Accuracy and precision of sonographic fetal weight estimation in Sweden

Abstract

Introduction

Fetal growth assessment by ultrasound is an essential part of modern obstetric care. The formula by Persson and Weldner for estimated fetal weight (EFW), used in Sweden since decades, has not yet been evaluated. The objective of this study was to evaluate accuracy and precision of the formula by Persson and Weldner, and to compare it to two other formulae using biparietal diameter instead of head circumference.

Material and methods

The study population consisted of 31 521 singleton pregnancies delivered at 22⁺⁰ gestational weeks or later, with an ultrasound EFW performed within 2 days before delivery, registered in the Swedish Pregnancy Register between 2014 and 2021. Fetal biometric ultrasound measurements were used to calculate EFW according to the formulae by Persson and Weldner, Hadlock 2 and Shepard. Bland–Altman analysis, systematic error (mean percentage error), random error (standard deviation [SD] of mean percentage error), proportion of weight estimates within ±10% of birthweight, and proportion with underestimated and overestimated weight was calculated. Moreover, calculations were made after stratification into small, appropriate, and large for gestational age (SGA, AGA and LGA), respectively, and gestational age at examination.

Results

For the formula by Persson and Weldner, MPE was −2.7 (SD 8.9) and the proportion of EFW within ±10% from actual birthweight was 76.0%. MPE was largest for fetuses estimated as severe SGA (<3rd percentile, −5.4) and for the most preterm fetuses (<24 weeks, −5.4). For Hadlock 2 and Shepard's formulae, MPE were 3.9 (SD 8.9) and 3.4 (SD 9.7), respectively, and the proportions of EFW within ±10% from actual birthweight were 69.4% and 67.1%, respectively. MPE was largest for fetuses estimated as severe LGA (>97th percentile), 7.6 and 9.4, respectively.

Conclusions

The recommended Swedish formula by Persson and Weldner is generally accurate for fetal weight estimation. The systematic underestimation of EFW and random error is largest in extreme preterm and estimated SGA‐fetuses, which is of importance in clinical decision making. The accuracy of EFW with the formula by Persson and Weldner is as good as or better than Hadlock 2 and Shepard's formulae.

The currently used Swedish formula to estimate fetal growth by sonography generally estimates fetal weight accurately. Accuracy is lower in extreme preterm and SGA‐fetuses. The formula is as good as or superior to other formulae including biparietal diameter instead of head circumference.

Abbreviations

AC

abdominal circumference

APE

absolute percentage error

BPD

biparietal diameter

CI

confidence interval

EFW

estimated fetal weight

FL

femur length

FPR

false positive rate

GA

gestational age

HC

head circumference

ISUOG

International Society of Ultrasound in Obstetrics and Gynecology

MAD

mean abdominal diameter

MPE

mean percentage error

SD

standard deviation

SPR

Swedish Pregnancy Register

Key message.

The currently used Swedish formula to estimate fetal growth by sonography generally estimates fetal weight accurately. Accuracy is lower in extreme preterm and SGA‐fetuses. The formula is as good as or superior to other formulae including biparietal diameter instead of head circumference.

1. INTRODUCTION

Deviant fetal growth is associated with a variety of adverse outcomes that may affect mother's and infant's short‐ and long‐term health. ¹ , ² , ³ Apart from being the single largest risk factor for stillbirth, fetal growth restriction is associated with perinatal morbidity, metabolic health issues, and impaired cognitive development. ³ , ⁴

Fetal growth assessment is an essential part of modern obstetric care. Ultrasonic fetal weight estimation allows identification of fetuses with insufficient or accelerated growth. Correct identification of deviant growth enables individualized clinical management, whereas incorrect identification implies increased risks of both absent and unnecessary interventions.

Several mathematical formulae are used to predict fetal weight based on ultrasonic fetal biometry. Biometric measurements included in the formulae are biparietal diameter (BPD), head circumference (HC), mean abdominal diameter (MAD), abdominal circumference (AC), and femur length (FL). The predictive value of estimated fetal weight (EFW) depends on the quality of the biometric measurements, the performance of the chosen formula, and the cutoff levels for classification of a fetus as small or large for gestational age (SGA and LGA, respectively).

In Sweden, all units that perform obstetric ultrasound scans use the formula for EFW published by Persson and Weldner in 1986, ⁵ in line with the recommendations from the Swedish Society of Obstetrics and Gynecology. ⁶ A majority of the internationally used formulae has been evaluated regarding the accuracy and precision of EFW. ⁷ , ⁸ , ⁹ However, to our knowledge, the formula by Persson and Weldner has not been evaluated before beyond Persson and Weldner's initial publication, ⁵ a comparative study including 176 women with a prolonged pregnancy, ¹⁰ and in a computer model estimating the frequency of macrosomia using different formulae. ¹¹ Since HC is not routinely measured in Sweden, it is only possible to compare the performance of the formula by Persson and Weldner with other formulae using BPD, MAD, AC, and/or FL.

Our primary aim was to evaluate accuracy and precision of the formula by Persson and Weldner for EFW in a large cohort of pregnancies using the new Swedish intrauterine reference ranges for estimated fetal weight. ¹² Our secondary aims were to compare the accuracy and precision of the formula by Persson and Weldner, the Hadlock 2 formula ¹³ (including BPD, AC and FL), and Shepard's formula ¹⁴ (including BPD and AC), and to evaluate if the accuracy and precision differed in preterm, term, and post‐term gestations, and for fetuses estimated as SGA, appropriate for gestational age (AGA), and LGA, respectively.

2. MATERIAL AND METHODS

This retrospective cohort study assessed ultrasound and obstetric data prospectively registered in the Swedish Pregnancy Register (SPR) between 2014 and 2021. The study population consisted of women who gave birth at 22⁺⁰ completed gestational weeks (154 gestational days) or later, after a singleton pregnancy with an ultrasound fetal weight estimation performed within 2 days before delivery. For comparison, we assessed maternal, newborn, and delivery characteristics of all singleton births registered in the SPR during the study years.

The Swedish Pregnancy Register is a national quality register. It collects data on pregnancy and childbirth, starting at the first antenatal visit in early pregnancy and ending at the follow‐up visit to maternity care at around 8–16 weeks postpartum. The majority of data is transferred directly from the electronic medical records. SPR included approximately 95% of all births in Sweden (19 of 21 regions) during the study period. It includes demographic, reproductive and maternal health data, information on prenatal diagnostics, and pregnancy outcomes for mothers and newborns.

The Swedish recommendations for fetal biometric assessment follow the recommendations of the International Society of Ultrasound in Obstetrics and Gynecology (ISUOG) regarding planes and caliper placing. ⁶ , ¹⁵ It is not routine to assess HC or AC in Sweden. Accordingly, the vast majority of fetal biometric measurements registered in the SPR do not include HC or AC measured by ellipse facility.

The data from the SPR included 33 705 weight estimations in pregnancies registered as singleton births. In cases where multiple weight estimations were registered (n = 163), only the last weight estimation was included in the analyses. In the remaining 33 542 pregnancies, we excluded stillbirths (n = 106), cases with missing birthweight (n = 168), missing or erroneous (extreme outliers) ultrasound measurements (n = 6), and pregnancies that were initially multiple but with early intrauterine demise of at least one fetus (n = 12). Further, newborns diagnosed with structural anomalies or chromosome aberrations (defined as ICD‐codes Q00–Q99) were excluded (n = 1729). The final study population consisted of 31 521 pregnancies.

During the study period, 811 493 singleton pregnancies were registered with delivery at 22⁺⁰ gestational weeks or later in the SPR. Stillbirths (n = 2786) were excluded, as were cases with missing birthweight (n = 2712) or birthweight exceeding ±5 standard deviations (SD) (n = 1080). The final reference population consisted of 804 915 pregnancies.

At the first antenatal visit, all women are interviewed regarding their obstetric history, present smoking habits, country of birth, and highest degree of formal education. Weight is measured in light indoor clothes. Maternal height is self‐reported. Standardized birthweight for gestational age (GA) in percentiles was classified according to the new Swedish sex‐specific reference ranges for fetal growth. ¹² If sex was missing in the register, the sex‐neutral reference was used. Maternal and newborn characteristics were grouped according to Supporting Information in Table S1.

Each measure of EFW was standardized for gestational age (standardized EFW), classified in percentiles according to the sex‐neutral intrauterine reference, ¹² and defined as: severe SGA (<3rd percentile); mild SGA (3rd to <10th percentile); AGA (10th–90th percentile); mild LGA (>90th–97th percentile); and severe LGA (>97th percentile).

Descriptive statistics were used for maternal, newborn and sonographic examination characteristics. No statistical tests were performed to evaluate potential differences between the study and reference populations, as the study population also was a part of the reference population.

Fetal biometric ultrasound measurements were used to calculate EFW according to the formula by Persson and Weldner, Hadlock's second, and Shepard's formulae (Table S2). ⁵ , ¹³ , ¹⁴ The formula π∙MAD was used to calculate AC.

The accuracy and predictive performance of EFW was assessed in several ways:

1. To estimate the systematic error, we used Bland and Altman analysis ¹⁶ to calculate the mean difference of birthweight and EFW with 95% confidence interval (CI), and construct limits of agreement at ±1.96 SD, where 95% of the differences between birthweight and EFW are found.

2. Mean percentage error (MPE), which also represents the systematic error, was calculated as the mean of the percentage errors ([EFW‐birthweight]/EFW)·100. This method of calculating MPE expressed as a percentage of EFW, was previously proposed by Edwards et al., ¹⁷ and adopted by for example Anderson et al., ¹⁸ and Scioscia et al., ¹⁹ who highlighted that ultrasonographic estimation represents the actual relevant information to clinicians for decision making (clinical management). Thus, EFW, not birthweight, was used as the denominator.

3. Random error, which represents the non‐systematic part of the prediction error, was calculated as the SD of the MPE.

4. Absolute percentage error (APE), or the mean of the absolute value of the deviation of EFW from birthweight, that is, deviation regardless of the direction (positive or negative), which integrates systematic and random errors.

5. Proportion of estimates where EFW was within ±10% and ±20% of the actual birthweight, and the proportion with under‐ and overestimated weight.

6. Predictive performance of the formulae was estimated as sensitivity, false positive rate (FPR), positive predictive value and negative predictive value for diagnosis of severe SGA, mild SGA, mild LGA, and severe LGA at birth.

Last, the cohort was stratified according to standardized EFW, and MPE was calculated. The cohort was then stratified according to GA at examination, and MPE and proportion of estimates within ±10% of actual birthweight was calculated. Finally, the diagnostic accuracy of severe SGA in the different strata of GAs were evaluated by generating receiver‐operating characteristics (ROC) curves. The area under the ROC curve (AUC) was used to quantify the discriminatory ability of each formula to predict severe SGA and severe LGA for the different weight estimation formulae.

Statistical analyses were performed using IBM Corp. Released 2017. IBM SPSS Statistics for Windows, Version 25.0. Armonk, NY: IBM Corp.

2.1. Ethics statement

The study was approved by the National Ethical Review Board (Diary nuo. 2021‐03123) on July 14, 2021. All procedures involving humans were carried out in accordance with the ethical standards of the 1964 Helsinki declaration. All registry data was merged and pseudonymized before delivery to the research group. Thereby, informed consent was not required.

3. RESULTS

Maternal and newborn characteristics of the study population and the reference population are presented as Supporting Information in Table S1. There were obvious differences between the study and the reference populations in proportions of preterm birth (15.5% vs 4.2%, respectively) and severe SGA at birth (24.8% vs 9.0%, respectively).

The sonographic fetal weight estimations were performed within 1 day of delivery in 60.7% of the cases (Table 1). The prevalence of estimated severe SGA (<3rd percentile) was very high regardless of the used fetal weight estimation formula; 26.6% using the formula by Persson and Weldner, 16.0% using Hadlock 2, and 17.9% using Shepard's formula. Also, the prevalence of estimated severe LGA (>97th percentile) was high: 4.9% for Persson and Weldner, 9.4% for Hadlock 2, and 10.3% for Shepard's formula.

TABLE 1.

Characteristics of study population and sonography (n = 31 521).

	N	%
Days between sonography and birth
0	4714	15.0
1	14 421	45.8
2	12 386	39.3
Standardized estimated fetal weight (Persson and Weldner's formula) a
Severe SGA	8382	26.6
Mild SGA	4542	14.4
AGA	15 405	48.9
Mild LGA	1649	5.2
Severe LGA	1543	4.9
Standardized estimated fetal weight (Hadlock 2 formula) a
Severe SGA	5035	16.0
Mild SGA	3365	10.7
AGA	17 961	57.0
Mild LGA	2206	7.0
Severe LGA	2954	9.4
Standardized estimated fetal weight (Shepard's formula) a
Severe SGA	5658	17.9
Mild SGA	3585	11.4
AGA	16 874	53.5
Mild LGA	2171	6.9
Severe LGA	3233	10.3

Abbreviations: AGA, appropriate for gestational age; LGA, large for gestational age; SGA, small for gestational age.

^a Severe SGA defined as estimated fetal weight (EFW) <3rd percentile, mild SGA as EFW 3rd to 9.9th percentile, AGA as EFW 10th to 90th percentile, mild LGA as EFW 90. 1st to 97th percentile, and severe LGA as EFW >97th percentile for gestational age according to (11).

The Bland and Altman analysis showed large variations between formulae in mean differences of EFW and birthweight; for the formula by Persson and Weldner −70.0 g, (95% CI: −73.1 to −66.9) with limits of agreement 481.5; −621.6, for Hadlock 2141.7 g, (95% CI: 125.9 to 133.0) with limits of agreement 716.7; −433.3, and for Shepard's formula 129.5 g, (95% CI: 138.5 to 145.0) with limits of agreement 753.8;–494.8 (Figure 1A–C).

FIGURE 1.

(A–C) Bland and Altman plot showing agreement between birthweight and estimated fetal weight using the Persson and Weldner's ⁵ (A), Hadlock 2 ¹² (B) and Shepard's ¹³ (C) formulae with 95% confidence interval. Full lines represent mean and limits of agreement at ±1.96 SD.

The MPE was negative for the formula by Persson and Weldner, −2.7 (SD 8.9), indicating a systematic underestimation of EFW. For the other formulae, the MPE was positive; for Hadlock 2 3.9 (SD 8.9) and for Shepard's formula 3.4 (SD 9.7), indicating a systematic overestimation of EFW.

Using the formula by Persson and Weldner, 76.0% of EFWs were within ±10% of birthweight, compared to 69.4% for Hadlock 2, and 67.1% for Shepard's formula. The sensitivity and the FPR for identifying fetuses as severe SGA was highest for Persson and Weldner (78.6%, FPR 9.5%), compared to Hadlock 2 (56.0%, FPR 2.8%) and Shepard (59.5%, FPR 4.3%). The sensitivity for identifying fetuses as severe LGA was lowest for Persson and Weldner (64.8%, FPR 2.1%), and similar for Hadlock 2 (83.4%, FPR 5.9%) and Shepard (84.4%, FPR 6.8%). All measures of accuracy are displayed in Table 2.

TABLE 2.

Accuracy and precision of sonographic weight estimation.

	Fetal weight estimation formula
	Persson and Weldner	Hadlock 2	Shepard
Mean percentage error ± SD, %	−2.7 ± 8.9	3.9 ± 8.9	3.4 ± 9.7
Absolute percentage error ± SD, %	7.2 ± 6.0	7.9 ± 5.8	8.2 ± 6.2
Proportion of estimated fetal weight within:
±10% from birthweight, n (%)	23 953 (76.0)	21 875 (69.4)	21 157 (67.1)
±20% from birthweight, n (%)	30 807 (97.7)	30 018 (95.2)	29 670 (94.1)
Underestimation of fetal weight, n (%)	18 901 (60.0)	9620 (30.5)	10 599 (33.6)
Overestimation of fetal weight, n (%)	12 620 (40.0)	21 901 (69.5)	20 922 (66.4)
Severe SGA a
Sensitivity, %	78.6	56.0	59.5
FPR, %	9.5	2.8	4.3
PPV, %	73.2	86.8	82.1
NPV, %	92.8	87.0	87.8
Severe LGA b
Sensitivity, %	64.8	83.4	84.4
FPR, %	2.1	5.9	6.8
PPV, %	59.7	40.1	37.1
NPV, %	98.3	99.2	99.2

Abbreviations: FPR, false positive rate; NPV, negative predictive value; PPV, positive predictive value.

^a Estimated fetal weight < 3rd percentile for gestational age according to (11).

^b Estimated fetal weight > 97th percentile for gestational age according to (11).

Figure 2 shows MPE by standardized EFW. For Persson and Weldner, EFW was underestimated for fetuses estimated as severe or mild SGA and AGA, and the underestimation was largest for severe SGA, −5.4%. For fetuses estimated as mild or severe LGA, EFW was overestimated; MPE 1.6 in both cases. For Hadlock 2 and Shepard, EFW was close to the actual birthweight for fetuses estimated as severe SGA; MPE 0.7 and −0.7, respectively, but for fetuses estimated as severe LGA, the EFW was overestimated; MPE 7.6 and 9.4, respectively.

FIGURE 2.

Mean percentage error (MPE) of fetal weight estimation for gestational age. Estimated fetal weight was calculated using Persson and Weldner's (full line), Hadlock 2 (dashed), and Shepard's (dotted) formulae.

Table 3 shows accuracy of fetal weight estimation after stratification for GA. For Persson and Weldner, the proportion with correct estimations increased with GA. MPE was largest in the most preterm pregnancies (−5.4) and smallest in late term or postterm pregnancies (−2.4 and −1.9, respectively). For Hadlock 2, the MPE showed a U‐shaped pattern, with the lowest MPE at 24 to 36 gestational weeks (1.0–1.6), and the highest for postterm pregnancies (5.1). For Shepard, no obvious pattern in MPE by GA was observed. The highest MPE was seen in the late term and postterm gestations. For all three formulae, random error (SD of MPE) was largest in the most preterm and smallest in term and postterm pregnancies.

TABLE 3.

Accuracy and precision of sonographic weight estimation by gestational age at ultrasound.

Gestational age at ultrasound	Prevalence, n (%)	Mean percentage error ± SD, %	Estimated weight within ± 10% of birthweight, n (%)
Weight estimation using the Persson and Weldner formula
22+0 to 23+6	76 (0.2)	−5.4 ± 12.7	48 (63.2)
24+0 to 27+6	336 (1.1)	−4.1 ± 10.9	221 (65.8)
28+0 to 31+6	789 (2.5)	−2.9 ± 10.5	534 (67.7)
32+0 to 36+6	4073 (12.9)	−4.1 ± 9.5	2839 (69.7)
37+0 to 38+6	7408 (23.5)	−3.1 ± 8.8	5570 (75.2)
39+0 to 40+6	11 568 (36.7)	−2.4 ± 8.7	8795 (76.0)
≥41+0	7271 (23.1)	−1.9 ± 8.8	5565 (76.6)
Weight estimation using Hadlock 2 formula
22+0 to 23+6	76 (0.2)	2.7 ± 11.7	55 (72.4)
24+0 to 27+6	336 (1.1)	1.0 ± 10.6	229 (68.4)
28+0 to 31+6	789 (2.5)	1.3 ± 10.3	529 (67.1)
32+0 to 36+6	4073 (12.9)	1.6 ± 9.5	2945 (72.3)
37+0 to 38+6	7408 (23.5)	3.5 ± 8.7	5273 (71.2)
39+0 to 40+6	11 563 (36.7)	4.5 ± 8.6	7893 (68.2)
≥41+0	7271 (23.1)	5.1 ± 8.8	4766 (65.6)
Weight estimation using Shepard's formula
22+0 to 23+6	76 (0.2)	2.2 ± 13.1	48 (63.2)
24+0 to 27+6	336 (1.1)	2.7 ± 12.4	204 (60.7)
28+0 to 31+6	789 (2.5)	3.1 ± 11.8	471 (59.8)
32+0 to 36+6	4073 (12.9)	2.2 ± 10.5	2688 (66.0)
37+0 to 38+6	7408 (23.5)	3.1 ± 9.6	4983 (67.3)
39+0 to 40+6	11 563 (36.7)	3.6 ± 9.3	7752 (67.0)
≥41+0	7271 (23.1)	4.2 ± 9.4	4754 (65.4)

Figures [Link], [Link] presents the discriminatory ability of the weight estimation formulas to predict severe SGA and LGA in the entire cohort of all gestational ages, and stratified for the gestational age groups.

4. DISCUSSION

In this study of 31 521 fetal weight estimations performed within 2 days before birth, it was found that the recommended Swedish weight estimation formula by Persson and Weldner accurately estimates fetal weight in a majority of fetuses. Fetal weight is marginally underestimated, with more pronounced underestimation and larger fraction with clinically relevant deviation from the true birthweight with decreasing GA and if the fetus was estimated as SGA. In contrast, Hadlock 2 and Shepard's formula not only overestimate fetal weight, but overestimate fetal weight to a higher degree than the formula by Persson and Weldner underestimates it. This was especially pronounced for LGA‐fetuses and in term and postterm gestations. Measures of spread, or random error, were similar. The prediction within ±10% of the actual birthweight was higher for the formula by Persson and Weldner, indicating a higher accuracy compared to the formulae by Hadlock and Shephard.

Numerous studies have evaluated fetal weight estimation using different formulae. Large efforts have been made to develop a formula that more accurately estimates fetal weight than previously published formulae. Nevertheless, the more recently developed formulae rarely outdo the formula by Hadlock including HC, AC and FL, with or without BPD (Hadlock 3 and 4), published in 1985. ⁸ , ⁹ , ²⁰ , ²¹ , ²² , ²³ , ²⁴ Since our registry data did not include measures of HC, it was only possible to compare the performance of the formula by Persson and Weldner to other formulae including BPD instead of HC.

To our knowledge, this is the first study that evaluates the performance of the formula by Persson and Weldner from viability to postterm gestation. The results are of clear clinical importance, since the formula is currently unanimously used for EFW in Sweden. Compared with the MPE of 0.14 for Hadlock 3 (incorporating HC, AC and FL, and using pooled data from seven studies), ⁸ our MPE of −2.7 the formula by Persson and Weldner is somewhat less accurate. However, systematic and random errors of a formula are highly dependent on the study population and the method used for calculating errors, which makes direct comparisons between different studies difficult. ⁷

Consistent with our results, the random error (SD) in studies of accuracy in fetal weight estimation rarely falls below 7%, and is often as high as 10%–15%. ⁷ The large spread is further demonstrated by the high proportion of fetal weight estimates exceeding ±10% of the true birthweight, which even for the most accurate formulae is close to 20%, ²⁵ and thereby only slightly better than the formula by Persson and Weldner.

Several studies, all of them with a substantially smaller study population than ours, have evaluated MPE and APE for Hadlock 2 and/or Shepard's formulae. ⁹ , ²⁰ , ²¹ , ²² While three and (for APE all four) of these studies found MPEs and APEs that were comparable or slightly higher than our results, only one study, including 495 fetal weight estimations, found substantially lower MPE for both formulae compared to our results. ²⁰ Thus, considering our results, with lower MPE (±SD) and APE (±SD) for the formula by Persson and Weldner, this formula should be considered as more accurate than Hadlock 2 and Shepard.

In line with our results, larger deviations are usually seen for newborns with low or high birthweight. ²⁴ , ²⁶ , ²⁷ , ²⁸ , ²⁹ , ³⁰ In a computer model, Mongelli and Benzie ¹¹ found the formula by Persson and Weldner to estimate a significantly lower fraction of fetuses as macrosomic compared with Hadlock 2 and Shepard. The change in body composition throughout pregnancy may, at least in part, explain the imprecision in fetal weight estimation in early GAs. The commonly used weight estimation formulae neither account for changes in proportions throughout gestation, nor for body composition in growth restricted or macrosomic fetuses. Specific formulae have been evolved for EFW in suspected SGA or LGA fetuses, which were superior to published general formulas when applied in a retrospective design. ²³ However, when a two‐stage model was used that mimicked a clinical setting, the accuracy of birthweight prediction was not improved compared with using Hadlock 3 for all. ⁹

The main strength of the study is the very large cohort, which is, to our knowledge, substantially larger than any previous cohort in this field. The cohort size allowed us to perform stratified assessments of the accuracy of EFW, and even to split the group of extreme preterm fetuses and look separately into week 22–23, which to our knowledge has not been done before. The group of fetuses near viability is increasingly relevant to study in detail since interventions on fetal indication are presently performed more often, and in Sweden full efforts are made to save newborns from gestational week 22⁺⁰ to 23⁺⁰. ³¹ Thus, awareness of potential errors in estimating fetal growth at this extreme gestational age is of substantial clinical importance. Another strength of the study is that all fetal weight estimations were performed within 2 days before delivery. The narrow interval between sonography and birth ensures minimal error due to fetal growth. Moreover, in this study, MPE is determined by EFW and GA at ultrasound, not birthweight and GA at birth, which is the point in time when clinical decisions are made. In line with previous studies, we believe that this approach better reflects the clinical situation than if birthweight and GA at birth are retrospectively assessed. ¹⁷ , ¹⁸ , ¹⁹ In several earlier studies, the chosen methodology is not clearly described, which can render difficulties in the comparison to earlier research.

A major limitation of the study is the absence of measurements of HC in our data, which excludes comparisons with highly ranked formulae, such as Hadlock 3 and 4, ¹³ Ott's, ³² Combs', ³³ and the INTERGROWTH‐21st ³⁴ formulae. Furthermore, the Swedish routine is to measure MAD rather than the more commonly used AC. Thus, the two‐diameters method was used to calculate AC, quite contrary to the recommendations of ISUOG, which might introduce systematic measurement error. ⁶ , ¹⁵ , ³⁵ Another limitation is the selection bias, where pregnancies at high risk of aberrant growth are more likely to be included as third trimester estimation of fetal weight is not part of the routine care in Sweden.

We found a small systematic underestimation of fetal weight when using the Swedish formula by Persson and Weldner that is larger than reported in studies evaluating Hadlock 3 and 4, while measures of spread, or random error, seem to be comparable to Hadlock 3 and 4. In normally grown fetuses and around term, where most fetal weight estimations are performed, the underestimation is not likely of clinical importance. However, it is essential to keep the uncertainty and potential error of fetal weight estimation in mind when clinical decisions are based on EFW. Knowledge of the weaknesses of the chosen formula, especially in the extremes of standardized EFW and GA, is of great importance in clinical decision making. Moreover, the difference in MPE of 6.6% (−2.7% for Persson and Weldner and +3.9% for Hadlock 2) indicates that it is not possible to compare fetal weight estimations straight off between different formulae.

It is important to be aware of the inaccuracy of fetal weight estimation in fetuses with low standardized EFW and in preterm fetuses near the limit of viability. Underestimation of fetal weight, as seen in our study, might lead to renounced intervention if the fetus is considered too small to survive. When using Persson and Weldner's formula in clinical practice, special attention should be given to the fact that EFW is less accurate in extreme preterm and estimated SGA‐fetuses. Our results do not advocate a switch to Hadlock 2 or Shepard's formulae for fetal weight estimation in Sweden.

5. CONCLUSION

Estimation of fetal weight using the recommended Swedish formula for EFW by Persson and Weldner systematically but marginally underestimates fetal weight, with similar spread as shown in earlier studies of fetal weight estimation accuracy. For a majority of fetuses, the formula by Persson and Weldner is accurate, but in one‐fourth the deviation between EFW and true birthweight is clinically relevant and exceeds 10%. The error is largest for fetuses estimated as SGA and at low GA. In general, the formula by Persson and Weldner is as good as or more accurate and precise than Hadlock 2 and Shepard's formula.

AUTHOR CONTRIBUTIONS

MG came up with the original idea. LL performed validation and cleaning of data and statistical analyses. LL and MG wrote the first draft of the manuscript. All authors contributed to study conception and design, analytic plan, interpretation of data and critical revision and approval of the manuscript.

FUNDING INFORMATION

The study was funded by the Swedish Infant Death Foundation (Spädbarnsfonden).

CONFLICT OF INTEREST STATEMENT

None.

Supporting information

Figure S1. Receiver‐operating characteristics (ROC) curves for prediction of severe small for gestational age newborn (birthweight <3rd percentile) using Persson and Weldner’s formula for estimated fetal weight in a) all gestational ages; b) gestational week 22–23; c) gestational week 24–27; d) gestational week 28–31; e) gestational week 32–36; f) gestational week 37–38; g) gestational week 39–40; and h) gestational week 41 and later. AUC, area under curve.

Figure S2. Receiver‐operating characteristics (ROC) curves for prediction of severe large for gestational age newborn (birthweight >97th percentile) using Persson and Weldner’s formula for estimated fetal weight in a) all gestational ages; b) gestational week 22–23; c) gestational week 24–27; d) gestational week 28–31; e) gestational week 32–36; f) gestational week 37–38; g) gestational week 39–40; and h) gestational week 41 and later. AUC, area under curve.

Figure S3. Receiver‐operating characteristics (ROC) curves for prediction of severe small for gestational age newborn (birthweight <3rd percentile) using Hadlock’s second formula for estimated fetal weight in a) all gestational ages; b) gestational week 22–23; c) gestational week 24–27; d) gestational week 28–31; e) gestational week 32–36; f) gestational week 37–38; g) gestational week 39–40; and h) gestational week 41 and later. AUC, area under curve.

Figure S4. Receiver‐operating characteristics (ROC) curves for prediction of severe large for gestational age newborn (birthweight >97th percentile) using Hadlock’s second formula for estimated fetal weight in a) all gestational ages; b) gestational week 22–23; c) gestational week 24–27; d) gestational week 28–31; e) gestational week 32–36; f) gestational week 37–38; g) gestational week 39–40; and h) gestational week 41 and later. AUC, area under curve.

Figure S5. Receiver‐operating characteristics (ROC) curves for prediction of severe small for gestational age newborn (birthweight <3rd percentile) using Shepard’s formula for estimated fetal weight in a) all gestational ages; b) gestational week 22–23; c) gestational week 24–27; d) gestational week 28–31; e) gestational week 32–36; f) gestational week 37–38; g) gestational week 39–40; and h) gestational week 41 and later. AUC, area under curve.

Figure S6. Receiver‐operating characteristics (ROC) curves for prediction of severe large for gestational age newborn (birthweight >97th percentile) using Shepard’s formula for estimated fetal weight in a) all gestational ages; b) gestational week 22–23; c) gestational week 24–27; d) gestational week 28–31; e) gestational week 32–36; f) gestational week 37–38; g) gestational week 39–40; and h) gestational week 41 and later. AUC, area under curve.

Table S1. Maternal and infant characteristics.

Table S2. Formulae for calculation of estimated fetal weight.

Lindström L, Cnattingius S, Axelsson O, Granfors M. Accuracy and precision of sonographic fetal weight estimation in Sweden. Acta Obstet Gynecol Scand. 2023;102:699‐707. doi: 10.1111/aogs.14554