Ultrasound-derived fetal size nomogram for a sub-Saharan African population: a longitudinal study




To create a fetal size nomogram for use in sub-Saharan Africa and compare the derived centiles with reference intervals from developed countries.


Fetal biometric measurements were obtained at entry to antenatal care (11–22 weeks' gestation) and thereafter at 4-week intervals from pregnant women enrolled in a longitudinal ultrasound study in Kinshasa, Democratic Republic of Congo. The study population comprised 144 singleton gestations with ultrasound-derived gestational age within 14 days of the menstrual estimate. A total of 755 monthly ultrasound scans were included with a mean ± SD of 5 ± 1 (range, 2–8) scans per woman. Estimated fetal weight (EFW) was calculated at each ultrasound examination using the Hadlock algorithm. A general mixed-effects linear regression model that incorporated random effects for both the intercept and slope was fitted to log-transformed EFW to account for both mean growth and within-fetus variability in growth. Reference centiles (5th, 10th, 50th, 90th and 95th centiles) were derived from this model.


Nomograms derived from developed populations consistently overestimated the 50th centile EFW value for Congolese fetuses by roughly 5–12%. Differences observed in the 10th and 90th centiles were inconsistent between nomograms, but generally followed a pattern of overestimation that decreased with advancing gestational age.


In low-resource settings, endemic malaria and maternal nutritional factors, including low prepregnancy weight and pregnancy weight gain, probably lead to lower fetal weight and utilization of nomograms derived from developed populations is not appropriate. This customized nomogram could provide more applicable reference intervals for diagnosis of intrauterine growth restriction in sub-Saharan African populations. Copyright © 2009 ISUOG. Published by John Wiley & Sons, Ltd.


Ultrasound assessment of intrauterine growth is often used as a clinical tool to identify aberrant growth and to evaluate fetal response to maternal interventions. Fetal size nomograms are used to assess the estimated fetal weight (EFW) in pregnancies of known gestational age against a reference standard at a certain point in gestation. Conventionally, fetal weight estimates that fall below the 10th centile of the nomogram are suggestive of intrauterine growth restriction (IUGR)1. This definition is highly dependent on the origin of the reference population. Most currently available nomograms were derived from developed, primarily Caucasian, populations. Maternal race/ethnicity influence fetal growth2–4, as do maternal and environmental factors5. For example, maternal human immunodeficiency virus (HIV) and malaria infection, chronic undernutrition, and micronutrient deficiency are often endemic in low-resource populations and are associated with lower birth weight6, 7.

Although minimal data exist regarding in-utero fetal growth patterns in sub-Saharan Africa, mean birth weights are, on average, 400–600 g lower than in developed countries and rates of small-for-gestational age at birth are twofold to threefold higher7, 8. In-utero growth patterns would also be expected to vary among these populations, and fetal weight nomograms created from developed countries may not serve as appropriate benchmarks for identifying growth-restricted fetuses in low-resource populations. If a nomogram identifies an inappropriately large proportion of fetuses as IUGR, the clinical value of this tool in distinguishing fetuses that are truly growth compromised, and would benefit from maternal interventions such as nutritional supplementation or malaria treatment, is vastly reduced.

The purpose of this study was to develop a nomogram of fetal size for use in low-resource settings in sub-Saharan Africa. The derived reference intervals are compared with commonly used nomograms from developed countries to assess the applicability of such nomograms to low-resource populations.


Study population

The study population consisted of 182 women enrolled in a prospective longitudinal cohort study designed to understand the effects of maternal malaria and nutritional status on fetal growth. The study was conducted between May 2005 and May 2006 at Binza Maternity Hospital in Kinshasa, Democratic Republic of Congo. This hospital-based clinic serves a predominately urban low-income population. To be enrolled in the study, participants had to have a singleton pregnancy of less than 22 weeks' gestation with no ultrasound-detected fetal abnormality and no evidence of pre-existing hypertension (systolic ≥ 140 mmHg or diastolic ≥ 90 mmHg). Enrolled women returned for monthly follow-up visits during which maternal malaria status, maternal anthropometry and fetal biometry were assessed. The protocol called for intensive malaria surveillance and treatment. All women received two courses of presumptive malaria therapy with sulfadoxine–pyrimethamine between 16–27 weeks' and 28–32 weeks' gestation, regardless of malaria status. In addition, women were treated each time they had a positive malaria smear. At delivery, birth outcome, newborn anthropometry and delivery complications were recorded. All participants provided written informed consent to participate and ethical approval of the study protocol was obtained from the Institutional Review Boards of the University of North Carolina at Chapel Hill, North Carolina, USA and the University of Kinshasa's School of Public Health, Kinshasa, Democratic Republic of Congo. Further details of the study are described elsewhere9.

We excluded three stillbirths and 35 fetuses with uncertain dates (last menstrual period date differed from the ultrasound-derived date by greater than ± 14 days) leaving 144 fetuses for this analysis. Women with obstetric complications were not excluded in order to obtain a representative population.

Ultrasound measurements

All ultrasound examinations were performed using a GE LOGIQ Book (GE Healthcare, Chalfont St. Giles, Buckinghamshire, UK) ultrasound system by a single ultrasonographer (V.L.). Biparietal diameter (BPD), head circumference (HC), abdominal circumference (AC) and femur length (FL) were measured to estimate gestational age and fetal weight. For fetuses in the first trimester, the crown–rump length was used to estimate gestational age. Using standard techniques10, HC and BPD were measured from an image that displayed the fetal head in an axial plane that included the thalamus and cavum septi pellucidi. BPD was measured by placing the calipers from leading edge to leading edge (outer to inner skull) and HC using an ellipse trace of the outline of the fetal head. AC was measured from an image in which the junction of the umbilical vein and portal sinus was visible. The ellipse function was used to trace the extreme perimeter of the fetal abdomen. FL was measured along the long axis of the femur from outer to outer margin, including the femoral diaphysis and excluding the epiphyses.

These four biometric parameters were used to estimate gestational age (GA) at the first ultrasound (GA = 10.85 + 0.060 × HC × FL + 0.6700 × BPD + 0.1680 × AC)11. Ultrasound dating was chosen as it provides a better prediction of dates than does last menstrual period when used early in pregnancy12, 13 and because 15% of enrolled women could not provide a ‘certain’ last menstrual period date. Fetal weight in grams was calculated at all ultrasound examinations using the Hadlock algorithm: log(EFW) = 1.3596 − 0.00386 × AC × FL + 0.0064 × HC + 0.00061 × BPD × AC + 0.0424 × AC + 0.174 × FL14. This EFW algorithm was selected to facilitate comparability with previously published nomograms and because use of multiple biometric parameters provides more accurate fetal weight estimation14, 15. The signed percentage difference between the mean EFW obtained from the Hadlock algorithm and the mean actual birthweights at weeks 38, 39 and 40 was calculated to assess the accuracy of the Hadlock algorithm in predicting true fetal weight.

A 10% sample of archived ultrasound images was assessed for quality by a maternal–fetal medicine physician at the University of North Carolina at Chapel Hill, USA (A.H.). Some 92% of reviewed images were deemed adequate for clinical assessment, 7% of questionable quality and 1% of poor quality (not all biometry landmarks clearly visible, shadowing in the image or poor tracing of the length or circumference). Intraoperator variability in measuring fetal biometry was assessed in 10 patients. The Pearson's correlation (r) between two independent measurements on the same fetus was r = 0.99 for each of BPD, HC, and FL, and r = 0.98 for AC.

Statistical analysis

All analyses were performed in SAS version 9.1 (SAS Institute, Cary, NC, USA). Owing to sparse data in the first trimester and post-term, analyses were limited to 15–40 weeks' gestation. Gestational age (in days) was calculated according to the ultrasound-derived dates and centered at 105 days (15 weeks). The 50th and outer reference centiles (2.5th, 5th, 10th, 25th, 75th, 90th, 95th and 97.5th) were derived using a linear mixed-effect model approach16. This method accounts for variability in EFW at both the between-subject and within-subject levels, by incorporating subject-specific effects for the intercept and growth (slope) component. Briefly, EFW (Yij) was log-transformed to ensure normality and reduce heteroscedasticity of the residual of EFW. Gestational age (the time variable, Tij) was a best fitting second-degree fractional polynomial linearizing function17. Let Zij denote log(EFW) and Xij represent the fractional polynomial transformation of time. Then the mean (µij) and variance (equation image) of Zij at transformed time Xij are:

equation image

where equation image represents the (between-fetus) variance of the random intercepts, equation image represents the (between-fetus) variance of the random slopes, σmath image is the covariance between them, and equation image is the estimated within-fetus variance.

The reference intervals for the untransformed EFW (Yi), are calculated from the mean and variance above as exp(µij ± ϕσij) where σij is the standard error of Zij and ϕ is the standard normal deviation of the distribution function (±1.96 for the 2.5th and 97.5th centiles, ±1.645 for the 5th and 95th centiles, ±1.282 for the 10th and 90th centiles, ±0.674 for the 25th and 75th centiles, and 0 for the 50th centile). Further details are provided in Appendix S1 online.

The raw and studentized residual errors were visually inspected by plots of the errors against gestational age. Normality in the distribution of the errors was determined by visual inspection of plots and subsequently confirmed based on the Shapiro–Wilks test. Influence diagnostics, based on iteratively deleting each subject from the model, were used to identify fetuses that overinfluenced the estimates of the fixed effects and/or precision of the variance estimates on overall model fit. The influence diagnostics identified two fetuses as potential outliers in fetal weight. However, when these observations were removed during sensitivity analyses, no appreciable change in either the fixed- or random-effects estimates was observed, and so they were ultimately included in the model.

We compared our estimated reference intervals to those derived from three developed world populations18–20. These studies were selected because they applied criteria for certain gestational dating and used similar algorithms for estimating EFW as used in the present study. These studies represent both cross-sectional and longitudinal study designs, and applied a range of different statistical methods to calculate the outer reference ranges. For each gestational week, we calculated the percentage difference in EFW between the developed population reference value and the Congo nomogram as: [(EFWReferenceEFWCongo/EFWCongo) × 100]. Thus, differences greater than zero represent a higher EFW value in the developed population reference compared with the value from the Congo nomogram (overestimation), whereas differences below zero denote underestimation.


The Congo reference intervals were based on 144 singleton fetuses that underwent 755 ultrasound examinations. The mean ± SD number of scans per fetus was 5 ± 1 (range, 2–8) and the mean ± SD duration between scans was 29 ± 4 (range, 16–51) days. Maternal and neonatal characteristics of the study population are provided in Table 1. The mean ± SD maternal age at enrollment was 27 ± 6 (range, 18–43) years and 29% of women were primigravid. During pregnancy, 39% (n = 56) of women never had malaria, 38% (n = 54) were malaria positive only once, 18% (n = 26) had two positive smears and 6% (n = 8) had three or more positive smears. At baseline, 16% of women had either a body mass index <19.8 kg/m2 or a mid upper arm circumference <23 cm, which is indicative of underweight status in pregnant women21. Five women (3%) were HIV positive. There were 11 Cesarean births (8%) and 11 women (8%) with delivery complications (four premature rupture of the membrane, seven breech and/or prolonged or obstructed labor, four postpartum hemorrhages). Six infants (4%) were delivered preterm (before 37 weeks' gestation) and there were three early neonatal deaths (2%).

Table 1. Maternal and fetal characteristics of the study population (Kinshasa, Democratic Republic of Congo, 2005–2006; n = 144)
CharacteristicMean ± SD (range) or %
  • *

    Data available for 142 infants.

  • Data available for 137 infants. BMI, body mass index; HIV, human immunodeficiency virus; MUAC, mid upper arm circumference.

 Age at enrollment (years)27 ± 6 (18–43)
 BMI at enrollment (kg/m2)23.7 ± 3.7 (16.0–38.5)
 Height (cm)161.4 ± 7.4 (120.0–177.2)
 MUAC at enrollment (cm)26.5 ± 3.2 (20.4–38.5)
 Weight gain per month (kg)1.6 ± 1.5 (–2.3 to 6.6)
 Hematocrit at enrollment (%)33.5 ± 3.9 (18.0–47.5)
 Malaria parasitemia at26
 HIV positive3
 Birth weight (g)*3041 ± 413 (1500–4270)
 Birth length (cm)49.7 ± 1.9 (42.0–53.5)
 Birth head circumference (cm)34.1 ± 2.4 (22.0–38.0)
 Gestational age at birth (days)*275 ± 11 (226–305)
 Low birth weight (<2500 g)6
 Preterm birth (<37 weeks)4
 Female gender56

The Hadlock EFW algorithm provided a reasonable estimate of actual fetal weight for our population, with percentage differences between the predicted 50th centile EFW value and mean actual birth weight of 3.9%, 7.7% and 7.4% at 38, 39 and 40 weeks, respectively.

The distribution of ultrasound examinations and descriptive statistics of the EFW variable by gestational age are shown in Table 2. The best fitting fractional polynomial of gestational age was a quadratic polynomial defined as equation image. The fitted regression equations for the mean and variance of log transformed EFW were:

equation image
Table 2. Distribution of ultrasound examinations and descriptive statistics of estimated fetal weight according to gestational age (Kinshasa, Democratic Republic of Congo, 2005–2006; n = 755)
Gestational age (weeks)Number of observationsEstimated fetal weight (g)
Mean ± SDMinimumMaximumCV (%)
  1. CV, coefficient of variation (SD/mean), expressed as a percentage.

152128 ± 81221346.2
1613140 ± 121241558.3
1711166 ± 131521877.5
1818217 ± 191762518.6
1925271 ± 192312957.1
2029319 ± 202713536.3
2135382 ± 373204709.5
2249464 ± 383905608.1
2322535 ± 394726437.2
2430617 ± 525207428.5
2540745 ± 696259079.3
2638874 ± 8070010629.2
2732988 ± 8585611988.6
28361119 ± 10294413419.1
29331314 ± 125107615289.5
30401480 ± 126127517618.5
31291661 ± 2111351246612.7
32361813 ± 1811402220310.0
33402073 ± 2401615252411.6
34322226 ± 2341814292910.5
35302536 ± 2592034324510.2
36382666 ± 3081639322711.6
37342935 ± 2932301362210.0
38343152 ± 3242466397510.3
39213360 ± 3572875452910.6
4083296 ± 255287535497.7

Using these formulae, the estimated mean (50th centile) and outer 2.5th, 5th, 10th, 25th, 75th95th and 97.5th centiles were derived at each gestational age from 15 to 40 weeks (Table 3). Figure 1 shows these estimated centiles superimposed on a scatter plot of the raw EFW values. As expected, 11% of the raw EFW values fell below the 10th centile, 79% between the 10th and 90th outer centiles, and 10% above the 90th centile. Growth was continuously linear through term, with variance increasing with advancing gestational age. Raw and studentized residuals of EFW were obtained from the mixed models and plotted against gestational age (Figure 2). The residuals were symmetrically dispersed at zero, suggesting that the logarithmic transformation of EFW was adequate to meet the assumption of homoscedastic variance of the residual errors.

Figure 1.

The 5th, 10th, 50th, 90th and 95th estimated fetal weight centiles for gestational ages 15–40 weeks (Kinshasa, Democratic Republic of Congo, 2005–2006). Circles represent the raw values.

Figure 2.

Studentized residuals across gestational age from the fit of the regression model.

Table 3. In-utero fetal weight centiles according to week of gestation (Kinshasa, Democratic Republic of Congo, 2005–2006)
Gestational age (weeks)Estimated fetal weight centiles (g)

Comparison with nomograms from developed countries

Table 4 shows key study design features and sample size information for the three developed country nomograms used for comparison. Figure 3 shows the percentage difference in EFW when comparing the derived reference intervals with those of three previously published nomograms from developed populations. For the 50th centile, all three developed nomograms overestimated the Congolese values by roughly 5–12%, and the difference tended to be larger at earlier gestational ages. The 10th centile of Hadlock et al.18 only slightly underestimated the Congo 10th centile at early gestational ages, whereas the 90th centile consistently overestimated fetal weight early in gestation; both differences became less pronounced near term. Both the 10th and 90th centiles derived by Gallivan et al.19 consistently overestimated the corresponding Congo centiles. The nomogram by Johnsen et al.20 also consistently overestimated the inner and outer centiles; however, for both, the overestimation gradually decreased with advancing gestation.

Figure 3.

Percentage difference in 10th (a), 50th (b) and 90th (c) centiles of estimated fetal weight between the Kinshasa nomogram and Hadlock (equation image), Gallivan (equation image) and Johnsen (equation image) nomograms. Lines above zero indicate overestimation by the developed nomogram compared with the Kinshasa nomogram, whereas lines below zero indicate underestimation.

Table 4. Comparison of estimated fetal weight reference intervals from our study population in Kinshasa (Democratic Republic of Congo) and from previously published developed-population nomograms
ParameterPresent studyHadlock et al.18Gallivan et al.19Johnsen et al.20
  1. GA, gestational age.

PopulationDemocratic RepublicUSA, CaucasianUK, CaucasianNorway, 98%
 of Congo, African  European
Study designLongitudinalCross-sectionalLongitudinalLongitudinal
Number of women14439267635
Total number of scans7553924341795
Average number of scans per woman5163
Estimated fetal weight algorithmHadlock14Hadlock14Hadlock14Combs43
Estimated fetal weight (g)
 GA 20 weeks 
  10th centile288275289283
  50th centile315331344340
  90th centile345387410408
 GA 25 weeks 
  10th centile672652709717
  50th centile743785816835
  90th centile822918940972
 GA 30 weeks 
  10th centile1307129414211403
  50th centile1470155916141619
  90th centile1654182418341868
 GA 35 weeks 
  10th centile2137215423622242
  50th centile2442259526632593
  90th centile2790303630032998
 GA 40 weeks 
  10th centile2945300432403021
  50th centile3403361936333511
  90th centile3934423440744081

To place these differences in the context of clinical practice, between gestational ages 15 and 40 weeks, 15% of fetuses in this population had at least one ultrasound-derived EFW value below the 10th centile of the Congo nomogram. By comparison, 12%, 35% and 36% of fetuses would have been characterized as IUGR using the 10th centiles values of the Hadlock, Gallivan, and Johnsen nomograms, respectively. This would yield a false-positive rate (1 − specificity) of 0%, 23% and 24% for the three nomograms, when using the Congo-derived nomogram as the ‘gold standard’.


Using advanced longitudinal data analysis methods, we developed a fetal size nomogram using ultrasound-estimated fetal weight data from a resource-poor urban antenatal population in sub-Saharan Africa. As a clinical reference tool for detecting pathologically small fetuses, nomograms should be derived from unselected pregnancies that represent the normal pregnancy experience of the target population. In the context of sub-Saharan Africa, excluding women with poor nutritional status or antenatal infections from the study population would create an ‘artificially healthy’ pregnancy cohort, and produce mean and centile values that overestimate the true source population fetal weight. The prevalence of undernutrition and malaria parasitemia observed in this Kinshasa population is similar to that of other sub-Saharan African antenatal populations22, 23. However, the antenatal HIV prevalence in Kinshasa is among the lowest in the region24 and all enrolled women received monthly antenatal care including aggressive antimalarial treatment as part of the parent study. Thus, it should be noted that this population may represent a healthier pregnancy cohort than the general antenatal care population in Kinshasa or other urban low-resource settings, as evidenced by a slightly lower prevalence of low birth weight (<2500 g)25, 26.

Despite the high level of antenatal care received by this study population, we still found that the mean EFW (50th centile) in this Congo population was, on average, consistently lower than the 50th centile of nomograms derived from developed populations. Ultrasound studies conducted in Africa in the late 1980s showed that the 50th centiles of single biometric parameters, including BPD and AC, were also lower than those from European-derived nomograms27–30. We hypothesize that much of this difference is due to maternal health factors, including nutritional deficiencies and infections such as malaria and HIV. However, variations in study methodology, such as choice of algorithm used to estimate fetal weight and/or gestational age, may also have contributed to the observed differences in the 50th centiles31. For example, the use of ultrasound-derived gestational age, as opposed to last menstrual period-derived age, often shifts the mean gestational age of the population to the left (earlier) by approximately 2 weeks, resulting in lower mean EFW32, 33. Furthermore, the decision to classify gestational age in days or rounded weeks, rather than completed weeks34 avoids the ‘averaging’ of fetal weights within 6-day intervals which can shift the mean EFW for a given week.

Although the 50th centile of a nomogram provides an indication of the average fetal size achieved, it is values at the extreme (<5th, <10th, > 90th and > 95th centiles) of a nomogram that are most informative for clinical risk identification. As variability in fetal growth over time is one of the chief components in the determination of outer centiles, it is imperative that statistical techniques accurately model both the ‘average’ growth across fetuses and the biological variability within fetuses across the spectrum of gestational age. Previous attempts to develop fetal size nomograms from longitudinal ultrasound studies used methods applicable for cross-sectional data that ignore the high correlation among biometric parameters over time (gestational age) and assume that the fetal growth velocity, and the estimated model residual errors, are constant over time35–37. These assumptions are not appropriate for fetal growth data, which generally demonstrate a pattern of increased variability in EFW with advancing gestational age. Several studies, including the Gallivan study19 used as a comparison here, attempted to improve the statistical methods for analyzing longitudinal data by fitting a separate regression curve to each fetus and using the average variation (i.e. average of the individual regression coefficients) among these curves to derive the size centiles38, 39. This method, however, still leads to outer centiles that are too narrow as they account only for between-fetus variation. The mixed-effects model approach used in our study overcomes the above limitations by considering both the between- and within-fetus variation in the calculation of the reference intervals.

Our findings suggest that nomograms derived from developed populations consistently overestimate the 50th EFW centile for fetuses in low-resource settings, suggesting that a locally derived nomogram may be more appropriate for monitoring fetal growth. Findings with regard to the outer centiles were varied, with the Hadlock nomogram providing the closest approximation to the 10th and 90th centiles used to detect pathological growth restriction or macrosomia.

In low-resource settings, it is not typically feasible to implement the technologically advanced interventions that are routinely practised in Western societies to manage ‘high-risk’ pregnancies. However, maternal interventions, such as presumptive antimalarial treatment, bednet programs, nutrition supplementation and hypertension management, can be delivered effectively in sub-Saharan Africa and have been shown to improve outcomes at birth and during the early neonatal periods40–42. Owing to limited resources, even these interventions are often reserved for women at highest risk of poor birth outcome. Use of a single ultrasound scan, along with a locally relevant and well constructed fetal size nomogram, could prove a useful clinical tool in low-resource settings to identify these high-risk women. However, the potential of this nomogram to differentiate such high-risk women will not be fully realized until the lower-centile cut-off point that most accurately identifies fetuses which actually go on to have a poor birth outcome, such as low birth weight or early neonatal mortality, has been defined. Future research to identify this cut-off point is therefore an important next step in the implementation of fetal size assessment into obstetric practice and IUGR identification in low-resource settings.


The following supporting information may be found in the online version of this article:

Appendix S1 Formulae and details of calculation of the reference intervals for estimated fetal weight.


This study formed part of the doctoral dissertation of Sarah H. Landis, submitted to the University of North Carolina at Chapel Hill, USA. Dr Landis is indebted to the NEWAID Foundation, the Philanthropic Educational Organization (PEO) Foundation and the Graduate School of the University of North Carolina at Chapel Hill for predoctoral funding related to this project.

We acknowledge Odile Muniaka, Crispin Fela and Hélène Matondo for their hard work and dedication in carrying out the clinical aspects of the study; David Nanlele for administrative assistance; Paluku Kitsa for laboratory assistance; and Dr Luyeye Godefroid and the staff of Binza Maternity Hospital for assistance with recruitment and delivery procedures.