Discussion

Based on laboratory markers profiles from 12,715 women in early pregnancy, this study systematically identified multiple routine laboratory markers independently linked to the subsequent development of PE, providing fresh insights for early risk evaluation. For the first time, liver function indices (GGT, ALP, CHE), renal marker (UAR MoM), lipid-inflammatory indicators (TyG MoM, CRP), and micronutrient levels (Cu, Ca, Mg, Fe) were shown to be significantly correlated with the risk of subsequent PE development during the early stages of pregnancy. Based on these features, a machine learning model was established and combined with the FMF prediction algorithm, substantially enhancing the model's discriminative performance. Collectively, our findings emphasize that routinely available laboratory markers can serve as a practical, low-cost approach for identifying women at elevated risk of PE, particularly in primary or resource-limited settings.

Regarding demographic characteristics, women who developed PE were slightly older (mean age, 32 vs. 31 years) and had a higher median BMI (24.1 vs. 21.79 kg/m2) than controls, suggesting that advanced maternal age and obesity may jointly contribute to PE pathogenesis through impaired placental vascular remodeling and enhanced systemic inflammatory responses.17,18 Furthermore, higher rates of twin gestation (9.71%) and assisted conception (24.28%) were observed in the PE group, consistent with previous findings.19,20 The combined effects of increased placental mass in multiple pregnancies and hormonal alterations during IVF conception may exacerbate placental ischemia and hypoxia, thereby promoting PE development.21, 22, 23

In this study, women who developed PE exhibited abnormalities in several liver function markers during early pregnancy (< 14 weeks), suggesting that hepatic dysfunction may represent not only a concomitant alteration but also an early indicator or potential contributor to PE pathogenesis. Mild elevations in ALT and GGT were interpreted as evidence of subclinical hepatocellular injury or cholestasis, which may aggravate placental oxidative stress through the systemic release of mitochondrial reactive oxygen species reaching the placenta via the circulation.24, 25, 26, 27 In addition, elevated bile acid levels may disrupt placental development through multiple mechanisms: by activating the farnesoid X receptor (FXR) and its downstream effector, the small heterodimer partner (SHP), within the placenta, leading to reduced VEGF and PLGF expression and impaired angiogenesis;28, 29, 30 and by promoting mitochondrial membrane permeability via hydrophobic bile acids such as deoxycholic acid, thereby triggering chromogranin C release, apoptotic signaling, and trophoblast cell death.31,32 Concurrently, elevated UAR MoM (OR = 10.16) emerged as another significant indicator of PE risk, consistent with recent Mendelian randomization evidence linking early pregnancy hyperuricemia to a 1.21-fold increased risk of PE.33

Disordered lipid metabolism has been increasingly recognized as a key metabolic feature of PE. The TyG MoM (OR = 4.29) was identified as an independent risk factor, suggesting the contribution of insulin resistance to its pathogenesis. Hyperinsulinemia may induce vasoconstriction and impair placental perfusion by suppressing nitric oxide synthesis and enhancing endothelin-1 release, thereby exacerbating endothelial dysfunction.34,35 In parallel, elevated TyG levels may aggravate oxidative stress and systemic inflammation via NADPH oxidase activation, leading to excessive reactive oxygen species generation and increased secretion of pro-inflammatory cytokines such as IL-6 and tumor necrosis factor-α (TNF-α) from adipose tissue.36,37 Functional impairment of HDL reduces its anti-inflammatory and cholesterol efflux capacities, whereas oxidative modification of LDL and VLDL generates oxidized LDL (ox-LDL), which activates macrophages and promotes foam-cell formation, leading to atherosclerosis-like remodeling of placental vessels.38, 39, 40 Collectively, these findings indicate that insulin resistance and lipid disorders act synergistically accelerate vascular injury in PE.

Electrolyte disturbances were also observed, characterized by increased Na, K, Ca, and Cu levels and decreased Mg and Fe levels, reflecting oxidative stress and impaired placental perfusion. Interestingly, Mg concentrations below a defined threshold showed an inverse association with PE risk, suggesting that higher Mg levels may confer vascular protection by limiting calcium entry and alleviating smooth muscle tension through activation of transient receptor potential melastatin 7 (TRPM7) channels.41,42 Finally, CRP exhibited a dose-dependent biphasic association with PE: below 8.09 mg/L, each 1 mg/L increase corresponded to a higher PE risk (OR = 1.18), indicating its dual role as a marker of vascular inflammation and a mediator of endothelial dysfunction. Collectively, this study demonstrates that hepatic dysfunction, insulin resistance, lipid dysregulation, electrolyte imbalance, and inflammation converge in the pathogenesis of PE. These findings reinforce the notion that PE is not merely a placental disorder but a systemic condition arising from early metabolic, immune, and vascular dysfunctions.43,44 Consequently, redefining intervention windows and preventive strategies from a systemic perspective centered on these early alterations may help improve maternal outcomes.

In multicenter studies conducted among Asian populations (e.g., in China), the FMF competing risk model yields a detection rate of about 60%–82% for predicting preterm PE,45,46 which is marginally lower than that reported in Caucasian and African cohorts. This disparity was also observed in the present study. Previous research has reported systematic differences in the regression patterns of MoM values for biomarkers such as MAP and UtA-PI across ethnic groups. Specifically, compared with the FMF reference population, East Asian women generally exhibit steeper declines in MAP MoM and flatter trends in UtA-PI MoM,47,48 and these non-compensatory differences may partly explain the reduced predictive efficiency of the FMF model in Asian populations. Furthermore, the FMF model primarily incorporates biophysical, biochemical, and ultrasonographic parameters in early pregnancy, while largely neglecting maternal factors such as inflammation, oxidative stress, and metabolic dysfunction—conditions more common in older, overweight, or IVF conception, particularly among Asian women. In this study, integrating routine laboratory markers, including liver and renal function tests, electrolytes, and lipid profiles, broadened the model's dimensional scope without increasing cost, providing a more practical and cost-effective screening strategy for resource-limited settings. For preterm PE, integrating conventional laboratory markers with the FMF model (MAP, PLGF, and UtA-PI) substantially enhanced predictive performance. The combined model achieved an AUC of 0.954 (95% CI, 0.914–0.969), significantly higher than the FMF model alone (AUC = 0.854, p < 0.001), with the detection rate rising to 92.5% from 65.0% at a 14% false-positive rate. For term PE, the logistic regression model yielded an AUC of 0.913 (95% CI, 0.887–0.931), outperforming the FMF model (AUC = 0.778, p < 0.001), and sensitivity increased to 81.1% from 47.7% at the same false-positive rate. SHAP analysis further revealed that inflammatory and metabolic markers contributed most strongly to model output, addressing the FMF model's limited responsiveness to such abnormalities and offering a plausible explanation for the superior predictive capacity of the integrated approach.

To implement the combined model in low-resource settings, we propose a minimal test set that includes key laboratory markers (GGT, ALP, UAR MoM, TyG MoM, CRP, Ca, and Mg) alongside FMF model parameters (maternal risk factors, MAP, PLGF, and UtA-PI). This streamlined panel balances strong predictive power with cost-effectiveness, making it feasible in settings with limited infrastructure. Clinically, the model can guide early management by stratifying risk. High-risk women identified through this model can be prioritized for closer monitoring, early initiation of low-dose aspirin (if appropriate), and more frequent assessment of blood pressure and fetal growth. Identifying specific metabolic issues, such as an elevated TyG index or low Mg, may prompt targeted interventions like dietary counseling or nutrient supplementation. A phased screening approach allows adaptive resource allocation: primary care settings can perform initial assessments, referring high-risk cases to higher-level facilities for confirmatory tests. This ensures efficient use of resources and timely interventions. The model can also be deployed via mobile platforms, enabling accessible, low-cost risk assessment without specialized equipment.

cfDNA TSS profiling in mid-pregnancy revealed extensive metabolic and inflammatory disturbances preceding the clinical onset of preterm PE. Functional enrichment analyses demonstrated coordinated down-regulation of hepatic detoxification pathways, including cytochrome P450 metabolism, glucuronidation, bile acid synthesis, and aspirin metabolism, suggesting reduced maternal capacity for endogenous toxin clearance and aligning with the preventive efficacy of early aspirin administration. Simultaneous up-regulation of IL-1 and IL-6 signaling suggested a systemic inflammatory milieu that may contribute to endothelial dysfunction and placental injury. Suppression of glycolytic and glucose handling genes, along with paradoxical activation of insulin signaling components (e.g., IGF1R, IRS2), implied early insulin resistance with compensatory transcriptional responses. These transcriptomic signatures were consistent with routine laboratory markers abnormalities observed in individuals who later developed PE, including elevated GGT, increased TyG index, and subclinical inflammation. Collectively, cfDNA TSS profiling enabled early detection of multisystem dysregulation, offering mechanistic insights into PE pathogenesis and a noninvasive means for early risk stratification and intervention.

Although retrospective, this multicenter study with a large cohort provides meaningful insights. Nonetheless, several limitations should be noted. The study cohort is drawn exclusively from Chinese pregnant women, and laboratory marker distributions and reference intervals may vary substantially across populations due to ethnic, genetic, dietary, and environmental differences. This variability underscores the challenges of transferring the model to different populations. The inconsistent performance of traditional tools such as the FMF algorithm in non-European (e.g., Asian) populations highlights the necessity for population-tailored or globally adaptable frameworks to achieve equitable, context-aware risk prediction in diverse clinical settings. Consequently, the risk estimates and biomarker thresholds identified in our model may not directly apply to non-Chinese populations without further calibration. External validation using independent prospective cohorts, particularly from diverse communities and primary care settings, is essential to confirm the model's generalizability and real-world applicability. Additionally, all laboratory markers were assessed at a single time point; future work should determine whether longitudinal trajectories, particularly in later trimesters, enhance predictive accuracy. Despite these limitations, the pathophysiological pathways emphasized by our model—liver dysfunction, dyslipidemia, systemic inflammation, and insulin resistance—are widely recognized in the pathogenesis of PE. Therefore, while specific thresholds may need adjustment, the overall approach of integrating routine laboratory markers with clinical parameters is likely to remain relevant across different populations.

In this study of 12,715 pregnancies, early metabolic, inflammatory, and laboratory markers, including liver enzymes, cholesterol, CRP, TyG index, and UAR, were significantly linked to PE risk, with several showing non-linear threshold effects. Machine learning models constructed from these routine laboratory indicators outperformed the traditional FMF model in predicting both preterm and term PE. SHAP analysis further confirmed the interpretability of the model and the relevance of key features. Overall, integrating routine laboratory parameters into machine learning frameworks may improve early PE screening, enable individualized interventions, and benefit maternal–fetal outcomes. Future studies should focus on validating this approach across diverse populations and evaluating its feasibility in clinical implementation.