns and reproductive efficiency among groups of differing socioeconomic status and ethnic origins. Am J Obstet Gynecol (1967), 97(5):608-624. Hepburn M ,Rosenberg K. An audit of the detection and management of small-forgestational age babies. Br J Obstet Gynaecol (1986), 93:212-216. Heymsfield SB, Wang J, Lichtman S et al. Body composition in elderly subjects: a critical appraisal of clinical methodology. Am J Clin Nutr (1989), 50:1167-1175. Hill LM, Breckle R, Wolfgram KR, et al. Evaluation of three methods for estimating fetal weight. J Clin Ultrasound (1986),14:171-178. Hill RM, Verniaud WM, Deter RL, Tennyson LM, Rettig GM, Zion TE, Vorderman AL, Helms PG, McCulley LB, Hill LL.The effect of intrauterine malnutrition on the term infant. Acta Paediatr Scand (1984), 73:482-487. Holmes GE, Miller HC, Hassanein K,Lansky SB, Goggin JE.Postnatal somatic growth in infants with atypical fetal growth patterns. Am J Dis Child (1977), 131:1078-1083. Hughes K, Tan NR, Lun KC. Low birthweight of live singletons in Singapore. Int J Epidemiol (1984), 13:465-471. Issel EP. Der einfluss des mathematischen modells und die anzahl der messtrecken auf die genauigkeit der fetalen gewichtsschatzung mittels ultraschall. Ultraschall in Med.(1991), 12:111-118. Jeanty P, Cantraine F, Romero R. A longitudinal study of fetal weight growth. J Ultrasound Med (1984), 3:321-328. Jensen OH, Larsen S. Evaluation of symphysis-fundus measurements and weighing during pregnancy. Acta Obstet Gynaecol Scand (1991),70:13-16. Johnson RW, Toshach C, Saginaw M. Estimation of fetal weight using longitudinal mensuration. Am J Obstet Gynecol (1954), 68(3):891-896.

163

Johnson JWC, Longmate JA, Frentzen B. Excessive maternal weight and pregnancy outcome. Am J Obstet Gynecol (1992),167:353-72. Jones RAK, Robertson NRC. Small for dates babies:are they really a problem? Arch Dis Child (1986), 61:877-880. Karlberg P,Niklasson A, Ericson A et al. A methodology for evaluating size at birth. Acta Paediatr Scand (1985), Suppl.319:26-37. Katz AI, Davison JM, Hayslett JP et al. Pregnancy in women with kidney disease. Kidney Int (1980),18:192-206. Keen DV, Pearse RG. Intrauterine growth curves:problems and limitations. Acta Paediatr Scand (1985), Suppl.319:52-54. Keirse M. Epidemiology and aetiology of the growth retarded baby. Clinics in Obstet & Gynaecol (1984), 11(2):415-433. Keirse M. Frequent prenatal ultrasound: time to think again. Lancet (1993), 342:878-9. Kessel E, Sastrawinata S, Mumford SD. Correlates of fetal growth and survival. Acta Paediatr Scand (1985), Suppl.319:120-127. Kirkup B, Welch G. 'Normal but dead':perinatal mortality in non-malformed babies of birth weight 2.5 Kg and over in the Northern Region in 1983. Br J Obstet Gynaecol (1990),97:381-392. Kuo H, Chang F, Wu C, Yao B, Liu C. The primary application of three-dimensional ultrasonography in obstetrics. Am J Obst Gynecol (1991),166:880-886. Kurtz AB, Wapner RJ, Kurtz RJ. Analysis of biparietal diameter as an accurate indicator of gestational age. J Clin Ultrasound (1980), 8:319-326. Kramer MS. Determinants of low birth weight: methodological assessment and metaanalysis. Bull WHO (1987), 65(5): 663-737. Kramer MS, Olivier M, McLean FH, Willis DM, Usher RH.Impact of intrauterine growth retardation and body proportionality on fetal and neonatal outcome. Pediatrics (1990),86:707-713. Kramer MS, McLean FH, Olivier M et al. Body proportionality and head and length 'sparing' in growth retarded neonates: a critical reappraisal. Pediatrics (1989), 4:717-723. Larsen T, Petersen S, Greisen G et al . Normal fetal growth evaluated by longitudinal ultrasound examinations. Early Human Development (1990), 24: 37-45. Lazar P, Dreyfus J, Papiernik-Berkhauer E. Individual correction of birth weight for parental stature with special reference to small-for-date and large-for-date infants.J Perinat Med (1975),3:242-247.

164

Lechtig A. Predicting risk of delivering low birthweight babies:which indicator is better? J Trop Ped (1988), 34:34-41. Leff M, Orleans M, Haverkamp AD, Baron AE, et al.The association of maternal low birthweight and infant low birthweight in a racially mixed population. Pediatric and Perinatal Epidemiology (1992), 6:51 -61. Lee JN, Chard T. Determination of biparietal diameter in the second trimester as a predictor of intrauterine growth retardation. Int J Gynaecol Obstet (1983),21:213-215. Lemoine P, Harousseau H, Borteyru J. Children of alcoholic parents. Observed anomalies (127 cases). Quest Med (1968), 21:476-482. Lessoway VA, Schulzer M, Wittmann BK. Sonographic measurement of the fetal femur: factors affecting accuracy. J Clin Ultrasound (1990),18:471-476. Lindgren R, Selbing A, Leander E. Which fetal growth charts should be used? Acta Obstet Gynecol Scand (1988),67:683-687. Lindhard A, Nielsen PV, Mouritsen LA, et al. The implications of introducing the symphyseal-fundal-height measurement . A prospective randomised controlled trial. Br J Obstet Gynaecol (1990), 97:675-680. Loeffler FE. Clinical foetal weight estimation. J Obstet Gynaecol Brit Cwlth (1967), 74:675-677. Long PA, Abell DA, Beischer NA. Fetal growth retardation and pre-eclampsia. Br J Obstet Gynaecol (1980), 87:13-18. Lubchenko LO, Hansman C, Dressler M et al. Intrauterine growth as estimated from liveborn birth-weight data at 24 to 42 weeks of gestation. Pediatrics (1963), 32:793-800. Lucas A, Morley R. Birthweight ratio. Arch Dis Child (1991), 66:1099 -1104. Lucas A, Cole J, Candy. Birthweight centiles in preterm infants reappraised. Early Human Development (1986), 13:313-322. Magnus P, Berg K, Bjerkedal T. Association of parental consanguinity with decreased birth weight and increased rate of early death and congenital malformation. Clinical Genetics (1985), 28:335-342. Mamelle N, Laumon B. Nouvelles limites de poids de naissance pour le diagnostic de retard de croissance intra-uterin. Rev Fr Gynecol Obstet (1989), 84(7-9):589-591. Mayhew T. Scaling placental oxygen diffusion to birthweight: studies on placentae from low- and high- altitude pregnancies. J Anat (1991), 175:187-194.

165

Mathai M, Jairaj P, Muthurathnam S. Screening for light-for-gestational-age infants: a comparison of three simple measurements. Br J Obstet Gynaecol (1987), 94:217-221.

Matsuda S, Sone T, Doi T et al. Seasonal variation of mean birth weight in Okinawa. Jap J Hygiene (1992),46(6):1063-1070. McDonald E. Mensuration of the child in the uterus with new methods. JAMA (1906),47:1979. McGregor JA, Leff M, Orleans M, Baron A. Fetal gender differences in preterm birth: findings in a North American cohort. Am J Perinatol (1992), 9(1):43-48. McKeown T, Gibson JR. Observations on all births (23970) in Birmingham. Br J Soc Med (1951), 5:98-112. McKeown T, Record RG. Influence of pre-natal environment on correlation between birth weight and parental height. Am J Human Genet (1954), 6:457-463. McLean FH, Boyd ME, Usher RH. Post-term infants: Too big or too small? Am J Obstet Gynecol (1991),164:619-624. Meadows NJ, Till J, Leaf A, Hughes E, Jani B, Larcher V.Screening for intrauterine growth retardation using ratio of mid-arm circumference to occipitofrontal circumference. BMJ (1986), 292:1039-1040. Meredith HV. Body weight at birth of viable human infants:a worldwide comparative treatise. Human Biology (1970),42:217-264. Merz E, Grussner A, Kern F. Mathematical modelling of fetal limb growth. J Clin Ultrasound (1989),17:179-185. Merz E, Lieser H, Schicketanz KH et al. Predicting fetal weight by ultrasound. Ultraschall (1988), 9:15-24. Miller HC, Hassanein K. Diagnosis of impaired fetal growth in newborn infants. Pediatrics (1971), 48(4):511-522 . Miller JM, Kissling GA, Brown HL, Gabert HA. Estimated fetal weight:applicability to small- and large-for-gestational -age fetus. J Clin Ultrasound (1988),16:95-97. Milner RDG, Richards B. An analysis of birthweight by gestational age of infants born in England and Wales 1967 to 1971. J Obstet Gynaecol Br Commw (1974), 81:956-967. Miller HC. Intrauterine infection. An unmet challenge. Am J Dis Child (1981), 135:944-948. Mongelli M, Gardosi J. Accuracy of determining 'post-term' pregnancy. Br J Obstet Gynaecol (1994),101(8):733-734.

166

Morley R, Brooke OG, Cole TJ, Powell R, Lucas A. Birth weight ratio and outcome in preterm infants. Archives of Diseases in Childhood (1990), 65:30-34. Morton NE. Empirical risks in consanguineous marriages: birth weight, gestation time, and measurement of infants. Am J Hum Genet (1958), 10:344-349. Morrison J, Williams GM, Najman JM, Andersen MJ. The influence of paternal height and weight on birthweight. Aust NZ J Obstet Gynaecol (1991), 31(2):12. Murao, F. Measurements of the fetal liver size, hormonal level and pregnancy outcome. Gynaecol Obstet Invest (1991), 32:153-156. Naeye RL. Cytomegalic inclusion disease: the fetal disorder. Am J Clin Pathol (1967), 47:738-744. Naeye RL. Malnutrition. Probable cause of fetal growth retardation. Arch Pathol (1965), 79:284-291. Naeye RL, Blanc W, Leblanc W, Khatamee MA. Fetal complications of maternal heroin addiction: abnormal growth, infections, and episodes of stress. J Pediatr (1973), 83:1055-61. Naeye RL, Tafari N. Biologic bases for international fetal growth curves. Acta Paediat Scand (1985), Suppl.319:164-169. Neilson JP. Routine serial symphysis-fundal height measurement. In: Pregnancy and Chilbirth Module (eds. Enkin MW, Keirse MJNC, Renfrew MJ, Neilson JP), ‘Cochrane Database of Systematic Reviews’: Review No. 06889, 24 March 1993. Published through ‘Cochrane Updates on Disk’, Oxford: Update Software, 1993, Disk Issue 2. Neilson JP, Munjanja SP, Whitfield CR. Screening for small for dates fetuses: a controlled trial. BMJ (1984), 289: 1179-1182. Newnham JP, Evans SF, Michael CA, Stanley FJ, Landau LI.Effects of frequent ultrasound during pregnancy: a randomised controlled trial. Lancet (1993), 342:887-891. Niklasson A, Ericson A, Fryer JG et al. An update of the Swedish reference standards for weight, length and head circumference at birth for given gestational age. Acta Paediatr Scand (1991),80:756-762. Nishida H, Sakamoto S, Sakanque M. New fetal growth curves for Japanese. Acta Paediatr Scand (1985), Suppl 319:62-67. Oakley JR, Parsons RJ, Whitelaw AGL. Standards for skinfold thickness in British newborn infants. Arch Dis Child (1977), 52:287-290. Ogunranti JO. Fundal height in normal pregnant Nigerian women: anthropometric gravidogram.

167

Int J Gynecol Obstet (1990),33:299-305. Okonofua FE, Ayangade SO, Ajibulu OA. Ultrasound measurement of fetal abdominal circumference and the ratio of biparietal diameter to transverse abdominal diameter in a mixed Nigerian population. Int J Obst Gyn (1988), 27:1-6. Osinusi BO. Ultrasound femur length as a means of assessing gestational age among Nigerians. West African Journal of Medicine (1990),9(2):116-119. Osinusi BO, Okubanjo DA. Ultrasonic fetal head circumference as a means of assessing gestational age in Nigerians. West African J Med (1990), 9(1):22-25. Ott, WJ. Accurate gestation dating. Obstet Gynecol (1985), 66(3): 311-315. Ott WJ. The diagnosis of altered fetal growth. Obstetrics and Gynecology Clinics of North America (1988),15(2): 237-263. Ott WJ, Doyle S, Flamm S, Wittman J. Accurate ultrasonic estimation of fetal weight.Prospective analysis of new ultrasonic formulas. Am J Perinatol (1986), 3(4):307-310. Ott, W. Intrauterine growth retardation and preterm delivery. Am J Obstet Gynecol (1993),168 (6):1: 1710-1717. Ouellette EM, Rosett HL, Rosman NP, Weiner L. Adverse effects on offspring of maternal alcohol abuse during pregnancy. N Eng J Med (1977), 297:528-530. Ounsted M, Ounsted C. Maternal regulation of intrauterine growth. Nature (1966), 3: 995-997. Ounsted MK. Maternal constraint of fetal growth in man. Dev Med Child Neurol (1965), 7:479-491. Ounsted MK. Unconstrained fetal growth in man. Dev Med Child Neurol (1966), 8:3-8.

Palmer J, Dillon-Baker C, Tecklin JS et al. Pregnancy in patients with cystic fibrosis. Ann Intern Med (1983), 99:596-600. Palo P, Erkkola R, Piiroinen O, Ruotsalainen P. Accuracy of ultrasonic fetal weight estimation and detection of small for gestational age fetuses. Am J Perinatol (1989), 6(4): 400-404. Parker AJ, Davies P, Mayho AM, Newton JR. The ultrasound estimation of sexrelated variations of intrauterine growth. Am J Obstet Gynecol (1984), 149(6):665-669. Pattinson RC, Norman K, Odendaal HJ. The role of Doppler velocimetry in the management of high risk pregnancies. Br J Obstet Gynaecol (1994),101(2):114-120.

168

Patterson RM, Prihoda TJ, Gibbs CE, Wood RC. Analysis of birthweight percentile as a predictor of perinatal outcome. Obstet Gynecol (1986), 68(4):459-463. Patterson RM, Pouliot MR. Neonatal morphometrics and perinatal outcome: who is growth retarded? Obstet Gynecol (1986), 68(4):691-693 Pearce JM, Campbell S. A comparison of symphysis-fundal height and ultrasound as screening tests for light-for-gestational age infants. Br J Obstet Gynaecol (1987), 94:100-104 Pearce JM, Campbell S.Intra-uterine growth retardation. In: The management of labour. John Studd, editor. Blackwell Scientific Publications, 1985. pp 68-91. Pearce JM, Campbell S. Ultrasonic monitoring of normal and abnormal fetal growth. In: Modern management of high-risk pregnancy. Laursen NH, editor.Plenum Medical Book Co. 1983. pp 57-100. Persson PH. Fetal growth curves. In: Fetal Growth. Sharp F, Milner RDG, Fraser RB, editors. RCOG Publications 1989. pp 13-25. Persson PH, Grennert L, Gennser G. Impact of fetal and maternal factors on the normal growth of the biparietal diameter. Acta Obstet Gynecol Scand (1978), Suppl 78:21-27. Persson B, Stangenberg M,Lunell NO, Brodin U, Holmberg NG, Vaclavinkova V. Prediction of size of infants at birth by measurement of the symphysis-fundus height. Br J Obstet Gynaecol (1986), 93:206-211 Persson PH ,Weldner BM. Reliability of ultrasound fetometry in estimating gestational age in the second trimester. Acta Obstet Gynecol Scand (1986),65:481-483. Persson PH, Weldner BM. Intrauterine weight curves obtained by ultrasound. Acta Obstet Gynecol Scand (1986), 65:169-173 Petersen S, Gotfredsen A, Knudsen FU. Lean body mass in small for gestational age and appropriate for gestational age infants. J Pediatrics (1988), 113 (5):886-889. Piwoz EG, Peerson JM, Brown KH. Potential for misclassification of infants' growth increments by using existing reference data. Am J Clin Nutr (1992), 56:58-64. Polani PE. Chromosomal and other genetic influences on birth weight variation. In: Size at Birth, Ciba Foundation Symposium No.27 . Elsevier, Amsterdam, 1974,12759. Powars D, Sandhu M, Niland-Weiss J, Johnson C, Bruce S, Manning PR. Pregnancy in sickle cell disease. Obstet Gynecol (1986), 67:217-228. Pschera H, Soderberg G. Estimation of fetal weight by external abdominal measurements. Acta Obstet Gynecol Scand (1984),63:175-179.

169

Quaranta P, Currell R. Prediction of small-for-dates infants by measurement of symphysial-fundal-height. Br J Obstet Gynaecol (1981),88:115-119. Ransome-Kuti, O. Intrauterine growth, birthweights and maturity of the African newborn. Acta Pediatr Scand (1985), Suppl.319:95-102. Rantakallio P. Fourteen year follow-up of children with normal and abnormal birth weight for their gestational age. Acta Paediatr Scand (1985),74: 62-69. Robinson HP, Fleming JEE. A critical evaluation of sonar 'crown-rump length' measurements. Br J Obstet Gynaecol 1975; 82:702-10. Robinson JS, Falconer J, Owens JA. Intrauterine growth retardation: clinical and experimental. Acta Paediatr Scand (1985), Suppl 319:135-142. Robson SC, Gallivan S, Walkinshaw SA, Vaughan J, Rodeck CH. Ultrasonic estimation of fetal weight: use of targeted formulas in small for gestational age fetuses. Obstet Gynecol (1993), 82(3):359-363. Rochelson B, Bracero LA, Porte J, Farmakides G.Diagnosis of IUGR as a two-step process with morphometric ultrasound and Doppler umbilical artery velocimetry. J Reprod Med (1992), 37(11):925-929. Roemer VM, Kieback DG, Buhler K. Gestationszeit und geburtsgewicht.3.Mitteilung: die plazenta und mutterliche kofaktoren. Z Geburtsh u Perinat (1991),195:195-208. Roemer VM, Kieback DG, Buhler K, Kohling U. Gestationsalter und geburtsgewicht.3.Mitteilung:ponderal-index, fetaler saure-basen-haushalt und hypotrophie. Z Geburtsh u Perinat (1991),195:239-249. Rogers MS, Chung TKH, Chang AMZ. Ultrasound fetal weight estimation: precision or guess-work? Aust NZ J Obstet Gynaecol (1993), 33(2):142-144. Rooth G, Meirik O, Karlberg P. Estimation of the 'normal' growth of Swedish infants at term. Acta Paediatr Scand (1985), Suppl 319:76-79. Rosenberg K, Grant JM, Tweedie I, Aitchison T. Measurement of fundal height as a screening test for fetal growth retardation. Br J Obstet Gynaecol (1982),89:447-450. Rosenberg SN, Verzo B,Engstrom JL et al. Reliability of length measurements for preterm infants. Neonatal Network (1992),11(2):23-27. Rossavik IK, Deter RL. Mathematical modelling of fetal growth:I.Basic principles. J Clin Ultrasound (1984),12:529-533.

170

Rossavik IK, Deter RL, Hadlock FP. Mathematical modelling of fetal growth.IV. Evaluation of trunk growth using the abdominal profile area. J Clin Ultrasound (1987),15:31-35. Rumbolz WL, McGoogan LS. Placental insufficiency and the small undernourished full term infant. Obstet Gynaecol (1953), 1(3):294-301. Sabbagha RE, Barton FB, Barton BA. Sonar biparietal diameter.I. Analysis of percentile growth differences in two normal populations using the same methodology. Am J Obstet Gynecol (1976), 126(4): 479-484. Sabbagha RE, Barton BA, Barton FB et al. Sonar biparietal diameter.II. Predictive of three fetal growth patterns leading to a closer assessment of gestational age and neonatal weight. Am J Obstet Gynecol (1976), 126(4): 485-490 Sabbagha RE, Hughey M, Dopp R. The assignment of growth-adjusted sonographic age (GASA): a simplified method. Obstet Gynecol (1978), 51:383-386. Sanderson DA, Wilcox MA, Johnson IR. The individualized birth weight ratio: a new method of identifying intrauterine growth retardation. Br J Obstet Gynaecol (1994),101:310-314. Sanderson DA, Wilcox MA, Johnson IR. Relative macrosomia identified by the individualized birth weight ratio. Acta Obstet Gynecol Scand (1994), 73:246-249. Sandmire HF. Whither ultrasonic prediction of fetal macrosomia? Obstet Gynecol (1993) 82:5, 860- 862. Sarmandal P, Bailey SM, Grant JM. A comparison of three methods of assessing inter-observer variation applied to ultrasonic fetal measurement in the third trimester. Br J Obstet Gynaecol (1989), 96:1261-1265. Sarmandal P, Grant JM. Effectiveness of ultrasound determination of fetal abdominal circumference and fetal ponderal index in the diagnosis of asymmetrical growth retardation. Br J Obstet Gynaecol (1990), 97:118-123. Sasanow SR, Georgieff MK, Pereira GR. Mid-arm circumference and mid-arm/head circumference ratios: Standard curves for anthropometric assessment of neonatal nutritional status. J Pediatrics (1986),109(2):311-315. Secher NJ, Hansen PK. A randomized study of fetal abdominal circumference and fetal weight estimation for detection of LGA infants in low risk pregnancies. Br J Obstet Gynaecol (1987), 94:105-109. Secher NJ, Hansen PK, Thomsen BL, Keiding N. Growth retardation in preterm infants. Br J Obstet Gynaecol (1987), 94:115-120. Secher NJ, Lundbye-Christensen S,Qvist I. An evaluation of clinical estimation of fetal weight and symphysis fundal distance for the detection of SGA infants.

171

Eur J Obstet Gynecol Reprod Biol (1990), 38:91-96. Seidman DS, Ever-Hadani P, Gale R. The effect of maternal weight gain in pregnancy on birth weight. Obstet Gynaecol (1989), 74(2): 240-246. Shalev E, Romano S, Weiner E et al.Predictive value of the femur length to abdominal circumference ratio in the diagnosis of intrauterine growth retardation. Isr J Med Sci (1991),27:131-133. Shami SA, Qadeer T, Schmitt LH et al. Consanguinity, gestational period and anthropometric measurements at birth in Pakistan. Ann Hum Biology (1991),18 (6):523-527. Shepard MJ, Richards VA, Berkowitz RL, et al. An evaluation of two equations for predicting fetal weight by ultrasound. Am J Obstet Gynecol (1982), 142:47-54. Shime J, Mocarski EJM, Hastings D, Webb GD, McLaughlin PR. Congenital heart disease in pregnancy: short and long-term implications. Am J Obstet Gynecol (1987), 156:313-322. Shinozuka N, Okai T, Kohzuma S et al. Formulas for fetal weight estimation by ultrasound measurements based on neonatal specific gravities and volumes. Am J Obstet Gynecol (1987), 157:1140-1145. Simmons K, Savage W, Nicholls B, Rao U. Fetal growth measured by ultrasound in Bengali women. J Obstet Gynaecol (1985), 5:233-236. Simon NV, Deter RL, Shearer DM. Prediction of normal fetal growth by the Rossavik growth model using two scans before 27 weeks, menstrual age. J Clin Ultrasound (1989),17:237-243. Simon NV, Levisky JS, Shearer DM et al. Influence of fetal growth patterns on sonographic estimation of fetal weight. J Clin Ultrasound (1987),15:376-383. Simon NV, O'Connor TJ, Shearer DM. Detection of intrauterine fetal growth retardation with abdominal circumference and estimated fetal weight using crosssectional growth curves. J Clin Ultrasound (1990),18: 685-690. Snow MHL. Effect of genome on size at birth. In: Fetal Growth. Eds. Sharp F, Fraser RB & Milner RDG. RCOG (1989). pp 3-12. Speert H. Essays in eponymy. New York: Macmillan, 1958. Spiegelberg O. Lehrbuch der geburtshulfe fur arzte und studierende. 3:126, aufl. neu bearbeitet von Max Wiener,Lahr, 1891; M. Schauenburg, 1891 . Spinnato JA, Allen RD, Mendenhall HW. Birth weight prediction from remote ultrasonographic examination. Am J Obstet Gynecol (1989), 161(3): 742-747. Spinnato JA, Allen RD, Mendenhall HW. Birth weight prediction from remote ultrasound examination.

172

Obstet Gynecol (1988), 71(3): 893-898. Stefos T, Deter RL. Individual growth curve standards for fetal head and abdominal circumferences: effect of the type of measurement on growth prediction. J Clin Ultrasound (1989),17:33-36. Stefos T, Deter RL, Simon NV. Effect of timing of initial scan and interval between scans on Rossavik growth model specification. J Clin Ultrasound 17:319-325. Stein Z, Susser M. The Dutch famine, 1944-1945, and the reproductive process. I. Effects on six indices at birth. Pediatr Res (1975), 9:70. Stewart HL, Duration of pregnancy and postmaturity. JAMA (1952), 148(13):1079-1083. Taylor DJ, Howie PW. Fetal growth achievement and neurodevelopmental disability. Br J Obstet Gynaecol (1989), 96:789-794. Thaler I, Bronstein M, Rubin AE. The course of pregnancy associated with bronchiectasis. Br J Obstet Gynaecol (1986), 93:1006-1008. Thompson HO, Casaceli C, Woods JR. Ultrasonic fetal weight estimation by an integrated computer-assisted system: Can each laboratory improve its accuracy? Am J Obstet Gynecol (1990), 163(3):986-994. Thomson AM, Billewicz WZ, Hytten FE. The assessment of fetal growth. J Obstet Gynaec Brit Cwlth (1968), 75:903-916. Ulizzi L,Terrenato L. Natural selection associated with birth weight.VI. Towards the end of the stabilizing component. Ann Hum Genet (1992),56:113-118. Van der Spuy ZM, Steer PJ, McCusker M, Steele SJ, Jacobs HS. Outcome of pregnancy in underweight women after spontaneous and induced ovulation. BMJ (1983), 296:962-964. Vialet R, Mbaye K, de Mouzon J, Spira A. Comparaison par echographie de la croissance foetale des enfants de meres africaines et europeennes. J Gynecol Obstet Biol Reprod (1988), 17:1003-1010. Voigt M, Eggers H, Jahrig K, Grauel .Beziehungen zwischen Alter, Paritat, Korpergewicht und -lange der Mutter und dem geburtsgewicht der neugeborenen. Zentralblatt fur Gynakologie (1989), 11(6):337-349. Waldenstrom U, Nilsson S, Fall O, Axelsson O, Eklund G, Lindberg S, Sjodin Y. Effects of routine one-stage ultrasound screening in pregnancy: a randomised controlled trial. Lancet (1988), ii: 585-588. Wallin A, Gyllensward A, Westin B. Symphysis-fundus measurements in prediction of fetal growth disturbances. Acta Obstet Gynaecol Scand (1981), 60:317-323.

173

Walter PR, Garin Y, Blot P. Placental pathologic changes in malaria. A histologic and ultrastructural study. Am J Pathol (1982), 109(3):330-340. Walton A, Hammond S Maternal effects on growth and conformation in Shire horse-Shetland pony crosses. Proceedings of the Royal Society (1938), 125B:311. Warburton D, Naylor AF. The effect of parity on placental weight and birthweight: An immunological phenomenon? A report of the collaborative study of cerebral palsy. Am J Hum Genetics (1971), 23:41-54. Warsof SL, Gohari P, Berkowitz RL, et al. The estimation of fetal weight by computer-assisted analysis. Am J Obstet Gynecol (1977),128:881-92. Weiner CP, Robinson D. Sonographic diagnosis of intrauterine growth retardation using the postnatal ponderal index and the crown-heel length as standards of diagnosis. Am J Perinatol(1989), 6(4):380-383. Westin B. Gravidogram and fetal growth. Comparison with biochemical supervision. Acta Obstet Gynecol Scand (1977), 56:273-282. Wharton BA. Sorrento studies of birthweight. Acta Paediatr Scand (1985), Suppl 319:170-179. Whittle MJ et al. Discussion. in :Fetal Growth. Sharp F, Fraser RB and Milner RDG, Editors.RCOG 1989. pp196-202. Wilcox AJ. Birth weight, gestation, and the fetal growth curve. Am J Obstet Gynecol (1981), 139 (8):863-867. Wilcox M, Gardosi J, Mongelli M, Ray C, Johnson I. Birth weight from pregnancies dated by ultrasonography in a multicultural British population. BMJ (1993), 307:588-91. Wilcox M, IR Johnson, Maynard P. Individualized birth weight ratio: a more logical outcome measure than birthweight alone. Br J Obstet Gynaecol (1993), 100:342-347. Williams RL, Creasy RK, Cunningham GC, et al. Fetal growth and perinatal viability in California. Obstet Gynecol (1982), 59(5):624 - 632. Winick M. Cellular changes during placental and fetal growth. Am J Obstet Gynecol (1971), 109:166-176. Witter FR, Luke B. The effect of maternal height on birth weight and length. Early Human Development (1991), 25 :181 - 186. Wolfe HM, Gross TL, Sokol RJ. Recurrent small for gestational age at birth: perinatal risks and outcomes. Am J Obstet Gynecol (1987), 157:288-293. Wolfe HM, Brans YW, Gross TL. Correlation of commonly used measures of intrauterine growth with estimated neonatal body fat.

174

Biol Neonate (1990), 57:167-171. Yip R. Altitude and birth weight. J Pediatrics (1987), 111(6):1:869-876. Yudkin PL, Aboualfa M, Eyre JA, Redman CWG,Wilkinson AR. Influence of elective preterm delivery on birthweight and head circumference standards. Arch Dis Child(1987), 62:24-29. Yudkin PL, Harlap S. High birthweight in an ethnic group of low socio-economic status. Br J Obstet Gynaecol (1983), 90:291-296. Yudkin PL, Aboualfa M, Eyre JA et al. Influence of elective preterm delivery on birthweight and head circumference standards.Archives of Disease in Childhood (1987), 62:24-29. Yudkin PL, Aboualfa M, Eyre JA , Redman CWG, Wilkinson AR. New birthweight and head circumference centiles for gestational ages 24 to 42 weeks.Early Human Development (1987), 15:45-52. Zador IE, Sokol RJ, Chik L. Interobserver variability: a source of error in obstetric ultrasound. J Ultrasound Med (1988), 7:245 - 249. Zweig MH, Campbell G. Receiver-operator characteristic plots: A fundamental evaluation tool in clinical medicine. Clin Chem (1993), 39(4): 561-577.

175