Study On Clinical Endpoint Prediction Based On Healthy Time Series Data

The analysis of maternal clinical endpoint has always been an urgently needed auxiliary diagnostic means.However,it is difficult to quickly analyze the risk factors that may affect the final birth outcome of pregnant women.With the rise of data science,especially the increasing maturity of deep learning technology,deep learning method is used to study the possible risk factors of pregnant women in the whole process from pregnancy to parturition based on electronic medical record data,which can facilitate the prediction of clinical endpoint of pregnant women.This paper aims to improve the existing depth model for the maternal birth control data with time series characteristics,and to realize the early prediction of postpartum hemorrhage and premature birth in the maternal clinical endpoint prediction.At present,most hospitals have established a complete medical electronic record system,and every electronic disease record and inspection record of each maternal check-up is kept in the database.Currently,the mining of electronic medical records has been widely carried out.Deep learning is an effective method to study such problems.Nowadays,there are two problems concerning the application of deep learning in electronic medical records.First,deep learning tends to have a complex structure and has many hyperparameters that directly affect the effect of the final model,while the adjustment of hyperparameters requires higher tuning techniques.The other is that the training and prediction process of depth model is in a "black box" due to the closure of its structure.The results obtained by the model cannot be traced back to the reasons for its generation.The results cannot be reasonably explained,while in medical research it is often desirable to explain the reasons for making decisions.In order to solve the problem that it is difficult to adjust the parameters of depth model,this paper proposes to use particle swarm optimization(PSO)to train the hyperparameters of depth model to find the optimal hyperparameters.In order to solve the problem that the result of depth model need to be explained,a hybrid model is adopted in this paper.On the way to guarantee the effect of the model,the interpretability of the traditional integrated model XGBoost results was used to construct the Xgboost-Lstm hybrid model,so as to realize the interpretability of the final model results and assist doctors in diagnosis.In this paper,the above two prediction models are applied to the study of fetal premature birth and postpartum hemorrhage in pregnant women.The experimental results show that,compared with the traditional prediction based on the cross-section data of specific time and the prediction results of a single traditional model,the model proposed in this paper has a higher prediction accuracy and more stable results,and can provide decision support for pregnant women in blood preparation during delivery and fetal health screening.