Development and Validation of a Small for Gestational Age Screening Model at 21–24 Weeks Based on the Real-World Clinical Data

Jing Gao, Zhongzhou Xiao, Chao Chen, Huwei Shi, Sen Yang, Lei Chen, Jie Xu, Weiwei ChengJournal of Clinical Medicine 2023 · DOI

TL;DR

This paper presents a novel nomogram-based screening model for Small for Gestational Age (SGA) infants, developed and validated using real-world clinical data from over 23,000 pregnancies. The model, designed for use in the mid-trimester (21-24 weeks), identifies key maternal and fetal factors to predict SGA risk, offering a more accurate and accessible tool for early intervention and improved pregnancy outcomes, particularly for high-risk preterm fetuses.

What Problem Does This Paper Address?

Small for Gestational Age (SGA) is a critical condition where a fetus's birth weight falls below the 10th percentile for its gestational age. This condition is associated with increased risks of perinatal morbidity and mortality, and can lead to long-term health issues in adulthood. Early and accurate screening for SGA is therefore highly desirable to allow for timely interventions and improved management of high-risk pregnancies. Traditional methods like maternal abdominal palpation are notoriously inaccurate. While ultrasound biometry and Doppler velocimetry have shown promise, there's a need for a comprehensive, widely applicable, and easily implemented screening tool that can be used earlier in pregnancy. This paper aims to fill that gap by developing and validating such a model for use between 21-24 weeks of gestation.

Methods

The researchers employed a retrospective observational study design. They utilized a large dataset of pregnant women and applied logistic regression analysis to identify independent risk factors for SGA. Specifically, they performed:

  • Univariate and Multivariate Logistic Regression: This was used to analyze the association between various maternal characteristics, laboratory test results, and sonographic parameters with SGA.
  • Backward Stepwise Selection: This statistical technique was used in the multivariate analysis to select the most significant predictors for the final model.
  • Nomogram Development: The identified significant predictors were used to construct a nomogram, a graphical tool that visually represents the probability of SGA based on individual patient data.
  • Model Validation: The nomogram's performance was assessed using both internal (training dataset) and external (validation dataset) validation. Key performance metrics included:
  • Discrimination: Measured by the Area Under the Curve (AUC) of the Receiver Operating Characteristic (ROC) curve.
  • Calibration: Assessed using the Hosmer-Lemeshow test and calibration curves to evaluate how well the predicted probabilities matched the observed outcomes.
  • Clinical Usefulness: Evaluated using Decision Curve Analysis (DCA) to determine the net benefit of using the model across different risk thresholds.
  • Subgroup Analysis: A post-hoc analysis was conducted specifically on the preterm subgroup of SGA fetuses to assess the model's performance in this high-risk population.

Datasets

The study utilized a real-world clinical dataset collected from the digital medical records system of the International Peace Maternity and Child Health Hospital in Shanghai.

  • Total Cohort: 30,757 pregnant women met the inclusion criteria.
  • Training Dataset: 11,746 women from January 2018 to December 2018.
  • Validation Dataset: 12,037 women from January 2019 to December 2019.

The data was nonrandomly classified based on the year of collection, which the authors argue is a better design for evaluating model performance against nonrandom variation between datasets.

Key Insights and Ideas

  • Mid-Trimester Screening is Feasible and Valuable: The study demonstrates that a robust SGA screening model can be developed and validated using data readily available during the mid-trimester ultrasound, a common point of care. This is earlier than many existing screening methods.
  • Multifactorial Nature of SGA Captured: The model successfully integrates a combination of maternal demographic factors (age, gravidity, parity, BMI), gestational age, and specific sonographic parameters (e.g., umbilical artery Doppler indices, abdominal circumference) to predict SGA. This highlights that SGA is not a single-cause condition.
  • Nomograms as Practical Tools: The development of a nomogram makes the predictive model accessible and interpretable for clinicians, translating complex statistical findings into a user-friendly visual aid for risk assessment.
  • Focus on High-Risk Subgroups: The model shows particular promise for identifying high-risk preterm SGA fetuses, suggesting its potential to guide more targeted and timely interventions for this vulnerable group.

Key Results

  • Model Performance: The developed SGA nomogram achieved an AUC of 0.70 (95% CI: 0.67–0.72) in the training dataset and 0.71 (95% CI: 0.68–0.73) in the validation dataset, indicating good discrimination ability.
  • Calibration: The model demonstrated favorable calibration, with the Hosmer-Lemeshow test yielding nonsignificant results (p = 0.84 in training, p = 0.39 in validation), suggesting good agreement between predicted and observed probabilities.
  • Predictive Factors: Twelve independent predictors were identified, including maternal age, gravidity, parity, BMI, gestational age, single umbilical artery (SUA), abdominal circumference (AC), humerus length (HL), anteroposterior trunk diameter (APTD), umbilical artery systolic/diastolic ratio (UA-S/D), transverse trunk diameter (TTD), and fasting plasma glucose (FPG). Notably, TTD was excluded due to correlation with APTD.
  • Preterm Subgroup Performance: For preterm SGA fetuses, the nomogram achieved a satisfactory performance with an average prediction rate of 86.3%.
  • Clinical Utility: Decision Curve Analysis suggested that the nomogram is more beneficial than "intervention for all" or "intervention for none" strategies when the threshold probability is between 4% and 15%.

Limitations and Future Work

  • Single-Center Study: The study was conducted at a single tertiary hospital, which may limit the generalizability of the findings to other populations, particularly those with different risk profiles or obstetric practices.
  • Retrospective Design: The retrospective nature of the study means that certain data points, such as gestational hypertension changes, drug treatments, and complications, could not be collected, potentially limiting the model's predictive power.
  • Exclusion of Fetal Sex: The study did not include fetal sex as a variable, as it is not revealed before delivery in their institution. This is a potential area for improvement.
  • Limited Variables: The authors acknowledge that the accuracy of the model could be improved by including more variables.
  • Future Directions: The authors suggest future studies should:
  • Incorporate more comprehensive data, including hypertension changes, drug treatments, and complications.
  • Utilize a prospective design to optimize the model.
  • Explore the inclusion of additional influencing factors.
  • Further investigate the model's utility in guiding timely diagnosis, intervention, and delivery for high-risk pregnancies.

Summary updated

Abstract

Our model is a reliable screening tool for SGA at 21-24 gestational weeks, especially for high-risk preterm fetuses. We believe that it will help clinical healthcare staff to arrange more comprehensive prenatal care examinations and, consequently, provide a timely diagnosis, intervention, and delivery.
BibTeX
@article{gao2023development,
  author = {Jing Gao and Zhongzhou Xiao and Chao Chen and Huwei Shi and Sen Yang and Lei Chen and Jie Xu and Weiwei Cheng},
  title = {{Development and Validation of a Small for Gestational Age Screening Model at 21–24 Weeks Based on the Real-World Clinical Data}},
  journal = {Journal of Clinical Medicine},
  year = {2023},
  doi = {10.3390/jcm12082993},
}
Figure