FCNN模型对妊娠早期高血压和子痫前期-子痫发病风险的预测价值  

Predictive Value of Fully Connected Neural Network Model in Early Pregnancy Hypertension and Pre-eclampsia/Eclampsia

作　　者：冯薇 伊诺 FENG Wei;YI Nuo(Department of Obstetrics and Gynecology,NO.12 Clinical College of Medicine,Capital Medical University Beijing Ditan Hospital,Capital Medical University,Beijing 100102,China)

机构地区：[1]首都医科大学第十二临床医学院、首都医科大学附属北京地坛医院妇产科,北京100102

出　　处：《转化医学杂志》2023年第4期180-184,共5页Translational Medicine Journal

基　　金：吴阶平医学基金会临床科研专项资助基金(320.6750.17190)。

摘　　要：目的分析全连接神经网络(FCNN)模型对妊娠早期高血压及子痫前期-子痫发病风险的预测价值。方法收集2017—2020年727例妊娠11~13+6周孕妇病历资料,包括妊娠期高血压及子痫前期-子痫发病的高危因素。检测孕妇外周血胎盘生长因子、妊娠相关血浆蛋白-A(PAPP-A)水平及双上肢血压数值。应用人工智能深度学习技术构建FCNN模型,采用受试者工作特征曲线分析其对妊娠期高血压及子痫前期-子痫发病的预测效能,并与美国妇产科医师学会(ACOG)、英国国家卫生与临床优化研究所(NICE)方法的预测效能比较,同时评估高危因素对疾病发病风险的影响。结果727例诊断妊娠期高血压或子痫前期-子痫73例,发病率为10.04%。FCNN模型对妊娠早期高血压及子痫前期-子痫发病的检出率为95.45%,假阳性率为30.61%,预测发病风险的曲线下面积(AUC)为0.826(95%CI:0.755,0.897)。ACOG方法的检出率为86.36%,假阳性率为63.27%,预测发病风险的AUC为0.612(95%CI:0.500,0.724)。NICE方法的检出率为59.09%,假阳性率为11.22%,预测发病风险的AUC为0.742(95%CI:0.614,0.869)。FCNN模型计算已知高危因素的影响系数,其中位列前5位的是PAPP-A、多胎妊娠、子痫前期家族史、糖尿病病史、妊娠间隔时间>10年,影响系数分别为8.30、7.74、7.45、7.12、7.00。结论FCNN模型对妊娠早期高血压及子痫前期-子痫发病的预测效能较好,优于ACOG、NICE方法。FCNN模型还能评估高危因素对此类疾病发生的影响水平,有望成为疾病早期预测的有益参考和补充。Objective To analyze the value of fully connected neural network(FCNN)model in predicting the risk of early pregnancy hypertension and pre-eclampsia/eclampsia.Methods The medical records of 727 pregnant women at 11-13+6 weeks of gestation between 2017 and 2020 were collected,including high risk factors for onset of pregnancy-induced hypertension(PIH)and pre-eclampsia/eclampsia.The levels of peripheral blood placental growth factor,pregnancy-associated plasma protein-A(PAPP-A)and blood pressure of both upper limbs on the day of peripheral blood collection were measured.Artificial intelligence deep learning technology was applied to construct the FCNN model,the receiver operating characteristic(ROC)curve was used to analyze its predictive efficacy for onset of PIH and pre-eclampsia/eclampsia,and the predictive efficacy was compared with the methods of the American College of Obstetricians and Gynecologists(ACOG)and the National Institute for Health and Clinical Excellence(NICE).In the meantime,the impact of high risk factors on the risk of disease occurrence was evaluated.Results Among the 727 patients,73 patients were diagnosed with PIH or pre-eclampsia/eclampsia during pregnancy,with an incidence of 10.04%.The detection rate and false positive rate of FCNN model for early pregnancy hypertension and pre-eclampsia/eclampsia were 95.45%and 30.61%respectively,and the area under the curve(AUC)for predicting the risk was 0.826(95%CI:0.755,0.897).The detection rate of ACOG method was 86.36%,the false positive rate was 63.27%,and the AUC for predicting the risk of disease was 0.612(95%CI:0.500,0.724).The detection rate of NICE method was 59.09%,the false positive rate was 11.22%,and the AUC for predicting the risk of disease was 0.742(95%CI:0.614,0.869).FCNN model was used to calculate the influence coefficients of known risk factors,among which the top 5 were PAPP-A,multiple pregnancy,family history of pre-eclampsia,history of diabetes,and gestational interval>10 years,with the influence coefficients of 8.30,7.74,7.45,7

关 键 词：妊娠期高血压 子痫前期 子痫 全连接神经网络 预测 高危因素 妊娠相关血浆蛋白-A 受试者工作特征曲线 

分 类 号：R714.246[医药卫生—妇产科学]