子宫肌瘤体积与胎儿分数：可解释建模新发现
本研究回顾性纳入单胎妊娠且NIPT结果正常、数据完整的孕妇，利用决策树、随机森林和XGBoost建模预测胎儿分数是否充足（≥4%）。XGBoost预测准确率最高（0.89），SHAP分析显示年龄、BMI和孕周影响最大，子宫肌瘤体积次之；在24例子宫肌瘤合并胎儿分数不足患者中，模拟缩小肌瘤体积常预测胎儿分数改善。
要点
- XGBoost预测胎儿分数是否充足的准确率最高，达0.89。
- 年龄、BMI和孕周对胎儿分数影响最大，子宫肌瘤体积次之。
- 模拟缩小子宫肌瘤体积常预测胎儿分数改善。
核心结论
- 子宫肌瘤体积是影响NIPT中胎儿分数水平的显著因素。
- 该预测模型有望在低胎儿分数给患者管理带来挑战时支持临床决策。
研究设计与样本
- 回顾性研究，纳入单胎妊娠、NIPT结果正常且数据完整的孕妇，样本来自一家三级医院。
- 分析母亲年龄、BMI、孕周、孕次、产次及子宫肌瘤体积等变量。
- 训练决策树、随机森林和XGBoost机器学习模型，将胎儿分数分类为充足（≥4%）或不足。
- 使用SHAP值解释表现最佳的XGBoost模型。
主要发现
- XGBoost达到最高预测准确率（0.89）。
- SHAP分析显示年龄、BMI和孕周影响最大，其次是子宫肌瘤体积。
- 对24例同时有子宫肌瘤和胎儿分数不足的患者进行情景模拟，缩小子宫肌瘤体积常预测胎儿分数改善。
关键数值与推荐
- 胎儿分数充足定义为≥4%。
- XGBoost预测准确率为0.89。
临床怎么用
- 可基于母亲年龄、BMI、孕周及子宫肌瘤体积等因素评估NIPT胎儿分数不足的风险。
- 对于子宫肌瘤合并低胎儿分数的孕妇，模型模拟提示缩小肌瘤体积可能改善胎儿分数，可辅助临床决策。
以上解读依据论文摘要（全文未开放获取或许可不允许翻译转载）。
原文
OBJECTIVE: This study aims to develop a predictive model to estimate the likelihood of achieving a sufficient fetal fraction (FF) for non-invasive prenatal testing (NIPT) based on maternal characteristics such as age, body mass index (BMI), gestational age, gravida, parity, and uterine leiomyoma volume, if present. METHOD: This retrospective study include singleton pregnancies with normal NIPT results and complete data from a tertiary hospital. Maternal and clinical variables are analyzed. Machine learning models (Decision Tree, Random Forest, XGBoost) are trained to classify FF as sufficient (≥ 4%) or insufficient. The best-performing model (XGBoost) is interpreted using SHAP values. Additionally, the impact of uterine leiomyoma volume on FF is demonstrated through scenario-based analyses derived from the model. RESULTS: XGBoost achieves the highest prediction accuracy (0.89). SHAP analysis shows that age, BMI, and gestational age are most influential, followed by uterine leiomyoma volume. Scenario-based simulations on 24 patients with both uterine leiomyoma and insufficient FF demonstrate that reducing uterine leiomyoma volume often led to a predicted improvement in FF. CONCLUSION: Uterine leiomyoma volume is identified as a significant factor influencing FF levels in NIPT. This predictive modeling has the potential to support clinical decision-making in cases where low FF poses challenges to effective patient management.
引用
- Uçar İ, Sarıyer G, Yaprak Uçar E, et al. Fetal Fraction and Uterine Leiomyoma Volume: New Insights From Interpretable Modeling. Prenat Diagn. 2026. PMID: 42025592. doi:10.1002/pd.70162
供执业医师工作参考，不替代临床判断；数值与分级以指南原文为准。