Using Artificial Neural Network To Predict Pregnancy Outcomes In Women With SLE

Objective: To compare the accuracy of logistic regression analysis and neural network in predicting pregnancy outcomes of pregnant SLE patients and to improve the traditional neural network model in order to increase the sensitivity of fetal loss prediction.Method: This study was a retrospective,single-center study that collected 469 pregnant SLE patients who had established archive and labored in the Department of Obstetrics and Gynecology,A Hospital affiliated to Shanghai Jiao Tong University School of Medicine from September 2011 to June 2018.Models for predicting pregnancy outcomes were established using binary logistic regression analysis and neural networks.Multilayer perceptron(MLP)and Radial basis function(RBF)were selected as neural network models.By integrating the thoughts of comparative and focused study into the artificial neural network,the improved model presented an effective algorithm aiming at imbalanced learning to predict fetal loss outcomes.Result: The logistic regression model had a prediction accuracy of 44.9% for fetal loss,97.3% for live birth and 91.3% for overall prediction in the validation,the area under the curve(AUC)of the ROC was 0.899.The MLP neural network model had a prediction accuracy of 42.9% for fetal loss,97.8% for live birth and 92.9% for overall prediction in the validation,and the area under the curve of the ROC(AUC)was 0.911.The RBF neural network model had a prediction accuracy of 11.8% for fetal loss,97.6% for live birth and 87.5% for overall prediction in the validation,the area under the curve(AUC)of the ROC was 0.880.The MLP neural network model has an advantage in predicting pregnancy outcomes for pregnant SLE patients.The improved neural network model got a prediction accuracy of 100% for fetal loss,82.7% for overall prediction in the internal validation,and got a prediction accuracy of 81.8% for fetal loss,75.5% for overall prediction in the external validation.The improved neural network model had a higher prediction accuracy for fetal loss and was more suitable for clinical application.Conclusion: The MLP neural network model is superior to the logistic regression model and the RBF neural network model in predicting pregnancy outcomes in pregnant SLE patients.However,the traditional MLP neural network model is less sensitive in predicting fetal loss in imbalanced pregnancy outcomes.By improving the neural network model,the sensitivity of predicting fetal loss was significantly improved,which could to assist healthcare providers to find potential fetal loss patients as much as possible and to make timely and accurate decisions to avoid dispensable fetal loss during expectation treatment.