A new predicting model of preeclampsia based on peripheral blood test value.

Objective To identify laboratory markers among platelet indices, coagulation parameters, blood lipid parameters, and liver/kidney function variables that can be used to predict preeclampsia. Patients and methods We studied records of 568 women with preeclampsia, gestational hypertension (GH), or normal term pregnancies hospitalized in the Obstetrics Department of the Fujian Maternal and Child Health Hospital from September 2014 to September 2018. We divided the patients' records into three groups (216 with preeclampsia, 136 with gestational hypertension, and 216 with normal pregnancies). We conducted retrospective analyses to compare variable measurements between the groups and find correlations. We looked into maternal pre-onset platelet indices, coagulation parameters (thrombin time [TT], fibrinogen [FIB]), biochemical parameters (total cholesterol [TC], triglycerides [TG], high-density lipoproteins [HDL], alanine transaminase [ALT], serum creatinine [CRE], blood urea nitrogen [BUN], uric acid [UA]), maternal complications, and perinatal outcomes. In addition to our statistical analysis, we trained a back-propagation (BP) neural network to identify the strongest predictors of preeclampsia. Results We found significant differences among the groups in terms of values for PLT, MPV, PDW, PLCR, TT, FIB, TG, LDH, BUN, and others. After adjusting for confounding factors in a multivariate ordered logistic regression model, we found that mean values for MPV, BUN, TG, and LDH can independently predict the risk of preeclampsia (the OR values were 1.858, 1.583, 1.104, and 1.020, respectively), the C-index (concordance statistic) was 0.73. Also, our BP neural network derived ALB, MPV, BUN, LDH and TG as the strongest predictors of preeclampsia. Conclusions MPV, TG, LDH, and BUN can help establish the risk for the development of preeclampsia to apply active measures and improve maternal and perinatal outcomes. The BP neural network can be used to study predictive models of preeclampsia.