
Comparison between using either sFlt-1/PlGF ratio or proposed panel of biomarkers. The latter is proposed by using statistical and machine learning methods. The levels of both sFlt-1 and PlGF are measured in pg/mL. sFlt-1 soluble fms-like tyrosine kinase-1, PlGF placental growth factor, PPV positive predictive value, NPV negative predictive value.
Hypertensive disorders of pregnancy (HDP) affect up to 10% of all pregnancy women. HDP encompasses a wide spectrum of disease severity from gestational hypertension (GH), preeclampsia (PE) to ante-partum eclampsia (APE). HDP is associated with a higher risk of cardiovascular diseases and heart failure, underscoring its impact on maternal health [1]. A BP target of below 140/90 mmHg is suggested for pregnant women with hypertension [2]. The pathophysiology related to HDP is the derailed process of immune tolerance and failed placentation [3]. Hypoxia and inflammation ensue the incomplete placental invasion and uterine spiral artery remodeling. The significant interplay among cytokines and overall inclination to proinflammatory status are important [4]. Specific biomarkers such as placental growth factor (PlGF) are already used for selecting patients at risk of preeclampsia [5]. Furthermore, studies succeeded in unveiling a combination of biomarkers that discriminate HDP from healthy pregnancy. Nonetheless, it is less effective for early detection of HDP by applying a panel of biomarkers, compared to simply setting cut-off values of each individual marker. The ratio of soluble fms-like tyrosine kinase 1 (sFlt-1) and PlGF was subsequently developed. Zeisler et al. reported use of sFlt-1:PIGF ratio predicts preeclampsia, with area under curve (AUC) value of 0.861, positive predictive value (PPV) of 36.7%, sensitivity of 66.2% and specificity of 83.1% [6]. Major society guidelines suggested use of sFlt-1:PIGF ratio for diagnosis of HDP [2, 7].
In the present work, Varghese et al. [8]. studied among 133 women in their late pregnancy, with or without having HDP. They utilized both statistical and machine learning (ML) methods for feature selection among a group of inflammatory (26 cytokines including interleukins, granulocyte colony stimulating factor, interferons, RANTES, vascular endothelial growth factor), angiogenic/anti-angiogenic (PlGF, sFlt-1, basic FGF) and endothelial dysfunction (nitric oxide, endothelin-1, superoxide dismutase activity) markers. Both methods were used to identify features that differentiate between antepartum eclampsia, preeclampsia and gestational hypertension versus healthy pregnancy (HP). Seven biomarkers including sFlt-1, PlGF and RANTES were significantly different in levels between HDP and its counterpart. Correlation analysis showed that regulatory interleukins (IL-9, IL-13), pro-angiogenic factors (basic FGF, PDGF-BB) and vasodilation factor (nitric oxide) were negatively correlated with laboratory data indicative of inflammation (white blood cells, neutrophils, LDH, uric acid, AST). Network analysis showed important interactions between VEGFR1, basic FGF, IL4, CCL1, eotaxin, endothelin-1 and RANTES by using STRING database. These markers were involved in chemotaxis of inflammatory cells, inflammatory response and phospholipase activity. Furthermore, the authors applied ML models such as support vector machine, logistic regression and stochastic gradient boosting for disease prediction. Totally 11–13 out of an overall 30 studied markers were selected by coefficient estimator object based on importance weights. Many of selected inflammatory and angiogenic/anti-angiogenic markers were prevalent for differentiating GH v. HP, GH v. PE and GH v. APE. It is fairly explainable considering that all markers are closely correlated in aforementioned network analysis. RANTES, a proinflammatory chemokine, was a commonly selected feature among various ML models for HDP versus healthy counterpart. It may serve as a pivotal marker in maternal plasma that denotes the progression to HDP from healthy state [9].
ML is widely used for disease forecasting with high accuracy. The authors proposed ML models that predicted the association of HDP with circulating protein markers in maternal plasma. Best ML model predicted the association of gestational hypertension with micro-F1 score of 0.87, area under curve (AUC) value of 0.80, accuracy of 87%, positive predictive value (PPV) of 81%, sensitivity of 92% and specificity of 66%. The results are compatible with previous finding by Yang et al. that the predictive value of ML models by using biomarkers is superior to using only epidemiological factors [10]. However, the AUC value herein achieved by the ML model is similar to what was achieved by the sFlt-1:PIGF ratio in predicting preeclampsia [6]. There are some limitations of this study. First, it is a cross-sectional study. Thus, findings cannot infer causality or predictability. Second, there was an imbalanced patient population. Patients with more severe forms of HDP are underrepresented. Third, the maternal blood samples were only taken once at the third trimester. The reproducibility of the findings could not be determined. Fourth, there is no external validation. In summary, this study suggested the association of HDP with a panel of circulating markers assessed by the ELISA assay. Given the limitations mentioned above, the jury is still out regarding the clinical applicability of the findings. It does shed light on the associations between circulating markers and various HDP.
References
Poudel K, Kobayashi S, Miyashita C, Ikeda-Araki A, Tamura N, Ait Bamai Y, et al. Hypertensive disorders during pregnancy (HDP), maternal characteristics, and birth outcomes among Japanese women: a Hokkaido study. Int J Environ Res Public Health. 2021;18:3342.
Wang TD, Chiang CE, Chao TH, Cheng HM, Wu YW, Wu YJ, et al. 2022 Guidelines of the Taiwan Society of Cardiology and the Taiwan Hypertension Society for the Management of Hypertension. Acta Cardiol Sin. 2022;38:225–325.
Gilbert JS, Gilbert SA, Arany M, Granger JP. Hypertension produced by placental ischemia in pregnant rats is associated with increased soluble endoglin expression. Hypertension. 2009;53:399–403.
Stefańska K, Zieliński M, Jankowiak M, Zamkowska D, Sakowska J, Adamski P, et al. Cytokine imprint in preeclampsia. Front Immunol. 2021;12:667841.
Rolnik DL, Wright D, Poon LC, O’Gorman N, Syngelaki A, de Paco Matallana C, et al. Aspirin versus placebo in pregnancies at high risk for preterm preeclampsia. N Engl J Med. 2017;377:613–22.
Zeisler H, Llurba E, Chantraine F, Vatish M, Staff AC, Sennström M, et al. Predictive value of the sFlt-1:PlGF ratio in women with suspected preeclampsia. N Engl J Med. 2016;374:13–22.
Stepan H, Herraiz I, Schlembach D, Verlohren S, Brennecke S, Chantraine F, et al. Implementation of the sFlt-1/PlGF ratio for prediction and diagnosis of pre-eclampsia in singleton pregnancy: implications for clinical practice. Ultrasound Obstet Gynecol. 2015;45:241–6.
Varghese B, Joy CA, Josyula JVN, Jangili S, Talukdar RK, Mutheneni SR, et al. Machine learning-based protein signatures for differentiating hypertensive disorders of pregnancy. Hypertens Res. 2023. https://doi.org/10.1038/s41440-023-01348-1.
Pengjie Z, Xionghui C, Yueming Z, Ting X, Na L, Jianying T, et al. LncRNA uc003fir promotes CCL5 expression and negatively affects proliferation and migration of trophoblast cells in preeclampsia. Pregnancy Hypertens. 2018;14:90–6.
Yang L, Sun G, Wang A, Jiang H, Zhang S, Yang Y, et al. Predictive models of hypertensive disorders in pregnancy based on support vector machine algorithm. Technol Health Care. 2020;28:181–6.
Author information
Authors and Affiliations
Corresponding author
Ethics declarations
Conflict of interest
The authors declare no competing interests.
Additional information
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
About this article
Cite this article
Pan, HY., Wang, TD. The association of a panel of circulating markers with hypertensive disorders of pregnancy: the jury is still out. Hypertens Res 46, 2759–2761 (2023). https://doi.org/10.1038/s41440-023-01445-1
Received:
Revised:
Accepted:
Published:
Version of record:
Issue date:
DOI: https://doi.org/10.1038/s41440-023-01445-1