基于人工神经网络的早发型子痫前期患者预测模型构建研究

Prediction of Early-Onset Preeclampsia Based on Artificial Neural Network

摘要: 目的：建立基于人工神经网络的子痫前期预测模型，为疾病早期筛查提供依据。方法：前瞻性收集2020年3月至2021年6月在青岛大学附属医院进行产前检查并分娩的741名孕妇资料，通过对所得资料进行单因素logistic回归以及多因素logistic回归筛选出子痫前期的独立危险因素。将所得独立危险因素运用人工神经网络算法进行预测模型的拟合，运用受试者工作特征曲线(ROC曲线)对模型进行评估。结果：所调查孕妇中共有71例(9.5%)发生早发型子痫前期，670例(90.5%)未发生早发型子痫前期，多因素logistic回归显示早发型子痫前期的独立危险因素有孕前BMI、孕次、孕前是否吸烟、孕前是否饮酒、平均动脉压(MAP)、葡萄糖、AST/ALT、血清游离三碘甲腺原氨酸(FT3)、甲胎蛋白(AFP)，9项指标；所得预测模型预测早发型PE的ROC曲线下面积为0.945。结论：基于人工神经网络的早发型PE预测模型不仅为早发型PE提供了理论和方法的支持，为疾病的早发现、早诊断、早治疗争取了时间，具有广阔的应用前景。

Abstract: Objective: To establish a prediction of early-onset preeclampsia based on artificial neural network, so as to provide a basis for the early screening of the disease. Methods: A prospective study was conducted on the data of 741 pregnant women who underwent prenatal examination and delivery in The Affiliated Hospital of Qingdao University from March 2020 to June 2021. Univariate logistic regression and multivariate logistic regression were used to screen the independent risk factors for preeclampsia. The independent risk factors were fitted to the prediction model by artificial neural network algorithm, and the receiver operating characteristic curve (ROC curve) was used to evaluate the model. Results: A total of 71 (9.5%) pregnant women developed early-onset preeclampsia, while 670 (90.5%) did not. Multivariate logistic regression showed that the independent risk factors for early-onset preeclampsia were BMI before pregnancy, pregnancy times, smoking before pregnancy, drinking before pregnancy, MAP, glucose, AST/ALT, serum free triiodothyronine (FT3), alpha-fetoprotein (AFP). The predictive model predicted that the area under ROC curve of early-onset PE was 0.945. Conclusion: The prediction model of early-onset PE based on artificial neural network not only provides theoretical and method support for early-onset PE, but also buys time for early detection, early diagnosis and early treatment of disease, which has broad application prospect.