The Analysis of Risk Factors for Recurrent Pregnancy Loss and Prediction of Pregnancy Loss Risk

Overview

Based on the comprehensive etiological screening results of patients with recurrent pregnancy loss, including basic characteristics, coagulation function indicators, autoimmune indicators, endocrine indicators, and gynecological ultrasound examination results, as well as the outcome of subsequent pregnancy after the patient's visit, analyze the independent risk factors affecting recurrent pregnancy loss, construct and validate an abortion risk prediction model to predict the risk of subsequent pregnancy loss in patients with recurrent pregnancy loss, and classify the patient's risk, Screening high-risk populations and guiding clinical early intervention and active treatment to improve pregnancy success rates.

1. Study population and follow-up: Patients were routinely taken history information at the initial consultation and underwent at least two etiological screenings at intervals of 4-6 weeks; during this period, sex hormone tests and uterine artery ultrasonography were performed 5-9 days after ovulation. Autoantibody tests are considered abnormal if they are positive on two or more occasions. Combined with the results of the two etiological screenings and the results of sex hormone and uterine artery ultrasound, anticoagulant, antiplatelet or immunosuppressant medication will be given for at least 2 months, and then coagulation and immunological indexes and uterine artery ultrasound will be carried out again to assess whether the indexes are improved or not, and the patients who have reached the standard of preparation for pregnancy will be given the appropriate preparation for pregnancy based on the programme of assisted reproduction or natural conception. If the patient is not pregnant after three months of pregnancy preparation, outpatient consultation is required to reassess and adjust the regimen; if the patient is pregnant, outpatient consultation is required to assess the post-pregnancy situation and adjust the regimen; after that, platelet aggregation rate (AA and ADP), D-dimer and fibrin degradation product (FDP) will be monitored every fortnight, and every four weeks, autoantibodies, coagulation, liver and renal routines, blood counts, thyroid function, and the corresponding gestational week will be monitored. Obstetric ultrasonography was performed to adjust the medication in time. If the patient had a spontaneous abortion confirmed by ultrasonography or histology after curettage before 28 weeks of gestation, including biochemical pregnancy and embryonic arrest, the pregnancy outcome was judged as pregnancy loss, and if the patient was followed up with an intrauterine viable pregnancy beyond 32 weeks, the pregnancy outcome was judged as pregnancy success, and the reproductive immunity clinic follow-up was then ended. In this study, only the first pregnancy outcome after the initial visit was collected, and follow-up ended if the patient had a pregnancy outcome event. Patients who did not have a pregnancy outcome event as of 31 December 2023 were excluded from this study, and those who had a pregnancy outcome event were included in the study by compiling the history information collected at the initial visit of the included patients, and by querying the medical record system and entering the etiological screening laboratory indexes and ultrasound results as well as the outcome of the first pregnancy after the visit. 2. Data collection: history collection at the initial consultation , outpatient medical record system query to collect laboratory indicators and ultrasound results, outpatient or telephone follow-up after the consultation of the outcome of the first pregnancy, "EpiData" software data entry; 3. Data processing: data cleaning to remove duplicates, interpolation of missing values, categorisation of variables uniquely hot coding, elimination of heterozygous ratio \<0.1 variables; characteristic Engineering descriptive statistics, correlation analysis, handling of outliers; data set division, the data were randomly divided into training set and test set according to the ratio of 7:3 according to the pregnancy outcome; 4. Predictive factor screening: t-test, analysis of variance (ANOVA), non-parametric test, chi-square test, and other analyses of the risk factors of miscarriage in patients with recurrent miscarriages; or according to the results of the analysis in combination with the results of previous studies and clinical expertise or LASSO regression (the last absolute shrinkage and the last absolute shrinkage and the last absolute shrinkage and the last absolute shrinkage). (least absolute shrinkage and selection operator, LASSO regression). Method 1: A one-way analysis of variance (ANOVA) was performed in the training set to screen for risk factors associated with miscarriage in patients with recurrent miscarriage. Two independent samples t-tests were used for continuous data, and Mann-Whitney tests were used for non-normally distributed data; for categorical data, Chi-square tests or Fisher's exact test (FET) were used. For categorical data, Chi-square tests or Fisher's exact tests were used. A two-sided p-value of less than 0.05 was considered statistically significant. Differential variables were then included in a multifactorial logistic analysis with stepwise regression to screen for independent risk factors predicting pregnancy loss in patients with recurrent miscarriage. Method 2: LASSO regression (least absolute shrinkage and selection operator) was used for feature selection in the training set. The basic principle is to introduce the L1 regularisation term on the basis of ordinary least squares to achieve feature selection and coefficient sparsification of the model by minimising the objective function, screening the important features related to the outcome variable, while setting the coefficients of irrelevant or redundant features to zero. During the fitting process, the sparsity of features is controlled by adjusting the regularisation parameter. Optimal regularisation parameters are found using methods such as cross-validation or grid search. Obtain the coefficients of all features based on the trained Lasso regression model. Sort the coefficients, in descending order of absolute value. Set a threshold to retain features with coefficients greater than the threshold. 5. Predictive model building: the training set data are taken to construct the model by machine learning methods such as logistic regression, K-nearest neighbour, decision tree, linear discriminant, neural network, random forest, support vector machine, gradient boosting, extreme gradient boosting, light gradient boosting or deep learning. 6. Internal validation method: k-fold cross-validation is used within the training set to compare the model performance, select the optimal model to adjust the hyper-parameters, and then test the generalisation ability of the model in the test set. 7. Comparison of model performance: calculate C-statistic (area under the curve AUC), accuracy, precision, recall, F1 score, and draw calibration curves, clinical decision curves and clinical impact curves to compare the prediction performance of different models. 8. Risk stratification: patients are classified into low-risk and high-risk according to the model, which is applied to clinical assessment of patients and pregnancy supervision and management. Risk stratification is proposed to construct a column-line diagram based on logistic regression and calculate the column-line diagram score for each patient, and determine the optimal score threshold based on the Youden index; patients lower than or equal to the optimal score threshold are classified as low-risk subgroups, and patients higher than the optimal score threshold are classified as high-risk subgroups. The Pearson chi-square test was used to determine the validity of risk stratification by comparing the differences in pregnancy outcomes between the low-risk and high-risk subgroups. 9. Model visualisation: column-line diagrams, risk score scales and "SHAP" were used to explain the model.