Risk Prediction Model of Preeclampsia

Research objects:

This is a prospective study. About 2000 pregnant women who will take regular prenatal examination in the Department of Obstetrics, Peking University Third Hospital. During 6-8 weeks of gestation, routine laboratory tests, such as liver function, were required before the establishment of obstetric records. The remain serum from routine laboratory tests will be collected and frozen at -80℃ for detection of biological markers after delivery.

Some routine laboratory tests will be carried out with the prenatal examination at 16-18 GWs、26-28 GWs、30-34GWs. The remain serum of the participants will be collected if the routine tests were done.

We will not draw extra blood samples from the participants.

Quality assurance plan:

1. Check the patient information and gestational age carefully to obtain the correct cases.
2. The samples of hemolysis, lipid turbidity and jaundice should be eliminated to prevent interference with the experimental results.
3. The serum was placed in a cryopreservation tube and immediately stored at -70℃.
4. Calibration and quality control should be carried out for each batch of testing. Record the results of quality control and start testing after control.

Data dictionary:

(1) General information of the research object: Data on risk factors for preeclampsia were collected at 6-8 weeks of gestation, including age, primipara or pluripara, multiple births, prepregnancy body mass index, preeclampsia history, basal systolic blood pressure, basal diastolic blood pressure, hypertension history, renal history, diabetes history, autoimmune history, etc. The above records will be obtained from the medical records system.

(3) Test results of routine laboratory tests: Laboratory test results, such as total cholesterol, triglycerides, high-density lipoprotein cholesterol, low density lipoprotein cholesterol and lipoprotein a and C reactive protein, alanine aminotransferase, aspartate aminotransferase, lactate dehydrogenase, urea, uric acid, creatinine and cystatine C, D-dimer, neutrophils and lymphocytes ratio, platelet and lymphocyte ratio and so on, the above test results can query from the electronic medical record system.

(4) Biological markers detection: After delivery, the biomarkers will be tested with the 6-8 GWs samples of the 2000 participants, such as the complement factor B, complement factor H, C3, complement C4, matrix metalloproteinases 7, placenta protein 13, soluble vascular endothelial growth factor receptor 1, placental growth factor, fibronectin, etc.

(5) Establishment of database: To input the above original data into the database.

Sample size: About 100 to 160 preeclampsia patients will be collected out of the 2000 participants accoeding to the he incidence of preeclampsia which is 3% to 8%.

The missing data will be reported as missing, unavailable, non-reported, uninterpretable, or considered missing because of data inconsistency or out-of-range results according to actual condition.

Statistical analysis plan:

By using univariate logistic regression model, maternal clinical data, routine laboratory tests and biological markers in early pregnancy were divided into two categories: "important indicators" and "general indicators".

The data set was divided into a training set and a test set in a 3:1 ratio for the training and testing of preeclampsia risk prediction model, respectively.

Samples of pregnant women without preeclampsia in the training set were evenly divided into three subsets A, B and C, and the sample set of preeclampsia patients in the training set was called set D.Build A deep learning model with two sets A and D, build A deep learning model with two sets B and D, and build A deep learning model with two sets C and D.These three models are successively referred to as Model 1, Model 2 and Model 3.

Model test method:

Substituting the data of each sample in the test set into the above three deep learning models, the three output values of each sample are obtained, and then the prediction of the type of each sample is obtained based on the average value of the three numbers. Then the prediction results are compared with the sample labels to evaluate the model.