LIAO Yuan, LIU Xing-hui, TAN Jing, et al.Development of a Predictive Model for Adverse Outcomes of Preeclampsia.Journal of Sichuan University (Medical Science Edition),2018;49(5):797-802
Development of a Predictive Model for Adverse Outcomes of Preeclampsia
Objective To determine factors associated with adverse outcomes of preeclampsia and develop a predictive model. Methods Clinical data of 2 532 patients with preeclampsia who were admitted to our hospital from 2005 to 2014 were extracted for the study. The patients were divided into two groups, including 990 (39.1%) with adverse outcomes and 1 542 (60.9%) without adverse outcomes. Factors associated with adverse outcomes were identified through univariate analyses. The predictive model was developed through multivariate logistic regression analyses using a randomly selected sample containing 80% of the cases. The remaining 20% of cases served for the purpose of validation and the establishment of the ROC curve. Results Primiparas, educational attainments, prenatal care, multiple births, edema, chest pain, dyspnea, dizziness, headache, blurred vision, intrahepatic cholestasis of pregnancy, gestational diabetes, cardiovascular disease, blood pressure, urine protein, liver and kidney functions were found to be associated with adverse outcomes of preeclampsia. Multiple births, edema, dyspnea, blurred vision, cardiovascular disease, liver and kidney functions entered into the logistic regression model (P<0.05). The Logit(P) model had a good fitness of data and 77.1% accuracy in predicting adverse outcomes. The area under the curve (AUC) of the ROC curve was 0.804 〔P<0.01, 95% confidence interval CI): 0.758 to 0.849〕. The highest sensitivity was achieved when the cut-off point set risk value at 0.300, with 58.6% patients having adverse outcomes representing 83.8% true positive rate and 46.8% false positive rate. Conclusion Adverse outcomes of preeclampsia can be predicted through multiple births, edema, dyspnea, blurred vision, cardiovascular disease, liver and kidney functions. Risk value ≥0.300 is recommended.
