Construction of a prediction model for adverse perinatal outcomes in fetal growth restriction based on a machine learning algorithm

Construction of a prediction model for adverse perinatal outcomes in fetal growth restriction based on a machine learning algorithm | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Construction of a prediction model for adverse perinatal outcomes in fetal growth restriction based on a machine learning algorithm Xiangli Meng, Lei Wang, Minghui Wu, Na Zhang, Xiaofei Li, Qingqing Wu This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-5698332/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Objective The objective of this study was to create and validate a machine learning (ML)-based model for predicting the adverse perinatal outcome (APO) in fetuses with a perinatal diagnosis of fetal growth restriction (FGR). Method This was a retrospective study of singleton gestations meeting the ISUOG-endorsed criteria for FGR from January 2021 and November 2023 at Beijing Obstetrics and Gynecology Hospital. The APO comprised one or more of: perinatal demise (stillbirth, immediate neonatal demise, or death before neonatal intensive care unit) discharge, cord arterial pH ≤ 7.10, and/or base excess ≥ 12, bronchopulmonary dysplasia, hypoxic ischemic encephalopathy, grade III-IV intraventricular hemorrhage, periventricular leukomalacia, seizures, necrotizing enterocolitis, and sepsis. Feature screening was performed using the random forest (RF), the Least Absolute Shrinkage and Selection Operator (LASSO) and logistic regression (LR). Subsequently, six ML methods, including Stacking, were used to construct models to predict the APO of FGR. Model’s performance was evaluated through indicators such as the area under the receiver operating characteristic curve (AUROC). The Shapley Additive exPlanation analysis was used to rank each model feature and explain the final model. Finally, we constructed a nomogram to make the predictive model results more readable. Results In total, a cohort of 411 non-anomalous singleton pregnancies with FGR were divided into a training set and a test set at a ratio of approximately 7:3. Among 16 candidate predictors (including maternal characteristics, maternal comorbidities, obstetric characteristics, and ultrasound parameters), the integration of RF, LASSO, and LR methodologies identified maternal pre-pregnancy body mass index, hypertensive disorders of pregnancy, gestational age at diagnosis of FGR, estimated fetal weight (EFW) z-score, EFW growth velocity, and abnormal umbilical artery Doppler as significant predictors. The Stacking model demonstrated a good performance in the test set (AUROC: 0.861,95% confidence interval (0.838–0.896)). The calibration curve and Hosmer-Lemeshow test demonstrated good calibration. Conclusions The ML algorithm was developed to possess the promising capacity of predicting APO in FGR at time of diagnosis. This approach may potentially improve early detection at high risk of APO in FGR. small-for-gestational age growth restriction perinatal morbidity and mortality machine learning predictive model nomogram Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 1. Introduction Fetal growth restriction (FGR) is a common pregnancy complication. It is associated with an increased risk of perinatal mortality and morbidity and long-term adverse infant outcomes, including cognitive impairment in childhood and non-communicable diseases in adulthood, such as hypertension, metabolic syndrome, insulin resistance, type-2 diabetes mellitus, coronary heart disease, and stroke ( 1 ) . Numerous studies have investigated predictive markers for the development of FGR, unfortunately, almost all models were built using ultrasound measurements taken at the time point closest to delivery or confounded the definitions of FGR and small for gestational age (SGA) ( 2 – 6 ) , but far fewer have studied the prediction of pregnancy outcome when at time of diagnosis or counseling. The inability to predict pregnancy outcomes leaves pregnant patients and their partners with a considerable burden of uncertainty. It also limits clinicians' abilities to personalize management and counseling, including about termination of the pregnancy. The challenges encountered by both clinicians and pregnant women with FGR can vary greatly based on the GA at diagnosis. Patient-tailored prediction of fetal and neonatal outcomes is attractive from a clinical point and to indicate intrauterine transfer to a hospital with appropriate facilities, such as a neonatal intensive care unit ( 7 ) . Hence, prediction models containing antenatal available predictors for fetal and neonatal outcomes can facilitate clinical decision-making and counseling in pregnancies complicated by FGR. Recently, machine learning (ML) algorithms have been used in various fields for clinical practice such as FGR diagnosis ( 8 – 11 ) . Various robust ML prediction models have been developed for obstetric adverse event ( 12 ) and complications prediction ( 13 ) . Traditional prediction models were constructed using a generalized linear model, which depends on the underlying assumption that risk factors have a linear correlation with outcomes ( 6 , 14 ) . While these models offer the advantages of being easy to code and fast to calculate, it may oversimplify the complex nonlinear interaction between variables. ML methods have been an alternative method to overcome current limitations. The aim, therefore, of this study was to use ML tools to develop a prediction model for predicting APO in FGR at time of diagnosis. 2. Methods 2.1 Study Population We performed a retrospective analysis of a cohort of ultrasound-dated singleton fetuses with a prenatal diagnosis of FGR at Beijing Obstetrics and Gynecology Hospital (BJOGH) between January 2021 and November 2023. This study included 411 pregnancies with an antenatal diagnosis of FGR based on the criteria endorsed by the International Society of Ultrasound in Obstetrics and Gynecology (ISUOG) ( 15 ) . The inclusion criteria was: singleton pregnancies with accurate GA (estimated based on the last menstrual period, corrected by the crown-rump length (between 9 + 0 and 14 + 0 weeks' gestation) if a discrepancy of > 7 days was present ( 16 ) ). Pregnancies with missing or incomplete sonographic records, and perinatal outcomes, and those with fetal congenital anomalies, abnormal chromosomes, or congenital infection were excluded. 2.2 Data Variables Ultrasound measurements. Fetal biometry, growth, and Doppler were measured following ISUOG guidelines ( 17 , 18 ) . Fetal biometry parameters (biparietal diameter, head circumference, abdominal circumference (AC), and femur length) and umbilical artery (UA) Doppler were measured systematically on all routine scans, at about 20–24, 28–32, 36–41 weeks’ gestation in BJOGH. For FGR and suspected FGR, prenatal surveillance based on local guidelines for FGR management ( 19 ) . In brief, monitoring was scheduled based on the severity of FGR and alterations in UA Doppler. Serial scans for interval growth are optimally performed at least 3 weeks after a preceding scan. At least weekly and often bi-weekly or daily if absent diastolic blood flow in the UA. Additionally, EFW was calculated using National Institute of Child Health and Human Development (NICHD)-Asian formula ( 20 ) . EFW percentiles were computed as previously reported for singleton fetal growth ( 20 ) and converted into z scores. UA pulsatility index (PI) centiles were obtained based on the Doppler charts of Ciobanu ( 21 ) . Abnormal UA Doppler findings were defined as UA raised PI (>95th centile for GA), absent or reversed end diastolic flow (AREDF) ( 22 ) . Fetal growth velocity. Growth velocity was computed as the estimated fetal growth rate (g/week) between the ultrasound at that GA and from the prior visit [ie, for 28–32 weeks GA: velocity = (EFW 32–28)/ (GA 32–28)]. We used the model developed by Grantz ( 23 ) et al. in the NICHD fetal growth study to calculate EFW velocity percentiles of a given fetus for any given set at diagnosis of FGR using a free online calculator ( https://www.nichd.nih.gov/fetalvelocitycalculator ). Percentile distributions for this difference were constructed on the log scale using the linear mixed models, detailed equations been previously reported ( 23 ) . These percentiles assessed the relative change in anthropometric measurement taken at any fixed set of 2 GA times. Five categories of information for participants were collected: (a) demographic characteristics, including maternal age, systolic blood pressure, diastolic blood pressure, maternal ethnicity (Han versus Minority), height, pre-pregnancy weight, pre-pregnancy body mass index (BMI) (kg/m 2 ); (b) maternal comorbidities, including hypertensive disorder of pregnancy (HDP) (yes/no), pregestational diabetes(yes/no), Hypothyroidism(yes/no), anemia(yes/no), antiphospholipid syndrome(yes/no); (c) obstetric characteristic, including parity (nullipara versus multipara), GA at diagnosis of FGR, delivery week, Caesarean section for non-reassuring fetal status; (d) neonatal characteristics, including infant sex (male/female), birthlength, birthweight, placental weight, birth weight placental weight ratio; (e) APO. APO was defined as any of the following: perinatal demise (stillbirth, immediate neonatal demise, or death before neonatal intensive care unit) discharge, cord arterial pH ≤ 7.10, and/or base excess ≥ 12, bronchopulmonary dysplasia, hypoxic ischemic encephalopathy, grade III-IV intraventricular hemorrhage, periventricular leukomalacia, seizures, necrotizing enterocolitis, and sepsis ( 24 ) . 2.3 Feature selection While useful and robust, traditional prediction methods are often limited to using a small number of predictors, and they typically assume that these predictors have a uniform and consistent effect across their entire range. The features should not be a larger number to avoid the over-fitting problems ( 25 ) . Therefore, three feature selection measures were applied in our research, including Random Forest (RF), least absolute shrinkage and selection operator (LASSO) and univariable logistic regression (LR) to improve model prediction performance and increase model stability. RF ( 26 ) was selected for its resistance to overfitting and its ability to handle many features. It provides feature importance scores, aiding in the understanding of their individual contributions to the model's predictive accuracy. Additionally, RF can handle both categorical and continuous data, making it versatile for various types of datasets. LASSO ( 27 ) can compress excessively small regression coefficients to zero by adding a penalty term to the model estimation, which acts as a variable filter while compressing the coefficients and can effectively avoid the problem of insufficient model probabilistic power due to overfitting. LR ( 28 ) was included for its simplicity and interpretability, being a prevalent tool for binary classification problems. It provides clear coefficients that indicate the relationship between each feature and the outcome, making it an ideal choice when the goal is not only to select features but also to understand the direction of their impact on the probability of the outcome. Given the distinct advantages of each of the three methods, we finally selected the features that were consistently selected by two of three as predictors for subsequent models (i.e., the intersection of variables screened by at least two methods). 2.4 Prediction models For the ML model development, we used a 2-step systematic framework comprising 5 widely applied independent ML methods and stacked ensemble-based ML models. These models were detailed elsewhere, and a summary provided in Additional File 1 (Prediction Models). Five independent ML methods included LR, Support Vector Machine (SVM), Neural network (Nnet), RF, and extreme gradient boosting (XGboost). The trained model was then validated on the test data set, with the output being evaluation indicators of the model’s performance. We then used a stacking algorithm. Following the super learner settle rules ( 29 ) , we selected the best three independent ML methods from above based on their area under the receiver operating characteristic curve (AUROC) as the base-level learners, with the LR model as meta-level learner. This study obtained the training and test sets through ten-fold cross-validation with 100 iterations. During each iteration, the entire dataset was divided into 10 parts randomly, with 7 parts selected as the training set and 3 as the test set. After each iteration, the dataset was re-divided, and the training and test sets were re-determined. This approach reduced the risk of overfitting to a particular training set. 2.5 Shapley Additive exPlanations interpretability analysis Shapley Additive exPlanations (SHAP) values were calculated for models to investigate feature importance and rank the importance of input features, thereby improving the transparency and interpretability of the models ( 30 ) . 2.6 Construction of nomogram To provide clinicians with an individualized tool to predict patient-specific probability for APO in FGR at time of diagnosis, a nomogram was developed based on the most satisfactory ML model. Each predictor was assigned a ‘point’ on the nomogram, based on its predictive ability for APO in FGR. The sum of points assigned to each predictor for antenatal diagnosis of FGR was then converted into an overall risk score for APO in FGR. The risk thresholds of APO in FGR (low, medium, high) were typically defined based on clinical relevance and the distribution of the nomogram scores. “Low risk”: Scores corresponding to a predicted probability of the outcome below 50%. Figure 1 explains the overall workflow of the proposed system more clearly. 2.6 Statistical analysis Statistical analyses were performed with R version 4.1.2 (Version 4.1.2, http://www.rproject.org ). Continuous variables with normal distribution are presented as mean ± standard deviation and were compared with the t-test. Continuous variables with skewed distributions are presented as medians with interquartile ranges and compared with the Mann–Whitney U test or the Kruskal–Wallis H test. Categorical variables are presented as numbers with percentages and compared with the chi-square test. Statistical significance was set at 2-tailed P < 0.05. The discrimination performance of the ML models was evaluated with several commonly used evaluation indicators, including the area under the receiver operating characteristic curve (AUROC), sensitivity, positive predictive value (PPV), negative predictive value (NPV), balanced accuracy and F1 score. The Brier score (ranging from 0 to 1) was used to calculate the difference between the estimated risk and the observed risk, with a value closer to 0 indicating better calibration, thus assessing model calibration. In addition, the calibration performance of the clinical prediction models was evaluated by the Hosmer–Lemeshow test, with P-values higher than 0.05 indicating a good fit between the model and the actual data ( 31 ) . The DeLong test was employed to determine whether there was a significant difference in the AUROC values of different models ( 32 ) . Decision curve analysis (DCA) and clinical impact curve (CIC) were conducted to show the net benefit of using a model at different thresholds to assess the clinical value of the model ( 33 ) . 3. Results 3.1 Sample characteristics A total of 411 FGR were divided into a training group consisting of 288 FGR and a test group consisting of 123 FGR, following a ratio of 7:3. Maternal characteristics The average age of women was 31.5 and 33.9 years in the absent and present APO groups, respectively. Women enrolled in present APO group had a higher pre-pregnancy weight and BMI than those participating in absent APO group. They also had a higher rate of HDP (77.6% versus 20.7%). Further maternal information can be found in Table 1 . Table 1 Maternal characteristics of pregnancies in fetuses with a perinatal diagnosis of FGR by adverse perinatal outcome. Adverse perinatal outcome P value Absent(n = 304) Present(n = 107) Maternal Characteristics Maternal age(years), mean (SD) 31.5 (4.0) 33.9 (4.2) < 0.001 SBP (mmHg), mean (SD) 121.1 (11.5) 133.0(17.1) < 0.001 DBP (mmHg), mean (SD) 74.7 (8.9) 82.6 (11.1) < 0.001 Height (cm), mean (SD) 161.7 (5.1) 161.3 (4.9) 0.634 Weight gain during pregnancy (Kg), mean (SD) 12.3(4.6) 10.2(4.4) < 0.001 Pre-pregnancy weight (Kg), mean (SD) 56.7 (9.3) 62.4 (12.7) < 0.001 BMI (kg/m 2 ), mean (SD) 21.6 (3.5) 24.0 (4.5) < 0.001 Maternal Comorbidities Hypertension related to pregnancy (%) 63 (20.7) 83(77.6) < 0.001 Pregestational diabetes, (%) 48 (15.8) 22 (20.6) 0.259 SBP: systolic blood pressure; DBP: diastolic blood pressure; BMI: body mass index; Obstetric and neonatal characteristics A comparison of obstetric and neonatal characteristics between the two groups is shown in Table 2 . The group with APO had lower GA at diagnosis of FGR (27.3 weeks versus 32.3 weeks). Moreover, pregnancies with APO tended to be a lower proportion of nulliparous (70.1% versus 81.9%). Table 2 Obstetric and neonatal characteristics of pregnancies in fetuses with a perinatal diagnosis of FGR by adverse perinatal outcome. Adverse perinatal outcome P value Absent(n = 304) Present(n = 107) Obstetric Characteristics Nulliparity 249(81.9) 75 (70.1) 0.010 GA at diagnosis of FGR (weeks), mean (SD) 32.3.(4.8) 27.3 (3.7) < 0.001 GA at delivery (weeks), mean (SD) 38.3 (1.7) 31.0 (3.9) < 0.001 Preterm (delivery weeks < 37 weeks) 29(9.5) 96(89.7) < 0.001 CS for non-reassuring fetal status (%) 66 (24.7) 55 (58.5) < 0.001 Neonatal Characteristics Male fetal sex (%) 103 (38.6) 45(47.9) 0.115 Birthlength (cm), mean (SD) 47.0(2.4) 37.1(5.6) < 0.001 Birthweight (g), mean (SD) 2415.9(296.2) 1203.0(521.9) < 0.001 Placental weight (g), mean (SD) 476.4(88.0) 309.4(118.0) < 0.001 Birthweight placental weight ratio 5.2(1.0) 3.9(1.2) < 0.001 Neonatal outcome Cord arterial PH of ≤ 7.10 (%) 1(0.3) 4(3.7) 0.006 Cord arterial base excess of ≥ 12, (%) 0 1(0.9) 0.091 Stillbirth, immediate neonatal demise, or death prior to NICU discharge, (%) 0 21(19.6) < 0.001 Bronchopulmonary dysplasia, (%) 0 14(13.1) < 0.001 Sepsis, (%) 0 11(10.3) < 0.001 Grade III-IV IVH, (%) 0 3(2.8) 0.003 Necrotizing enterocolitis, (%) 0 1(0.9) 0.091 GA: gestational age; CS cesarean section; NICU, neonatal intensive care unit. 3.2 Predictor screening The study encompassed 16 predictors related to maternal characteristics (maternal age, height, pre-pregnancy weight, pre-pregnancy BMI), maternal comorbidities (HDP, pregestational diabetes, hypothyroidism, anemia, antiphospholipid syndrome), obstetric characteristics (nulliparity, GA at diagnosis of FGR), ultrasound parameters (AC z score, EFW z score, AC growth velocity, EFW growth velocity, abnormal UA Doppler) (Table S1 ). Of these, HDP, AC z score, EFW z score, AC growth velocity, EFW growth velocity and UA Doppler were obtained at time of FGR diagnosis. Utilizing the RF algorithm, we identified 6 key factors by excluding variables with RF importance values of 0. These factors included abnormal UA Doppler, EFW z score, HDP, GA at diagnosis of FGR, AC z score EFW growth velocity (Fig. 2 A). LASSO regression was also applied to discern factors and LASSO coefficients set to 0 were exclude. This analysis identified factors such as maternal BMI, GA at diagnosis of FGR, EFW z score, EFW growth velocity, abnormal UA Doppler (Figs. 2 B). Univariate LR suggested 9 factors having statistically significant, such as abnormal UA Doppler, HDP, GA at diagnosis of FGR, AC growth velocity, EFW z score, maternal BMI, age, nulliparous, and EFW growth velocity (Fig. 2 C). We delineated a common subset of feature variables endorsed by at least two methods of the results from LR, LASSO regression, and RF algorithm screening. These selected features were ultimately utilized in the construction of the model and consisted of maternal pre-pregnancy BMI, HDP, abnormal UA Doppler, GA at diagnosis of FGR, EFW z score, EFW growth velocity. 3.3 Model performance In this study, the balanced accuracy, sensitivity, PPV, NPV, and F1 score of each model were computed (Table S2 and S3, Fig. 3 A, 3 B). Among independent ML models in the train set, the AUROC illustrated that RF and SVM had the best predictive performance, with AUC values of 1.000 and 0.909, respectively, followed by LR with AUC values of 0.906 and Nnet with AUROC values of 0.855. Stacking ensemble models were subsequently developed based on LR, SVM and RF. In the test set, the Stacking model had the best performance, with a highest AUC of 0.861 and a calibration degree of 0.111. The Stacking model had the balanced accuracy (0.850), sensitivity (0.884), NPV (0.954) and F1 score (0.754). The DeLong test showed that the AUROC of the Stacking model was higher than that of several other models, and the difference was statistically significant (P < 0.05). In summary, the Stacking model performed the best in the test set and is thus recommended as the preferred model for the prediction APO of FGR at time of diagnosis. In this study, we evaluated the prediction accuracy and calibration of the Stacking model by calibration curve analysis of the training and test sets (Figs. 3 C, 3 D), respectively. The predicted values closely correlate with the observed values, indicating good model calibration. The results of the Hosmer–Lemeshow test for the training set and the test set were P = 0.387 and P = 0.825, respectively, indicating that the model showed good fit. The DCA for the training and test set are depicted in Figs. 3 E and 3 F, respectively, and the clinical impact curves are presented in Figs. 3 G and 3 H, respectively. These curves show that the model yields great Net Benefit over a large threshold probability range. 3.4 Feature importance According to the SHAP values, the GA at diagnosis of FGR was the most important factor, followed by abnormal UA Doppler. See Fig. 4 A for more details on variable importance. The GA at diagnosis of FGR was positively correlated with the APO risk in FGR. In contrast, it was negatively correlated with variables including abnormal UA Doppler and HDP. See Fig. 4 B for correlations between variables and outcomes. 3.5 Construction of nomograms The predictive ability of each independent marker included in the Stacking model for APO in FGR is reflected in the nomogram (Fig. 5 A). Instructions on the use of the nomogram model are provided in Fig. 5 B. As an example, consider a pregnant woman with abnormal UA Doppler and no HDP, GA of 30 weeks at diagnosis of FGR, a pre-pregnancy BMI of 24 kg/m², and an EFW z-score and EFW growth velocity of -3 and 30 g/week, respectively. Her final risk score would be 33(abnormal UA) + 0 (no HDP) + 28.5 (GA at diagnosis of FGR) + 50 (EFW z score) + 8.5 (pre-pregnancy BMI) + 25 (EFW growth velocity) = 145, which corresponds to a medium risk stratification with an APO probability of 0.72. 4. Discussion 4.1 Main findings To our knowledge, this is the first ML model to predict APO in FGR at time of diagnosis. We identified six available predictive risk factors in routine prenatal care and used various ML algorithms to construct an APO risk prediction model for FGR at time of diagnosis. The Stacking model, which has an AUC of 0.861, is effective for predicting APO in FGR at the time of diagnosis. 4.2 Comparison with previous studies To date, studies on predicting the risk of APO in FGR have been reported. However, most of these studies use a single method (traditional LR) for model building ( 6 , 14 , 34 ) , which may not be capable of handling complex relationships and hence affect prediction performance. Combining clinical and ultrasound parameters with complex ML algorithms can facilitate the development of clinical prediction models ( 35 ) . Among the six ML models, the Stacking model had the highest AUC, sensitivity, with good F1 score, calibration, and net benefit. Stacking surpasses other models due to its nature as a heterogeneous ensemble method, which allows it to leverage the strengths of various base learners. Multiple studies have demonstrated that the Stacking method is very valuable for prediction models in the medical field ( 36 – 38 ) . In the present study, we employed the Stacking algorithm to develop a final model containing 6 features. These features can be easily acquired and evaluated during routine prenatal care, making this model a promising tool for effectively predicting the risk of APO in FGR at time of diagnosis. To date the majority of published data have demonstrated that a high BMI, HDP and early onset FGR are associated with a greater likelihood of APO ( 39 – 41 ) . These factors may associate with blood vessel generation and remodeling, thereby influencing the establishment and maintenance of a stable fetal supply of nutrition and oxygen through the placenta, leading to adverse effects on fetal growth and well-being due to placental insufficiency ( 42 ) . With regard to GA at diagnosis of FGR, APO were related to its etiology, which affected fetuses since the early stages of pregnancy ( 14 ) . These factors interact and result in a vicious cycle. Our data appears to corroborate higher BMI, HDP and early A at diagnosis of FGR are predictors for APO. Given that ultrasound assessment of biometry and Doppler velocimetry forms the mainstay of identification and monitoring of FGR, it is unsurprising that the EFW z score and UA category were validated as predictors of fetal or neonatal APO ( 7 ) . The predictor variables in our study reached similar conclusions. Alfirevic et al ( 43 ) reported that using of Doppler ultrasound on the UA in high-risk pregnancies reduced the risk of perinatal deaths by 29%. Baschat et al ( 3 ) also confirmed that UA Doppler allowed identification of fetuses at risk for adverse outcome but was a poor predictor of the fetal condition. Likewise, EFW/AC<3rd was associated with a higher risk of perinatal complications, it performed poorly as a standalone parameter in predicting adverse perinatal outcome with AUC of 0.60 ( 44 ) . Thus, ISUOG practice guideline recommended continuous measurements should be performed because they offer significant advantages over single-point measurements in determining true growth parameters and growth trajectories in assessment of fetal growth ( 17 ) . Previous studies have shown that fetal growth velocity was independently associated with perinatal outcomes and monitoring growth velocity may help identify at-risk pregnancies and prevent APO ( 45 , 46 ) . Poor growth might be an endpoint of several pathological changes ( 42 ) . Hence, assessing growth velocity might be a more appropriate marker of adverse outcomes. This study is the first to integrate growth velocity into prediction model, providing longitudinal growth information. Several recent studies have highlighted the potential utility of PlGF and sFlt-1/PIGF concentration to predict outcomes in small-for-gestational-age (SGA) and FGR pregnancies ( 47 – 49 ) . However, the serum biomarkers for FGR risk assessment were relatively limited, and they might increase the economic and psychological burden on pregnant women, so they should not be considered a priority. Our study combined various ultrasound parameters with clinical features significantly enhance the accuracy of predicting APO in FGR, achieving an AUC of 0.861. Possible reasons for this result might be that fetal hypoxic injury might vary in the target organs most affected and the timing of deterioration, which was why the combination of different fetal parameters continues to be the best model at this point ( 50 ) . 4.3 Strength and Limitations A major advantage of the present study is that the nomogram is a visualizable tool. By using point scales to assign scores to each predictive factor, the scores of all predictive factors are summed up to obtain a total score. Based on the total score, the corresponding risk value can be derived, and risk stratification can be performed, making it a convenient tool for clinical use. Another strength of the present study is that the predictive factors included in the model were all routine items obtained for patients during prenatal care and were easy to obtain, providing feasibility for the promotion and application of the model in clinical practice. However, we acknowledge some limitations of this study. Firstly, being a retrospective study, there was a potential for omitted data and selection bias to impact the results. Secondly, from an ML perspective, the small size of the dataset, which posed the risk of overfitting the obtained results. To mitigate the risk of overfitting, this research employed a cross-validation methodology that went beyond mere partitioning of the overall dataset. Thirdly, the small sample size of this study and the fact that the sample was collected from a single center may limit the generalizability of the findings. However, Beijing, as the capital of China, has a relatively diverse population, including people from all over the country. This may to some extent mitigate some of the limitations of single-center studies. In addition, APO are not only outcomes of FGR but also include adverse outcomes of prematurity, leading to intervention bias. To address this, we will continue to explore ways to mitigate this bias in future research. Finally, due to the small sample size, we were unable to differentiate between early-onset FGR and late-onset FGR. In the future, we will conduct multicenter studies with larger sample sizes and develop predictive models targeted at each subtype. 5. Conclusion We proved that the Stacking model, composed of six factors readily available in daily clinical practice at the time of FGR diagnosis, can accurately predict APO in FGR and is expected to be an effective assistant tool for predict APO of FGR at the time of diagnosis. This nomogram may be used for personalized counseling, allocation of resources, risk stratification and management of pregnancies complicated by FGR. Abbreviations ML machine learning APO Adverse perinatal outcome FGR Fetal growth restriction RF Random forest LASSO Least Absolute Shrinkage and Selection Operator LR Logistic regression EFW Estimated fetal weight SGA Small for gestational age BJOGH Beijing Obstetrics and Gynecology Hospital ISUOG International Society of Ultrasound in Obstetrics and Gynecology GA Gestational age AC Abdominal circumference UA Umbilical artery NICHD National Institute of Child Health and Human Development PI Pulsatility index AREDF Absent or reversed end diastolic flow BMI Body mass index HDP Hypertensive disorder of pregnancy SVM Support Vector Machine Nnet Neural network XGboost Extreme gradient boosting AUROC Area under the receiver operating characteristic curve PPV Positive predictive value NPV negative predictive value DCA Decision curve analysis CIC Clinical impact curve SHAP Shapley Additive exPlanations Declarations Ethics approval and consent to participate This study was approved by the Beijing Obstetrics and Gynecology Hospital, Maternal and Child Health Centre, Capital Medical University Ethics Committee (NO.2023-KY-079-01). Given the retrospective nature of the study and the use of de-identified patient data, the requirement for informed consent was waived by the Beijing Obstetrics and Gynecology Hospital, Maternal and Child Health Centre, Capital Medical University Ethics Committee. The study was conducted in accordance with the ethical standards of the Declaration of Helsinki and its later amendments. Consent for publication Not applicable. Availability of data and material The datasets generated or analyzed during this study are available from the corresponding author on reasonable request. Competing Interests The authors have no relevant financial or non-financial interests to disclose. Funding This study was supported by the National Key Research and Development Program of China (2022YFC2703300, 2022YFC2703301). Beijing Municipal Health Commission, demonstration construction project of Clinical Research ward (NO: BCRW202109). Authors’ Contributions All authors (Xiangli Meng, Lei Wang, Minghui Wu, Na Zhang, Xiaofei Li and Qingqing Wu) contributed to the study conception and design. Material preparation, data collection and analysis were performed by Xiangli Meng, Lei Wang. The first draft of the manuscript was written by Xiangli Meng and all authors (Xiangli Meng, Lei Wang, Minghui Wu, Na Zhang, Xiaofei Li and Qingqing Wu) commented on previous versions of the manuscript. All authors (Xiangli Meng, Lei Wang, Minghui Wu, Na Zhang, Xiaofei Li and Qingqing Wu) read and approved the final manuscript. Acknowlegements Not applicable. References Barker DJ, Osmond C, Forsén TJ, Kajantie E, Eriksson JG. Trajectories of growth among children who have coronary events as adults. N Engl J Med. 2005;353(17):1802–9. Baião AER, de Carvalho PRN, Moreira MEL, de Sá RAM, Junior SCG. Predictors of perinatal outcome in early-onset fetal growth restriction: A study from an emerging economy country. Prenat Diagn. 2020;40(3):373–9. Baschat AA, Gembruch U, Weiner CP, Harman CR. Qualitative venous Doppler waveform analysis improves prediction of critical perinatal outcomes in premature growth-restricted fetuses. Ultrasound Obstet Gynecol. 2003;22(3):240–5. Bilardo CM, Wolf H, Stigter RH, Ville Y, Baez E, Visser GH, Hecher K. Relationship between monitoring parameters and perinatal outcome in severe, early intrauterine growth restriction. Ultrasound Obstet Gynecol. 2004;23(2):119–25. Hernandez-Andrade E, Crispi F, Benavides-Serralde JA, Plasencia W, Diesel HF, Eixarch E, et al. Contribution of the myocardial performance index and aortic isthmus blood flow index to predicting mortality in preterm growth-restricted fetuses. Ultrasound Obstet Gynecol. 2009;34(4):430–6. Odibo AO, Goetzinger KR, Cahill AG, Odibo L, Macones GA. Combined sonographic testing index and prediction of adverse outcome in preterm fetal growth restriction. Am J Perinatol. 2014;31(2):139–44. Spencer R, Maksym K, Hecher K, Maršál K, Figueras F, Ambler G et al. Maternal PlGF and umbilical Dopplers predict pregnancy outcomes at diagnosis of early-onset fetal growth restriction. J Clin Invest. 2023;133(18). Pini N, Lucchini M, Esposito G, Tagliaferri S, Campanile M, Magenes G, Signorini MG. A Machine Learning Approach to Monitor the Emergence of Late Intrauterine Growth Restriction. Front Artif Intell. 2021;4:622616. Sufriyana H, Wu YW, Su EC. Prediction of Preeclampsia and Intrauterine Growth Restriction: Development of Machine Learning Models on a Prospective Cohort. JMIR Med Inf. 2020;8(5):e15411. Teng LY, Mattar CNZ, Biswas A, Hoo WL, Saw SN. Interpreting the role of nuchal fold for fetal growth restriction prediction using machine learning. Sci Rep. 2022;12(1):3907. Ananth CV, Brandt JS. Fetal growth and gestational age prediction by machine learning. Lancet Digit Health. 2020;2(7):e336–7. Kuhle S, Maguire B, Zhang H, Hamilton D, Allen AC, Joseph KS, Allen VM. Comparison of logistic regression with machine learning methods for the prediction of fetal growth abnormalities: a retrospective cohort study. BMC Pregnancy Childbirth. 2018;18(1):333. Mustafa HJ, Kalafat E, Prasad S, Heydari MH, Nunge RN, Khalil A. Machine learning prediction of hypertension and diabetes in twin pregnancies using characteristics at prenatal care entry: a nationwide study. Ultrasound Obstet Gynecol; 2024. Punyapet P, Suwanrath C, Chainarong N, Sawaddisan R, Vichitkunakorn P. Predictors of adverse perinatal outcomes in fetal growth restriction using a combination of maternal clinical factors and simple ultrasound parameters. Int J Gynaecol Obstet. 2023;162(2):669–75. Lees CC, Stampalija T, Baschat A, da Silva Costa F, Ferrazzi E, Figueras F, et al. ISUOG Practice Guidelines: diagnosis and management of small-for-gestational-age fetus and fetal growth restriction. Ultrasound Obstet Gynecol. 2020;56(2):298–312. Committee opinion. 611: method for estimating due date. Obstet Gynecol. 2014;124(4):863–6. Salomon LJ, Alfirevic Z, Da Silva Costa F, Deter RL, Figueras F, Ghi T, et al. ISUOG Practice Guidelines: ultrasound assessment of fetal biometry and growth. Ultrasound Obstet Gynecol. 2019;53(6):715–23. Bhide A, Acharya G, Baschat A, Bilardo CM, Brezinka C, Cafici D, et al. ISUOG Practice Guidelines (updated): use of Doppler velocimetry in obstetrics. Ultrasound Obstet Gynecol. 2021;58(2):331–9. Fetal Medicine Subgroup SoPM, Chinese Medical Association, Subgroup O. Society of Obstetrics and Gecology, Chinese Medical Association. Expert consensus on fetal growth restriction. Chin J Perinat Med. 2019;22(06):361–80. Buck Louis GM, Grewal J, Albert PS, Sciscione A, Wing DA, Grobman WA, et al. Racial/ethnic standards for fetal growth: the NICHD Fetal Growth Studies. Am J Obstet Gynecol. 2015;213(4):449. .e1-.e41 . Ciobanu A, Wright A, Syngelaki A, Wright D, Akolekar R, Nicolaides KH. Fetal Medicine Foundation reference ranges for umbilical artery and middle cerebral artery pulsatility index and cerebroplacental ratio. Ultrasound Obstet Gynecol. 2019;53(4):465–72. Rocha AS, Andrade ARA, Moleiro ML, Guedes-Martins L. Doppler Ultrasound of the Umbilical Artery: Clinical Application. Rev Bras Ginecol Obstet. 2022;44(5):519–31. Grantz KL, Kim S, Grobman WA, Newman R, Owen J, Skupski D, et al. Fetal growth velocity: the NICHD fetal growth studies. Am J Obstet Gynecol. 2018;219(3):285. .e1-.e36 . Healy P, Gordijn SJ, Ganzevoort W, Beune IM, Baschat A, Khalil A et al. A Core Outcome Set for the prevention and treatment of fetal GROwth restriction: deVeloping Endpoints: the COSGROVE study. Am J Obstet Gynecol. 2019;221(4):339.e1-.e10. Goldstein BA, Navar AM, Carter RE. Moving beyond regression techniques in cardiovascular risk prediction: applying machine learning to address analytic challenges. Eur Heart J. 2016;38(23):1805. Breiman L. Random Forests. Mach Learn. 2001;45(1):5–32. R T. Regression Shrinkage and Selection via the Lasso. J Royal Stat Society: Ser B Stat Methodol. 1996;73(3):273–82. Goldstein R. Regression Methods in Biostatistics: Linear, Logistic, Survival and Repeated Measures Models. Technometrics. 2006;48:149–50. Naimi AI, Balzer LB. Stacked generalization: an introduction to super learning. Eur J Epidemiol. 2018;33(5):459–64. Lundberg SMaL S-I. A Unified Approach to Interpreting Model Predictions. Proceedings of the 31st International Conference on Neural Information Processing Systems (NIPS 2017). Long Beach, CA, USA. 2017:4768-77. Kramer AA, Zimmerman JE. Assessing the calibration of mortality benchmarks in critical care: The Hosmer-Lemeshow test revisited. Crit Care Med. 2007;35(9):2052–6. DeLong ER, DeLong DM, Clarke-Pearson DL. Comparing the areas under two or more correlated receiver operating characteristic curves: a nonparametric approach. Biometrics. 1988;44(3):837–45. Steyerberg EW, Vickers AJ, Cook NR, Gerds T, Gonen M, Obuchowski N, et al. Assessing the performance of prediction models: a framework for traditional and novel measures. Epidemiology. 2010;21(1):128–38. WHO Child Growth. Standards based on length/height, weight and age. Acta Paediatr Suppl. 2006;450:76–85. Goecks J, Jalili V, Heiser LM, Gray JW. How Machine Learning Will Transform Biomedicine. Cell. 2020;181(1):92–101. Wang X, Ren H, Ren J, Song W, Qiao Y, Ren Z, et al. Machine learning-enabled risk prediction of chronic obstructive pulmonary disease with unbalanced data. Comput Methods Programs Biomed. 2023;230:107340. Wang J, Chen H, Wang H, Liu W, Peng D, Zhao Q, Xiao M. A Risk Prediction Model for Physical Restraints Among Older Chinese Adults in Long-term Care Facilities: Machine Learning Study. J Med Internet Res. 2023;25:e43815. Sun T, Liu J, Yuan H, Li X, Yan H. Construction of a risk prediction model for lung infection after chemotherapy in lung cancer patients based on the machine learning algorithm. Front Oncol. 2024;14:1403392. Yang Z, Feng G, Gao X, Yan X, Li Y, Wang Y et al. Maternal adiposity and perinatal and offspring outcomes: an umbrella review. Nat Hum Behav. 2024. Webster K, Fishburn S, Maresh M, Findlay SC, Chappell LC. Diagnosis and management of hypertension in pregnancy: summary of updated NICE guidance. BMJ. 2019;366:l5119. Figueras F, Gratacós E. Update on the diagnosis and classification of fetal growth restriction and proposal of a stage-based management protocol. Fetal Diagn Ther. 2014;36(2):86–98. Di Martino DD, Avagliano L, Ferrazzi E, Fusè F, Sterpi V, Parasiliti M et al. Hypertensive Disorders of Pregnancy and Fetal Growth Restriction: Clinical Characteristics and Placental Lesions and Possible Preventive Nutritional Targets. Nutrients. 2022;14(16). Alfirevic Z, Stampalija T, Dowswell T. Fetal and umbilical Doppler ultrasound in high-risk pregnancies. Cochrane Database Syst Rev. 2017;6(6):Cd007529. Meler E, Martinez-Portilla RJ, Caradeux J, Mazarico E, Gil-Armas C, Boada D, et al. Severe smallness as predictor of adverse perinatal outcome in suspected late small-for-gestational-age fetuses: systematic review and meta-analysis. Ultrasound Obstet Gynecol. 2022;60(3):328–37. Hugh O, Gardosi J. Fetal weight projection model to define growth velocity and validation against pregnancy outcome in a cohort of serially scanned pregnancies. Ultrasound Obstet Gynecol. 2022;60(1):86–95. Bommarito PA, Cantonwine DE, Stevens DR, Welch BM, Davalos AD, Zhao S, et al. An application of group-based trajectory modeling to define fetal growth phenotypes among small-for-gestational-age births in the LIFECODES Fetal Growth Study. Am J Obstet Gynecol. 2023;228(3):334. .e1-.e21 . Sharp A, Jackson R, Cornforth C, Harrold J, Turner MA, Kenny L, et al. A prediction model for short-term neonatal outcomes in severe early-onset fetal growth restriction. Eur J Obstet Gynecol Reprod Biol. 2019;241:109–18. Rodríguez-Calvo J, Villalaín C, Gómez-Arriaga PI, Quezada MS, Herraiz I, Galindo A. Prediction of perinatal survival in early-onset fetal growth restriction: role of placental growth factor. Ultrasound Obstet Gynecol. 2023;61(2):181–90. Mendoza M, Hurtado I, Bonacina E, Garcia-Manau P, Serrano B, Tur H, et al. Individual risk assessment for prenatal counseling in early-onset growth-restricted and small-for-gestational-age fetuses. Acta Obstet Gynecol Scand. 2021;100(3):504–12. Lees CC, Marlow N, van Wassenaer-Leemhuis A, Arabin B, Bilardo CM, Brezinka C, et al. 2 year neurodevelopmental and intermediate perinatal outcomes in infants with very preterm fetal growth restriction (TRUFFLE): a randomised trial. Lancet. 2015;385(9983):2162–72. Additional Declarations No competing interests reported. Supplementary Files SupplementaryMaterial.docx Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-5698332","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":443256810,"identity":"5a616cd7-a711-44ee-b168-c5d27db88457","order_by":0,"name":"Xiangli Meng","email":"","orcid":"","institution":"Department of Ultrasound, Beijing Obstetrics and Gynecology Hospital, Capital Medical University, Beijing Maternal and Child Health Care Hospital, Beijing, China","correspondingAuthor":false,"prefix":"","firstName":"Xiangli","middleName":"","lastName":"Meng","suffix":""},{"id":443256813,"identity":"b62cdd0d-5b15-4027-b9c9-c9e195c02740","order_by":1,"name":"Lei Wang","email":"","orcid":"","institution":"Beijing Obstetrics and Gynecology Hospital, Capital Medical University, Beijing Maternal and Child Health Care Hospital, Beijing, China","correspondingAuthor":false,"prefix":"","firstName":"Lei","middleName":"","lastName":"Wang","suffix":""},{"id":443256814,"identity":"ffa96a42-cf4c-45b2-8181-46f6fed3b65f","order_by":2,"name":"Minghui Wu","email":"","orcid":"","institution":"Beijing Obstetrics and Gynecology Hospital, Capital Medical University, Beijing Maternal and Child Health Care Hospital, Beijing, China","correspondingAuthor":false,"prefix":"","firstName":"Minghui","middleName":"","lastName":"Wu","suffix":""},{"id":443256815,"identity":"ad466755-ec9b-4506-bdd1-8d1f49863766","order_by":3,"name":"Na Zhang","email":"","orcid":"","institution":"Department of Ultrasound, Beijing Obstetrics and Gynecology Hospital, Capital Medical University, Beijing Maternal and Child Health Care Hospital, Beijing, China","correspondingAuthor":false,"prefix":"","firstName":"Na","middleName":"","lastName":"Zhang","suffix":""},{"id":443256816,"identity":"0ae9bccd-d244-4621-9bb0-a7d4cdac0343","order_by":4,"name":"Xiaofei Li","email":"","orcid":"","institution":"Department of Ultrasound, Beijing Obstetrics and Gynecology Hospital, Capital Medical University, Beijing Maternal and Child Health Care Hospital, Beijing, China","correspondingAuthor":false,"prefix":"","firstName":"Xiaofei","middleName":"","lastName":"Li","suffix":""},{"id":443256817,"identity":"38bcea88-dab6-4f1d-92ba-b786095b1404","order_by":5,"name":"Qingqing Wu","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA6ElEQVRIiWNgGAWjYDACZhBhAGYdgAolEK2FDaaUkBYE4DEgTgvfcd7DL94U3LHbcCPn44efOYcZ+NlzDBh+7sCtRfIwX5rlHINnyRtu5G6W7N12mEGy540BY+8Z3FoMDvOYGfMYHE42u5G7jZkRqMXgRo4BM2MbUVpynoG12BOhxfgxUIsdUAsbxBYJAlokgbYwzjE4nGB/5pkx0C/pPBJnnhUc7MWjhe/8GeMPb/4ctpdsT3744ec2azn+9uSND37i0cJwgIFNgoeBIbEByueBCOIDBxiYPwCV2eNVNApGwSgYBSMbAADomVNVa4b13wAAAABJRU5ErkJggg==","orcid":"","institution":"Department of Ultrasound, Beijing Obstetrics and Gynecology Hospital, Capital Medical University, Beijing Maternal and Child Health Care Hospital, Beijing, China","correspondingAuthor":true,"prefix":"","firstName":"Qingqing","middleName":"","lastName":"Wu","suffix":""}],"badges":[],"createdAt":"2024-12-23 09:23:42","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-5698332/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-5698332/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":80755537,"identity":"4d76b01f-f6e2-4de1-8840-e3d012a2e2d5","added_by":"auto","created_at":"2025-04-16 17:41:45","extension":"jpeg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":287146,"visible":true,"origin":"","legend":"\u003cp\u003eResearch flowchart. FGR, fetal growth restriction; RF, random forest; LASSO, the least absolute shrinkage and selection operator; SVM, Support Vector Machine; Nnet, Neural network; LR, Logistic Regression; the area under the receiver operating characteristic curve (AUROC); DCA, decision curve analysis; CIC, clinical impact curve.\u003c/p\u003e","description":"","filename":"Figure1.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-5698332/v1/16339d16306e3c3f256737d6.jpeg"},{"id":80756518,"identity":"d3d94d2f-5f53-4848-9fd6-a561eee3bb6e","added_by":"auto","created_at":"2025-04-16 17:57:45","extension":"jpg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":830025,"visible":true,"origin":"","legend":"\u003cp\u003ePredictor screening results. A: factor screening based on RF; B: factor screening based on the LASSO regression; C: factor screening based on univariate logistic regression.\u003c/p\u003e\n\u003cp\u003eRF, Random Forest; LASSO, the least absolute shrinkage and selection operator; UA, umbilical artery; EFW, estimated fetal weight; AC, abdominal circumstance; HDP, hypertensive disorder of pregnancy; BMI, body mass index.\u003c/p\u003e","description":"","filename":"Figure2.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5698332/v1/943762b0a81ac0631183a6f2.jpg"},{"id":80755542,"identity":"0a2e87ab-e100-473c-873d-557e702b7892","added_by":"auto","created_at":"2025-04-16 17:41:45","extension":"jpg","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":1565860,"visible":true,"origin":"","legend":"\u003cp\u003ePerformance of Stacking model predicting APO in FGR at time of diagnosis in the training and test sets. ROC curve analysis (A, B) using six ML algorithms in the training and test sets, calibration curve analysis (C, D), DCA curves (E, F) and CICs predicting APO in FGR at time of diagnosis using Stacking algorithm in the training and test sets.\u003c/p\u003e\n\u003cp\u003eAPO, adverse perinatal outcome; FGR, fetal growth restriction; ROC, receiver operating characteristic curve; ML, machine learning; LR: logistic regression; SVM: support vector machine; XGboost, extreme gradient boosting; RF: random forest; Nnet: neural network; DCA, decision curve analysis; CIC, clinical impact curve.\u003c/p\u003e","description":"","filename":"Figure3.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5698332/v1/83feb0b54cf123988d02e000.jpg"},{"id":80755867,"identity":"90c7bf4d-f71e-4ef7-91bd-33d08c688482","added_by":"auto","created_at":"2025-04-16 17:49:45","extension":"jpg","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":946035,"visible":true,"origin":"","legend":"\u003cp\u003eStacking model explanation by the SHapley Additive exPlanation (SHAP) method. (A) SHAP summary bar plot. This plot evaluates the contribution of each feature to the model using mean SHAP values, displayed in descending order. (B) SHAP summary dot plot. Each dot represents a patient’s SHAP value for a given feature, with a higher value appearing greener, and a lower value being bluer. UA, umbilical artery; EFW, estimated fetal weight; BMI, body mass index. FGR, fetal growth restriction.\u003c/p\u003e","description":"","filename":"Figure4.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5698332/v1/b233a52bd17512e3aeb231b0.jpg"},{"id":80755868,"identity":"ad6d6c03-74f3-4700-88c2-79ce9af6a9cb","added_by":"auto","created_at":"2025-04-16 17:49:45","extension":"jpg","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":1457468,"visible":true,"origin":"","legend":"\u003cp\u003eA: Nomogram prediction model for APO in FGR at time of diagnosis. B: Example of how to use nomogram prediction model in FGR at time of diagnosis. Dashed arrows indicate relative score of each variable, and solid arrow shows total score and corresponding risk value.\u003c/p\u003e\n\u003cp\u003eFGR, fetal growth restriction; UA, umbilical artery; HDP, hypertensive disorder of pregnancy; EFW, estimated fetal weight; BMI, body mass index.\u003c/p\u003e","description":"","filename":"Figure5.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5698332/v1/b572755f6e3da9c82529ab25.jpg"},{"id":83453374,"identity":"5c1f3f94-26fb-43b6-a9d1-b28cdda0de47","added_by":"auto","created_at":"2025-05-26 13:48:11","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":3573434,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-5698332/v1/14d85c18-c32a-44fd-925d-4574e4d0da7b.pdf"},{"id":80755866,"identity":"ac43e36b-a929-49c3-83e1-4af1f6c9abf0","added_by":"auto","created_at":"2025-04-16 17:49:45","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":52367,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryMaterial.docx","url":"https://assets-eu.researchsquare.com/files/rs-5698332/v1/a572c65dd65c84e48ab7f208.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Construction of a prediction model for adverse perinatal outcomes in fetal growth restriction based on a machine learning algorithm","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003eFetal growth restriction (FGR) is a common pregnancy complication. It is associated with an increased risk of perinatal mortality and morbidity and long-term adverse infant outcomes, including cognitive impairment in childhood and non-communicable diseases in adulthood, such as hypertension, metabolic syndrome, insulin resistance, type-2 diabetes mellitus, coronary heart disease, and stroke \u003csup\u003e(\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e)\u003c/sup\u003e. Numerous studies have investigated predictive markers for the development of FGR, unfortunately, almost all models were built using ultrasound measurements taken at the time point closest to delivery or confounded the definitions of FGR and small for gestational age (SGA) \u003csup\u003e(\u003cspan additionalcitationids=\"CR3 CR4 CR5\" citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e)\u003c/sup\u003e, but far fewer have studied the prediction of pregnancy outcome when at time of diagnosis or counseling. The inability to predict pregnancy outcomes leaves pregnant patients and their partners with a considerable burden of uncertainty. It also limits clinicians' abilities to personalize management and counseling, including about termination of the pregnancy. The challenges encountered by both clinicians and pregnant women with FGR can vary greatly based on the GA at diagnosis. Patient-tailored prediction of fetal and neonatal outcomes is attractive from a clinical point and to indicate intrauterine transfer to a hospital with appropriate facilities, such as a neonatal intensive care unit \u003csup\u003e(\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e)\u003c/sup\u003e. Hence, prediction models containing antenatal available predictors for fetal and neonatal outcomes can facilitate clinical decision-making and counseling in pregnancies complicated by FGR.\u003c/p\u003e \u003cp\u003eRecently, machine learning (ML) algorithms have been used in various fields for clinical practice such as FGR diagnosis \u003csup\u003e(\u003cspan additionalcitationids=\"CR9 CR10\" citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e)\u003c/sup\u003e. Various robust ML prediction models have been developed for obstetric adverse event \u003csup\u003e(\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e)\u003c/sup\u003e and complications prediction \u003csup\u003e(\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e)\u003c/sup\u003e. Traditional prediction models were constructed using a generalized linear model, which depends on the underlying assumption that risk factors have a linear correlation with outcomes \u003csup\u003e(\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e, \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e)\u003c/sup\u003e. While these models offer the advantages of being easy to code and fast to calculate, it may oversimplify the complex nonlinear interaction between variables. ML methods have been an alternative method to overcome current limitations. The aim, therefore, of this study was to use ML tools to develop a prediction model for predicting APO in FGR at time of diagnosis.\u003c/p\u003e"},{"header":"2. Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003e2.1 Study Population\u003c/h2\u003e \u003cp\u003eWe performed a retrospective analysis of a cohort of ultrasound-dated singleton fetuses with a prenatal diagnosis of FGR at Beijing Obstetrics and Gynecology Hospital (BJOGH) between January 2021 and November 2023.\u003c/p\u003e \u003cp\u003eThis study included 411 pregnancies with an antenatal diagnosis of FGR based on the criteria endorsed by the International Society of Ultrasound in Obstetrics and Gynecology (ISUOG) \u003csup\u003e(\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e)\u003c/sup\u003e. The inclusion criteria was: singleton pregnancies with accurate GA (estimated based on the last menstrual period, corrected by the crown-rump length (between 9\u0026thinsp;+\u0026thinsp;0 and 14\u0026thinsp;+\u0026thinsp;0 weeks' gestation) if a discrepancy of \u0026gt;\u0026thinsp;7 days was present \u003csup\u003e(\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e)\u003c/sup\u003e). Pregnancies with missing or incomplete sonographic records, and perinatal outcomes, and those with fetal congenital anomalies, abnormal chromosomes, or congenital infection were excluded.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003e2.2 Data Variables\u003c/h2\u003e \u003cp\u003e\u003cb\u003eUltrasound measurements.\u003c/b\u003e Fetal biometry, growth, and Doppler were measured following ISUOG guidelines \u003csup\u003e(\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e, \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e)\u003c/sup\u003e. Fetal biometry parameters (biparietal diameter, head circumference, abdominal circumference (AC), and femur length) and umbilical artery (UA) Doppler were measured systematically on all routine scans, at about 20\u0026ndash;24, 28\u0026ndash;32, 36\u0026ndash;41 weeks\u0026rsquo; gestation in BJOGH. For FGR and suspected FGR, prenatal surveillance based on local guidelines for FGR management \u003csup\u003e(\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e)\u003c/sup\u003e. In brief, monitoring was scheduled based on the severity of FGR and alterations in UA Doppler. Serial scans for interval growth are optimally performed at least 3 weeks after a preceding scan. At least weekly and often bi-weekly or daily if absent diastolic blood flow in the UA. Additionally, EFW was calculated using National Institute of Child Health and Human Development (NICHD)-Asian formula \u003csup\u003e(\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e)\u003c/sup\u003e. EFW percentiles were computed as previously reported for singleton fetal growth \u003csup\u003e(\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e)\u003c/sup\u003e and converted into z scores. UA pulsatility index (PI) centiles were obtained based on the Doppler charts of Ciobanu \u003csup\u003e(\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e)\u003c/sup\u003e. Abnormal UA Doppler findings were defined as UA raised PI (\u0026gt;95th centile for GA), absent or reversed end diastolic flow (AREDF) \u003csup\u003e(\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e)\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003e \u003cb\u003eFetal growth velocity.\u003c/b\u003e Growth velocity was computed as the estimated fetal growth rate (g/week) between the ultrasound at that GA and from the prior visit [ie, for 28\u0026ndash;32 weeks GA: velocity = (EFW 32\u0026ndash;28)/ (GA 32\u0026ndash;28)]. We used the model developed by Grantz \u003csup\u003e(\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e)\u003c/sup\u003e et al. in the NICHD fetal growth study to calculate EFW velocity percentiles of a given fetus for any given set at diagnosis of FGR using a free online calculator (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.nichd.nih.gov/fetalvelocitycalculator\u003c/span\u003e\u003cspan address=\"https://www.nichd.nih.gov/fetalvelocitycalculator\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e). Percentile distributions for this difference were constructed on the log scale using the linear mixed models, detailed equations been previously reported \u003csup\u003e(\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e)\u003c/sup\u003e. These percentiles assessed the relative change in anthropometric measurement taken at any fixed set of 2 GA times.\u003c/p\u003e \u003cp\u003eFive categories of information for participants were collected: (a) demographic characteristics, including maternal age, systolic blood pressure, diastolic blood pressure, maternal ethnicity (Han versus Minority), height, pre-pregnancy weight, pre-pregnancy body mass index (BMI) (kg/m\u003csup\u003e2\u003c/sup\u003e); (b) maternal comorbidities, including hypertensive disorder of pregnancy (HDP) (yes/no), pregestational diabetes(yes/no), Hypothyroidism(yes/no), anemia(yes/no), antiphospholipid syndrome(yes/no); (c) obstetric characteristic, including parity (nullipara versus multipara), GA at diagnosis of FGR, delivery week, Caesarean section for non-reassuring fetal status; (d) neonatal characteristics, including infant sex (male/female), birthlength, birthweight, placental weight, birth weight placental weight ratio; (e) APO. APO was defined as any of the following: perinatal demise (stillbirth, immediate neonatal demise, or death before neonatal intensive care unit) discharge, cord arterial pH\u0026thinsp;\u0026le;\u0026thinsp;7.10, and/or base excess\u0026thinsp;\u0026ge;\u0026thinsp;12, bronchopulmonary dysplasia, hypoxic ischemic encephalopathy, grade III-IV intraventricular hemorrhage, periventricular leukomalacia, seizures, necrotizing enterocolitis, and sepsis \u003csup\u003e(\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e)\u003c/sup\u003e.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec5\" class=\"Section2\"\u003e \u003ch2\u003e2.3 Feature selection\u003c/h2\u003e \u003cp\u003eWhile useful and robust, traditional prediction methods are often limited to using a small number of predictors, and they typically assume that these predictors have a uniform and consistent effect across their entire range. The features should not be a larger number to avoid the over-fitting problems \u003csup\u003e(\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e)\u003c/sup\u003e. Therefore, three feature selection measures were applied in our research, including Random Forest (RF), least absolute shrinkage and selection operator (LASSO) and univariable logistic regression (LR) to improve model prediction performance and increase model stability. RF \u003csup\u003e(\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e)\u003c/sup\u003e was selected for its resistance to overfitting and its ability to handle many features. It provides feature importance scores, aiding in the understanding of their individual contributions to the model's predictive accuracy. Additionally, RF can handle both categorical and continuous data, making it versatile for various types of datasets. LASSO \u003csup\u003e(\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e)\u003c/sup\u003e can compress excessively small regression coefficients to zero by adding a penalty term to the model estimation, which acts as a variable filter while compressing the coefficients and can effectively avoid the problem of insufficient model probabilistic power due to overfitting. LR \u003csup\u003e(\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e)\u003c/sup\u003e was included for its simplicity and interpretability, being a prevalent tool for binary classification problems. It provides clear coefficients that indicate the relationship between each feature and the outcome, making it an ideal choice when the goal is not only to select features but also to understand the direction of their impact on the probability of the outcome. Given the distinct advantages of each of the three methods, we finally selected the features that were consistently selected by two of three as predictors for subsequent models (i.e., the intersection of variables screened by at least two methods).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec6\" class=\"Section2\"\u003e \u003ch2\u003e2.4 Prediction models\u003c/h2\u003e \u003cp\u003eFor the ML model development, we used a 2-step systematic framework comprising 5 widely applied independent ML methods and stacked ensemble-based ML models. These models were detailed elsewhere, and a summary provided in Additional File 1 (Prediction Models). Five independent ML methods included LR, Support Vector Machine (SVM), Neural network (Nnet), RF, and extreme gradient boosting (XGboost). The trained model was then validated on the test data set, with the output being evaluation indicators of the model\u0026rsquo;s performance. We then used a stacking algorithm. Following the super learner settle rules \u003csup\u003e(\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e)\u003c/sup\u003e, we selected the best three independent ML methods from above based on their area under the receiver operating characteristic curve (AUROC) as the base-level learners, with the LR model as meta-level learner. This study obtained the training and test sets through ten-fold cross-validation with 100 iterations. During each iteration, the entire dataset was divided into 10 parts randomly, with 7 parts selected as the training set and 3 as the test set. After each iteration, the dataset was re-divided, and the training and test sets were re-determined. This approach reduced the risk of overfitting to a particular training set.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec7\" class=\"Section2\"\u003e \u003ch2\u003e\u003cb\u003e2.5\u003c/b\u003e Shapley Additive exPlanations interpretability analysis\u003c/h2\u003e \u003cp\u003eShapley Additive exPlanations (SHAP) values were calculated for models to investigate feature importance and rank the importance of input features, thereby improving the transparency and interpretability of the models \u003csup\u003e(\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e)\u003c/sup\u003e.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003e2.6 Construction of nomogram\u003c/h2\u003e \u003cp\u003eTo provide clinicians with an individualized tool to predict patient-specific probability for APO in FGR at time of diagnosis, a nomogram was developed based on the most satisfactory ML model. Each predictor was assigned a \u0026lsquo;point\u0026rsquo; on the nomogram, based on its predictive ability for APO in FGR. The sum of points assigned to each predictor for antenatal diagnosis of FGR was then converted into an overall risk score for APO in FGR. The risk thresholds of APO in FGR (low, medium, high) were typically defined based on clinical relevance and the distribution of the nomogram scores. \u0026ldquo;Low risk\u0026rdquo;: Scores corresponding to a predicted probability of the outcome below \u0026lt;\u0026thinsp;20%. \u0026ldquo;Medium risk\u0026rdquo;: Scores corresponding to a predicted probability between 20\u0026ndash;50%. \u0026ldquo;High risk\u0026rdquo;: Scores corresponding to a predicted probability above \u0026gt;\u0026thinsp;50%. Figure\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e explains the overall workflow of the proposed system more clearly.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec9\" class=\"Section2\"\u003e \u003ch2\u003e2.6 Statistical analysis\u003c/h2\u003e \u003cp\u003eStatistical analyses were performed with R version 4.1.2 (Version 4.1.2, \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://www.rproject.org\u003c/span\u003e\u003cspan address=\"http://www.rproject.org\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e). Continuous variables with normal distribution are presented as mean\u0026thinsp;\u0026plusmn;\u0026thinsp;standard deviation and were compared with the t-test. Continuous variables with skewed distributions are presented as medians with interquartile ranges and compared with the Mann\u0026ndash;Whitney U test or the Kruskal\u0026ndash;Wallis H test. Categorical variables are presented as numbers with percentages and compared with the chi-square test. Statistical significance was set at 2-tailed P\u0026thinsp;\u0026lt;\u0026thinsp;0.05.\u003c/p\u003e \u003cp\u003eThe discrimination performance of the ML models was evaluated with several commonly used evaluation indicators, including the area under the receiver operating characteristic curve (AUROC), sensitivity, positive predictive value (PPV), negative predictive value (NPV), balanced accuracy and F1 score. The Brier score (ranging from 0 to 1) was used to calculate the difference between the estimated risk and the observed risk, with a value closer to 0 indicating better calibration, thus assessing model calibration. In addition, the calibration performance of the clinical prediction models was evaluated by the Hosmer\u0026ndash;Lemeshow test, with P-values higher than 0.05 indicating a good fit between the model and the actual data \u003csup\u003e(\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e)\u003c/sup\u003e. The DeLong test was employed to determine whether there was a significant difference in the AUROC values of different models \u003csup\u003e(\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e)\u003c/sup\u003e. Decision curve analysis (DCA) and clinical impact curve (CIC) were conducted to show the net benefit of using a model at different thresholds to assess the clinical value of the model \u003csup\u003e(\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e)\u003c/sup\u003e.\u003c/p\u003e \u003c/div\u003e"},{"header":"3. Results","content":"\u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003e3.1 Sample characteristics\u003c/h2\u003e \u003cp\u003eA total of 411 FGR were divided into a training group consisting of 288 FGR and a test group consisting of 123 FGR, following a ratio of 7:3.\u003c/p\u003e \u003cp\u003e \u003cb\u003eMaternal characteristics\u003c/b\u003e \u003c/p\u003e \u003cp\u003eThe average age of women was 31.5 and 33.9 years in the absent and present APO groups, respectively. Women enrolled in present APO group had a higher pre-pregnancy weight and BMI than those participating in absent APO group. They also had a higher rate of HDP (77.6% versus 20.7%). Further maternal information can be found in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eMaternal characteristics of pregnancies in fetuses with a perinatal diagnosis of FGR by adverse perinatal outcome.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"4\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003eAdverse perinatal outcome\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eP value\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAbsent(n\u0026thinsp;=\u0026thinsp;304)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003ePresent(n\u0026thinsp;=\u0026thinsp;107)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"4\" nameend=\"c4\" namest=\"c1\"\u003e \u003cp\u003eMaternal Characteristics\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMaternal age(years), mean (SD)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e31.5 (4.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e33.9 (4.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSBP (mmHg), mean (SD)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e121.1 (11.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e133.0(17.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDBP (mmHg), mean (SD)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e74.7 (8.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e82.6 (11.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHeight (cm), mean (SD)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e161.7 (5.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e161.3 (4.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.634\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eWeight gain during pregnancy (Kg), mean (SD)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e12.3(4.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e10.2(4.4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePre-pregnancy weight (Kg), mean (SD)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e56.7 (9.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e62.4 (12.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBMI (kg/m\u003csup\u003e2\u003c/sup\u003e), mean (SD)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e21.6 (3.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e24.0 (4.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"4\" nameend=\"c4\" namest=\"c1\"\u003e \u003cp\u003eMaternal Comorbidities\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHypertension related to pregnancy (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e63 (20.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e83(77.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePregestational diabetes, (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e48 (15.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e22 (20.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.259\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"4\"\u003eSBP: systolic blood pressure; DBP: diastolic blood pressure; BMI: body mass index;\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cb\u003eObstetric and neonatal characteristics\u003c/b\u003e \u003c/p\u003e \u003cp\u003eA comparison of obstetric and neonatal characteristics between the two groups is shown in Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e. The group with APO had lower GA at diagnosis of FGR (27.3 weeks versus 32.3 weeks). Moreover, pregnancies with APO tended to be a lower proportion of nulliparous (70.1% versus 81.9%).\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eObstetric and neonatal characteristics of pregnancies in fetuses with a perinatal diagnosis of FGR by adverse perinatal outcome.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"4\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003eAdverse perinatal outcome\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eP value\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAbsent(n\u0026thinsp;=\u0026thinsp;304)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003ePresent(n\u0026thinsp;=\u0026thinsp;107)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"4\" nameend=\"c4\" namest=\"c1\"\u003e \u003cp\u003eObstetric Characteristics\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNulliparity\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e249(81.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e75 (70.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.010\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGA at diagnosis of FGR (weeks), mean (SD)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e32.3.(4.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e27.3 (3.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGA at delivery (weeks), mean (SD)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e38.3 (1.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e31.0 (3.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePreterm (delivery weeks\u0026thinsp;\u0026lt;\u0026thinsp;37 weeks)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e29(9.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e96(89.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCS for non-reassuring fetal status (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e66 (24.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e55 (58.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"4\" nameend=\"c4\" namest=\"c1\"\u003e \u003cp\u003eNeonatal Characteristics\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMale fetal sex (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e103 (38.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e45(47.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.115\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBirthlength (cm), mean (SD)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e47.0(2.4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e37.1(5.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBirthweight (g), mean (SD)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e2415.9(296.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1203.0(521.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePlacental weight (g), mean (SD)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e476.4(88.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e309.4(118.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBirthweight placental weight ratio\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e5.2(1.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e3.9(1.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"4\" nameend=\"c4\" namest=\"c1\"\u003e \u003cp\u003eNeonatal outcome\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCord arterial PH of \u0026le;\u0026thinsp;7.10 (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1(0.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e4(3.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.006\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCord arterial base excess of \u0026ge;\u0026thinsp;12, (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1(0.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.091\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eStillbirth, immediate neonatal demise, or death prior to NICU discharge, (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e21(19.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBronchopulmonary dysplasia, (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e14(13.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSepsis, (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e11(10.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGrade III-IV IVH, (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e3(2.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.003\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNecrotizing enterocolitis, (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1(0.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.091\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"4\"\u003eGA: gestational age; CS cesarean section; NICU, neonatal intensive care unit.\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003e3.2 Predictor screening\u003c/h2\u003e \u003cp\u003eThe study encompassed 16 predictors related to maternal characteristics (maternal age, height, pre-pregnancy weight, pre-pregnancy BMI), maternal comorbidities (HDP, pregestational diabetes, hypothyroidism, anemia, antiphospholipid syndrome), obstetric characteristics (nulliparity, GA at diagnosis of FGR), ultrasound parameters (AC z score, EFW z score, AC growth velocity, EFW growth velocity, abnormal UA Doppler) (Table \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e). Of these, HDP, AC z score, EFW z score, AC growth velocity, EFW growth velocity and UA Doppler were obtained at time of FGR diagnosis. Utilizing the RF algorithm, we identified 6 key factors by excluding variables with RF importance values of 0. These factors included abnormal UA Doppler, EFW z score, HDP, GA at diagnosis of FGR, AC z score EFW growth velocity (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eA). LASSO regression was also applied to discern factors and LASSO coefficients set to 0 were exclude. This analysis identified factors such as maternal BMI, GA at diagnosis of FGR, EFW z score, EFW growth velocity, abnormal UA Doppler (Figs.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eB). Univariate LR suggested 9 factors having statistically significant, such as abnormal UA Doppler, HDP, GA at diagnosis of FGR, AC growth velocity, EFW z score, maternal BMI, age, nulliparous, and EFW growth velocity (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eC). We delineated a common subset of feature variables endorsed by at least two methods of the results from LR, LASSO regression, and RF algorithm screening. These selected features were ultimately utilized in the construction of the model and consisted of maternal pre-pregnancy BMI, HDP, abnormal UA Doppler, GA at diagnosis of FGR, EFW z score, EFW growth velocity.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003e3.3 Model performance\u003c/h2\u003e \u003cp\u003eIn this study, the balanced accuracy, sensitivity, PPV, NPV, and F1 score of each model were computed (Table S2 and S3, Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eA, \u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eB). Among independent ML models in the train set, the AUROC illustrated that RF and SVM had the best predictive performance, with AUC values of 1.000 and 0.909, respectively, followed by LR with AUC values of 0.906 and Nnet with AUROC values of 0.855. Stacking ensemble models were subsequently developed based on LR, SVM and RF. In the test set, the Stacking model had the best performance, with a highest AUC of 0.861 and a calibration degree of 0.111. The Stacking model had the balanced accuracy (0.850), sensitivity (0.884), NPV (0.954) and F1 score (0.754). The DeLong test showed that the AUROC of the Stacking model was higher than that of several other models, and the difference was statistically significant (P\u0026thinsp;\u0026lt;\u0026thinsp;0.05). In summary, the Stacking model performed the best in the test set and is thus recommended as the preferred model for the prediction APO of FGR at time of diagnosis.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eIn this study, we evaluated the prediction accuracy and calibration of the Stacking model by calibration curve analysis of the training and test sets (Figs.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eC, \u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eD), respectively. The predicted values closely correlate with the observed values, indicating good model calibration. The results of the Hosmer\u0026ndash;Lemeshow test for the training set and the test set were \u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.387 and \u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.825, respectively, indicating that the model showed good fit.\u003c/p\u003e \u003cp\u003eThe DCA for the training and test set are depicted in Figs.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eE and \u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eF, respectively, and the clinical impact curves are presented in Figs.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eG and \u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eH, respectively. These curves show that the model yields great Net Benefit over a large threshold probability range.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003e3.4 Feature importance\u003c/h2\u003e \u003cp\u003eAccording to the SHAP values, the GA at diagnosis of FGR was the most important factor, followed by abnormal UA Doppler. See Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eA for more details on variable importance. The GA at diagnosis of FGR was positively correlated with the APO risk in FGR. In contrast, it was negatively correlated with variables including abnormal UA Doppler and HDP. See Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eB for correlations between variables and outcomes.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec15\" class=\"Section2\"\u003e \u003ch2\u003e3.5 Construction of nomograms\u003c/h2\u003e \u003cp\u003eThe predictive ability of each independent marker included in the Stacking model for APO in FGR is reflected in the nomogram (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eA). Instructions on the use of the nomogram model are provided in Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eB. As an example, consider a pregnant woman with abnormal UA Doppler and no HDP, GA of 30 weeks at diagnosis of FGR, a pre-pregnancy BMI of 24 kg/m\u0026sup2;, and an EFW z-score and EFW growth velocity of -3 and 30 g/week, respectively. Her final risk score would be 33(abnormal UA)\u0026thinsp;+\u0026thinsp;0 (no HDP)\u0026thinsp;+\u0026thinsp;28.5 (GA at diagnosis of FGR)\u0026thinsp;+\u0026thinsp;50 (EFW z score)\u0026thinsp;+\u0026thinsp;8.5 (pre-pregnancy BMI)\u0026thinsp;+\u0026thinsp;25 (EFW growth velocity)\u0026thinsp;=\u0026thinsp;145, which corresponds to a medium risk stratification with an APO probability of 0.72.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e"},{"header":"4. Discussion","content":"\u003cdiv id=\"Sec17\" class=\"Section2\"\u003e \u003ch2\u003e4.1 Main findings\u003c/h2\u003e \u003cp\u003eTo our knowledge, this is the first ML model to predict APO in FGR at time of diagnosis. We identified six available predictive risk factors in routine prenatal care and used various ML algorithms to construct an APO risk prediction model for FGR at time of diagnosis. The Stacking model, which has an AUC of 0.861, is effective for predicting APO in FGR at the time of diagnosis.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec18\" class=\"Section2\"\u003e \u003ch2\u003e4.2 Comparison with previous studies\u003c/h2\u003e \u003cp\u003eTo date, studies on predicting the risk of APO in FGR have been reported. However, most of these studies use a single method (traditional LR) for model building \u003csup\u003e(\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e, \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e, \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e)\u003c/sup\u003e, which may not be capable of handling complex relationships and hence affect prediction performance. Combining clinical and ultrasound parameters with complex ML algorithms can facilitate the development of clinical prediction models \u003csup\u003e(\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e)\u003c/sup\u003e. Among the six ML models, the Stacking model had the highest AUC, sensitivity, with good F1 score, calibration, and net benefit. Stacking surpasses other models due to its nature as a heterogeneous ensemble method, which allows it to leverage the strengths of various base learners. Multiple studies have demonstrated that the Stacking method is very valuable for prediction models in the medical field \u003csup\u003e(\u003cspan additionalcitationids=\"CR37\" citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e)\u003c/sup\u003e. In the present study, we employed the Stacking algorithm to develop a final model containing 6 features. These features can be easily acquired and evaluated during routine prenatal care, making this model a promising tool for effectively predicting the risk of APO in FGR at time of diagnosis.\u003c/p\u003e \u003cp\u003eTo date the majority of published data have demonstrated that a high BMI, HDP and early onset FGR are associated with a greater likelihood of APO \u003csup\u003e(\u003cspan additionalcitationids=\"CR40\" citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e41\u003c/span\u003e)\u003c/sup\u003e. These factors may associate with blood vessel generation and remodeling, thereby influencing the establishment and maintenance of a stable fetal supply of nutrition and oxygen through the placenta, leading to adverse effects on fetal growth and well-being due to placental insufficiency \u003csup\u003e(\u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e42\u003c/span\u003e)\u003c/sup\u003e. With regard to GA at diagnosis of FGR, APO were related to its etiology, which affected fetuses since the early stages of pregnancy \u003csup\u003e(\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e)\u003c/sup\u003e. These factors interact and result in a vicious cycle. Our data appears to corroborate higher BMI, HDP and early A at diagnosis of FGR are predictors for APO.\u003c/p\u003e \u003cp\u003eGiven that ultrasound assessment of biometry and Doppler velocimetry forms the mainstay of identification and monitoring of FGR, it is unsurprising that the EFW z score and UA category were validated as predictors of fetal or neonatal APO \u003csup\u003e(\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e)\u003c/sup\u003e. The predictor variables in our study reached similar conclusions. Alfirevic et al \u003csup\u003e(\u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e43\u003c/span\u003e)\u003c/sup\u003e reported that using of Doppler ultrasound on the UA in high-risk pregnancies reduced the risk of perinatal deaths by 29%. Baschat et al \u003csup\u003e(\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e)\u003c/sup\u003e also confirmed that UA Doppler allowed identification of fetuses at risk for adverse outcome but was a poor predictor of the fetal condition. Likewise, EFW/AC\u0026lt;3rd was associated with a higher risk of perinatal complications, it performed poorly as a standalone parameter in predicting adverse perinatal outcome with AUC of 0.60 \u003csup\u003e(\u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e44\u003c/span\u003e)\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eThus, ISUOG practice guideline recommended continuous measurements should be performed because they offer significant advantages over single-point measurements in determining true growth parameters and growth trajectories in assessment of fetal growth \u003csup\u003e(\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e)\u003c/sup\u003e. Previous studies have shown that fetal growth velocity was independently associated with perinatal outcomes and monitoring growth velocity may help identify at-risk pregnancies and prevent APO \u003csup\u003e(\u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e45\u003c/span\u003e, \u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e46\u003c/span\u003e)\u003c/sup\u003e. Poor growth might be an endpoint of several pathological changes \u003csup\u003e(\u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e42\u003c/span\u003e)\u003c/sup\u003e. Hence, assessing growth velocity might be a more appropriate marker of adverse outcomes. This study is the first to integrate growth velocity into prediction model, providing longitudinal growth information.\u003c/p\u003e \u003cp\u003eSeveral recent studies have highlighted the potential utility of PlGF and sFlt-1/PIGF concentration to predict outcomes in small-for-gestational-age (SGA) and FGR pregnancies \u003csup\u003e(\u003cspan additionalcitationids=\"CR48\" citationid=\"CR47\" class=\"CitationRef\"\u003e47\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR49\" class=\"CitationRef\"\u003e49\u003c/span\u003e)\u003c/sup\u003e. However, the serum biomarkers for FGR risk assessment were relatively limited, and they might increase the economic and psychological burden on pregnant women, so they should not be considered a priority. Our study combined various ultrasound parameters with clinical features significantly enhance the accuracy of predicting APO in FGR, achieving an AUC of 0.861. Possible reasons for this result might be that fetal hypoxic injury might vary in the target organs most affected and the timing of deterioration, which was why the combination of different fetal parameters continues to be the best model at this point \u003csup\u003e(\u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e50\u003c/span\u003e)\u003c/sup\u003e.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec19\" class=\"Section2\"\u003e \u003ch2\u003e4.3 Strength and Limitations\u003c/h2\u003e \u003cp\u003eA major advantage of the present study is that the nomogram is a visualizable tool. By using point scales to assign scores to each predictive factor, the scores of all predictive factors are summed up to obtain a total score. Based on the total score, the corresponding risk value can be derived, and risk stratification can be performed, making it a convenient tool for clinical use. Another strength of the present study is that the predictive factors included in the model were all routine items obtained for patients during prenatal care and were easy to obtain, providing feasibility for the promotion and application of the model in clinical practice.\u003c/p\u003e \u003cp\u003eHowever, we acknowledge some limitations of this study. Firstly, being a retrospective study, there was a potential for omitted data and selection bias to impact the results. Secondly, from an ML perspective, the small size of the dataset, which posed the risk of overfitting the obtained results. To mitigate the risk of overfitting, this research employed a cross-validation methodology that went beyond mere partitioning of the overall dataset. Thirdly, the small sample size of this study and the fact that the sample was collected from a single center may limit the generalizability of the findings. However, Beijing, as the capital of China, has a relatively diverse population, including people from all over the country. This may to some extent mitigate some of the limitations of single-center studies. In addition, APO are not only outcomes of FGR but also include adverse outcomes of prematurity, leading to intervention bias. To address this, we will continue to explore ways to mitigate this bias in future research. Finally, due to the small sample size, we were unable to differentiate between early-onset FGR and late-onset FGR. In the future, we will conduct multicenter studies with larger sample sizes and develop predictive models targeted at each subtype.\u003c/p\u003e \u003c/div\u003e"},{"header":"5. Conclusion","content":"\u003cp\u003eWe proved that the Stacking model, composed of six factors readily available in daily clinical practice at the time of FGR diagnosis, can accurately predict APO in FGR and is expected to be an effective assistant tool for predict APO of FGR at the time of diagnosis. This nomogram may be used for personalized counseling, allocation of resources, risk stratification and management of pregnancies complicated by FGR.\u003c/p\u003e"},{"header":"Abbreviations","content":"\u003cdiv class=\"DefinitionList\"\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003eML\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003emachine learning\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003eAPO\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eAdverse perinatal outcome\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003eFGR\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eFetal growth restriction\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003eRF\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eRandom forest\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003eLASSO\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eLeast Absolute Shrinkage and Selection Operator\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003eLR\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eLogistic regression\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003eEFW\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eEstimated fetal weight\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003eSGA\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eSmall for gestational age\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003eBJOGH\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eBeijing Obstetrics and Gynecology Hospital\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003eISUOG\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eInternational Society of Ultrasound in Obstetrics and Gynecology\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003eGA\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eGestational age\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003eAC\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eAbdominal circumference\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003eUA\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eUmbilical artery\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003eNICHD\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eNational Institute of Child Health and Human Development\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003ePI\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003ePulsatility index\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003eAREDF\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eAbsent or reversed end diastolic flow\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003eBMI\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eBody mass index\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003eHDP\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eHypertensive disorder of pregnancy\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003eSVM\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eSupport Vector Machine\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003eNnet\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eNeural network\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003eXGboost\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eExtreme gradient boosting\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003eAUROC\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eArea under the receiver operating characteristic curve\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003ePPV\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003ePositive predictive value\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003eNPV\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003enegative predictive value\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003eDCA\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eDecision curve analysis\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003eCIC\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eClinical impact curve\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003eSHAP\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eShapley Additive exPlanations\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003c/div\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eEthics approval and consent to participate\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study was approved by the Beijing Obstetrics and Gynecology Hospital, Maternal and Child Health Centre, Capital Medical University Ethics Committee (NO.2023-KY-079-01). Given the retrospective nature of the study and the use of de-identified patient data, the requirement for informed consent was waived by the Beijing Obstetrics and Gynecology Hospital, Maternal and Child Health Centre, Capital Medical University Ethics Committee. The study was conducted in accordance with the ethical standards of the Declaration of Helsinki and its later amendments.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent for publication\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAvailability of data and material\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe datasets generated or analyzed during this study are available from the corresponding author on reasonable request.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting Interests\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors have no relevant financial or non-financial interests to disclose.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study was supported by the National Key Research and Development Program of China (2022YFC2703300, 2022YFC2703301). Beijing Municipal Health Commission, demonstration construction project of Clinical Research ward (NO: BCRW202109).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthors\u0026rsquo; Contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAll authors (Xiangli Meng, Lei Wang, Minghui Wu, Na Zhang, Xiaofei Li and Qingqing Wu) contributed to the study conception and design. Material preparation, data collection and analysis were performed by Xiangli Meng, Lei Wang. The first draft of the manuscript was written by Xiangli Meng and all authors (Xiangli Meng, Lei Wang, Minghui Wu, Na Zhang, Xiaofei Li and Qingqing Wu) commented on previous versions of the manuscript. All authors (Xiangli Meng, Lei Wang, Minghui Wu, Na Zhang, Xiaofei Li and Qingqing Wu) read and approved the final manuscript.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowlegements\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eBarker DJ, Osmond C, Fors\u0026eacute;n TJ, Kajantie E, Eriksson JG. Trajectories of growth among children who have coronary events as adults. N Engl J Med. 2005;353(17):1802\u0026ndash;9.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBai\u0026atilde;o AER, de Carvalho PRN, Moreira MEL, de S\u0026aacute; RAM, Junior SCG. Predictors of perinatal outcome in early-onset fetal growth restriction: A study from an emerging economy country. Prenat Diagn. 2020;40(3):373\u0026ndash;9.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBaschat AA, Gembruch U, Weiner CP, Harman CR. Qualitative venous Doppler waveform analysis improves prediction of critical perinatal outcomes in premature growth-restricted fetuses. Ultrasound Obstet Gynecol. 2003;22(3):240\u0026ndash;5.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBilardo CM, Wolf H, Stigter RH, Ville Y, Baez E, Visser GH, Hecher K. Relationship between monitoring parameters and perinatal outcome in severe, early intrauterine growth restriction. Ultrasound Obstet Gynecol. 2004;23(2):119\u0026ndash;25.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHernandez-Andrade E, Crispi F, Benavides-Serralde JA, Plasencia W, Diesel HF, Eixarch E, et al. Contribution of the myocardial performance index and aortic isthmus blood flow index to predicting mortality in preterm growth-restricted fetuses. Ultrasound Obstet Gynecol. 2009;34(4):430\u0026ndash;6.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eOdibo AO, Goetzinger KR, Cahill AG, Odibo L, Macones GA. Combined sonographic testing index and prediction of adverse outcome in preterm fetal growth restriction. Am J Perinatol. 2014;31(2):139\u0026ndash;44.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSpencer R, Maksym K, Hecher K, Marš\u0026aacute;l K, Figueras F, Ambler G et al. Maternal PlGF and umbilical Dopplers predict pregnancy outcomes at diagnosis of early-onset fetal growth restriction. J Clin Invest. 2023;133(18).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePini N, Lucchini M, Esposito G, Tagliaferri S, Campanile M, Magenes G, Signorini MG. A Machine Learning Approach to Monitor the Emergence of Late Intrauterine Growth Restriction. Front Artif Intell. 2021;4:622616.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSufriyana H, Wu YW, Su EC. Prediction of Preeclampsia and Intrauterine Growth Restriction: Development of Machine Learning Models on a Prospective Cohort. JMIR Med Inf. 2020;8(5):e15411.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTeng LY, Mattar CNZ, Biswas A, Hoo WL, Saw SN. Interpreting the role of nuchal fold for fetal growth restriction prediction using machine learning. Sci Rep. 2022;12(1):3907.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAnanth CV, Brandt JS. Fetal growth and gestational age prediction by machine learning. Lancet Digit Health. 2020;2(7):e336\u0026ndash;7.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKuhle S, Maguire B, Zhang H, Hamilton D, Allen AC, Joseph KS, Allen VM. Comparison of logistic regression with machine learning methods for the prediction of fetal growth abnormalities: a retrospective cohort study. BMC Pregnancy Childbirth. 2018;18(1):333.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMustafa HJ, Kalafat E, Prasad S, Heydari MH, Nunge RN, Khalil A. Machine learning prediction of hypertension and diabetes in twin pregnancies using characteristics at prenatal care entry: a nationwide study. Ultrasound Obstet Gynecol; 2024.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePunyapet P, Suwanrath C, Chainarong N, Sawaddisan R, Vichitkunakorn P. Predictors of adverse perinatal outcomes in fetal growth restriction using a combination of maternal clinical factors and simple ultrasound parameters. Int J Gynaecol Obstet. 2023;162(2):669\u0026ndash;75.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLees CC, Stampalija T, Baschat A, da Silva Costa F, Ferrazzi E, Figueras F, et al. ISUOG Practice Guidelines: diagnosis and management of small-for-gestational-age fetus and fetal growth restriction. Ultrasound Obstet Gynecol. 2020;56(2):298\u0026ndash;312.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCommittee opinion. 611: method for estimating due date. Obstet Gynecol. 2014;124(4):863\u0026ndash;6.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSalomon LJ, Alfirevic Z, Da Silva Costa F, Deter RL, Figueras F, Ghi T, et al. ISUOG Practice Guidelines: ultrasound assessment of fetal biometry and growth. Ultrasound Obstet Gynecol. 2019;53(6):715\u0026ndash;23.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBhide A, Acharya G, Baschat A, Bilardo CM, Brezinka C, Cafici D, et al. ISUOG Practice Guidelines (updated): use of Doppler velocimetry in obstetrics. Ultrasound Obstet Gynecol. 2021;58(2):331\u0026ndash;9.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFetal Medicine Subgroup SoPM, Chinese Medical Association, Subgroup O. Society of Obstetrics and Gecology, Chinese Medical Association. Expert consensus on fetal growth restriction. Chin J Perinat Med. 2019;22(06):361\u0026ndash;80.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBuck Louis GM, Grewal J, Albert PS, Sciscione A, Wing DA, Grobman WA, et al. Racial/ethnic standards for fetal growth: the NICHD Fetal Growth Studies. Am J Obstet Gynecol. 2015;213(4):449. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e.e1-.e41\u003c/span\u003e\u003cspan address=\"http://.e1-.e41\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCiobanu A, Wright A, Syngelaki A, Wright D, Akolekar R, Nicolaides KH. Fetal Medicine Foundation reference ranges for umbilical artery and middle cerebral artery pulsatility index and cerebroplacental ratio. Ultrasound Obstet Gynecol. 2019;53(4):465\u0026ndash;72.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRocha AS, Andrade ARA, Moleiro ML, Guedes-Martins L. Doppler Ultrasound of the Umbilical Artery: Clinical Application. Rev Bras Ginecol Obstet. 2022;44(5):519\u0026ndash;31.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGrantz KL, Kim S, Grobman WA, Newman R, Owen J, Skupski D, et al. Fetal growth velocity: the NICHD fetal growth studies. Am J Obstet Gynecol. 2018;219(3):285. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e.e1-.e36\u003c/span\u003e\u003cspan address=\"http://.e1-.e36\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHealy P, Gordijn SJ, Ganzevoort W, Beune IM, Baschat A, Khalil A et al. A Core Outcome Set for the prevention and treatment of fetal GROwth restriction: deVeloping Endpoints: the COSGROVE study. Am J Obstet Gynecol. 2019;221(4):339.e1-.e10.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGoldstein BA, Navar AM, Carter RE. Moving beyond regression techniques in cardiovascular risk prediction: applying machine learning to address analytic challenges. Eur Heart J. 2016;38(23):1805.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBreiman L. Random Forests. Mach Learn. 2001;45(1):5\u0026ndash;32.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eR T. Regression Shrinkage and Selection via the Lasso. J Royal Stat Society: Ser B Stat Methodol. 1996;73(3):273\u0026ndash;82.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGoldstein R. Regression Methods in Biostatistics: Linear, Logistic, Survival and Repeated Measures Models. Technometrics. 2006;48:149\u0026ndash;50.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNaimi AI, Balzer LB. Stacked generalization: an introduction to super learning. Eur J Epidemiol. 2018;33(5):459\u0026ndash;64.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLundberg SMaL S-I. A Unified Approach to Interpreting Model Predictions. Proceedings of the 31st International Conference on Neural Information Processing Systems (NIPS 2017). Long Beach, CA, USA. 2017:4768-77.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKramer AA, Zimmerman JE. Assessing the calibration of mortality benchmarks in critical care: The Hosmer-Lemeshow test revisited. Crit Care Med. 2007;35(9):2052\u0026ndash;6.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDeLong ER, DeLong DM, Clarke-Pearson DL. Comparing the areas under two or more correlated receiver operating characteristic curves: a nonparametric approach. Biometrics. 1988;44(3):837\u0026ndash;45.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSteyerberg EW, Vickers AJ, Cook NR, Gerds T, Gonen M, Obuchowski N, et al. Assessing the performance of prediction models: a framework for traditional and novel measures. Epidemiology. 2010;21(1):128\u0026ndash;38.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWHO Child Growth. Standards based on length/height, weight and age. Acta Paediatr Suppl. 2006;450:76\u0026ndash;85.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGoecks J, Jalili V, Heiser LM, Gray JW. How Machine Learning Will Transform Biomedicine. Cell. 2020;181(1):92\u0026ndash;101.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWang X, Ren H, Ren J, Song W, Qiao Y, Ren Z, et al. Machine learning-enabled risk prediction of chronic obstructive pulmonary disease with unbalanced data. Comput Methods Programs Biomed. 2023;230:107340.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWang J, Chen H, Wang H, Liu W, Peng D, Zhao Q, Xiao M. A Risk Prediction Model for Physical Restraints Among Older Chinese Adults in Long-term Care Facilities: Machine Learning Study. J Med Internet Res. 2023;25:e43815.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSun T, Liu J, Yuan H, Li X, Yan H. Construction of a risk prediction model for lung infection after chemotherapy in lung cancer patients based on the machine learning algorithm. Front Oncol. 2024;14:1403392.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYang Z, Feng G, Gao X, Yan X, Li Y, Wang Y et al. Maternal adiposity and perinatal and offspring outcomes: an umbrella review. Nat Hum Behav. 2024.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWebster K, Fishburn S, Maresh M, Findlay SC, Chappell LC. Diagnosis and management of hypertension in pregnancy: summary of updated NICE guidance. BMJ. 2019;366:l5119.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFigueras F, Gratac\u0026oacute;s E. Update on the diagnosis and classification of fetal growth restriction and proposal of a stage-based management protocol. Fetal Diagn Ther. 2014;36(2):86\u0026ndash;98.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDi Martino DD, Avagliano L, Ferrazzi E, Fus\u0026egrave; F, Sterpi V, Parasiliti M et al. Hypertensive Disorders of Pregnancy and Fetal Growth Restriction: Clinical Characteristics and Placental Lesions and Possible Preventive Nutritional Targets. Nutrients. 2022;14(16).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAlfirevic Z, Stampalija T, Dowswell T. Fetal and umbilical Doppler ultrasound in high-risk pregnancies. Cochrane Database Syst Rev. 2017;6(6):Cd007529.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMeler E, Martinez-Portilla RJ, Caradeux J, Mazarico E, Gil-Armas C, Boada D, et al. Severe smallness as predictor of adverse perinatal outcome in suspected late small-for-gestational-age fetuses: systematic review and meta-analysis. Ultrasound Obstet Gynecol. 2022;60(3):328\u0026ndash;37.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHugh O, Gardosi J. Fetal weight projection model to define growth velocity and validation against pregnancy outcome in a cohort of serially scanned pregnancies. Ultrasound Obstet Gynecol. 2022;60(1):86\u0026ndash;95.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBommarito PA, Cantonwine DE, Stevens DR, Welch BM, Davalos AD, Zhao S, et al. An application of group-based trajectory modeling to define fetal growth phenotypes among small-for-gestational-age births in the LIFECODES Fetal Growth Study. Am J Obstet Gynecol. 2023;228(3):334. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e.e1-.e21\u003c/span\u003e\u003cspan address=\"http://.e1-.e21\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSharp A, Jackson R, Cornforth C, Harrold J, Turner MA, Kenny L, et al. A prediction model for short-term neonatal outcomes in severe early-onset fetal growth restriction. Eur J Obstet Gynecol Reprod Biol. 2019;241:109\u0026ndash;18.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRodr\u0026iacute;guez-Calvo J, Villala\u0026iacute;n C, G\u0026oacute;mez-Arriaga PI, Quezada MS, Herraiz I, Galindo A. Prediction of perinatal survival in early-onset fetal growth restriction: role of placental growth factor. Ultrasound Obstet Gynecol. 2023;61(2):181\u0026ndash;90.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMendoza M, Hurtado I, Bonacina E, Garcia-Manau P, Serrano B, Tur H, et al. Individual risk assessment for prenatal counseling in early-onset growth-restricted and small-for-gestational-age fetuses. Acta Obstet Gynecol Scand. 2021;100(3):504\u0026ndash;12.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLees CC, Marlow N, van Wassenaer-Leemhuis A, Arabin B, Bilardo CM, Brezinka C, et al. 2 year neurodevelopmental and intermediate perinatal outcomes in infants with very preterm fetal growth restriction (TRUFFLE): a randomised trial. Lancet. 2015;385(9983):2162\u0026ndash;72.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"small-for-gestational age, growth restriction, perinatal morbidity and mortality, machine learning, predictive model, nomogram","lastPublishedDoi":"10.21203/rs.3.rs-5698332/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-5698332/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eObjective\u003c/h2\u003e \u003cp\u003eThe objective of this study was to create and validate a machine learning (ML)-based model for predicting the adverse perinatal outcome (APO) in fetuses with a perinatal diagnosis of fetal growth restriction (FGR).\u003c/p\u003e\u003ch2\u003eMethod\u003c/h2\u003e \u003cp\u003eThis was a retrospective study of singleton gestations meeting the ISUOG-endorsed criteria for FGR from January 2021 and November 2023 at Beijing Obstetrics and Gynecology Hospital. The APO comprised one or more of: perinatal demise (stillbirth, immediate neonatal demise, or death before neonatal intensive care unit) discharge, cord arterial pH\u0026thinsp;\u0026le;\u0026thinsp;7.10, and/or base excess\u0026thinsp;\u0026ge;\u0026thinsp;12, bronchopulmonary dysplasia, hypoxic ischemic encephalopathy, grade III-IV intraventricular hemorrhage, periventricular leukomalacia, seizures, necrotizing enterocolitis, and sepsis. Feature screening was performed using the random forest (RF), the Least Absolute Shrinkage and Selection Operator (LASSO) and logistic regression (LR). Subsequently, six ML methods, including Stacking, were used to construct models to predict the APO of FGR. Model\u0026rsquo;s performance was evaluated through indicators such as the area under the receiver operating characteristic curve (AUROC). The Shapley Additive exPlanation analysis was used to rank each model feature and explain the final model. Finally, we constructed a nomogram to make the predictive model results more readable.\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e \u003cp\u003eIn total, a cohort of 411 non-anomalous singleton pregnancies with FGR were divided into a training set and a test set at a ratio of approximately 7:3. Among 16 candidate predictors (including maternal characteristics, maternal comorbidities, obstetric characteristics, and ultrasound parameters), the integration of RF, LASSO, and LR methodologies identified maternal pre-pregnancy body mass index, hypertensive disorders of pregnancy, gestational age at diagnosis of FGR, estimated fetal weight (EFW) z-score, EFW growth velocity, and abnormal umbilical artery Doppler as significant predictors. The Stacking model demonstrated a good performance in the test set (AUROC: 0.861,95% confidence interval (0.838\u0026ndash;0.896)). The calibration curve and Hosmer-Lemeshow test demonstrated good calibration.\u003c/p\u003e\u003ch2\u003eConclusions\u003c/h2\u003e \u003cp\u003eThe ML algorithm was developed to possess the promising capacity of predicting APO in FGR at time of diagnosis. This approach may potentially improve early detection at high risk of APO in FGR.\u003c/p\u003e","manuscriptTitle":"Construction of a prediction model for adverse perinatal outcomes in fetal growth restriction based on a machine learning algorithm","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-04-16 17:41:40","doi":"10.21203/rs.3.rs-5698332/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"1af7115f-4041-4de1-a3d0-1fd37ce37244","owner":[],"postedDate":"April 16th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2025-05-26T13:39:53+00:00","versionOfRecord":[],"versionCreatedAt":"2025-04-16 17:41:40","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-5698332","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-5698332","identity":"rs-5698332","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}