Sonographic birth-weight prediction in obese patients using the gestation-adjusted prediction method

Authors

  • L. L. Thornburg,

    Corresponding author
    1. Division of Maternal Fetal Medicine, Department of Obstetrics and Gynecology, University of Rochester Strong Memorial Hospital, Rochester, NY, USA
    • 601 Elmwood Avenue, Department of Ob/Gyn, Box 668, Rochester, NY 16420, USA
    Search for more papers by this author
  • C. Barnes,

    1. Division of Maternal Fetal Medicine, Department of Obstetrics and Gynecology, University of Rochester Strong Memorial Hospital, Rochester, NY, USA
    Search for more papers by this author
  • J. C. Glantz,

    1. Division of Maternal Fetal Medicine, Department of Obstetrics and Gynecology, University of Rochester Strong Memorial Hospital, Rochester, NY, USA
    Search for more papers by this author
  • E. K. Pressman

    1. Division of Maternal Fetal Medicine, Department of Obstetrics and Gynecology, University of Rochester Strong Memorial Hospital, Rochester, NY, USA
    Search for more papers by this author

Abstract

Objectives

Ultrasound birth-weight prediction may be more accurate if assessed at 34 to 36 + 6 weeks' gestation and extrapolated using the gestation-adjusted projection (GAP) method than if done at term. Because ultrasound is less accurate in women with elevated body mass index (BMI), we assessed the accuracy of GAP birth-weight prediction in obese as compared to non-obese women.

Methods

We performed a retrospective review of 1382 women with singleton pregnancies who had undergone fetal ultrasound examination at between 34 + 0 and 36 + 6 weeks, subclassified by pre-pregnancy BMI. Analysis of variance was used to compare predicted and actual birth weight.

Results

1025 controls and 357 obese women were included. The obese women were divided by BMI: 159 in Class I (BMI, 30–34.9 kg/m2); 105 in Class II (BMI, 35–40 kg/m2) and 93 in Class III (BMI > 40 kg/m2). Mean systematic (percent) birth-weight prediction error was within 4% for all groups, with a 95% error range between − 5% and + 5%. The GAP method was able to predict actual birth weight within 20% for all groups in over 90% of cases. For all groups, the GAP method correctly excluded the presence of macrosomia with ≥ 90% specificity. Negative likelihood ratios for the prediction of macrosomia were between 0.4 and 0.6 for all groups, regardless of obesity.

Conclusions

The GAP method of birth-weight prediction using ultrasound measurement at 34 to 36 + 6 weeks predicts birth weight within 20% error in over 90% of cases, and is able to exclude macrosomia with over 90% accuracy regardless of maternal BMI. Copyright © 2008 ISUOG. Published by John Wiley & Sons, Ltd.

Introduction

Obesity has reached epidemic proportions in the USA, with 66% of Americans obese or overweight as of 20021, and almost 5% morbidly obese, with a body mass index (BMI) over 40 kg/m2. Obesity is a risk factor for almost all obstetric complications as well as fetal macrosomia, or birth weight over 4000 g, which is associated with both neonatal and maternal morbidity2. Accurate prediction of birth weight can be especially difficult in obese women, since increased maternal BMI obscures visualization of fetal anatomy by ultrasound, regardless of gestational age3–6. Using the gestation-adjusted projection (GAP) method, birth weight is predicted by extrapolation of Brenner's median fetal weights from sonographic measurements remote from term7. There are many methods of birth-weight prediction, but this method has been shown to be accurate8. It may be better than term estimations of birth weight secondary to improved visualization at earlier gestation, since at term there is significant deterioration of resolution as the fluid to fetus ratio decreases, bony structures become increasingly calcified, and the vertex descends in the pelvis, making measurements of head circumference and biparietal diameter more difficult5. The GAP method has been validated in women with gestational and pregestational diabetes as an accurate prediction method for birth-weight estimation9. What is not known is whether the accuracy of this birth-weight prediction method extends to obese parturients, especially those with morbid obesity in whom visualization may be severely limited. In this study we sought to determine the accuracy of the GAP method of birth-weight prediction in obese parturients.

Patients & Methods

We retrospectively identified all women with singleton pregnancies who underwent ultrasound assessment with fetal biometric measurements at between 34 + 0 and 36 + 6 weeks' gestation at the University of Rochester from May 1994 to July 2000. All indications for ultrasound were included; patients were included in the study if outcome data of birth weight, gestational age at delivery, and BMI data were available. This study was exempt from IRB approval as a retrospective study by the Research Subjects' Review Board at the University of Rochester. The patients were divided into a total of four study groups based on BMI as determined by self-reported pre-pregnancy height and weight. The control group comprised women with a BMI of < 30.0. The study groups were divided by class of obesity into: Class I, BMI 30–34.9 kg/m2; Class II, BMI 35–40 kg/m2; and Class III, BMI > 40 kg/m2.

All ultrasound examinations were performed by experienced sonographers using standard techniques10. The studies were completed on one of the following ultrasound units: ATL 5000, ATL 3500, ATL 3000, GE 400, AI 5200, HDI Ultramark 9 or an ATL Ultramark 4 (ATL, Phillips Medical Systems, Best, the Netherlands; GE, GE Medical Systems, Fairfield, CT; HDI, Phillips Medical Systems, Best, the Netherlands; AI, Dornier Medical Systems, Phoenix, AZ, USA). Measurements used for calculation included the biparietal diameter (BPD), abdominal circumference (AC) and femur length (FL). If more than one measurement was obtained during an ultrasound examination, the measurements deemed best by the sonographer were chosen. Estimated fetal weight was calculated using the formula of Hadlock et al.11: log10weight = 1.335 − 0.0034AC × FL + 0.0316BPD + 0.0457AC + 0.1623FL.

Data were collected on fetal biometry, estimated fetal weight, gestational age at the time of ultrasound, gestational age at delivery, birth weight, maternal height and pre-pregnancy weight (as self-reported either at the time of ultrasound examination or at delivery), and latency from time of the ultrasound scan to time of delivery. Analysis of variance and Chi-square as appropriate were used to compare these data. Gestational age was based on available early ultrasound or menstrual dating. Sonographic dating was chosen when there were disagreements between menstrual and ultrasound dating of more than 8% for a given gestational age12. The birth weight for each patient was predicted from the results of the ultrasound scan using the GAP method and Brenner's median fetal weights for gestational age7, 13. In brief, the GAP method works by using the ratio between the estimated fetal weight and Brenner's median fetal weight for the gestational age at which the sonogram was performed. This ratio is then multiplied by the Brenner's median birth weight at the gestational age of delivery to give the predicted birth weight. Therefore, the estimated birth weight is calculated using the following formula: birth weight at delivery = median fetal weight at gestational age at delivery × (estimated fetal weight by ultrasound/median fetal weight at gestational age of ultrasound). Because Brenner's median weights were originally recorded in whole weeks, a curve was fitted to the original data and the medians were interpolated to estimate the median birth weight at each day of gestation8, 13.

Birth-weight errors were determined by calculating the difference between the predicted birth weight and the actual birth weight. This difference, as a percent of actual birth weight, was used to determine percent birth-weight error. Means of error and percent error were compared with analysis of variance (ANOVA), with post-hoc Bonferroni testing to assess the differences between groups. Single-sample t-testing was used to compare percent errors in all groups to zero, and Levene's test was used to compare homogeneity of variance between groups. Linear regression analysis was used to assess the relationship between BMI and predicted birth weight and BMI and actual birth weight for each of the four groups. The constants and slopes of the resultant equations were compared using T distributions (slope or constant difference/SE of slopes or constants). In addition, the ability of the GAP method to correctly predict birth weight to within 10, 15 and 20% of actual birth weight was assessed, as well as the ability of the GAP method to correctly identify those patients with macrosomic fetuses weighing more than 4000 g. SPSS 13 for Mac was used for statistical analysis.

Results

A total of 357 obese women and 1025 controls were included, with the distribution of obese patients as follows: 159 in Class I (BMI, 30–34.9 kg/m2); 105 in Class II (BMI, 35–40 kg/m2) and 93 in Class III (BMI > 40 kg/m2). Of those with Class III obesity, BMI ranged from 40 to 58 kg/m2, with 16 (17%) patients having a BMI > 50 kg/m2. The control and study groups did not differ with respect to age, median parity, gestational age at the time of the ultrasound scan, gestational age at delivery, or latency until delivery. A higher percentage of the women in the control group were nulliparous and Caucasian. The percentage of diabetic patients did not increase as maternal BMI increased. Birth weight, however, as well as percentage of patients with macrosomia (birth weight > 4000 g), did increase with increasing maternal BMI (Table 1).

Table 1. Comparison of characteristics between obese and non-obese patients
CharacteristicControls BMI < 30 kg/m2 (n = 1025)Obese womenP
Class I BMI 30–34.9 kg/m2 (n = 159)Class II BMI 35–40 kg/m2 (n = 105)Class III BMI > 40 kg/m2 (n = 93)
  • Data are presented as mean ± SD, median (range) or n (%).

  • *

    Analysis of variance.

  • Chi-square. BMI, body mass index; GA, gestational age; NS, not significant.

BMI (kg/m2)22.7 ± 3.432.4 ± 1.437.5 ± 1.545.7 ± 4.5< 0.0001*
Age (years)26.7 ± 6.927.5 ± 6.327.3 ± 5.527.5 ± 5.0NS*
Parity1 (0–8)1 (0–6)1 (0–9)1 (0–7)NS
Nulliparous431 (42)55 (35)33 (31)20 (22)0.0002
GA at sonogram (weeks)35.5 ± 0.935.5 ± 0.835.5 ± 0.835.5 ± 0.9NS*
GA at delivery (weeks)38.9 ± 1.638.9 ± 1.838.9 ± 1.739.0 ± 1.7NS*
Latency until delivery (days)23.7 ± 12.324.0 ± 12.323.6 ± 12.524.5 ± 12.8NS*
Birth weight (g)3202 ± 5063436 ± 6643454 ± 6133610 ± 700< 0.0001*
Macrosomic71 (6.9)29 (18)15 (14)27 (29)< 0.0001
Ethnicity
 White/Hispanic628 (61)92 (58)51 (49)47 (51)0.02
 Black365 (36)64 (40)54 (51)45 (48) 
 Asian23 (2)1 (1)00 
 Other9 (1)2 (1)01 (1) 
Smoker298 (29)39 (25)28 (27)27 (29)NS
Diabetic197 (19)21 (13)13 (12)19 (20)NS

Mean birth-weight error varied significantly between obese patients and normal weight controls (P < 0.0001), with birth weight underestimated in those patients with Class III obesity as compared to controls and those patients with Class I and Class II obesity (Table 2). Post-hoc Bonferroni testing indicated that these differences were between those patients with Class III obesity compared to all other groups, and that control, Class I and Class II obese patients did not differ from each other. Mean percent birth-weight error also significantly differed between all four groups (P < 0.001), with post-hoc testing indicating that this difference was between controls and patients with Class III obesity. Student's t-test indicated that the mean percent errors differed significantly from zero (P < 0.05) for all groups except Class III, although the mean percent error was within ± 5% for all the groups. The standard deviations of percent errors were between 9.6 and 11.9 for all the groups, with Levene's test for homogeneity of variance indicating significant differences between groups (P < 0.05). Overall, these results indicate that the GAP birth-weight prediction method performed equally well in controls and in patients with Class I and Class II obesity, but less well in those with a BMI over 40 kg/m2 (Class III obesity).

Table 2. Birth weight (BW) prediction errors
Error in BW predictionControls BMI < 30 kg/m2 (n = 1025)Obese women
Class I BMI 30–34.9 kg/m2 (n = 159)Class II BMI 35–40 kg/m2 (n = 105)Class III BMI > 40 kg/m2 (n = 93)
  • *

    Estimated 95% CI expressing the expected level of error in birth-weight prediction.

  • 95% CI for the estimation of the mean error. BMI, body mass index; SEM, standard error of the mean.

BW error (g)
 Mean ± SD (95% CI)*115.2 ± 30675.5 ± 33979.6 ± 364− 48.1 ± 442
 (−485 to 715)(−589 to 740)(−634 to 793)(−914 to 818)
 Mean ± SEM (95% CI)115.2 ± 9.475.5 ± 26.979.6 ± 35.5− 48.1 ± 45.8
 (96.5–133.9)(22.8–128.2)(10.0–149.2)(−137.9 to 41.7)
BW error (%)
 Mean ± SD (95% CI)*4.2 ± 9.63.2 ± 103.1 ± 10.6− 1.9 ± 11.9
 (−14.6 to 23.0)(−16.4 to 22.8)(−17.7 to 23.9)(−25.2 to 21.4)
 Mean ± SEM (95% CI)4.2 ± 0.33.2 ± 0.83.1 ± 1.0− 1.9 ± 1.2
 (3.6–4.8)(1.6–4.8)(1.1–5.1)(−4.3 to 0.5)

Linear regression equations relating BMI to birth weight were compared between the four groups. The slopes were very similar in all groups, although the constants differed. The slopes ranged between 0.83 and 0.96, while the constants ranged between 74 and 451. The biggest differences were between the control and Class III obese patients, again indicating that the predicted birth-weight error is greatest in the most obese patients.

The GAP method was able to correctly predict birth weight within 20% of the actual birth weight in over 90% of cases for all the groups. Additionally, the GAP method was able to correctly predict birth weight within 15% in over 80% of cases within all groups, and within 10% in over 59% of cases for all groups (Table 3). The ability to predict birth weight within each percentile did not differ between the groups.

Table 3. Percentage of correct birth weight predictions
Birth-weight errorControls BMI < 30 kg/m2 (n = 1025)Obese womenP*
Class I BMI 30–34.9 kg/m2 (n = 159)Class II BMI 35–40 kg/m2 (n = 105)Class III BMI > 40 kg/m2 (n = 93)
  • All data given as n (%).

  • *

    Chi-square.

± 20%966 (94)151 (95)98 (93)87 (94)0.94
± 15%876 (85)134 (84)90 (86)76 (82)0.74
± 10%692 (68)108 (68)74 (70)55 (59)0.34

Finally, for the ability of the GAP method to correctly predict macrosomia, sensitivity ranged from 48 to 60% and positive predictive value from 45 to 70% for all groups (Table 4). GAP method specificity ranged from 90 to 95% and negative predictive value from 81 to 96% for all the groups, indicating that the GAP method had only moderate ability in predicting macrosomia but good accuracy in excluding macrosomia. Positive likelihood ratios varied from 10.7 to 5.3 between the controls and those patients with Class III obesity, but negative likelihood ratios were equal for all the groups.

Table 4. Prediction of macrosomia by gestation-adjusted projection method
Prediction of macrosomiaControls BMI < 30 kg/m2 (n = 1025)Obese women
Class I BMI 30–34.9 kg/m2 (n = 159)Class II BMI 35–40 kg/m2 (n = 105)Class III BMI > 40 kg/m2 (n = 93)
Sensitivity (%)55556048
Specificity (%)95959091
Positive predictive value (%)45705368
Negative predictive value (%)96909381
Positive likelihood ratio10.710.26.25.3
Negative likelihood ratio0.50.50.40.6

Discussion

Evaluation of different methods of birth-weight prediction can be difficult. Differences between the predicted birth weight and the actual birth weight do not always accurately reflect the level of error, since the greatest errors are often in those at the extremes of birth weight. Therefore, the difference as a percent of actual birth weight is more useful in assessing errors in birth-weight prediction. Positive prediction errors indicate an overestimation of birth weight while negative prediction errors indicate an underestimation of birth weight. The percent difference between predicted and actual birth weight is an estimate of systematic errors within the birth-weight prediction method used, with the 95% confidence interval allowing evaluation of the precision of this estimate. The standard deviation of the percent birth-weight error is an estimate of the random prediction error, and gives information on the variability of individual percent deviations within the group, with 95% of subjects falling within twice the standard deviation.

The GAP method tended to overestimate birth weight in control as well as Class I and Class II obese patients while underestimating birth weight in Class III obese patients, as seen by negative values in birth weight mean and percent errors. This may be owing to increasing maternal obesity, or it may be owing to relatively larger infant weights in the patients with Class III obesity. There is some evidence that Hadlock's estimated fetal weight formulae may become less accurate above fetal weights of 4500 g14. Since the infants born to patients with Class III obesity were significantly larger than those in the other three groups, it is possible that this is the reason for the error in this group. Regardless of the statistically significant differences in the percent birth weight error between the groups, the overall mean systematic error (percent birth weight error) of the GAP method is within 4%, with a 95% error range between − 5% and + 5%. T-testing indicates that this error is significantly different from zero, zero being the ideal error for any birth-weight prediction method. The random error of the test (standard deviation of the percent birth weight error) is between 9.6% and 11.9%, indicating that 95% of patients' individual percent birth-weight errors would fall between 19.2% and 23.8% (twice the standard deviation). Levene's test for homogeneity of variance indicates that these percent birth weight errors differed between groups.

Overall, the GAP method is able to predict birth weight within 20% in over 90% of cases for both the obese and non-obese parturient, regardless of division by class of obesity. The systemic prediction errors for the GAP method did vary between groups, from 4.2% for controls to − 1.9% for those with Class III obesity. However, there were no statically significant differences between controls and those with Class I or II obesity. For individual patients, twice the standard deviation, or 20% random error could be expected at any maternal BMI.

Finally, although birth-weight prediction is important, predication of macrosomia is the primary clinical use of many birth-weight prediction methods. While the sensitivity and positive predictive values for the prediction of macrosomia by the GAP method were low, the specificity and negative predictive value of the GAP method were between 80 and 90% for all groups, indicating that the GAP method was able to correctly exclude macrosomia, which is the more important clinical measure when counseling patients. Positive likelihood ratios indicate that the GAP method was much more likely to positively identify the macrosomic fetus in the normal-weight patient than the morbidly obese (Class III) patient. However, the equal negative likelihood ratios for all groups indicate that the GAP method was able to exclude macrosomia equally well in all groups.

The strengths of our study include large numbers of very obese women, as well as an ethnically mixed representation in all study groups. The main weakness of this study is its retrospective nature. The patients referred for ultrasound scans at between 34 and 36 + 6 weeks may not be representative of the general population. There may be an over-representation of chronic disease within all groups, the patients having indications for periodic ultrasound assessment of fetal weight and thus being at higher risk for abnormal fetal growth. Another possible reason for over-representation of obese women in our study is that the assessment of fundal height and the performance of Leopold's maneuvers to determine fetal position may be difficult in such women, and therefore their healthcare providers may be more likely to refer them for ultrasound assessment of fetal growth. Also, BMI measurements used in our study were by self-report, which may not be accurate. However, one would assume that patients would be more likely to minimize their weight and maximize their height; therefore the actual BMIs could be even higher than those reported. Finally, BMI alone may not be a predictor of the difficulty of obtaining sonographic measurements since the distributions of body fat and abdominal adiposity may not be equal despite equal BMIs. However, for patients with Class III obesity, the mean pre-pregnancy weight was more than 120 kg, and we can therefore be assured that the degree of abdominal adiposity is significant.

Furthermore, those patients with Class III obesity had infants with much larger birth weight and higher rates of macrosomia, and Class III included a larger proportion of multiparous women, which is not unexpected since obesity is closely associated with increasing birth weight and parity15. An unexpected finding was that the rates of pre-existing diabetes did not vary with the degree of maternal obesity, but this may be a reflection of the young age of these women, who had not yet had time to develop Type II diabetes.

Overall, we have found that the GAP method performs equally well for most obese women as for normal controls, although it may become less accurate in patients with Class III obesity. Although there are statistical differences in the systematic errors of our test to predict birth weight in the most obese patients, the overall clinical error of the test is within 20% regardless of maternal BMI, and the clinician can be assured that this method is within 20% of actual birth weight in over 90% of cases. This method has previously been shown to be an accurate predictor of term birth weight, and has now been validated in both diabetic and obese patients. Although other methods of birth-weight prediction remote from term have shown equal results in both the second and third trimesters, the GAP method represents a particularly easy method of birth-weight prediction that is readily accessible to the clinician16. The GAP method may represent the best method for term birth-weight prediction, and further assessment of this method in comparison to other birth-weight prediction methods is warranted. Whether any method of birth-weight prediction can alter fetal outcome remains to be seen.

Ancillary