Research On Prediction Model Of Risk Of Reproductive Chlamydia Trachomatis In Pregnant Women Based On BP Neural Network

ObjectiveThe purpose of this research is to establish a system of predictors of Chlamydia trachomatis infection in pregnant women,and to construct a predictive model for the risk of Chlamydia trachomatis infection in pregnant women based on BP neural network algorithm to provide a tool for population screening.MethodsThe pool of relevant entries was screened based on the literature review induction,and then 15 experts in the relevant fields were invited to fill in the consultation questionnaire using the Delphi method.SPSS 26.0 was used to analyse the authority and positivity of the expert group,the degree of agreement with the indicators and the importance rating by calculating the positive coefficient,the authority coefficient,the coordination coefficient,the full score ratio,the arithmetic mean,the standard deviation,the coefficient of variation,the number of expert opinions proposed and the consistency of opinions.The indicators were screened by a combination of the cut-off value screening method and the expert scoring method to establish a system of predictors of Chlamydia trachomatis infection in the pregnant population.After pre-processing such as database cleaning,factor analysis,normalization,data set partitioning and training set balancing,MATLAB was used as the model training environment to build and validate a model for predicting the risk of Chlamydia trachomatis infection of pregnant women based on the traditional logistic regression algorithm and BP neural network algorithm,and then the performance of the two models was compared by sensitivity,specificity,positive predictive value,negative predictive value,Youden’s Index,F1 score,G-mean and AUC.ResultsAfter three rounds of expert consultation,a system of predictors of risk for Chlamydia trachomatis infection in the pregnant population was established which consists of five primary indicators and 22 secondary indicators.Then a questionnaire was designed to collect a total of 1904 valid samples containing chlamydial nucleic acid results(including 106 samples from Chlamydia trachomatis-infected pregnant women)from two regions.After data pre-processing,the training set used for modelling consisted of 2880 data(including 1442 pregnant women with chlamydia)and the validation set consisted of 361 data(including 21 pregnant women with chlamydia).The final overall accuracy,sensitivity,specificity,positive predictive value,negative predictive value,Youden’s Index,F1 score,G-mean and AUC of the BP neural network prediction model(The network structure is 8-15-13-2)on the training set were 98.5%,98.6%,98.3%,98.3%,98.6%,0.969,0.984,0.984 and 0.986,respectively;on the validation set were 98.7%,76.2%,100%,100%,98.6%,0.762,0.865,0.873 and 0.949,respectively.All the indicators are at a higher level compared with the prediction model of the logistic regression algorithm,and the BP neural network prediction model constructed in this paper has the ability to better identify whether pregnant women are infected with Chlamydia trachomatis.ConclusionIn this study,a system of predictors of Chlamydia trachomatis infection in pregnant women was constructed.The better positive coefficients of experts,authority coefficients and opinion coordination coefficients indicated the reliability of the consultation results and the validity of the index system.Compared with the traditional logistic regression model,the BP neural network model constructed for predicting the risk of Chlamydia trachomatis infection in the pregnant women has better predictive performance and is valuable.The external validation of the model and the application of its prediction results to the Internet for risk judgement are the focus of the next step.