Dynamic risk prediction models for different subtypes of hypertensive disorders in pregnancy

Xinyu Zhang; Qi Xu; Lin Yang; Ge Sun; Guoli Liu; Cuiting Lian; Ziwei Li; Dongmei Hao; Yimin Yang; Xuwen Li

doi:10.3389/fsurg.2022.1005974

Dynamic risk prediction models for different subtypes of hypertensive disorders in pregnancy

Front Surg. 2022 Oct 26:9:1005974. doi: 10.3389/fsurg.2022.1005974. eCollection 2022.

Authors

Xinyu Zhang¹, Qi Xu², Lin Yang¹, Ge Sun¹, Guoli Liu², Cuiting Lian¹, Ziwei Li¹, Dongmei Hao¹, Yimin Yang¹, Xuwen Li¹

Affiliations

¹ Faculty of Environment and Life, Beijing University of Technology, Beijing, China.
² Department of Obstetrics, Peking University People's Hospital, Beijing, China.

Abstract

Background: Hypertensive disorders in pregnancy (HDP) are diseases that coexist with pregnancy and hypertension. The pathogenesis of this disease is complex, and different physiological and pathological states can develop different subtypes of HDP.

Objective: To investigate the predictive effects of different variable selection and modeling methods on four HDP subtypes: gestational hypertension, early-onset preeclampsia, late-onset preeclampsia, and chronic hypertension complicated with preeclampsia.

Methods: This research was a retrospective study of pregnant women who attended antenatal care and labored at Beijing Maternity Hospital, Beijing Haidian District Maternal and Child Health Hospital, and Peking University People's Hospital. We extracted maternal demographic data and clinical characteristics for risk factor analysis and included gestational week as a parameter in this study. Finally, we developed a dynamic prediction model for HDP subtypes by nonlinear regression, support vector machine, stepwise regression, and Lasso regression methods.

Results: The AUCs of the Lasso regression dynamic prediction model for each subtype were 0.910, 0.962, 0.859, and 0.955, respectively. The AUC of the Lasso regression dynamic prediction model was higher than those of the other three prediction models. The accuracy of the Lasso regression dynamic prediction model was above 85%, and the highest was close to 92%. For the four subgroups, the Lasso regression dynamic prediction model had the best comprehensive performance in clinical application. The placental growth factor was tested significant (P < 0.05) only in the stepwise regression dynamic prediction model for early-onset preeclampsia.

Conclusion: The Lasso regression dynamic prediction model could accurately predict the risk of four HDP subtypes, which provided the appropriate guidance and basis for targeted prevention of adverse outcomes and improved clinical care.

Keywords: dynamic prediction model; hypertensive disorders in pregnancy; lasso regression; modeling method; risk factor analysis; subtype.