Diagnostic accuracy of ultrasound-based fetal weight estimation and prediction modeling for low birth weight: A retrospective cohort study in Northwest Ethiopia

Abstract

Objective

To evaluate the accuracy of ultrasound-based fetal weight estimation and to develop and internally validate a prediction model for low birth weight.

Methods

A retrospective cohort study was conducted among 269 individuals.

Results

The accuracy of ultrasound weight estimation was 72.9%, with 27.1% of estimates falling outside ±10% of actual birth weight. The Pearson correlation for estimated fetal weight and actual birth weight was 0.92. Using LASSO regression eight predictors were selected and entered in the multivariable analysis of which gestational age, estimated fetal weight, timing of estimation, and placental location were found significant. The original model demonstrated strong discrimination of AUC of 95.9% 95%CI (93.1-98.0%), and good calibration. Internal validation via 1000-bootstrap resampling yielded a bias-corrected AUC of 94.3%, indicating preserved predictive performance with minimal optimism. An integer-based risk score was derived to enhance clinical usability. The risk score model achieved acceptable discrimination (AUC = 76.8%, 95% CI: 70.0–83.0%) and good calibration (p = 0.301). Agreement between the full prediction model and the simplified risk score was fair (Cohen’s Kappa = 0.353, p < 0.0001), indicating that the simplified score is best suited as a pragmatic bedside risk-stratification tool rather than a direct substitute for the full model. Women were stratified into low (0–10 points), medium (11–21), and high-risk (22–36) groups corresponding to increasing predicted probabilities of low birth weight (3.2%–96.3%).

Conclusion

Ultrasound-based fetal weight estimation using the Hadlock-III formula indicated acceptable accuracy and strong correlation with actual birth weight. We developed and internally validated prediction model for low birth weight with excellent discrimination, good calibration, and predictive value beyond estimated fetal weight alone. A risk-score tool was created to facilitate rapid bedside risk stratification. However, given the single-center nature of the study and the absence of external validation, further multicenter validation is required before broader clinical implementation.

Keywords

ultrasound estimated fetal weight accuracy risk prediction low birth weight

Introduction

Estimating fetal weight is one of the most important components of the follow-up of pregnant women. Accurately estimating fetal weight (EFW) is critically important for assessing fetal growth, deciding the mode of delivery, and counseling parents on perinatal morbidity and mortality, It is also important for management of labor abnormalities¹ and for prevention of intrapartum and postnatal complications. Especially, in high risk pregnancies, estimating fetal weight is very important.^2–4

Before the introduction of ultrasound, abdominal palpation and other physical examination oriented fetal weight estimation was being practiced. Currently, estimating fetal weight is a standard of practice and is highly dependent on ultrasound.⁵

Studies show that compared to clinical estimation, ultrasound based fetal weight estimation strongly predicts actual birth weight.^6–9 Nevertheless, there are reports that show systematic and random errors in ultrasound estimated fetal weight. The accuracy of birth weight estimation varied from institution to institution and from formulas to formulas.^10–15

Several regression formulas have been developed different scientific organizations. Standard measurement of head circumference, biparietal diameter, abdominal circumference and femoral length are common measurements used to calculate the fetal weight.^1,16,17 Various sonographic formulae have been compared to determine the most accurate method for estimating fetal weight, with Hadlock IV and Vintzileos formulas showing significant associations with accurate birth weight prediction¹⁶ and these formulas are recommended by the international society of ultrasound in obstetrics and gynecology (ISUOG).¹⁸ However, there is no superior model prescribed for every setting and population.¹⁹

In addition to the formula being used,^10,16 there are also other different factors that affect the accuracy of ultrasound based estimated birth weight. These factors include; maternal age,¹¹ maternal BMI,^20–22 position,²³ neonatal weight and length,²⁴ time interval between sonographic weight estimation and delivery,^25,26 fetal gender,^15,26 presentation,^24,26,27 gestational age,^28,29 amniotic fluid^15,18,30 and the experience of sonographers.^15,26

To improve neonatal health in addition to understanding the role of US based birth weight estimation; predicting low birth weight based on easy to assess maternal, pregnancy and neonatal related factors, and identifying the added diagnostic value US based fetal weight is quite essential for early decision making for clinicians.³¹ In the Ethiopian context, despite the routine fetal weight estimation before admission, there is no published evidence on how accurate our fetal weight estimation is with respect to actual birth weight; moreover, there are limited studies on prediction of low birth weight for early stratification of pregnancies for risk tailored interventions. Therefore, this study aimed to assess the accuracy of ultrasound birth weight estimation and develop a predictive model for low birth weight among neonates in Tibebe Gion specialized hospital (TGSH).

Method and materials

Study design, area and period

This is a retrospective cohort study conducted at Tibebe Ghion Specialized hospital in Bahir Dar city; we used hospital based patient record data of pregnant individuals followed during antenatal care and delivery. Mothers who gave birth at Tibebe Ghion Specialized Hospital between January–July 2024 were our source population. Following the receipt of our ethical permission, data collection was performed in August 2024.

Bahir Dar is the capital city of Amhara national regional state which is located 565 km from Addis Ababa. The city has two governmental specialized hospitals, one primary hospital, four private hospitals and ten health centers. Tibebe Ghion specialized hospital is a university hospital under college of medicine and health sciences of Bahir Dar university which has a large maternal service area with labor and delivery ward, high risk ward, maternity ward and gynecology ward allocating 70 beds in patient and 12 outpatients’ rooms. Twenty actively working senior consultants, 49 residents, and 20 midwives are currently (at the time of data collection) giving the service.

Population

All mothers who gave birth in Tibebe Ghion specialized hospital during the study period and with inclusion criteria of having ultrasound fetal weight estimation within 7 days before birth were the study population. Mothers with stillbirth and structural anomalies involving the head circumference, biparietal diameter, femoral length and abdominal circumference were excluded from the study.

Study variables

Outcome variables: The primary outcome for the prediction model was low birth weight (LBW), defined as birth weight <2500 grams (yes/no), measured using the actual birth weight at delivery.

For the diagnostic accuracy analysis, accuracy of ultrasound-based fetal weight estimation was defined as an estimated fetal weight within ±10% of the actual birth weight; estimates outside this range were classified as inaccurate.

LBW was defined according to the WHO classification as birth weight < 2500 g, irrespective of gestational age. Birth weight was abstracted from delivery records within one hour of birth³²

Predictor variables: comorbidities, fetal presentation, gestational age at delivery, fetal growth restriction, macrosomia, timing of weight estimation, amniotic fluid volume, location of the placenta, maternal age, sex of the fetus, and side of the spine.

Operational and standard definition

Timing of fetal weight estimation is the time from estimation to the time of delivery.^11,20

Accurate weight estimation is considered when the estimated fetal weight is within 10 % of the actual birth weight.^15–17,33 Estimated fetal weight is the most recent ultrasound estimation of fetal weight prior to the delivery.

Intrapartum ultrasound was defined as an ultrasound assessment conducted after the onset of regular painful contractions with cervical dilatation ≥4 cm (active phase of labor). When fetal head descent limited visualization of standard BPD or HC planes, sonographers optimized probe angulation and maternal positioning and relied on abdominal parameters (AC and FL) when necessary, in accordance with established obstetric sonography practice^12,34,35

Birth weight: is neonatal weight measured immediately after delivery using a weight scale.

Ultrasound model: in our hospital we use a Samsung HS40 model curvilinear transducer.

Ultrasound examinations were performed by both senior resident physicians enrolled in the obstetrics (who are 3 years and more residency training) and gynecology specialty training program or by senior physicians (specialists/consultants). The secondary dataset did not include granular operator-level experience indicators such as years of practice, the number of ultrasound examinations performed, or certification status. However, all staff performing measurements had formal ultrasound training and followed standardized hospital protocols for fetal biometry assessment.

Sample size determination and sampling procedures

The sample size was computed using sample size determination in a population survey in Epi info version 7.2.6.0 software by considering confidence level of 95%, and the proportion of weight estimated within 10% from Uganda using Hadlock was 78.4%. So, the total sample size was 286 including the 10% non-response rate.¹⁸ Based on the previous 6 months before January 1st, there were 2320 women who gave birth in the database, and since sample size calculated is 286, every 8th mother was taken using systematic random sampling procedure.

Data collection tool and procedure

A checklist was used to extract the data from the study participants, medical records (case notes, ultrasound evaluation, delivery summaries, operation notes, and discharge summaries) and last ultrasound evaluation. Location of the placenta, sex, side of spine, presentation and other variables were taken from the ultrasound evaluation form. The actual birth weight was taken from the delivery summaries measured using birth weight scale. Data was collected using epicollect5 software. The supervisor and main investigator examined and checked the checklist for completeness, accuracy, and consistency, and corrective steps were made. To assure the quality of the data training was given to data collectors and supervisors on how to retrieve and extract data from selected medical charts. The supervisor and investigators closely followed the data collection process. The fetal weight estimation by Hadlock III was used. The Hadlock III formula was selected because it was the routine fetal weight estimation method used at Tibebe Ghion Specialized Hospital during the study period and was the formula recorded in the ultrasound reports available in the secondary dataset. Since this study relied on retrospective medical record review, alternative formulas such as Hadlock IV or INTERGROWTH-21st could not be consistently reconstructed from the available data.

In the routine care health care in hospitals in Ethiopia, ultrasound examinations were performed by either resident physicians enrolled in the obstetrics and gynecology specialty training program or by senior physicians (specialists/consultants).

Data management and analysis

After data collection, the data was downloaded into Excel, where it was checked for completeness and coded. The dataset was then exported as a CSV file into R and Statistical Package for the Social Sciences (SPSS) version 27 for analysis. To explain the study population in relation to relevant variables, descriptive statistics such as frequencies and percentages were calculated.

The accuracy analysis was performed using contingency tables to compare ultrasound-estimated birth weights (EFW) with actual birth weights (BW) for inaccurate vs. accurate. The frequency distribution of differences between EFW and BW within 10% of the discrepancies was computed.

A Correlation and Agreement Analysis was also conducted using the mean percentage error (MPE) which is ((EFW-BW)/BW) *100, and mean absolute percentage error (MAPE) which is (│EFW-BW│/BW) *100. MPE was considered as marker for the systematic error, and the corresponding standard deviation and MAPE as marker for the random error, respectively. The 95% confidence intervals (CI) for MPE were used to identify significant differences from zero. The Pearson correlation coefficients were calculated between ultrasound-estimated and actual birth weights to evaluate correlation between the two. In addition, the Bland-Altman Plot was generated to assess agreement between ultrasound estimates and actual birth weights in terms of bias, lines of agreement, systematic trend and outliers.

We applied Least Absolute Shrinkage and Selection Operator (LASSO) regression to identify relevant predictors and reduce model overfitting. A total of approximately 18 candidate predictors were initially entered in LASSO model development with 76 low birth weight events available for analysis. Although this is below commonly recommended thresholds for stable prediction model development, LASSO penalization was applied to reduce dimensionality and minimize overfitting by shrinking less informative coefficients toward zero. Predictors with non-zero coefficients were retained for multivariable logistic regression. Variable retention was guided primarily by predictive contribution identified through LASSO penalization and overall model performance rather than solely by statistical significance in the multivariable regression model.^36,37 Predictors considered clinically relevant were retained to preserve the interpretability and bedside applicability of the simplified risk score. The regression coefficients with their 95% confidence intervals and p-values were reported for the models.

LASSO predictor selection and penalty tuning were performed once on the full development dataset prior to bootstrap internal validation. The selected predictors were subsequently entered into the multivariable logistic regression model, and bootstrap resampling (1000 repetitions with replacement) was applied to estimate optimism-corrected model performance. Therefore, the bootstrap validation assessed the finalized model and did not repeat the complete variable selection process within each bootstrap replicate. Bootstrap resampling was selected for internal validation because it is considered one of the most efficient approaches for estimating optimism and evaluating prediction model stability in logistic regression models.^38,39

Model performance was evaluated using the area under the receiver operating characteristic curve (AUC) for discrimination, calibration plots, calibration slope, calibration-in-the-large, Brier score, and decision curve analysis (DCA). The bootstrap procedure and corresponding estimates were re-computed and verified after correcting an earlier reporting inconsistency identified during manuscript revision. Comparisons between the apparent model and the internally validated bootstrap model were performed using different performance metrics including AUC, Dxy, calibration slope, calibration-in-the-large, and Brier score to assess optimism and loss of predictive accuracy.

Decision curve analysis (DCA) was used to evaluate the potential clinical and public health utility of the model by estimating standardized net benefit across a range of threshold probabilities (0–1).

We developed an easy to use risk score based on the regression coefficients obtained from the multivariable regression modeling. The coefficients were divided into a scaling factor of 0.2 and rounded to the nearest integer. We determined the total score for each individual by assigning the points for each variable present and adding them up. For simple interpretation in a clinical setting, we categorized the total risk score into three based on the risk distribution in the population and practical sensibility. In addition, we developed a graph showing risk distribution based on the risk level of the population given their predicted risk and an easy to use table containing predictors along with their risk score.

This study was conducted and reported in accordance with the TRIPOD (Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis) and The STROBE (Strengthening the Reporting of Observational Studies in Epidemiology) statement.^40,41 Completed STROBE and TRIPOD checklists are provided as supplementary materials.

Ethical clearance

Ethical clearance was obtained from the institutional review board (IRB) of Bahir Dar University, college of medicine and health science on July 22, 2024, with meeting ID: 004/2024, and protocol number: 1071/2024. The IRB determined that the study was collected in a fully anonymized manner based on secondary data; as a result the need for participant consent was waived by the ethics committee. The Ethical permission was communicated to Tibebe Ghion Specialized hospital before the commencement of the data collection. All patient data were fully de-identified prior to analysis. No personally identifiable information was accessed or included in the dataset to ensure confidentiality and privacy. The study was conducted in accordance with the ethical principles of the Helsinki Declaration of 1975, as revised in 2024.

Results

Demographic, obstetric, medical and weight estimation related factors

A total of 2320 women gave birth were identified in the database, 43 were excluded because of missing important information including estimated birth weight estimation, therefore, 269 were included in our final analysis (Figure 1). The mean maternal age of the study population was 28.55±5.45 years. The mean gestational age at delivery was 38.3±2.54 weeks.

Figure 1.

Flowchart of the study participants.

About 89% of the fetal weight estimations were done by the residents, 67.7% of the estimations were done in the antepartum period before the onset of labor and 78.8% of the estimations were within 3 days prior to the delivery. While the mean of estimated fetal weight was 2918.4±629.54gm, the mean of actual birth weight was 2844.9±695.65 gm.

Majority (78.1%) of the individuals had a normal amniotic fluid status at time of fetal weight estimation and 13% of them had non cephalic presentation including breech and transverse lie. The back of fetus at time of ultrasound was lateral in 48.7% of the cases and 8.9% of them had the back anterior. In 53.5% of the mothers the placenta was posterior during ultrasound evaluation. Male sex was 56.9% newborns and 82% of the deliveries were at term.

Most ultrasound examinations were performed within 3 days of delivery (212/269, 78.8%), whereas 57 (21.2%) were performed more than 3 days but within 7 days before delivery. Preterm pregnancies accounted for 49 (18.2%) of the study population, while 220 (81.8%) were term or post-term deliveries. Greater variability between ultrasound-estimated fetal weight and actual birth weight was more commonly observed among pregnancies with longer ultrasound-to-delivery intervals and among preterm deliveries, suggesting that timing of assessment and gestational age may contribute to increased measurement variability (Table 1).

Table 1.

Maternal demographic, obstetric, medical and weight estimation related factors among mother who gave birth in Tibebe Ghion Specialized hospital, August, 2024.

Variables	Categories	Outcome variable		Total
Variables	Categories	Within 10% of ABW n (%)	Outside 10% of ABW n (%)	Total
Professional involved	Resident	176 (65.4)	65 (24.2)	241 (89.6)
Professional involved	Senior	20 (7.4)	8 (3)	28 (10.4)
Medical comorbidity	No	169 (62.8)	61 (22.7)	230 (85.5)
Medical comorbidity	Yes	27 (10)	12 (4.5)	39 (14.5)
Timing of estimation	Antepartum	133 (49.4)	49 (18.2)	182 (67.7)
Timing of estimation	Intrapartum	63 (23.4)	24 (8.9)	87 (32.3)
Time interval	Within 3 days	155 (57.6)	57 (21.2)	212 (78.8)
Time interval	>3 by ≤7 days	41 (15.2)	16 (5.9)	57 (21.2)
Amniotic fluid status	Normal	156 (58)	54 (20.1)	210 (78.1)
Amniotic fluid status	Abnormal	40 (14.9)	19 (7.1)	59 (21.9)
Fetal presentation	Cephalic	173 (64.3)	61 (22.7)	234 (87)
Fetal presentation	Non cephalic	23 (8.6)	12 (4.5)	35 (13)
Placental location	Anterior	87 (32.3)	38 (14.1)	125 (46.5)
Placental location	Posterior	109 (40.5)	35 (13)	144 (53.5)
Fetal sex	Male	113 (42)	40 (14.9)	153 (56.9)
Fetal sex	Female	83 (30.9)	33 (12.3)	116 (43.1)
Side of the back	Anterior	64 (23.8)	24 (8.9)	88 (32.7)
	Lateral	96 (35.7)	35 (13)	131 (48.7)
	Posterior	36 (13.4)	14 (5.2)	50 (18.6)
GA at delivery	<37 weeks	30 (11.2)	19 (7.1)	49 (18.2)
	≥37 weeks	166 (61.7)	54 (20.1)	220 (81.8)

Continuous variables mean (SD)
	Mean	SD	Minimum	Maximum
Maternal age in years	28.55	5.44	18	45
Parity	2.49	1.62	1	9
GA at delivery in weeks	38.33	2.54	30	42
EFW in grams	2918.4	629.54	1356	4500
Actual birth weight	2844.9	695.65	1200	4500

Accuracy of fetal weight estimation

The accuracy of ultrasound fetal weight estimation using Hadlock III formula was found to be 169 (72.9%) and 73 (27.1%) of ultrasound estimations were outside the margin of ± 10% of the actual birth weight.

The ultrasound estimated fetal weight within 7 days of delivery versus actual birth weight showed a Pearson correlation of 0.92 which showed that fetal weight estimated is highly correlated (P<0.0001). As shown in table 2 the mean percentage error of ultrasound fetal weight estimation shows the average percentage by which the estimated fetal weight deviation from the actual birth weight was 3.9%. The number denotes that there is overestimation as the number is positive. The mean absolute percentage error was 8.2% which means about the average magnitude of errors as a percentage of the actual values regardless of the direction. The mean error, which means the average of the differences between the estimated fetal weight and actual birth weight, was 73.5gm while the magnitude error without considering the direction (MAE) was 210.97 gm (Table 2). Moreover, regarding measurement agreement, most observations fell within the 95% limits of agreement, and no clear systematic pattern was observed across the measurement range. However, the limits of agreement were relatively wide, and several observations fell outside these limits (Figure 2). These findings suggest that although ultrasound-estimated fetal weight showed reasonable agreement with actual birth weight at the population level, clinically important individual-level discrepancies remained in some pregnancies. The observed outliers were more common among preterm pregnancies and among ultrasounds performed more than 3 days before delivery, suggesting increased variability in fetal weight estimation under these conditions.

Table 2.

Measurements of accuracy metrics and correlation of fetal weight estimation and actual birth weight.

Measurement	Value
Mean Percentage Error (MPE)	3.9%
Mean Absolute Percentage Error (MAPE)	8.23%
Mean Error (ME) (Bias)	73.5 gm
Mean Absolute Error (MAE)	210.97
Estimation within ± 10% of ABW (Accurate)	72.9%
Estimation outside ± 10% of ABW (Inaccurate)	27.1 %
Pearson correlation	0.922, (0.9, 0.94)**

**Statistically significant with P-value of < 0.001.

Figure 2.

Bland Altman plot for the average differences and means of estimated fetal weight and actual birth weight.

Variable selection using LASSO regression for predictive modeling

Before performing multivariable logistic regression, we applied Least Absolute Shrinkage and Selection Operator (LASSO) regression to identify the most relevant predictors and reduce model over fitting. A total of 18 candidate predictors were included in the LASSO regression analysis. Among these, 8 predictors had non-zero coefficients and were retained for inclusion in the multivariable logistic regression model. The variables selected by LASSO regression for the final logistic regression model were: maternal age, gestational age at delivery, estimated fetal weight, time interval, time of estimation, placenta location, professional level, and education level. The optimal lambda value (λ) that minimized the cross-validated error and was used to select these predictors was 0.0111, at standard error of 0.0404.

Out of the eight selected variables included in the multivariable regression model, Gestational age at delivery, estimated fetal weight, Timing of estimation, and Placental location were retained as important predictors in the multivariable model for the occurrence of low birth weight. In contrast, the Profession involved in fetal weight estimation was showed a marginally significant contribution for the prediction of low birth weight (Table 3).

Table 3.

Multivariable logistic regressions for the prediction of low birth weight.

Variable	Beta	Lower 95% CI	Upper 95% CI	P-value
Intercept	13.595095234	8.029	18.572	0.000
Maternal age <20 years	1.0105264	-0.992	3.005	0.313
Maternal age 20-34 years	-0.063832054	-1.44	1.386	0.929
Gestational age Preterm	1.796168128	0.42	3.31	0.013
Estimated Fetal weight	-0.005967325	-0.008	-0.004	0.000
Timing estimation = Intrapartum	1.406564094	0.263	2.648	0.020
Time interval 3-7 days	0.678114432	-0.497	1.875	0.258
Posterior Placenta location	1.106547459	0.116	2.171	0.033
Professional = Consultant	1.46941134	-0.758	3.495	0.167
No formal education	0.798	-0.849	2.626	0.365
Primary education	-0.744690344	-1.983	2.113	0.959
Secondary education	0.792029261	-0.074	3.489	0.078

This study used lrm in rms package regression based on datadist () to penalize our original model. The model demonstrated strong discriminative performance, which was further evaluated against alternative approaches (see below). The area under the ROC curve (AUC) was 95.9%, 95%CI: (93.1 - 98.0%) (Figure 3), indicating outstanding discrimination ability between low birth weight and non-low birth weight outcomes. The model’s discrimination index (Dxy) was 0.917, supporting strong concordance between predicted and observed classifications. The model explained a substantial proportion of variability in the outcome, with a Nagelkerke R² of 0.760. The Brier score was 0.064, suggesting high overall accuracy and good calibration of predicted probabilities. Additionally, the likelihood ratio chi-square test statistic was 202.4, with a p-value < 0.0001, indicating that the model provides a significantly better fit than a null model with no predictors.

Figure 3.

Receiver operating characteristics curve (ROC) for prediction of low birth weight original model.

The calibration performance

The calibration of our original predictive model was assessed using the GIVITI calibration belt method in R. The calibration plot demonstrated a close agreement between the predicted probabilities and the observed probabilities. The graph closely follows the 45 degree reference line, indicating that the model is well-calibrated. The 95% confidence belt mostly included the diagonal line of perfect calibration, suggesting no significant evidence of miscalibration across the range of predicted risks. This result supports the reliability of the model in estimating the actual probability of low birth weight outcomes within the study population. In addition p-value on the curve was 0.371, showing there is no difference between predicted and observed probabilities (Figure 4).

Figure 4.

Calibration plot for the prediction of low birth weight for the original model.

This graph illustrates the model’s effectiveness in distinguishing between pregnancies that developed low birth weight and those normal birth weights. The density plot reveals strong class separation between positive and negative cases. The model assigns high probabilities to positive cases (low birth weight), with the corresponding density curve peaking sharply near 0.00 and tapering off rapidly beyond 0.25. In contrast, negative cases (normal birth weight) show a peak around 0.87, with minimal density below 0.50. The limited overlap between the two curves indicates that the model effectively distinguishes between the two outcomes, suggesting good discriminative performance. These findings support the model’s utility in identifying individuals at risk for low birth weight based on predicted probability thresholds (Figure 5).

Figure 5.

Density plot for predicting low birth weight.

Incremental predictive value beyond estimated fetal weight

Because estimated fetal weight (EFW) is strongly correlated with actual birth weight, we conducted additional analyses to evaluate whether the multivariable model provides predictive value beyond a simple EFW-based classification.

First, we assessed the performance of EFW alone using a clinically relevant threshold of <2500g. This approach demonstrated good discrimination with an AUC of 83.6%, sensitivity of 69.7%, specificity of 97.4%, positive predictive value (PPV) of 91.4%, and negative predictive value (NPV) of 89.1%.

Second, we evaluated a model excluding EFW, which retained acceptable discrimination (AUC 85%), though with lower sensitivity (56.6%) and similarly high specificity (96.9%).

In comparison, the full multivariable model achieved substantially improved discrimination (AUC = 95.9%) and higher sensitivity (81.6%) while maintaining excellent specificity (96.9%).

These findings indicate that although estimated fetal weight is a strong predictor, comparative analyses demonstrated that the multivariable model substantially improves discrimination beyond EFW alone, indicating that additional maternal and clinical factors contribute meaningful incremental predictive value (Table 4).

Table 4.

Incremental predictive value beyond estimated fetal weight.

Model	AUC	Sensitivity	Specificity	Pos pred value	Neg pred value
Full model (with EFW)	0.959	0.816	0.969	0.912	0.930
EFW <2500 g only model	0.836	0.697	0.974	0.914	0.891
Model WITHOUT EFW	0.846	0.566	0.969	0.878	0.850

Bootstrap based internal validation for the prediction of low birth weight

Internal validation was done using bootstrapping method, 1000 iterations were performed to avoid over optimistic results and to make our model applicable and capture the unseen data. The bootstrap resampling analysis result showed that original predictive performance of the model was optimistic and decreased after correction for over fitting. After bootstrap validation, the discrimination ability of the model was reduced. The bias-corrected discrimination index (Dxy) declined from its original value of 0.917 to 0.887, indicating some loss of concordance between predicted and observed outcomes due to over fitting. Correspondingly, the bias-corrected AUC decreased from 95.9% in the original model to approximately 94.3% after internal validation, still reflecting very good discrimination. The explained variance (Nagelkerke R²) was also reduced from 0.760 in the apparent model to 0.703 after bootstrap correction, suggesting that the model explains slightly less variability in the outcome when accounting for optimism. The Brier score also increased from 0.064 (original model) to 0.078 after internal validation, indicating slightly reduced overall accuracy of predicted probabilities but remaining within an acceptable range for clinical prediction models. The optimism-corrected calibration slope after bootstrap validation was 0.84, indicating moderate overfitting and some shrinkage of predictor effects after internal validation. Calibration-in-the-large was −0.11, suggesting minimal systematic overestimation of predicted probabilities. The calibration plot of the validated model (Figure 6) clearly indicates the results after bootstrap are in good range. Taken together, internal validation indicated that the model retained good discrimination and moderate calibration, though the bootstrap optimism-corrected performance metrics reflect meaningful over fitting in the apparent model (Table 5).

Figure 6.

Bootstrap plot bias corrected calibration curve for low birth weight for the internally validated model.

Table 5.

Bootstrap bias-corrected coefficients with 95% CI (1000 resamples) for low birth weight data.

Variable	Beta (bootstrap)	Lower 95% CI	Upper 95% CI
Intercept	13.595095234	5.864445903	21.325744565
Maternal age <20 years	1.0105264	-2.148791016	4.169843817
Maternal age 20-34 years	-0.063832054	-1.714480051	1.586815943
Gestational age Preterm	1.796168128	-1.597218961	5.189555216
Estimated Fetal weight	-0.005967325	-0.009218641	-0.002716009
Timing estimation = Intrapartum	1.406564094	-0.118043768	2.931171956
Time interval 3-7 days	0.678114432	-0.73449325	2.090722115
Posterior placenta location	1.106547459	-0.432285799	2.645380716
Professional = Consultant	1.46941134	-1.741626025	4.680448705
Primary education	-0.744690344	-2.856323642	1.366942953
Secondary education	0.792029261	-0.771747569	2.355806091
College and above	-0.797767051	-2.822606859	1.227072758

Decision curve analysis (DCA)

The Decision Curve Analysis (DCA) was evaluated for our model to show the clinical utility and net benefit of prediction model for low birth weight (LBW). The finding illustrated that clinically meaningful threshold probabilities ranged from 10% to 30%. The decision curve analysis demonstrated that the full model provided higher net benefits across a wide range of threshold probabilities compared to the “treat all” or “treat none” strategies. The graph illustrates that at threshold of 0.1 probabilities the net benefit for the full model was 0.895, compared to 0.718 for treating all and 0 for treating none. Moreover, at 0.3 probabilities the net benefit of the full model was about 0.78, whereas the net benefit of for treating all and for treating none was 0. This suggests that using the full model for decision-making improves clinical utility by reducing unnecessary interventions while identifying true low birth weight cases (Figure 7).

Figure 7.

A plot of decision curve analysis illustrating the standard net benefit of the model against threshold probability and corresponding cost-benefit ratio of low birth weight model.

Risk score model to predict low birth weight

To construct a clinically usable risk score, the regression coefficients from the final multivariable model were rescaled. While the traditional approach divides all β coefficients by the smallest significant β, this produced excessively large point totals that were less practical for bedside use. Therefore, we applied a scaling factor of 0.2 to generate concise, interpretable integer point values while maintaining the relative weighting across predictors. Then after each predictor was assigned points based on its relative weight to predict low birth weight. Then for each individual, the risk score was computed as the sum of assigned points corresponding to their risk factors. Next, this total risk score treated as predictor, and risk regression model was developed. Then the accuracy of the model was evaluated using ROC with its AUC and calibration plot with its p-value. The risk score model has obtained an AUC of 76.8%, 95%CI: 70.2-83.0) showing an acceptable range of discrimination for decision making (Figure 8). The calibration plot demonstrated good agreement between predicted and observed probabilities, with no evidence of miscalibration (p = 0.301, GIVITI calibration belt) (Figure 9). Then risk categorization technique was done based on the optimal cut point of the predicted probability (threshold probability 0.22). These thresholds were derived from the predicted probability distribution observed within the development dataset and selected to enhance practical clinical interpretability and bedside applicability; therefore, they should be interpreted cautiously until externally validated. Less than 0.22 (low risk), from 0.22 to 0.75 (medium risk), and above 0.75 (high risk) a corresponding scores of risk points to these threshold probability of identified risk points table was developed. Based on the risk score classification, when a woman has less than 11 score, she can have 3.2% to 17.5% probability of lbw which is low chance a chance of developing low birth weight baby, implying that routine care may be sufficient. A woman having score from 11 to 21 corresponds to chance of low birth weight is 20.4% to 62% which is medium risk, should be monitored more closely, as their probability approaches 1 in 2. Moreover, woman with score of 22 to 36 points has a chance of 66.2% to 96.3% of lbw which is high risk, identifies pregnancies where more than two in three are predicted to result in LBW, warranting targeted interventions and intensive follow-up (Table 6).

Figure 8.

Receiver operating characteristics curve (ROC) for risk score model to predict low birth weight.

Figure 9.

Model calibration for the risk model for predicting low birth weight.

Table 6.

Risk classification based on risk score and predicted probability for low birth weight.

Risk group	Total score range	Predicted probability
Low risk	0-10	0.032-0.175
Medium risk	11-21	0.204-0.62
High risk	22-36	0.662-0.963

In addition, risk factor, points table for an easy use was developed, and risk group chart, enabling clinicians to quickly classify patients was prepared (Table 7). Moreover, risk curve graph was developed and illustrated a monotonic increase in LBW probability as the total risk score increased. Shaded zones (green, yellow, red) highlighted the probability for easier interpretation (Figure 10).

Table 7.

Friendly low birth weight risk score sheet.

Risk factor	Beta coefficient	Score points	Tick
Maternal age less than 20 years	1.192	5	☐
Gestational age Preterm	0.755	9	☐
Time of estimation_Intrapartum	2.015	7	☐
Time interval 3- 7 days	0.732	3	☐
Posterior placenta location	1.469	6	☐
Senior Professional involved	2.298	7	☐
No formal educ_level	0.912	4	☐
Secondary Educ_level	1.655	8	☐

Figure 10.

Risk curve with shaded risk zones for predicting low birth weight.

Examples how to use this risk score model; assume that a 19-year-old primigravida at 37 weeks gestation presented for assessment during early labor. An intrapartum ultrasound was performed while the cervix was 4 cm dilated. The placenta was posterior, the ultrasound occurred within 3–7 days of delivery, and the scan was performed by a senior clinician. She also had no formal education. Based on these characteristics, the patient accumulated 5 points for maternal age, 7 points for intrapartum scanning, 3 points for the short scan-to-delivery interval, 6 points for posterior placentation, 7 points for senior professional involvement, and 4 points due to absence of formal education resulting in a total score of 32. This score corresponds to a predicted probability of low birth weight between 20.4% and 62%, placing her in the medium-risk category. Women in this group require closer surveillance, timely decision-making, and preparation for possible neonatal complications.

Another example how to use the model is a 22-year-old primigravida who presented in active labor at 34 weeks of gestation. The placenta was posterior, the ultrasound was performed intrapartum, the scan-to-delivery interval was between 3 and 7 days, and the examination was conducted by a senior clinician. She had no formal education. In this scenario, the patient accumulated 9 points for preterm gestation, 7 points for intrapartum ultrasound, 3 points for the short scan-to-delivery interval, 6 points for posterior placenta location, 7 points for assessment by a senior professional, and 4 points for lack of formal education, giving a total score of 36. This score corresponded to a predicted probability of low birth weight between 66.2% and 96.3%, categorizing her as high-risk. Such patients require intensive monitoring, preparedness for neonatal resuscitation, and early involvement of specialized neonatal care.

Agreement between the original prediction model and the simplified risk score

To evaluate the consistency between the original prediction model and the simplified integer-based risk score, both categorical and continuous agreement were evaluated. Categorical agreement across the three risk groups (low, moderate, high) yielded a Cohen’s Kappa value of 0.353 (p < 0.0001). Cohen’s Kappa of 0.353 indicates only fair agreement between the original model and the simplified risk score, suggesting that substantial individual-level misclassification remained after model simplification despite preservation of acceptable overall discrimination. This reflects the expected loss of information resulting from coefficient scaling and conversion of a continuous prediction model into a simplified integer-based bedside scoring system.

A Bland–Altman plot between predicted probabilities of the original model and risk score model was drawn, the plot demonstrated a mean bias close to zero with acceptable limits of agreement, indicating that the score-derived predicted probabilities reasonably approximate those produced by the full original model (Figure 11).

Figure 11.

Bland Altman plot agreement between predicted probabilities of original model and risk score model.

Discussion

Accurate estimation of fetal weight using prenatal ultrasound plays a vital role in optimizing antenatal care and guiding timely interventions, thereby improving the management of high-risk pregnancies and reducing perinatal complications. The use of prenatal ultrasound is expanding globally, including in low-income countries.⁴² Early identification and stratification of high-risk pregnancies are essential for minimizing adverse outcomes and the development of predictive models and producing easy to use tools such as risk scores can support clinical decision-making.⁴³

This study aimed to assess the accuracy of ultrasound estimated fetal weight in relation to actual birth weight, and develop risk prediction models for low birth weight based on maternal, pregnancy, and fetal parameters at TGSH. Accordingly in this study, the accuracy of the ultrasound in estimating fetal weight using Hadlock III formula as compared to the actual birth weight was 72.9 %. This is in line with a study conducted in Nigeria and Portugal where the Hadlock formula demonstrated an accuracy of 82.3% and 78% respectively.^44,45 However, as compared to studies done in Italy, Nigeria, Sudan and Pakistan there is a better accuracy in this study. Moreover, the accuracy in the current study is slightly lower than previous studies conducted in New Zealand, India and Turkey.^11,15,24

The difference in accuracy is acceptable as there are differences in level of experience for sonographers, in our study 89% of estimated fatal weights were performed by resident doctors who are on training that might decrease the level of accuracy. In addition, type of instrument and the formula used for fetal weight calculation can contribute to the variations.⁴⁶ Moreover, lower accuracy of ultrasound-based fetal weight estimation in our study (72.9%) can be explained by the high burden of preterm birth in Amhara region, where our study was conducted. Multiple studies have showed that the prevalence of preterm birth in Amhara region is substantially higher than global averages, where studies from the region report preterm birth rates ranging from 11.4% to 15.1%.^47,48

In our sample, 18.2% of births occurred before 37 weeks of gestation. Because ultrasound fetal weight estimation formulas such as Hadlock were calibrated primarily in term or near-term fetuses in high-income populations, accuracy tends to decrease among smaller fetuses and in populations with high rates of growth restriction or prematurity. This is consistent with evidence showing that fetal biometry becomes less precise as gestational age decreases, due to greater biological variability and difficulty in obtaining standard measurement planes in preterm fetuses.^49,50

The weight estimation inaccuracies were positive mean percentage error (3.9%) and with bias of 73.5gms which implied that ultrasound fetal weight estimations were overestimated. The absolute percentage error and mean absolute error were 8.3% and 210.97gm respectively. The mean absolute error of this study is consistent with studies done in Tunisia in which the mean absolute error was found to be 200gm. The mean absolute percentage error was almost similar from the study in Pakistan and India where mean absolute percentage error was 8.2%. Similar study from Germany showed that the absolute mean error of 261.5gm which is almost similar to our study.^24,51–53 In this study, the estimated fetal weight was highly correlated with actual birth weight with a high correlation coefficient which is incongruent with a study done in Germany and Nigeria.^4,53

Although the Bland–Altman analysis demonstrated relatively small overall systematic bias between ultrasound-estimated fetal weight and actual birth weight, the relatively wide limits of agreement and the presence of several outlying observations indicate that clinically important individual-level discrepancies remain possible.

Greater variability was observed particularly among preterm pregnancies and among ultrasounds performed more than 3 days before delivery, suggesting that gestational maturity and timing between ultrasound assessment and delivery may influence estimation precision. Similar observations have been reported in previous studies.⁴⁹ Our finding that greater variability in ultrasound fetal weight estimation was observed among preterm pregnancies and among scans performed more than 3 days before delivery is therefore consistent with existing evidence.^49,54 Several studies have reported that the accuracy of ultrasound-estimated fetal weight tends to decrease with increasing time interval between ultrasound examination and delivery because continued fetal growth after assessment may widen the discrepancy between estimated and actual birth weight.⁵⁴ Similarly, preterm pregnancies are often associated with altered fetal growth patterns, reduced soft tissue mass and technical challenges in biometric measurements, all of which may increase estimation variability.⁵⁵ A systematic review by Dudley reported that the precision of sonographic fetal weight estimation is influenced by gestational age and timing of assessment, with wider limits of agreement observed in some preterm pregnancies and when the interval between ultrasound examination and delivery increases.⁴⁹ Likewise, Colman et al. demonstrated that shorter scan-to-delivery intervals improve the accuracy of ultrasound fetal weight estimation, particularly near term40. These findings support our observation that gestational maturity and the timing of ultrasound assessment may substantially influence fetal weight estimation precision.¹²

In the case of this study, clinically, these findings imply that while ultrasound-estimated fetal weight may perform reasonably well for population-level screening and risk stratification. Caution is required when making individualized clinical decisions near the low birth weight diagnostic threshold, where modest estimation errors may alter classification and subsequent management.⁴⁹

In our risk prediction model for low birth weight, we entered eight prognostic determinants that were retained from LASSO regression. The predictors were maternal age, gestational age at delivery, estimated fetal weight, timing estimation, time interval, location of placenta, professional involved in the estimation, and maternal educational level. This model achieved an AUC of 95.9%, and calibration plot of p-value 0.371; this model can effectively distinguishes between LBW and normal birth weight cases, and the calibration plot suggests a good agreement between observed and predicted probabilities. This level of accuracy and calibration is crucial for early intervention and management of potential LBW cases, which are associated with increased neonatal morbidity and mortality. Similar studies have reported comparable levels of model accuracy; including a study utilized a super learner ensemble approach incorporating maternal nutritional consumption produced an AUC of 96%.⁵⁶ And another study data-mining model using decision trees and SMOTE for handling imbalanced datasets have demonstrated high AUC values achieving an AUC of 90.3%.⁵⁷

An important methodological consideration in this study is the role of estimated fetal weight (EFW) within the prediction model, given its strong biological and statistical relationship with actual birth weight. To address this, we conducted additional analyses comparing the full model with both an EFW-only approach and a model excluding EFW.

The EFW-only model demonstrated good discrimination (AUC = 84%), but with lower sensitivity (69.7%), indicating that a substantial proportion of low birth weight cases would be missed when relying solely on a fixed EFW threshold. In contrast, the multivariable model substantially improved discrimination (AUC = 95.9%) and increased sensitivity to 81.6%, suggesting enhanced ability to identify at-risk pregnancies. Furthermore, a model excluding EFW retained acceptable discrimination (AUC ≈ 85%), highlighting the independent predictive contribution of maternal, obstetric, and contextual variables. Together, these findings indicate that the proposed model is not merely a restatement of fetal weight estimation, but rather integrates complementary sources of information to achieve superior predictive performance, particularly in identifying borderline or clinically ambiguous cases.

To assess model stability and mitigate optimism bias, we performed 1000 bootstrap iterations; a gold-standard approach for internal validation in predictive modeling.⁵⁸ Although the overall predictive performance remained strong after bootstrap validation, several bootstrap-corrected 95% confidence intervals for individual predictors crossed zero, suggesting sensitivity of some coefficient estimates to sampling variability. In contrast, estimated fetal weight remained consistently associated with low birth weight after optimism correction, indicating greater stability of this predictor. These findings suggest that while several predictors contribute collectively to model performance, some individual effects should be interpreted with caution given the relatively limited number of events available for model development. Moreover, the validated model retained a high AUC of 94.3%, only slightly reduced from the original 95.9%, confirming excellent discriminative ability after optimism correction.

We performed decision curve analysis for our model; it provided greater net benefit than the “treat all” or “treat none” strategies across a wide range of threshold probabilities, demonstrating its clinical utility in improving identification of low birth weight cases while minimizing unnecessary interventions. Furthermore, we developed a simplified risk score for predicting low birth weight (LBW) by converting regression coefficients into integer points after scaling and rounding. The total score for each woman, obtained by summing assigned points for her risk factors, was used as a predictor in a risk regression model. The risk score model demonstrated acceptable discrimination (AUC = 76.8%, 95% CI: 70.2–83.0). The drop in the discrimination performance AUC 95.9% in the original model vs. AUC 76.8% in risk score model occurred because of the computational process in changing regression coefficients into score points, and then employing a regression model for the point score resulted in the loss of accuracy. This trade-off is well-recognized in clinical prediction modeling, where simplified tools aim to improve usability at the expense of some predictive precision.³⁹ In addition, the risk score model got very good calibration (p = 0.301), indicating agreement between predicted and observed outcomes. Based on an optimal probability threshold of 0.22, the study subjects were stratified into low (<11 points), medium (11–21 points), and high (22–36 points) risk groups, corresponding to increasing probabilities of LBW from 3.2% up to 96.3%. A user-friendly points table, classification chart, and risk curve were developed to facilitate clinical application and enable rapid identification of women at higher risk who may benefit from closer monitoring or targeted interventions. Although the simplified risk score maintained acceptable discrimination, the relatively low Cohen’s Kappa value indicates that agreement with the original model was only fair, implying that some patients may be classified differently at the individual level after score simplification. Therefore, the simplified score should primarily be considered a pragmatic bedside risk stratification tool rather than a direct substitute for the full multivariable prediction model.

Compared with traditional international low-birth-weight prediction models that rely primarily on maternal Sociodemographic characteristics and gestational age, the present model incorporates a richer set of clinically relevant predictors selected through LASSO regression.^59,60 Our model integrates both maternal and fetal domains; including ultrasound estimated fetal weight, placental location, timing of assessment, scan-to-delivery interval, and operator level. These variables reflect real-time fetal growth, placental function, and measurement context, enabling a more precise estimation of LBW risk. The strong performance of our model AUC of 94.3% after bootstrap correction and good calibration demonstrates its clear incremental value over simpler models that lack fetal biometric inputs.

Furthermore, unlike earlier LBW prediction tools derived solely from antenatal characteristics, our model benefits from the predictive power of ultrasound-based fetal weight estimation and incorporates methodological rigor through penalization and bootstrap validation. The inclusion of EFW and placental characteristics allows the model to account for population specific fetal growth patterns, which are particularly important in Ethiopian settings where the prevalence of preterm birth, fetal growth restriction, and placental abnormalities is higher than in many high-income regions.^48,61

The use of LASSO regression ensures that only the most predictive and non-redundant variables were retained, minimizing overfitting and improving generalizability.⁶² This contrasts with traditional models that often lack formal variable selection procedures and rely on subjective inclusion of predictors. Additionally, the development of a simplified integer-based risk-score tool extends the model’s practical utility by enabling frontline clinicians to estimate LBW risk quickly without computational support. Overall, the combination of rigorous statistical modeling, inclusion of fetal and intrapartum factors, and translation into a practical bedside tool distinguishes our model from prior LBW prediction models and provides meaningful clinical advantage in resource-limited settings.

Despite several strengths, this study has some limitations. First, because secondary data were used, some potentially important predictors were unavailable for analysis. Second, the study was conducted at a single tertiary care center, which may limit the generalizability of the findings to other regions of Ethiopia and to other low-income settings with different patient characteristics, clinical practices, and healthcare resources. Future studies should therefore include multicenter and more heterogeneous populations to externally validate the model across different settings.

The use of secondary data also limited the availability of detailed operator-level information, such as years of experience, number of ultrasound examinations performed, and certification status, all of which may influence the accuracy of fetal weight estimation. In addition, although estimated fetal weight (EFW) was an important predictor, its inclusion in the model may introduce partial dependency with the outcome. However, comparative analyses demonstrated that the model provides meaningful predictive improvement beyond EFW alone.

Another limitation relates to the relatively low events-per-variable (EPV) ratio during model development. Although penalized regression using LASSO and bootstrap internal validation were applied to reduce overfitting, the limited EPV may still affect model stability and contribute to optimism in predictive performance estimates. Recent methodological guidance has highlighted that low EPV may increase instability and optimism in prediction models, particularly when predictor-selection procedures are used, reinforcing the need for external validation in independent populations.^36,37 Consistent with this concern, several bootstrap-corrected confidence intervals crossed zero, suggesting that some individual predictor effects were sensitive to sampling variability despite the model maintaining strong overall predictive performance.

A further limitation relates to the internal validation strategy. Predictor selection using LASSO regression was performed before bootstrap validation rather than being repeated within each bootstrap replicate. Consequently, variable-selection uncertainty was not fully propagated during optimism estimation, and the predictive performance estimates may therefore remain somewhat optimistic. Similarly, because the simplified risk-score thresholds were derived and evaluated within the development dataset, their performance characteristics may differ when applied in independent populations. Therefore, external validation in larger, independent, and multicenter populations is required before widespread clinical implementation of either the full prediction model or the simplified risk-score tool.

Conclusion

The emphasis of this study was the clinical importance of birth weight estimation using prenatal ultrasound in resource limited settings to facilitate timely identification of risk pregnancies for low birth weight to improve perinatal outcomes. The Hadlock III formula revealed acceptable accuracy in estimating fetal weight, with performance metrics comparable to international benchmarks. In this study, we developed and internally validated a multivariable prediction model for low birth weight (LBW) using routinely collected clinical and ultrasound data. We identified eight predictors using LASSO based data-driven variable selection and penalized logistic regression. Our model demonstrated excellent apparent discrimination (AUC 95.9%) and good calibration (p-value = 0.371), and retained very good performance after bootstrap correction (AUC 0.94, calibration slope 0.79) affirming its utility in stratifying the risk of low birth weight.

We translated the model into an integer-based risk score that provides rapid bedside estimation of LBW risk to enhance clinical use. Based on the risk score for practical use we stratified pregnant women into low, medium, and high-risk groups, thereby facilitating timely decision making activities by healthcare providers. The developed model demonstrated promising predictive performance and may support risk stratification of low birth weight in similar resource-limited settings. However, given the single-center nature of the study, the reliance on secondary data and the absence of external validation, multicenter external validation is required before broader clinical implementation.

Supplemental material

Supplemental material - Diagnostic accuracy of ultrasound-based fetal weight estimation and prediction modeling for low birth weight: A retrospective cohort study in Northwest Ethiopia

Supplemental material for Diagnostic accuracy of ultrasound-based fetal weight estimation and prediction modeling for low birth weight: A retrospective cohort study in Northwest Ethiopia by Amsalu Worku Mekonin, Sintayehu Debas, Haile Mekonnen Fenta, Yihun Miskir Wubie, Asres Zegeye, Zelalem Alamrew Anteneh in Science Progress

Supplemental material

Supplemental material - Diagnostic accuracy of ultrasound-based fetal weight estimation and prediction modeling for low birth weight: A retrospective cohort study in Northwest Ethiopia

Footnotes

Acknowledgment

We extend our sincere gratitude to Bahir Dar University for granting ethical clearance to conduct this study. We are also deeply thankful to the administration office of Tibebe Ghion Specialized Hospital for providing the official permission letter. Our appreciation further goes to the data collectors and Card Room staff members for their invaluable cooperation and support throughout the data collection process. We acknowledge the use of Microsoft Copilot to improve the clarity, grammar, and readability of the manuscript and to identify typographical errors. This tool was not used for data analysis, statistical modeling, interpretation of findings, or any aspect of study design.

ORCID iD

Zelalem Alamrew Anteneh

Author contributions

AWM- conceptualized the study, designed the study, supervised the data collection, performed data analysis, interpreted the findings, drafted the manuscript, and approved the manuscript.

SD- involved designing the study, validated data analysis, interpreted the findings, edited and reviewed the manuscript.

HMF- involved data analysis, interpreted the findings, and reviewed the manuscript.

YMW- involved in the data analysis, interpreted the findings, and drafted the manuscript.

AZ- validation of data analysis, interpreted the findings, and reviewed the manuscript.

ZAA- conceptualized the study, designed the study, supervised the data collection, performed data analysis, interpreted the findings, drafted the manuscript, and approved the manuscript.

All authors had thoroughly read and approved the manuscript.

Funding

The authors received no financial support for the research, authorship, and/or publication of this article.

Declaration of conflicting interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Data Availability Statement

Anonymized full dataset underlying the study is available within the paper and its Supporting information files.*

Supplemental material

Supplemental material for this article is available online.

References

Mirghani

Weerasinghe

Ezimokhai

, et al. Ultrasonic estimation of fetal weight at term: An evaluation of eight formulae. The journal of obstetrics and gynaecology research 2005; 31: 409–413. https://doi.org/10.1111/j.1447-0756.2005.00311.x

Pario

Pasha

Anwar

, et al. Actual Birth Weight Versus Sonographic Estimation of Fetal Weight.

Preyer

Husslein

Concin

, et al. Fetal weight estimation at term - ultrasound versus clinical examination with Leopold's manoeuvres: a prospective blinded observational study. BMC Pregnancy Childbirth 2019; 19: 122. https://doi.org/10.1186/s12884-019-2251-5

Eze

Ohagwu

Abonyi

, et al. Reliability of sonographic estimation of fetal weight: A study of three tertiary hospitals in Nigeria. Saudi journal of medicine & medical sciences 2017; 5: 38–44. https://doi.org/10.4103/1658-631X.194256

Parvathavarthini

Santhanalakshmi

Prasad

. Comparative study of various methods of fetal weight estimation at term pregnancy in a tertiary hospital in Kanchipuram, Tamil Nadu, India. International Journal of Reproduction, Contraception, Obstetrics and Gynecology 2018; 7: 1602–1608.

Eze

Abonyi

Njoku

, et al. Correlation of ultrasonographic estimated fetal weight with actual birth weight in a tertiary hospital in Lagos, Nigeria. African health sciences 2015; 15: 1112–1122. https://doi.org/10.4314/ahs.v15i4.9

Simms-Stewart

Hunter

Fletcher

, et al. Comparison of ultrasonographic estimated fetal weight and actual birthweight performed by residents in training at the university hospital of the west indies. The West Indian medical journal 2013; 62: 829–832. https://doi.org/10.7727/wimj.2012.265

Gurung

Shrestha

Gauchan

, et al. Comparison of Actual Birth Weight with the Ultrasonographic and Clinical Estimation of Fetal Birth Weight: A Prospective Study. Nepalese Journal of Radiology 2022; 12: 8–12. https://doi.org/10.3126/njr.v12i1.46796

Rosen

Gold-Zamir

Lopian

, et al. Accuracy of sonographic fetal weight estimation and prediction of birth-weight discordance in twin pregnancy: large single-center study. Ultrasound in obstetrics & gynecology : the official journal of the International Society of Ultrasound in Obstetrics and Gynecology 2023; 62: 821–828. https://doi.org/10.1002/uog.26277

10.

Lindström

Cnattingius

Axelsson

, et al. Accuracy and precision of sonographic fetal weight estimation in Sweden. Acta Obstetricia et Gynecologica Scandinavica 2023; 102: 699–707. https://doi.org/10.1111/aogs.14554

11.

Benson‐Cooper

Tarr

Kelly

, et al. Accuracy of ultrasound in estimating fetal weight in New Zealand. Australasian Journal of Ultrasound in Medicine 2021; 24: 13–19. https://doi.org/10.1002/ajum.12239

12.

Colman

Maharaj

Hutton

, et al. Reliability of ultrasound estimation of fetal weight in term singleton pregnancies. The New Zealand Medical Journal (Online) 2006; 119: U2146.

13.

Ridha

Bergin

Kelly

, et al. Accuracy of ultrasound in the estimation of customised birth weight in a public hospital service. Journal of Medical Imaging and Radiation Oncology 2022; 66: 1044–1051. https://doi.org/10.1111/1754-9485.13408

14.

Acharya

Tiwari

. Accuracy of prediction of birth weight by fetal ultrasound. Journal of KIST Medical College 2020; 2: 11–14. https://doi.org/10.3126/jkistmc.v2i2.33566

15.

Konwar

Basumatary

Dutta

, et al. Accuracy of fetal weight estimation by ultrasonographic evaluation in a northeastern region of India. International Journal of Biomaterials 2021; 2021: 9090338. https://doi.org/10.1155/2021/9090338

16.

Milner

Arezina

. The accuracy of ultrasound estimation of fetal weight in comparison to birth weight: A systematic review. Ultrasound 2018; 26: 32–41. https://doi.org/10.1177/1742271X17732807

17.

Barel

Vaknin

Tovbin

, et al. Assessment of the accuracy of multiple sonographic fetal weight estimation formulas: a 10‐year experience from a single center. Journal of Ultrasound in Medicine 2013; 32: 815–823. https://doi.org/10.7863/ultra.32.5.815

18.

Sereke

Omara

Bongomin

, et al. Prospective verification of sonographic fetal weight estimators among term parturients in Uganda. BMC Pregnancy and Childbirth 2021; 21: 175. https://doi.org/10.1186/s12884-021-03645-4

19.

Ashrafganjooei Tn

Eshrati

Babapoor

. Accuracy of ultrasound, clinical and maternal estimates of birth weight in term women. Eastern Mediterranean Health Journal 2010; 16(3): 16.

20.

Aksoy

Karadağ

Öİ

, et al. Influence of maternal body mass index on sonographic fetal weight estimation prior to scheduled delivery. Journal of Obstetrics and Gynaecology Research 2015; 41: 1556–1561. https://doi.org/10.1111/jog.12755

21.

Mohamed Adam

Tambul

Mohamed

YMY

, et al. The accuracy of ultrasound in estimation of fetal weight, 2013.

22.

Paganelli

Soncini

Comitini

, et al. Sonographic fetal weight estimation in normal and overweight/obese healthy term pregnant women by gestation-adjusted projection (GAP) method. Archives of gynecology and obstetrics 2016; 293: 775–781. https://doi.org/10.1007/s00404-015-3910-z

23.

Krispin

Dreyfuss

Fischer

, et al. Significant deviations in sonographic fetal weight estimation: causes and implications. Archives of Gynecology and Obstetrics 2020; 302: 1339–1344. https://doi.org/10.1007/s00404-020-05732-x

24.

Tas

Kir

Yilmaz

, et al. Accuracy of sonographic fetal weight estimation in full-term singleton pregnant women. Pakistan Journal of Medical Sciences 2019; 35: 34–38. https://doi.org/10.12669/pjms.35.1.373

25.

Heer

Kümper

Vögtle

, et al. Analysis of factors influencing the ultrasonic fetal weight estimation. Fetal diagnosis and therapy 2008; 23: 204–210. https://doi.org/10.1159/000116742

26.

Rutherford

Phelan

Smith

, et al. The four-quadrant assessment of amniotic fluid volume: an adjunct to antepartum fetal heart rate testing. Obstetrics & Gynecology 1987; 70: 353–356.

27.

Shmueli

Aviram

Bardin

, et al. Effect of fetal presentation on sonographic estimation of fetal weight according to different formulas. International Journal of Gynecology & Obstetrics 2017; 137: 234–240. https://doi.org/10.1002/ijgo.12152

28.

Tuten

Guralp

Gok

, et al. Factors Affecting Ultrasonographic Fetal Weight Estimation Accuracy in Low Birth Weight Newborns. Gynecology Obstetrics & Reproductive Medicine 2022; 28: 23–30.

29.

Dudley

. A review of ultrasound fetal weight estimation in the early prediction of low birthweight. Ultrasound 2013; 21: 181–186. https://doi.org/10.1177/1742271x13505259

30.

Ashwal

Hiersch

Melamed

, et al.

Does the level of amniotic fluid have an effect on the accuracy of sonographic estimated fetal weight at term?

The Journal of Maternal-Fetal & Neonatal Medicine 2015; 28: 638–642. https://doi.org/10.3109/14767058.2014.929113

31.

Sanchez-Martinez

Marti-Castellote

Hoodbhoy

, et al. Prediction of low birth weight from fetal ultrasound and clinical characteristics: a comparative study between a low- and middle-income and a high-income country. BMJ global health 2024; 9: e016088. https://doi.org/10.1136/bmjgh-2024-016088

32.

World Health Organization . Low birth weight. https://www.who.int/data/nutrition/nlis/info/low-birth-weight (accessed 1 April 2026).

33.

Hiwale

. A Systematic Evaluation of Ultrasound-based Fetal Weight Estimation Models on Indian Population. J Med Ultrasound 2017; 25: 201–207. https://doi.org/10.1016/j.jmu.2017.07.001

34.

Khalil

Sotiriadis

D'Antonio

, et al. ISUOG Practice Guidelines: performance of third-trimester obstetric ultrasound scan. Ultrasound in obstetrics & gynecology : the official journal of the International Society of Ultrasound in Obstetrics and Gynecology 2024; 63: 131–147. https://doi.org/10.1002/uog.27538

35.

World Health Organization . Imaging ultrasound before 24 weeks of pregnancy: WHO antenatal care recommendations for a positive pregnancy experience. technical brief. 22, 2022.

36.

Riley

Ensor

Snell

KIE

, et al. Calculating the sample size required for developing a clinical prediction model. BMJ (Clinical research ed) 2020; 368: m441. https://doi.org/10.1136/bmj.m441

37.

van Smeden

Moons

de Groot

, et al. Sample size for binary logistic prediction models: Beyond events per variable criteria. Statistical methods in medical research 2019; 28: 2455–2474. https://doi.org/10.1177/0962280218784726

38.

Harrell

Jr . Regression Modeling Strategies: With Applications to Linear Models, Logistic Regression, and Survival Analysis. 2nd ed. Springer, 2015.

39.

Steyerberg

Harrell

Jr Borsboom

, et al. Internal validation of predictive models: efficiency of some procedures for logistic regression analysis. Journal of clinical epidemiology 2001; 54: 774–781. https://doi.org/10.1016/s0895-4356(01)00341-9

40.

Moons

Altman

Reitsma

, et al. Transparent Reporting of a multivariable prediction model for Individual Prognosis or Diagnosis (TRIPOD): explanation and elaboration. Annals of internal medicine 2015; 162: W1–W73. https://doi.org/10.7326/m14-0698

41.

von

Altman

Egger

, et al. The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement: guidelines for reporting observational studies. Journal of clinical epidemiology 2008; 61: 344–349. https://doi.org/10.1016/j.jclinepi.2007.11.008

42.

Stewart

Navarro

Kambala

, et al. Trends in ultrasound use in low and middle income countries: a systematic review. International Journal of Maternal and Child Health and AIDS 2020; 9: 103–120. https://doi.org/10.21106/ijma.294

43.

Heestermans

Payne

Kayode

, et al. Prognostic models for adverse pregnancy outcomes in low-income and middle-income countries: a systematic review. BMJ global health 2019; 4: e001759. https://doi.org/10.1136/bmjgh-2019-001759

44.

Ezeugo

Agboghoroma

Jimoh

. Comparison of clinical and ultrasonographic estimation of foetal weight at term and their correlation with birth weight. African journal of reproductive health 2021; 25: 108–117. https://doi.org/10.29063/ajrh2021/v25i4.12

45.

Veríssimo

CRI

. Estimativa ponderal fetal ecográfica na gravidez de termo, 2009.

46.

Seman

Adem

Disasa

, et al. Development of birth weight estimation model for Ethiopian population from sonographic evaluation. BMC Pregnancy and Childbirth 2023; 23: 850. https://doi.org/10.1186/s12884-023-06145-9

47.

Adugna

. Prevalence and associated risk factors of preterm birth among neonates in referral hospitals of Amhara Region, Ethiopia. PloS one 2022; 17: e0276793. https://doi.org/10.1371/journal.pone.0276793

48.

Chan

Goddard

FGB

Hunegnaw

, et al. Estimates of Stillbirths, Neonatal Mortality, and Medically Vulnerable Live Births in Amhara, Ethiopia. JAMA Network Open 2022; 5: e2218534. https://doi.org/10.1001/jamanetworkopen.2022.18534%JJAMANetworkOpen

49.

Dudley

. A systematic review of the ultrasound estimation of fetal weight. Ultrasound in obstetrics & gynecology : the official journal of the International Society of Ultrasound in Obstetrics and Gynecology 2005; 25: 80–89. https://doi.org/10.1002/uog.1751

50.

Ioannou

Talbot

Ohuma

, et al. Systematic review of methodology used in ultrasound studies aimed at creating charts of fetal size. BJOG: an international journal of obstetrics and gynaecology 2012; 119: 1425–1439. https://doi.org/10.1111/j.1471-0528.2012.03451.x

51.

Dimassi

Douik

Ajroudi

, et al.

Ultrasound fetal weight estimation: how accurate are we now under emergency conditions?

Ultrasound in Medicine & Biology 2015; 41: 2562–2566. https://doi.org/10.1016/j.ultrasmedbio.2015.05.020

52.

Hiwale

Firtion

. Analysis of factors influencing accuracy of ultrasound-based fetal weight estimation. Indian Journal of Radiology and Imaging 2020; 30: 156–162. https://doi.org/10.4103/ijri.IJRI_167_19

53.

Stubert

Peschel

Bolz

, et al. Accuracy of immediate antepartum ultrasound estimated fetal weight and its impact on mode of delivery and outcome-a cohort analysis. BMC pregnancy and childbirth 2018; 18: 1–8. https://doi.org/10.1186/s12884-018-1772-7

54.

Peregrine

O'Brien

Jauniaux

. Clinical and ultrasound estimation of birth weight prior to induction of labor at term. Ultrasound in obstetrics & gynecology : the official journal of the International Society of Ultrasound in Obstetrics and Gynecology 2007; 29: 304–309. https://doi.org/10.1002/uog.3949

55.

Elenis

Wikström

Simic

. Accelerated fetal growth in early pregnancy and risk of preterm birth: a prospective cohort study. BMC Pregnancy Childbirth 2020; 20: 764. https://doi.org/10.1186/s12884-020-03458-x

56.

Khadidos

Saleem

Selvarajan

, et al. Ensemble machine learning framework for predicting maternal health risk during pregnancy. Scientific reports 2024; 14: 21483. https://doi.org/10.1038/s41598-024-71934-x

57.

Hange

Selvaraj

Galani

, et al. A Data-Mining Model for Predicting Low Birth Weight with a High AUC. In: Lee

(ed) Computer and Information Science. Springer International Publishing, 2018, pp. 109–121.

58.

Harrell

Jr . Regression Modeling Strategies: With Applications to Linear Models, Logistic and Ordinal Regression, and Survival Analysis. Springer, 2015.

59.

Kramer

. Determinants of low birth weight: methodological assessment and meta-analysis. Bulletin of the World Health Organization 1987; 65: 663–737.

60.

McCowan

Thompson

Taylor

, et al. Prediction of Small for Gestational Age Infants in Healthy Nulliparous Women Using Clinical and Ultrasound Risk Factors Combined with Early Pregnancy Biomarkers. PloS one 2017; 12: e0169311. https://doi.org/10.1371/journal.pone.0169311

61.

Blencowe

Cousens

Oestergaard

, et al. National, regional, and worldwide estimates of preterm birth rates in the year 2010 with time trends since 1990 for selected countries: a systematic analysis and implications. Lancet (London, England) 2012; 379: 2162–2172. https://doi.org/10.1016/s0140-6736(12)60820-4

62.

Tibshirani

. Regression Shrinkage and Selection Via the Lasso. Journal of the Royal Statistical Society: Series B (Methodological) 2018; 58: 267–288. https://doi.org/10.1111/j.2517-6161.1996.tb02080

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.03 MB

0.22 MB