Maternal plasma cytokines and the subsequent risk of uterine atony and postpartum hemorrhage

Abstract

Objectives: To determine whether the maternal plasma concentrations of cytokines are higher in pregnant women with postpartum hemorrhage (PPH) compared to pregnant women without PPH.

Methods: A retrospective case-control study included 36 women with PPH and 72 matched controls. Cases and controls were matched for gestational age at delivery, labor status, delivery route, parity, and year of sample collection. Maternal plasma samples were collected up to 3 days prior to delivery. Comparison of the plasma concentrations of 29 cytokines was performed by using linear mixed-effects models and included adjustment for covariates and multiple testing. A false discovery rate adjusted p-value <0.1 was used to infer significance. Random forest models with evaluation by leave-one-out and 9-fold cross-validation were used to assess the combined value of the proteins in predicting PPH.

Results: Concentrations of interleukin (IL)-16, IL-6, IL-12/IL-23p40, monocyte chemotactic protein 1 (MCP-1), and IL-1β were significantly higher in PPH than in the control group. This difference remained significant after adjustment for maternal age, clinical chorioamnionitis, and preeclampsia. Multi-protein random forest proteomics models had moderate cross-validated accuracy for prediction of PPH [area under the ROC curve, 0.69 (0.58-0.81) by leave-one-out cross validation and 0.73 (0.65-0.81) by 9-fold cross-validation], and the inclusion of clinical and demographic information did not increase the prediction performance.

Conclusions: Pregnant women with severe PPH had higher median maternal plasma concentrations of IL-16, IL-6, IL-12/IL-23p40, MCP-1, and IL-1β than patients without PPH. These cytokines could serve as biomarkers or their pathways may be therapeutic targets.

Keywords: CCL2; IL-12/IL-23p40; IL-16; IL-1β; IL-6; MCP-1; biomarker; inflammation; interleukin (IL); maternal proteins.

Figure 1:

Maternal plasma protein concentrations in patients without postpartum hemorrhage (PPH) (controls) (n=72), and with PPH (n=36). Data are shown in boxplots. Logarithmic scales were used to improve normality of the distribution of the variables. p-values shown were derived using linear mixed-effect models to account for the random matching blocks. IL-1β does not show a normal distribution; for it, a non-parametric Wilcoxon test yielded a p-value of p=0.018.

Figure 2:

Comparison of differential cytokine concentrations in  postpartum hemorrhage (PPH) between this study and Jiang et al. The y-axis represents the fold changes (log, base 2, thereof) in concentration between PPH and controls. Positive values represent increased concentration in cases. The x-axis represents the late third trimester logistic model coefficients for the association between PPH and one log unit increase in concentration (Extracted from Table 5 in Jiang et al.). Positive x-axis values correspond to an increase in log-odds of PPH with increasing concentration of the protein.

Figure 3:

Receiver Operating Characteristic (ROC) curve for prediction of postpartum hemorrhage. Leave-one-out (A) and 9-fold cross validated (B) predictions were obtained using the full set of proteins only (black line) or proteins and additional demographic and clinical variables (see Materials and methods) (red line). Area under the ROC curve (AUC) is given with 95% confidence intervals.

Figure 4:

Variable importance plot for prediction of postpartum hemorrhage. The mean decrease in Gini Index represents a measure of the relative importance of the variables in the multi-variate random forest model used to predict postpartum hemorrhage.