论文部分内容阅读
目的以子痫前期分度为例,比较反向传播(BP)神经网络模型与传统Logistic回归模型在复杂性疾病中的拟合效果及诊断效能。方法 2008年1月至2013年12月中山大学附属第一医院孕妇区收治的145例符合子痫前期诊断标准及实验入选标准的患者,其中轻度87例、重度58例;随机分组后,分成127例训练集和18例测试集以建立妊娠晚期子痫前期分度诊断的Logistic回归模型及BP神经网络诊断模型,并对模型的诊断效能进行比较,探讨子痫前期孕前和孕期高危因素。结果入选模型的变量有妊娠早期纤维蛋白原(Fbg)、血小板计数(PLT)、平均血小板体积(MPV)以及妊娠晚期尿蛋白;训练集中BP神经网络模型的一致率为80.30%,灵敏度为74.50%,特异度为84.21%,均高于Logistic回归模型的74.80%、58.82%、82.89%,且差异均有统计学意义(P<0.05)。BP神经网络模型的ROC曲线下面积为(0.887±0.029),大于Logistic回归模型的(0.823±0.036);在最佳诊断值界值的BP神经网络Youden index为67.0%,仍高于Logistic回归模型的54.1%,差异均有统计学意义(P<0.05)。结论 BP神经网络模型在妊娠晚期子痫前期分度中的拟合效果优于Logistic回归,更适合用于复杂性疾病多因素分析的研究;妊娠早期Fbg、PLT、MPV以及妊娠晚期尿蛋白与妊娠晚期子痫前期分度有关。
Objective To take the sub-index of preeclampsia as an example to compare the fitting effect and diagnostic efficacy of BP neural network model and traditional Logistic regression model in complex diseases. Methods From January 2008 to December 2013, 145 patients who met the diagnostic criteria of preeclampsia and the experimental inclusion criteria were enrolled in the First Affiliated Hospital of Sun Yat-sen University, 87 patients were mild and 58 were severe. After randomization, 127 training sets and 18 test sets to establish the Logistic regression model and BP neural network diagnosis model of the third trimester of pregnancy diagnosis of preeclampsia and to compare the diagnostic efficacy of the model to explore the risk factors of preeclampsia before pregnancy and during pregnancy. Results Fetal fibrinogen (Fbg), platelet count (PLT), mean platelet volume (MPV), and third trimester urinary protein in early pregnancy were significantly higher than those in the first trimester of pregnancy. The concordance rate of trained BP neural network model was 80.30% and the sensitivity was 74.50% , Specificity was 84.21%, which were all higher than 74.80%, 58.82% and 82.89% of Logistic regression model, and the differences were statistically significant (P <0.05). The area under the ROC curve of BP neural network model was (0.887 ± 0.029), which was higher than that of Logistic regression model (0.823 ± 0.036). The Youden index of BP neural network with the best diagnostic value was 67.0%, which was still higher than that of Logistic regression model 54.1%, the differences were statistically significant (P <0.05). Conclusion The BP neural network model is better than Logistic regression in the third trimester of pregnancy and is more suitable for multivariate analysis of complex diseases. Fbg, PLT, MPV in early pregnancy and urine protein in pregnancy and pregnancy Late stage of pre-eclampsia related.