Application of Early Pregnancy Plasma AEA Levels Combined with Clinical and Laboratory Indicators in Predicting the Risk of Preeclampsia Based on Machine Learning: A Retrospective Study

目的：安非他命（AEA）与子痫前期的早期不良事件密切相关。本研究旨在调查妊娠早期孕妇血浆中AEA的水平，并分析其在预测先兆子痫风险方面的价值。方法：共纳入200例子痫前期孕妇和200例正常孕妇。收集妊娠早期（6-12周）的外周血样本。通过灵敏度、特异性、阳性预测值、阴性预测值计算和受试者工作特征 （ROC） 曲线分析来评估机器学习模型的预测性能。结果：子痫前期孕妇外周血AEA水平为4.09±1.66 pmol/ml，低于正常组的5.67±1.30 pmol/ml（P<0.001）。相关性分析表明，AEA与子痫前期呈负相关（R=-0.47，P<0.001）。包含 AEA 的预测模型显示最佳曲线下面积 （AUC） 为 0.95，优于没有 AEA 的模型，后者的最佳 AUC 为 0.923。结论：利用血浆AEA水平和临床指标建立的预测模型可作为预测子痫前期的有效工具。它可以区分最早可能在怀孕三个月后发生子痫前期的患者。关键词：机器学习;安非他命;内源性大麻素;先兆子痫;预测模型

Purpose: Amphetamine (AEA) is closely associated with early adverse events of preeclampsia. This study aims to investigate the plasma levels of AEA in pregnant women during early pregnancy and analyze its value in predicting the risk of preeclampsia. Methods: A total of 200 pregnant women with preeclampsia and 200 normal pregnant women were enrolled. Peripheral blood samples were collected during early pregnancy (6-12 weeks). The predictive performance of the machine learning model was evaluated by calculating sensitivity, specificity, positive predictive value, negative predictive value and performing receiver operating characteristic (ROC) curve analysis. Results: The peripheral blood AEA level of pregnant women with preeclampsia was 4.09±1.66 pmol/ml, which was lower than 5.67±1.30 pmol/ml in the normal control group (P<0.001). Correlation analysis showed that AEA was negatively correlated with preeclampsia (R=-0.47, P<0.001). The predictive model incorporating AEA exhibited an optimal area under the curve (AUC) of 0.95, which outperformed the model without AEA, whose optimal AUC was 0.923. Conclusions: The predictive model established based on plasma AEA levels and clinical indicators can serve as an effective tool for predicting preeclampsia. It can distinguish patients who may develop preeclampsia as early as three months after pregnancy. Keywords: Machine learning; Amphetamine; Endocannabinoid; Preeclampsia; Predictive model