Organ-system-based subclassification of preeclampsia using machine learning predicts pregnancy outcomes

Yanhong Xu; Yizheng Zu; Xiaosi Lu; Yiping Wang; Jiaying Zheng; Xia Xu; Jianying Yan

doi:10.1186/s12884-025-07892-7

. 2025 Jul 24;25:787. doi: 10.1186/s12884-025-07892-7

Organ-system-based subclassification of preeclampsia using machine learning predicts pregnancy outcomes

Yanhong Xu ^1,^#, Yizheng Zu ^2,^#, Xiaosi Lu ¹, Yiping Wang ¹, Jiaying Zheng ¹, Xia Xu ^1,^3,^4,^✉, Jianying Yan ^1,^3,^4,^✉

PMCID: PMC12288313 PMID: 40707898

Abstract

Background

Preeclampsia (PE) is a complex disorder with significant variability in organ involvement. It remains unclear whether machine learning can identify organ-system-based subclasses of PE. This study aimed to identify novel subclasses of PE using organ-system biomarkers to improve pregnancy outcome prediction.

Methods

We retrospectively analyzed clinical data from PE patients at Fujian Maternity and Child Health Hospital. K-means clustering was applied using organ system function indicators, with 10-fold cross-validation. Functional indicators and pregnancy outcomes were compared across subclasses. Heatmap and sankey diagrams were used to reveal the distribution of patients across combined organ system clusters.

Results

The analysis included 7,531 PE patients treated between 2013 and 2023. 10-fold cross-validation confirmed clustering robustness with mean ARI of 0.8806 ± 0.0099 and NMI of 0.7800 ± 0.0123. Three heart function clusters were identified using five indicators, with H-Cluster 1 showing the poorest heart function and the highest complication rates. Five kidney clusters were determined from ten indicators. K-Cluster 1 and K-Cluster 5 showed distinct biomarker patterns but similar complication rates (P > 0.05). Liver function analysis using thirteen indicators revealed four clusters. L-Cluster 1 exhibited elevated liver enzymes and bilirubin with higher severe PE and intrahepatic cholestasis rates, whereas L-Cluster 3 had lower protein levels but higher anemia, fetal distress and hemorrhage incidence (P < 0.05). Five coagulation clusters were determined from nine indicators, showing significant differences in indicators and complication rates (P < 0.05). Heatmap and sankey diagram analyses revealed significant overlap between high-risk clusters, with the most frequent combination being H-Cluster 1, K-Cluster 1, L-Cluster 1 and C-Cluster 5.

Conclusions

Machine learning identified distinct PE subclasses based on organ system dysfunction patterns, each demonstrating unique pregnancy outcomes. This suggests potential clinical utility of computational approaches for PE subclassification and generates hypotheses for further investigation of its biological mechanisms.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12884-025-07892-7.

Keywords: Preeclampsia, Subclassification, Pregnancy outcome, Cluster analysis

Introduction

Preeclampsia (PE) is an idiopathic pregnancy disorder that poses a serious threat to both pregnant people and their infants. The most effective known treatment for PE is the termination of pregnancy, which often results in adverse neonatal outcomes [1, 2].

PE arises from disruption of the delicate physiological balance of pregnancy, involving complex interactions among multiple organ systems, including the heart, kidneys, liver, and coagulation system [3, 4]. Coronary microvascular dysfunction is a primary cause of heart dysfunction in women with PE [5]. Research has shown a strong association between PE and cardiovascular disease, including pulmonary arterial hypertension, cardiomyopathy, and ischaemic heart disease [6, 7]. Kidney dysfunction in PE is characterized by proteinuria, acute kidney injury, and chronic kidney disease, often resulting from endothelial damage and podocyte loss [8, 9]. The liver is affected, with HELLP syndrome (hemolysis, elevated liver enzymes and low platelets) representing a severe complication of PE [10]. The coagulation system is also significantly affected by PE. The elevated hypercoagulable state in PE patients can lead to placental fibrin deposition and microthrombus formation [11].

Given the heterogeneity of PE, there is growing recognition that it may encompass multiple disorders with distinct underlying mechanisms. An increasing number of studies support the classification and stratification of PE, which may enable more focused and precise medical interventions [12, 13]. However, the disaggregation of PE based solely on individual indicators does not capture the complexity of organ system damage. An optimal classification system should be able to target different disease mechanisms, utilize measurable clinical indicators, guide targeted treatment, and predict pregnancy outcomes. This study specifically addresses this gap through comprehensive cluster analysis. This knowledge gap necessitates innovative analytical approaches capable of integrating clinical data while preserving biological interpretability. It is a challenge particularly suited to modern computational methods [14, 15].

Machine learning algorithms, such as cluster analysis, offer a powerful solution by objectively detecting phenotypic patterns in complex clinical data. These methods can enhance prognostic assessment and disease management by uncovering patterns in datasets [16–18]. To date, few studies have applied machine learning methods to evaluate pregnancy prognosis in PE.

This study aimed to employ cluster analysis to identify PE subclasses based on functional indicators from the heart, kidney, liver, and coagulation systems. We sought to improve early clinical assessment of pregnancy prognosis and facilitate individualized management of PE.

Methods

Study design and participants

This retrospective study was approved by the Ethics Committee of Fujian Maternal and Child Health Hospital. The number of ethics approval was 2021KLRD631.

A total of 7,531 patients were collected. All included participants were diagnosed with PE. The criteria for diagnosing PE and severe PE (sPE) were established according to the American College of Obstetricians and Gynecologists (ACOG) 2019 guidelines [19]. Eligibility criteria for participants were as follows: (1) singleton pregnancies; (2) women with regular antenatal examination and delivery at our hospital. The exclusion criteria were as follows: (1) chronic hypertension (defined as hypertension diagnosed prior to pregnancy or before 20 weeks of gestation); (2) Severe pre-existing organ dysfunction including heart failure, kidney failure, liver cirrhosis, or disseminated intravascular coagulation; (3)history of malignancy, mental disease, or systemic infection; (4) aneuploidy or significant fetal abnormalities; (5) incomplete clinical data.

Data collection

Basic demographic data and laboratory tests were collected. Demographic data included age, body mass index (BMI), gravidity, parity, gestational week at PE diagnosis, history of assisted reproductive technology, uterine scarring, pregestational diabetes, thyroid disorders, autoimmune diseases, and medication use (including aspirin, antispasmodic drug, diuretic drug). Laboratory tests included routine clinical indicators to ensure generalizability across medical institutions. They were systematically collected during the pre-delivery period using standardized protocols. This timing was selected to capture the most clinically relevant organ dysfunction profiles for delivery planning. Heart function indicators included creatine kinase (CK), creatine kinase isoenzyme (CK-MB), lactic dehydrogenase (LDH), α-hydroxybutyric dehydrogenase (HBDH), c-reactive protein (CRP). Kidney function indicators included creatinine (CREA), carbamide (UREA), uric acid (UA), carbon dioxide combining power (CO2CP), cystatin (Cys-C), urine specific gravity (SG), microalbuminuria (MAU), urine pathologic cast count (U-CAST), urine white blood cell count (U-WBC), urine red blood cell count (U-RBC). Liver function indicators included alanine aminotransferase (ALT), aspartate aminotransferase (AST), glutamyltranspeptidase (GGT), alkaline phosphatase (ALP), cholinesterase (CHE), total bile acid (TBA), total bilirubin (TBIL), direct bilirubin (DBIL), indirect bilirubin (IBIL), total protein (TP), albumin (ALB), globulin (GLB), albumin/globulin ratio (A/G). Coagulation function indicators included platelet count (PLT), activated partial thrombin time (APTT), D-dimer (D2), fibrin degradation products (FDP), fibrinogen (FBG), prothrombin activity (PTA), international normalized ratio (INR), prothrombin time (PT), thrombin time (TT). These laboratory indicators were selected to comprehensively evaluate multi-organ dysfunction in PE. Heart parameters reflect myocardial injury and stress (CK, CK-MB, LDH, HBDH) and inflammation (CRP). Kidney parameters assess glomerular filtration (CREA, Cys-C), tubular function (UA, CO2CP), and urinary abnormalities (MAU for early glomerular damage, U-CAST for tubular pathology, U-WBC/U-RBC for urinary tract integrity). Liver parameters capture hepatocellular injury (ALT, AST), cholestasis (TBA, ALP), and synthetic function (TP, ALB). Coagulation parameters evaluate thrombotic risk (PLT, D2) and hemostatic capacity (PT, APTT) [20–25].

Data on the diagnosis of pregnancy complications and neonatal outcomes were also collected. The pregnancy complications variables included anemia of pregnancy, gestational diabetes mellitus (GDM), intrahepatic cholestasis of pregnancy (ICP), oligohydramnion, placental abruption, spontaneous premature rupture of membranes (sPROM), fetal distress, postpartum hemorrhage (PPH). The newborn outcome included prematurity, low birth weight (LBW), neonatal intensive care unit (NICU) admission and 5 min Apgar score ≤ 7.

K-means cluster analysis

K-means clustering was chosen for its computational efficiency with large datasets and ability to generate clinically interpretable, size-balanced clusters, which aligns with our goal of identifying distinct patient subgroups [26]. K-means cluster analysis was performed using the k-means function in R. Specifically, we conducted four separate cluster analyses. Each analysis used a different set of indicators: five heart function indicators, ten kidney function indicators, thirteen liver function indicators, and nine coagulation function indicators.

The optimal number of clusters (k) was determined using the elbow method (minimizing within-cluster variance), silhouette coefficient (maximizing inter-cluster separation), and Calinski-Harabasz criterion (optimizing cluster compactness versu dispersion). To ensure the robustness and stability of clustering algorithm, we performed 10-fold cross-validation, calculating the Adjusted Rand Index (ARI) and Normalized Mutual Information (NMI). The ARI quantifies the similarity between clustering results across folds, while the NMImeasures the mutual dependence between cluster assignments.

All missing values in continuous measures were imputed using the Multiple Imputation by Chained Equations (MICE) method with predictive mean matching. No categorical variables required imputation as they were completely recorded. All continuous variables were Z-score standardized (mean = 0, SD = 1) prior to clustering analysis to ensure equal weighting of different measurement scales. Categorical variables were encoded prior to clustering. Principal Component Analysis (PCA) and Uniform Manifold Approximation and Projection (UMAP) were used for dimensionality reduction and visualization.

Statistical analysis

Statistical analyses were performed with SPSS 26.0, Graphpad Prism 9 and R 4.2. All continuous variables were assessed for normality using Shapiro-Wilk tests and visual inspection of Q-Q plots. Continuous variables were presented as means with standard deviation (SD) and analyzed using analysis of variance (ANOVA) or non-parametric tests, as appropriate. For post-hoc comparisons between clusters, the least-significant-difference method was applied with false discovery rate correction for multiple testing. Qualitative variables were presented as percentages and analyzed by chi-square or Fisher’s exact test with Bonferroni adjustment where applicable. The P-value < 0.05 was considered statistically significant.

Results

K-means cluster result

The analysis included 7,531 PE patients treated between January 2013 and December 2023 after data cleansing. The missing data rate for each variable was different, but all were under 15%. The missing rate of heart function indicators reached up to 13.1% at most. The cohort had a mean age of 30.90 ± 4.96 years, mean BMI of 22.66 ± 3.63 kg/m², and mean gestational week at PE diagnosis of 37.21 ± 3.38. Notably, 46.7% had gravidity ≥ 2, 10.6% had parity ≥ 2, 8.5% conceived via assisted reproductive technology, 20.7% had uterine scarring, 1.7% had pregestational diabetes, 7.3% had thyroid disorders, and 2.8 had autoimmune diseases. Medication exposure included aspirin (26.0%), antispasmodic drug (73.6%), and diuretic drug (25.6%).

The optimal cluster numbers (k = 3 for heart, k = 5 for kidney, k = 4 for liver, k = 5 for coagulation) were determined through elbow method, silhouette coefficient, and Calinski-Harabasz criterion analyses (Supplementary Fig. S1). The mean ARI was 0.8806 ± 0.0099, and the mean NMI was 0.7800 ± 0.0123, indicating consistency and stability of the clustering results. After applying the K-means clustering procedure with these predetermined k values, the 7,531 specimens were divided into three clusters (H-Cluster 1, H-Cluster 2, H-Cluster 3) based on heart function indicators, five clusters (K-Cluster 1, K-Cluster 2, K-Cluster 3, K-Cluster 4, K-Cluster 5) based on kidney function indicators, four clusters (L-Cluster 1, L-Cluster 2, L-Cluster 3, L-Cluster 4) based on liver function indicators, five clusters (C-Cluster 1, C-Cluster 2, C-Cluster 3, C-Cluster 4, C-Cluster 5) based on coagulation function indicators (Fig. 1).

Fig. 1 — UMAP plots of different system function cluster analysis. a Cluster analysis based on heart function indicators. b Cluster analysis based on kidney function indicators. c Cluster analysis based on liver function indicators. d Cluster analysis based on coagulation function indicators. H-Cluster heart function clusters (1–3). K-Cluster kidney function clusters (1–5). L-Cluster liver function clusters (1–4). C-Cluster coagulation function clusters (1–5)

Function characteristics

By comparing the standardized Z-scores means for each cluster, we examined quantitative differences between the clusters (Fig. 2). Significant differences were observed in all five heart function indicators across the three clusters (P < 0.05). Compared with the other two clusters, H-Cluster 1 exhibited the highest level of CK, CK-MB, LDH, HBDH and CRP, indicating the poorest heart function in this cluster.

There were statistically significant differences in the ten kidney function indicators among the five clusters (P < 0.05). K-Cluster 1 was characterized by the highest levels of UREA, UA, Cys-C and MAU. K-Cluster 2 exhibited the lowest levels of UREA, UA, Cys-C and MAU. The SG level of K-Cluster 3 was higher than that of the other clusters. All kidney function indicators in K-Cluster 4 were relatively low, except for the U-RBC. Additionally, K-Cluster 5 showed the most significantly elevated CREA levels.

There were statistically significant differences in the thirteen liver function indicators among the four clusters (P < 0.05). L-Cluster 1 exhibited elevated levels of liver enzymes (ALT, AST and GGT), elevated metabolic indicators (TBA, TBIL, DBIL, IBIL), and decreased liver synthesis function indicators (TP, ALB and GLB). L-Cluster 2 had the lowest ALP level and the highest CHE, TP and ALB levels. All indicators of L-Cluster 3 were at low levels, particularly TP, ALB and GLB. L-Cluster 4 had the highest ALP level.

There were statistically significant differences in the nine coagulation function indicators among the five clusters (P < 0.05). C-Cluster 1 was characterized by the highest level of INR and the lowest level of PTA. C-Cluster 2 was characterized by the highest level of PTA, and the lowest level of INA and PT. C-Cluster 5 was characterized by the highest level of APTT and PT.

CK creatine kinase; CK-MB creatine kinase isoenzyme; LDH lactic dehydrogenase; HBDH α-hydroxybutyric dehydrogenase; CRP c-reactive protein; CREA creatinine; UREA carbamide; UA uric acid; CO2CP carbon dioxide combining power; Cys-C cystatin; SG urine specific gravity; MAU microalbuminuria; U-CAST urine pathologic cast count; U-WBC urine white blood cell count; U-RBC urine red blood cell count; ALT alanine aminotransferase; AST aspartate ami-notransferase; GGT glutamyltranspeptidase; ALP alkaline phosphatase; CHE cho-linesterase; TBA total bile acid; TBIL total bilirubin; DBIL direct bilirubin; IBIL indirect bilirubin; TP total protein; ALB albumin; GLB globulin; A/G albu-min/globulin ratio; PLT platelet count; APTT activated partial thrombin time; D2 D-dimer; FDP fibrin degradation products; FBG fibrinogen; PTA prothrombin activity; INR international normal-ized ratio; PT prothrombin time; TT thrombin time. H-Cluster heart function clusters (1–3). K-Cluster kidney function clusters (1–5). L-Cluster liver function clusters (1–4). C-Cluster coagulation function clusters (1–5).

Pregnancy outcomes

Three heart function clusters showed distinct pregnancy outcomes (Table 1). H-Cluster 1 exhibited the highest incidence of sPE (44.5% vs. 32.6% in H-Cluster 2 and 30.3% in H-Cluster 3), anemia (26.6% vs. 19.6–19.9%), and PPH (4.9% vs. 2.0-2.8%) (P < 0.05). Neonates in H-Cluster 1 had the highest rates of prematurity (30.5% vs. 20.3–22.6% in other clusters), LBW (26.0% vs. 20.2–20.5%), and NICU admission (18.5% vs. 14.1–15.0%) (P < 0.05).

Table 1.

Pregnancy outcomes among heart function clusters

	H-Cluster 1 (N = 2596)	H-Cluster 2 (N = 3072)	H-Cluster 3 (N = 1863)	χ²	P
sPE	1154 (44.5)	1000 (32.6)^a	565 (30.3)^a	122.189	< 0.001
Anemia	691 (26.6)	601 (19.6)^a	370 (19.9)^a	47.728	< 0.001
GDM	702 (27.0)	874 (28.5)	528 (28.3)	1.587	0.452
ICP	51 (2.0)	874 (28.5)	52 (2.8)	3.310	0.191
oligohydramnion	158 (6.1)	181 (5.9)	97 (5.2)	1.639	0.441
Placental abruption	104 (4.0)	101 (3.3)	77 (4.1)	0.441	0.217
sPROM	558 (21.5)	520 (16.9)^a	578 (31.0)^ab	134.936	< 0.001
Fetal distress	366 (14.1)	463 (15.1)	242 (13.0)	< 0.001	0.124
PPH	127 (4.9)	61 (2.0) ^a	53 (2.8)^a	39.378	< 0.001
Prematurity	793 (30.5)	625 (20.3)^a	421 (22.6)^a	83.793	< 0.001
LBW	675 (26.0)	622 (20.2)^a	382 (20.5)^a	31.469	< 0.001
NICU admission	481 (18.5)	433 (14.1)^a	280 (15.0)^a	21.993	< 0.001
5 min Apgar score ≤ 7	28 (1.1)	30 (1.0)	16 (0.9)	0.540	0.763

Open in a new tab

Data were expressed as the number (%)

H-Cluster heart function clusters (1–3), sPE Severe preeclampsia, GDM Gestational diabetes mellitus, ICP Intrahepatic cholestasis of pregnancy, sPROM Spontaneous premature rupture of membranes, PPH Postpartum hemorrhage, LBW Low birth weight, NICU Neonatal intensive care unit

^acompared with Cluster 1, the difference was statistically significant

^bcompared with Cluster 2, the difference was statistically significant

Five kidney function clusters demonstrated significant differences in pregnancy outcomes (Table 2). K-Cluster 1 had the highest incidence of sPE (64.2% vs. 23.5–30.9% in K-Clusters 2–4), anemia (27.0% vs. 19.5–21.3%), ICP (3.7% vs. 1.6–2.2%), and fetal distress (17.9% vs. 12.1–14.1%) (P < 0.05). K-Cluster 5 also showed elevated sPE rates (66.0%), though not significantly different from K-Cluster 1 (P > 0.05). Neonates in K-Cluster 1 had the worst outcomes, with 46.4% prematurity, 43.6% LBW, and 27.8% NICU admission rates (P < 0.05).

Table 2.

Pregnancy outcomes among kidney function clusters

	K-Cluster 1 (N = 1829)	K-Cluster 2 (N = 1835)	K-Cluster 3 (N = 1884)	K-Cluster 4 (N = 1936)	K-Cluster 5 (N = 47)	χ²	P
sPE	1174 (64.2)	431 (23.5)^a	582 (30.9)^ab	501 (25.9)^ac	31 (66.0)^bcd	880.035	< 0.001
Anemia	494 (27.0)	358 (19.5)^a	402 (21.3)^a	401 (20.7)^a	7 (14.9)	37.008	< 0.001
GDM	495 (27.1)	507 (27.6)	549 (29.1)	539 (27.8)	14 (29.8)	2.222	0.695
ICP	68 (3.7)	29 (1.6)^a	42 (2.2)^a	34 (1.8)^a	0 (0.0)	-	< 0.001
oligohydramnion	112 (6.1)	96 (5.2)	104 (5.5)	122 (6.3)	2 (4.3)	2.806	0.591
Placental abruption	91 (5.0)	56 (3.1)^a	60 (3.2)^a	74 (3.8)	1 (2.1)	12.143	0.016
sPROM	282 (15.4)	416 (22.7)^a	414 (22.0)^a	536 (27.7)^abc	8 (17.0)	83.836	< 0.001
Fetal distress	328 (17.9)	239 (13.0)^a	266 (14.1)^a	234 (12.1)^a	4 (8.5)	31.317	< 0.001
PPH	63 (3.4)	61 (3.3)	54 (2.9)	60 (3.1)	3 (6.4)	2.723	0.605
Prematurity	848 (46.4)	301 (16.4)^a	330 (17.5)^a	350 (18.1)^a	10 (21.3)^a	632.201	< 0.001
LBW	798 (43.6)	248 (13.5)^a	319 (16.9)^ab	304 (15.7)^a	10 (21.3)^a	642.109	< 0.001
NICU admission	508 (27.8)	191 (10.4)^a	248 (13.2)^ab	239 (12.3)^a	8 (17.0)^a	263.745	< 0.001
5 min Apgar score ≤ 7	27 (1.5)	15 (0.8)	12 (0.6)^a	18 (0.9)	2 (4.3)^bcd	12.638	0.013

Open in a new tab

Data were expressed as the number (%)

K-Cluster kidney function clusters (1–5), sPE Severe preeclampsia, GDM Gestational diabetes mellitus, ICP Intrahepatic cholestasis of pregnancy, sPROM Spontaneous premature rupture of membranes, PPH Postpartum hemorrhage, LBW Low birth weight, NICU Neonatal intensive care unit

^acompared with Cluster 1, the difference was statistically significant

^bcompared with Cluster 2, the difference was statistically significant

^ccompared with Cluster 3, the difference was statistically significant

^dcompared with Cluster 4, the difference was statistically significant

Four liver function clusters revealed divergent outcomes (Table 3). L-Cluster 1 had the highest sPE (45.7% vs. 25.2–41.8% in other clusters) and ICP rates (5.2% vs. 0.7–3.3%), while L-Cluster 3 showed higher anemia (27.0% vs. 22.9% in L-Cluster 1), fetal distress (16.8% vs. 13.8%), and PPH (4.8% vs. 3.4%) (P < 0.05). Neonates in L-Cluster 4 had the most favorable outcomes, with the lowest prematurity (12.5%), LBW (13.1%), and NICU admission (11.1%) rates (P < 0.05).

Table 3.

Pregnancy outcomes among liver function clusters

	L-Cluster 1 (N = 1456)	L-Cluster 2 (N = 1476)	L-Cluster 3 (N = 2131)	L-Cluster 4 (N = 2468)	χ²	P
sPE	665 (45.7)	541 (36.7)^a	891 (41.8)^ab	622 (25.2)^abc	215.215	< 0.001
Anemia	333 (22.9)	297 (20.1)	297 (20.1)	456 (18.5)^ac	52.808	< 0.001
GDM	435 (29.9)	401 (27.2)	575 (27.0)	693 (28.1)	4.143	0.246
ICP	75 (5.2)	49 (3.3)^a	32 (1.5)^ab	17 (0.7)^abc	94.169	< 0.001
oligohydramnion	90 (6.2)	67 (4.5)	129 (6.1)	150 (6.1)	5.288	0.152
Placental abruption	54 (3.7)	62 (4.2)	96 (4.5)	70 (2.8)^bc	9.923	0.019
sPROM	304 (20.9)	345 (23.4)	518 (24.3)^a	489 (19.8)^bc	16.184	0.001
Fetal distress	201 (13.8)	211 (14.3)	357 (16.8)^ab	302 (12.2)^c	19.377	< 0.001
PPH	50 (3.4)	39 (2.6)	103 (4.8)^ab	49 (2.0) ^ac	31.847	< 0.001
Prematurity	489 (33.6)	323 (21.9)^a	718 (33.7)^b	309 (12.5)^abc	360.056	< 0.001
LBW	397 (27.3)	327 (22.2)^a	631 (29.6)^b	324 (13.1)^abc	206.333	< 0.001
NICU admission	257 (17.7)	235 (15.9)	429 (20.1)^b	273 (11.1)^abc	75.243	< 0.001
5 min Apgar score ≤ 7	21 (1.4)	11 (0.7)	25 (1.2)	17 (0.7)	7.002	0.072

Open in a new tab

Data were expressed as the number (%)

L-Cluster liver function clusters (1–4), sPE Severe preeclampsia, GDM Gestational diabetes mellitus, ICP Intrahepatic cholestasis of pregnancy, sPROM Spontaneous premature rupture of membranes, PPH Postpartum hemorrhage, LBW Low birth weight, NICU Neonatal intensive care unit

^acompared with Cluster 1, the difference was statistically significant

^bcompared with Cluster 2, the difference was statistically significant

^ccompared with Cluster 3, the difference was statistically significant

Five coagulation clusters varied significantly in pregnancy outcomes (Table 4). C-Cluster 1 had the highest PPH incidence (5.1% vs. 1.9–3.6% in other clusters) (P < 0.05). Neonates in C-Cluster 3 exhibited the best outcomes, with lower prematurity (18.4%) and LBW (16.4%) rates (P < 0.05).

Table 4.

Pregnancy outcomes among coagulation function clusters

	C-Cluster 1 (N = 1613)	C-Cluster 2 (N = 1552)	C-Cluster 3 (N = 1585)	C-Cluster 4 (N = 1378)	C-Cluster 5 (N = 1403)	χ²	P
sPE	543 (33.7)	664 (42.8)^a	431 (27.2)^ab	431 (27.2)^ab	554 (39.5)^abc	98.438	< 0.001
Anemia	369 (22.9)	257 (16.6)^a	317 (20.0)^ab	330 (23.9)^bc	389 (27.7)^acd	60.888	< 0.001
GDM	466 (28.9)	445 (28.7)	462 (29.1)	352 (25.5)	379 (27.0)	6.813	0.146
ICP	42 (2.6)	42 (2.7)	29 (1.8)	51 (3.7) ^c	9 (0.6^abcd	32.613	< 0.001
oligohydramnion	92 (5.7)	100 (6.4)	78 (4.9)	74 (5.4)	92 (6.6)	5.391	0.250
Placental abruption	70 (4.3)	60 (3.9)	59 (3.7)	51 (3.7)	42 (3.0)	3.853	0.426
sPROM	351 (21.8)	298 (19.2)	382 (24.1)^b	318 (23.1)^b	307 (21.9)	12.163	0.016
Fetal distress	198 (12.3)	271 (17.5)^a	214 (13.5)^b	190 (13.8)^b	198 (14.1)^b	19.262	0.001
PPH	82 (5.1)	29 (1.9) ^a	34 (2.1) ^a	46 (3.3)^abc	50 (3.6)^abc	33.736	< 0.001
Prematurity	369 (22.9)	432 (27.8)^a	292 (18.4)^ab	321 (23.3)^bc	425 (30.3)^acd	69.936	< 0.001
LBW	327 (20.3)	444 (28.6)^a	260 (16.4)^ab	333 (24.2)^abc	315 (22.5)^bc	74.070	< 0.001
NICU admission	223 (13.8)	277 (17.8)^a	234 (14.8)^b	233 (16.9)^a	227 (16.2)	12.275	0.015
5 min Apgar score ≤ 7	14 (0.9)	20 (1.3)	18 (1.1)	11 (0.8)	11 (0.8)	3.144	0.534

Open in a new tab

Data were expressed as the number (%)

C-Cluster coagulation function clusters (1–5), sPE Severe preeclampsia, GDM Gestational diabetes mellitus, ICP Intrahepatic cholestasis of pregnancy, sPROM Spontaneous premature rupture of membranes, PPH Postpartum hemorrhage, LBW Low birth weight, NICU Neonatal intensive care unit

^acompared with Cluster 1, the difference was statistically significant

^bcompared with Cluster 2, the difference was statistically significant

^ccompared with Cluster 3, the difference was statistically significant

^dcompared with Cluster 4, the difference was statistically significant

Organ function interactions

We generated a heatmap to visualize the distribution of patients across combined clusters of heart, kidney, liver and coagulation functions (Fig. 3). The heatmap revealed that the predominant cluster combination was H-Cluster 1, K-Cluster 1, L-Cluster 1, and C-Cluster 5, comprising 135 patients (1.8% of cohort). This was followed by combinations of H-Cluster 2 with K-Cluster 3, L-Cluster 4 and C-Cluster 3 (133 patients, 1.7%) and H-Cluster 2 with K-Cluster 4, L-Cluster 4 and C-Cluster 3 (108 patients, 1.4%).

Fig. 3 — Patient distribution heatmap across different system function clusters. The x-axis showed combined Heart-Kidney Clusters. The y-axis showed combined Liver-Coagulation Clusters. Color intensity represented the number of patients in each combined cluster. H-Cluster heart function clusters (1–3). K-Cluster kidney function clusters (1–5). L-Cluster liver function clusters (1–4). C-Cluster coagulation function clusters (1–5)

The sankey diagrams demonstrated significant patient flow between high-risk clusters (Fig. 4). Notably, 45.9% of patients in K-Cluster 1 overlapped with H-Cluster 1, while 53.1% of K-Cluster 5 patients overlapped with H-Cluster 1. The L-Cluster 1 flowed to H-Cluster1 the most (46.9%), and the flow of L-cluster3 also went to H-Cluster1 the most (38.6%). We added heatmaps for all complications (Supplementary Fig. S2). Each heatmap showed the incidence of a complication under specific four cluster combinations, which can help understand the patterns of cluster - related complication incidences.

Discussion

Our study proposed novel hypotheses for PE classification by analyzing patient heterogeneity through machine learning. Patients with PE could be divided into different subclasses based on different system functional characteristics. These subclasses demonstrated varying associations with pregnancy outcomes.

PE is a multisystemic syndrome. Organ systems involved in the clinical presentation of PE are quite variable, including the heart, brain, kidney, liver, vascular, hemopoietic, coagulation, immune system and so on. The high heterogeneity of PE introduces new challenges for current clinical studies [13]. At present, the functional characteristics of PE are primarily assessed through the measurement of individual, commonly used indicators [23, 27]. Even if clinical assessment is combined, phenotypic data reflecting the unique functional status of PE patients may be overlooked. In this study, several variables representing heart, kidney, liver or coagulation function characteristics were selected, which were important for assessing the occurrence and severity of organism damage in patients with PE. Our results demonstrated that biomarkers affiliated with these systems might help identify distinct subclasses of PE.

Heart function monitoring is critical in PE management [6]. Similar to Li et al. [28], who classified PE patients into high-, medium-, and low-risk clusters based on postpartum hypertension, we identified three heart function subclasses. H-Cluster 1, with the poorest cardiac function, correlated with the worst pregnancy outcomes. However, the low incidence of sPROM in H-Cluster 1 might be due to early iatrogenic delivery. It should be noted that we included surrogate indicators for heart function monitoring. Future studies incorporating echocardiographic parameters could provide more straightforward cardiac assessment and validate our biomarker-based subclassification.

The kidney is one of the primary target organs injured in patients with PE [8], with proteinuria severity often indicating disease progression [25, 29]. We identified five subclasses in PE patients based on kidney biochemical characteristics. The K-Cluster 1 showed characteristic elevations in UREA, UA, Cys-C, and MAU, consistent with broader kidney dysfunction and correlating with poor outcomes. In contrast, K-Cluster 5 presented a distinct pattern with predominant CREA elevation alongside relatively preserved other markers, yet still associated with significant complications. This dissociation between CREA and other kidney parameters may reflect different underlying pathological processes, where K-Cluster 1 likely represents more global kidney impairment while K-Cluster 5 could indicate more selective injury or extrarenal influences on CREA metabolism. Further studies are needed to determine whether these clusters represent distinct PE phenotypes or different severity stages along the same pathological spectrum [30, 31].

The liver plays a vital role in metabolism, and liver function tests are routine in prenatal care [32]. Liver dysfunction in PE is associated with hypertensive disorders [10, 33]. Our study reported five subclasses in PE patients based on liver biomarkers. Compared with L-Cluster 3, L-Cluster 1 exhibited higher levels of liver transaminase indicators, metabolism indicators, and liver synthesis function indicators. While L-Cluster 1 exhibited more severe liver dysfunction, L-Cluster 3 was more frequently associated with malnutrition. In comparison of pregnancy outcomes, the incidence of sPE, ICP, fetal distress, and PPH were higher in L-Cluster 1 than in L-Cluster 3. The incidence of anemia was higher in L-Cluster 1 than in L-Cluster 3. These findings highlight that different PE subclasses cannot uniformly predict outcomes, and each liver function indicator provides unique clinical insights.

Coagulative status serve as valuable predictors for the onset and development of PE [34]. Our study observed five subclasses in PE patients based on multiple coagulation biomarkers. Compared to other clusters, C-Cluster 1 exhibited the lowest PTA and highest INR levels, correlating with the highest PPH incidence. This underscores the clinical importance of coagulation markers in managing PE-related hemorrhage risks.

The sankey diagrams and heatmap generated in this study indicated interactions between clusters of different organ systems. The observed clustering patterns provided novel insights into the systemic nature of PE. The frequent overlap between high-risk clusters suggested shared underlying mechanisms, such as endothelial dysfunction, oxidative stress, or inflammatory cascades, which may simultaneously impair multiple organ systems. For instance, the strong association between H-Cluster 1 and K-Cluster 1 could reflect a common pathway of vascular injury leading to both myocardial stress and glomerular damage [30, 35]. Clinically, these findings underscore the importance of comprehensive multi-organ assessment in PE management. Patients presenting with dysfunction in one high-risk cluster (e.g., severe heart impairment) should prompt closer evaluation of other systems.

This study’s cluster analysis reveals distinct PE subgroups with differential clinical trajectories, potentially informing future precision medicine approaches in PE management. Cluster analysis enabled unbiased integration of multiple variables, which may support exploratory strategies for risk stratification and intervention tailoring. For potential clinical consideration, the H-Cluster 1 profile, characterized by elevated cardiac biomarkers, might identify patients who could potentially benefit from extended postpartum cardiovascular surveillance, including serial echocardiograms and long-term blood pressure monitoring. Similarly, those with coagulation dysfunction might be candidates for evaluation of early prophylactic anticoagulation to potentially mitigate PPH risk, while severe heart and kidney involvement (H-Cluster 1, K-Cluster 1) underscores the need for NICU preparedness. Exploratory approaches such as automated alert systems for high-risk cluster profiles could be investigated to prompt timely clinical assessments, pending validation. Simplified decision-support tools based on discriminatory biomarkers may warrant evaluation for risk stratification, particularly in resource-constrained settings. These findings require prospective validation but may help generate hypotheses for bridging research and clinical practice, potentially contributing to more individualized management of PE patients.

This study has several limitations. The retrospective single-center design may introduce selection bias and limit generalizability, particularly regarding treatment protocols and population characteristics. The exclusive focus on pre-delivery biomarkers provided optimal data for peri-partum decision-making but precluded analysis of disease progression patterns. The lack of standardized measurement timepoints and potential confounding by clinical interventions may affect biomarker interpretation. We chose to analyze organ-specific dysfunction separately to capture distinct pathophysiological pathways, this precluded comparison with existing PE subtyping methods and limited identification of integrated phenotypes. The absence of external validation raises questions about the reproducibility of these clusters across diverse populations. Future studies should employ advanced integrative techniques that simultaneously incorporate biomarkers across all systems and include prospective multicenter validation to address these limitations. Additional constraints include: potential residual bias despite multiple imputation for missing data; unmeasured confounding from milder pre-existing conditions; and inability to fully account for treatment effects like aspirin use due to variable documentation. While these factors are inherent to retrospective designs, they highlight the need for cautious interpretation.

Conclusions

In conclusion, this exploratory study proposed a novel approach to subclassifying PE based on organ-specific functional characteristics. The identified subclasses highlight potential associations between organ dysfunction patterns and pregnancy outcomes. These findings warrant validation in prospective studies to assess their translational potential for personalized PE management.

Supplementary Information

Supplementary Material 1.^{(198KB, docx)}

Supplementary Material 2.^{(2MB, docx)}

Acknowledgements

The authors would like to thank all participants, doctors, and nurses at Fujian Medical University Fujian Maternity and Child Health Hospital for their contributions to this study.

Abbreviations

PE: Preeclampsia
sPE: Severe preeclampsia
GDM: Gestational diabetes mellitus
ICP: Intrahepatic cholestasis of pregnancy
sPROM: Spontaneous premature rupture of membranes
PPH: Postpartum hemorrhage
LBW: Low birth weight
NICU: Neonatal intensive care unit
ARI: Adjusted rand index
NMI: Normalized mutual information
PCA: Principal component analysis
UMPA: Uniform manifold approximation and projection

Authors’ contributions

XX and JYY conceived the study and designed the protocol. YHX and YZZ performed data management, analyses, and drafted the manuscript. YHX, YZZ, XX and JYY implemented the literature reviews and data interpretation. YHX, YZZ, XSL, YPW, JYZ were all involved in providing guidance on the clinical implementation and managed participant recruitment. XX and JYY provided critical revisions and supervised the manuscript. YHX and YZZ contributed equally to this work. All authors have read and agreed to the published version of the manuscript.

Funding

This work was funded by Joint Funds for the innovation of science and Technology, Fujian province (2020Y9134); Fujian Provincial Health Technology Project (2024ZD01005); Fujian Provincial Natural Science Foundation of China (2024Y0035); National Key Clinical Specialty Construction Program of China (Obstetric); Startup Fund for scientific research, Fujian Medical University (2024QH2047); Startup Fund for scientific research, Fujian Medical University (2020QH1206).

Data availability

The data is available from the corresponding authors upon reasonable request.

Declarations

Ethics approval and consent to participate

The study was conducted in accordance with the Declaration of Helsinki, and approved by the Ethics Committee of the Fujian Maternity and Child Health Hospital, Fuzhou, China. The number of the ethics approval is 2021KLRD631. The data were anonymous, and the requirement for informed consent was therefore waived.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.

Footnotes

Publisher’s Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Yanhong Xu and Yizheng Zu contributed equally to this work.

Contributor Information

Xia Xu, Email: xuxia0623@fjmu.edu.cn.

Jianying Yan, Email: yanjy2019@fjmu.edu.cn.

References

1.Ives CW, Sinkey R, Rajapreyar I, Tita ATN, Oparil S. Preeclampsia-Pathophysiology and clinical presentations: JACC State-of-the-Art review. J Am Coll Cardiol. 2020;76(14):1690–702. [DOI] [PubMed] [Google Scholar]
2.Tousty P, Fraszczyk-Tousty M, Ksel-Hryciow J, Loniewska B, Tousty J, Dzidek S, Michalczyk K, Kwiatkowska E, Cymbaluk-Ploska A, Torbe A, et al. Adverse neonatal outcome of pregnancies complicated by preeclampsia. Biomedicines. 2022;10(8):2048. [DOI] [PMC free article] [PubMed] [Google Scholar]
3.Shabani M, Eghbali M, Abiri A, Abiri M. Comprehensive microarray analysis of severe preeclampsia placenta to identify differentially expressed genes, biological pathways, hub genes, and their related non-coding RNAs. Placenta. 2024;155:22–31. [DOI] [PubMed] [Google Scholar]
4.Ramlakhan KP, Johnson MR, Roos-Hesselink JW. Pregnancy and cardiovascular disease. Nat Rev Cardiol. 2020;17(11):718–31. [DOI] [PubMed] [Google Scholar]
5.Honigberg MC, Economy KE, Pabon MA, Wang X, Castro C, Brown JM, Divakaran S, Weber BN, Barrett L, Perillo A, et al. Coronary microvascular function following severe preeclampsia. Hypertension. 2024;81(6):1272–84. [DOI] [PMC free article] [PubMed] [Google Scholar]
6.Inversetti A, Pivato CA, Cristodoro M, Latini AC, Condorelli G, Di Simone N, Stefanini G. Update on long-term cardiovascular risk after pre-eclampsia: a systematic review and meta-analysis. Eur Heart J Qual Care Clin Outcomes. 2024;10(1):4–13. [DOI] [PubMed] [Google Scholar]
7.Ramlakhan KP, Malhame I, Marelli A, Rutz T, Goland S, Franx A, Sliwa K, Elkayam U, Johnson MR, Hall R, et al. Hypertensive disorders of pregnant women with heart disease: the ESC EORP ROPAC registry. Eur Heart J. 2022;43(38):3749–61. [DOI] [PMC free article] [PubMed] [Google Scholar]
8.Andronikidi PE, Orovou E, Mavrigiannaki E, Athanasiadou V, Tzitiridou-Chatzopoulou M, Iatrakis G, Grapsa E. Placental and renal pathways underlying pre-eclampsia. Int J Mol Sci. 2024. 10.3390/ijms25052741. [DOI] [PMC free article] [PubMed] [Google Scholar]
9.Srialluri N, Surapaneni A, Chang A, Mackeen AD, Paglia MJ, Grams ME. Preeclampsia and long-term kidney outcomes: an observational cohort study. Am J Kidney Dis. 2023;82(6):698–705. [DOI] [PMC free article] [PubMed] [Google Scholar]
10.Naruse K. Liver biomarkers and preeclampsia: an easy-accessible prediction method of the disease. Hypertens Res. 2024;47(9):2610–1. [DOI] [PubMed] [Google Scholar]
11.Nieves-Colon MA, Badillo Rivera KM, Sandoval K, Villanueva Davalos V, Enriquez Lencinas LE, Mendoza-Revilla J, Adhikari K, Gonzalez-Buenfil R, Chen JW, Zhang ET, et al. Clotting factor genes are associated with preeclampsia in high-altitude pregnant women in the Peruvian Andes. Am J Hum Genet. 2022;109(6):1117–39. [DOI] [PMC free article] [PubMed] [Google Scholar]
12.Chaiworapongsa T, Romero R, Gotsch F, Suksai M, Gallo DM, Jung E, Krieger A, Chaemsaithong P, Erez O, Tarca AL. Preeclampsia at term can be classified into 2 clusters with different clinical characteristics and outcomes based on angiogenic biomarkers in maternal blood. Am J Obstet Gynecol. 2023;228(5):569. e561-569 e524. [DOI] [PMC free article] [PubMed] [Google Scholar]
13.Roberts JM, Rich-Edwards JW, McElrath TF, Garmire L, Myatt L, Global Pregnancy C. Subtypes of preeclampsia: recognition and determining clinical usefulness. Hypertension. 2021;77(5):1430–41. [DOI] [PMC free article] [PubMed] [Google Scholar]
14.Qi W, Zhu X, Wang B, Shi Y, Dong C, Shen S, Li J, Zhang K, He Y, Zhao M, et al. Alzheimer’s disease digital biomarkers multidimensional landscape and AI model scoping review. NPJ Digit Med. 2025;8(1): 366. [DOI] [PMC free article] [PubMed] [Google Scholar]
15.Sappia MS, de Korte CL, van Ginneken B, Ninalga D, Kondo S, Kasai S, Hirasawa K, Akumu T, Martin-Isla C, Lekadir K, et al. Acouslic-AI challenge report: fetal abdominal circumference measurement on blind-sweep ultrasound data from low-income countries. Med Image Anal. 2025;105: 103640. [DOI] [PubMed] [Google Scholar]
16.Fuest KE, Ulm B, Daum N, Lindholz M, Lorenz M, Blobner K, Langer N, Hodgson C, Herridge M, Blobner M, et al. Clustering of critically ill patients using an individualized learning approach enables dose optimization of mobilization in the ICU. Crit Care. 2023;27(1):1. [DOI] [PMC free article] [PubMed] [Google Scholar]
17.Pfeifer B, Sirocchi C, Bloice MD, Kreuzthaler M, Urschler M. Federated unsupervised random forest for privacy-preserving patient stratification. Bioinformatics. 2024;40(Supplement2):ii198–207. [DOI] [PMC free article] [PubMed] [Google Scholar]
18.Topaloglu F. A hybrid approach based on k-means and SVM algorithms in selection of appropriate risk assessment methods for sectors. PeerJ Comput Sci. 2024;10: e2198. [DOI] [PMC free article] [PubMed] [Google Scholar]
19.ACOG Practice Bulletin No. 202: gestational hypertension and preeclampsia. Obstet Gynecol. 2019;133(1):1. [DOI] [PubMed] [Google Scholar]
20.MacDonald TM, Walker SP, Hannan NJ, Tong S. Kaitu’u-Lino TJ: clinical tools and biomarkers to predict preeclampsia. EBioMedicine. 2022;75:103780. [DOI] [PMC free article] [PubMed] [Google Scholar]
21.Chen N, Xiao J, Luo Y, Fu K, Sun Z, Zhang X, Liu Y, Lu H, Ji X. The triglyceride-to-high-density lipoprotein cholesterol ratio is associated with an increased risk of peripartum cardiomyopathy. Front Endocrinol (Lausanne). 2024;15: 1447791. [DOI] [PMC free article] [PubMed] [Google Scholar]
22.Tsaousi M, Iliodromiti Z, Iacovidou N, Karapati E, Sulaj A, Tsantes AG, Petropoulou C, Boutsikou T, Tsantes AE, Sokou R. Hemostasis in neonates with perinatal hypoxia-laboratory approach: a systematic review. Semin Thromb Hemost. 2023;49(4):391–401. [DOI] [PubMed] [Google Scholar]
23.Kaminska J, Dymicka-Piekarska V, Tomaszewska J, Matowicka-Karna J, Koper-Lenkiewicz OM. Diagnostic utility of protein to creatinine ratio (P/C ratio) in spot urine sample within routine clinical practice. Crit Rev Clin Lab Sci. 2020;57(5):345–64. [DOI] [PubMed] [Google Scholar]
24.Gao YD, Ding M, Dong X, Zhang JJ, Kursat Azkur A, Azkur D, Gan H, Sun YL, Fu W, Li W, et al. Risk factors for severe and critically ill COVID-19 patients: a review. Allergy. 2021;76(2):428–55. [DOI] [PubMed] [Google Scholar]
25.Fang L, Shen B, Zhang H, Yin N, Cai J, Zhang J, Zhao H. Clinical features and prognosis of pregnancy-related renal damage and pregnancy after chronic kidney disease. BMC Pregnancy Childbirth. 2023;23(1):619. [DOI] [PMC free article] [PubMed] [Google Scholar]
26.Dalmaijer ES, Nord CL, Astle DE. Statistical power for cluster analysis. BMC Bioinformatics. 2022;23(1):205. [DOI] [PMC free article] [PubMed] [Google Scholar]
27.Shao H, Gao S, Dai D, Zhao X, Hua Y, Yu H. The association of antenatal D-dimer and fibrinogen with postpartum hemorrhage and intrauterine growth restriction in preeclampsia. BMC Pregnancy Childbirth. 2021;21(1):605. [DOI] [PMC free article] [PubMed] [Google Scholar]
28.Li YY, Cao J, Li JL, Zhu JY, Li YM, Wang DP, Liu H, Yang HL, He YF, Hu LY, et al. Screening high-risk population of persistent postpartum hypertension in women with preeclampsia using latent class cluster analysis. BMC Pregnancy Childbirth. 2022;22(1):687. [DOI] [PMC free article] [PubMed] [Google Scholar]
29.Okamoto T, Watanabe K, Banno T, Saitou T, Sugiura K, Iwasaki A, Matsushita H, Wakatsuki A. Amount of proteinuria as associated with severity classification of pregnant women with preeclampsia. Pregnancy Hypertens. 2022;29:30–5. [DOI] [PubMed] [Google Scholar]
30.Gembillo G, Visconti L, Giusti MA, Siligato R, Gallo A, Santoro D, Mattina A. Cardiorenal syndrome: new pathways and novel biomarkers. Biomolecules. 2021. 10.3390/biom11111581. [DOI] [PMC free article] [PubMed] [Google Scholar]
31.Freedman BS. Physiology assays in human kidney organoids. Am J Physiol Ren Physiol. 2022;322(6):F625–38. [DOI] [PMC free article] [PubMed] [Google Scholar]
32.Huang D, Lai H, Shi X, Jiang J, Zhu Z, Peng J, Zhang S, Chen Y, Yu H, Zhao Q, et al. Global temporal trends and projections of acute hepatitis E incidence among women of childbearing age: age-period-cohort analysis 2021. J Infect. 2024;89(4): 106250. [DOI] [PubMed] [Google Scholar]
33.Liu H, Zhang L, Ainiwan D, Chi P, Cheng H, Alifu X, Qiu Y, Huang Y, Zhuang Y, Zhou H, et al. Associations of maternal liver biomarkers in the first trimester with the risk of hypertensive disorders of pregnancy. Hypertens Res. 2024;47(8):2183–94. [DOI] [PubMed] [Google Scholar]
34.Alemayehu E, Mohammed O, Belete MA, Mulatie Z, Debash H, Gedefie A, Weldehanna DG, Eshetu B, Shibabaw A, Tekele SG, et al. Association of prothrombin time, thrombin time and activated partial thromboplastin time levels with preeclampsia: a systematic review and meta-analysis. BMC Pregnancy Childbirth. 2024;24(1):354. [DOI] [PMC free article] [PubMed] [Google Scholar]
35.Mendez-Barbero N, Oller J, Sanz AB, Ramos AM, Ortiz A, Ruiz-Ortega M, Rayego-Mateos S. Mitochondrial dysfunction in the cardio-renal axis. Int J Mol Sci. 2023. 10.3390/ijms24098209. [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Supplementary Material 1.^{(198KB, docx)}

Supplementary Material 2.^{(2MB, docx)}

Data Availability Statement

The data is available from the corresponding authors upon reasonable request.

[CR1] 1.Ives CW, Sinkey R, Rajapreyar I, Tita ATN, Oparil S. Preeclampsia-Pathophysiology and clinical presentations: JACC State-of-the-Art review. J Am Coll Cardiol. 2020;76(14):1690–702. [DOI] [PubMed] [Google Scholar]

[CR2] 2.Tousty P, Fraszczyk-Tousty M, Ksel-Hryciow J, Loniewska B, Tousty J, Dzidek S, Michalczyk K, Kwiatkowska E, Cymbaluk-Ploska A, Torbe A, et al. Adverse neonatal outcome of pregnancies complicated by preeclampsia. Biomedicines. 2022;10(8):2048. [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR3] 3.Shabani M, Eghbali M, Abiri A, Abiri M. Comprehensive microarray analysis of severe preeclampsia placenta to identify differentially expressed genes, biological pathways, hub genes, and their related non-coding RNAs. Placenta. 2024;155:22–31. [DOI] [PubMed] [Google Scholar]

[CR4] 4.Ramlakhan KP, Johnson MR, Roos-Hesselink JW. Pregnancy and cardiovascular disease. Nat Rev Cardiol. 2020;17(11):718–31. [DOI] [PubMed] [Google Scholar]

[CR5] 5.Honigberg MC, Economy KE, Pabon MA, Wang X, Castro C, Brown JM, Divakaran S, Weber BN, Barrett L, Perillo A, et al. Coronary microvascular function following severe preeclampsia. Hypertension. 2024;81(6):1272–84. [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR6] 6.Inversetti A, Pivato CA, Cristodoro M, Latini AC, Condorelli G, Di Simone N, Stefanini G. Update on long-term cardiovascular risk after pre-eclampsia: a systematic review and meta-analysis. Eur Heart J Qual Care Clin Outcomes. 2024;10(1):4–13. [DOI] [PubMed] [Google Scholar]

[CR7] 7.Ramlakhan KP, Malhame I, Marelli A, Rutz T, Goland S, Franx A, Sliwa K, Elkayam U, Johnson MR, Hall R, et al. Hypertensive disorders of pregnant women with heart disease: the ESC EORP ROPAC registry. Eur Heart J. 2022;43(38):3749–61. [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR8] 8.Andronikidi PE, Orovou E, Mavrigiannaki E, Athanasiadou V, Tzitiridou-Chatzopoulou M, Iatrakis G, Grapsa E. Placental and renal pathways underlying pre-eclampsia. Int J Mol Sci. 2024. 10.3390/ijms25052741. [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR9] 9.Srialluri N, Surapaneni A, Chang A, Mackeen AD, Paglia MJ, Grams ME. Preeclampsia and long-term kidney outcomes: an observational cohort study. Am J Kidney Dis. 2023;82(6):698–705. [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR10] 10.Naruse K. Liver biomarkers and preeclampsia: an easy-accessible prediction method of the disease. Hypertens Res. 2024;47(9):2610–1. [DOI] [PubMed] [Google Scholar]

[CR11] 11.Nieves-Colon MA, Badillo Rivera KM, Sandoval K, Villanueva Davalos V, Enriquez Lencinas LE, Mendoza-Revilla J, Adhikari K, Gonzalez-Buenfil R, Chen JW, Zhang ET, et al. Clotting factor genes are associated with preeclampsia in high-altitude pregnant women in the Peruvian Andes. Am J Hum Genet. 2022;109(6):1117–39. [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR12] 12.Chaiworapongsa T, Romero R, Gotsch F, Suksai M, Gallo DM, Jung E, Krieger A, Chaemsaithong P, Erez O, Tarca AL. Preeclampsia at term can be classified into 2 clusters with different clinical characteristics and outcomes based on angiogenic biomarkers in maternal blood. Am J Obstet Gynecol. 2023;228(5):569. e561-569 e524. [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR13] 13.Roberts JM, Rich-Edwards JW, McElrath TF, Garmire L, Myatt L, Global Pregnancy C. Subtypes of preeclampsia: recognition and determining clinical usefulness. Hypertension. 2021;77(5):1430–41. [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR14] 14.Qi W, Zhu X, Wang B, Shi Y, Dong C, Shen S, Li J, Zhang K, He Y, Zhao M, et al. Alzheimer’s disease digital biomarkers multidimensional landscape and AI model scoping review. NPJ Digit Med. 2025;8(1): 366. [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR15] 15.Sappia MS, de Korte CL, van Ginneken B, Ninalga D, Kondo S, Kasai S, Hirasawa K, Akumu T, Martin-Isla C, Lekadir K, et al. Acouslic-AI challenge report: fetal abdominal circumference measurement on blind-sweep ultrasound data from low-income countries. Med Image Anal. 2025;105: 103640. [DOI] [PubMed] [Google Scholar]

[CR16] 16.Fuest KE, Ulm B, Daum N, Lindholz M, Lorenz M, Blobner K, Langer N, Hodgson C, Herridge M, Blobner M, et al. Clustering of critically ill patients using an individualized learning approach enables dose optimization of mobilization in the ICU. Crit Care. 2023;27(1):1. [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR17] 17.Pfeifer B, Sirocchi C, Bloice MD, Kreuzthaler M, Urschler M. Federated unsupervised random forest for privacy-preserving patient stratification. Bioinformatics. 2024;40(Supplement2):ii198–207. [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR18] 18.Topaloglu F. A hybrid approach based on k-means and SVM algorithms in selection of appropriate risk assessment methods for sectors. PeerJ Comput Sci. 2024;10: e2198. [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR19] 19.ACOG Practice Bulletin No. 202: gestational hypertension and preeclampsia. Obstet Gynecol. 2019;133(1):1. [DOI] [PubMed] [Google Scholar]

[CR20] 20.MacDonald TM, Walker SP, Hannan NJ, Tong S. Kaitu’u-Lino TJ: clinical tools and biomarkers to predict preeclampsia. EBioMedicine. 2022;75:103780. [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR21] 21.Chen N, Xiao J, Luo Y, Fu K, Sun Z, Zhang X, Liu Y, Lu H, Ji X. The triglyceride-to-high-density lipoprotein cholesterol ratio is associated with an increased risk of peripartum cardiomyopathy. Front Endocrinol (Lausanne). 2024;15: 1447791. [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR22] 22.Tsaousi M, Iliodromiti Z, Iacovidou N, Karapati E, Sulaj A, Tsantes AG, Petropoulou C, Boutsikou T, Tsantes AE, Sokou R. Hemostasis in neonates with perinatal hypoxia-laboratory approach: a systematic review. Semin Thromb Hemost. 2023;49(4):391–401. [DOI] [PubMed] [Google Scholar]

[CR23] 23.Kaminska J, Dymicka-Piekarska V, Tomaszewska J, Matowicka-Karna J, Koper-Lenkiewicz OM. Diagnostic utility of protein to creatinine ratio (P/C ratio) in spot urine sample within routine clinical practice. Crit Rev Clin Lab Sci. 2020;57(5):345–64. [DOI] [PubMed] [Google Scholar]

[CR24] 24.Gao YD, Ding M, Dong X, Zhang JJ, Kursat Azkur A, Azkur D, Gan H, Sun YL, Fu W, Li W, et al. Risk factors for severe and critically ill COVID-19 patients: a review. Allergy. 2021;76(2):428–55. [DOI] [PubMed] [Google Scholar]

[CR25] 25.Fang L, Shen B, Zhang H, Yin N, Cai J, Zhang J, Zhao H. Clinical features and prognosis of pregnancy-related renal damage and pregnancy after chronic kidney disease. BMC Pregnancy Childbirth. 2023;23(1):619. [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR26] 26.Dalmaijer ES, Nord CL, Astle DE. Statistical power for cluster analysis. BMC Bioinformatics. 2022;23(1):205. [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR27] 27.Shao H, Gao S, Dai D, Zhao X, Hua Y, Yu H. The association of antenatal D-dimer and fibrinogen with postpartum hemorrhage and intrauterine growth restriction in preeclampsia. BMC Pregnancy Childbirth. 2021;21(1):605. [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR28] 28.Li YY, Cao J, Li JL, Zhu JY, Li YM, Wang DP, Liu H, Yang HL, He YF, Hu LY, et al. Screening high-risk population of persistent postpartum hypertension in women with preeclampsia using latent class cluster analysis. BMC Pregnancy Childbirth. 2022;22(1):687. [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR29] 29.Okamoto T, Watanabe K, Banno T, Saitou T, Sugiura K, Iwasaki A, Matsushita H, Wakatsuki A. Amount of proteinuria as associated with severity classification of pregnant women with preeclampsia. Pregnancy Hypertens. 2022;29:30–5. [DOI] [PubMed] [Google Scholar]

[CR30] 30.Gembillo G, Visconti L, Giusti MA, Siligato R, Gallo A, Santoro D, Mattina A. Cardiorenal syndrome: new pathways and novel biomarkers. Biomolecules. 2021. 10.3390/biom11111581. [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR31] 31.Freedman BS. Physiology assays in human kidney organoids. Am J Physiol Ren Physiol. 2022;322(6):F625–38. [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR32] 32.Huang D, Lai H, Shi X, Jiang J, Zhu Z, Peng J, Zhang S, Chen Y, Yu H, Zhao Q, et al. Global temporal trends and projections of acute hepatitis E incidence among women of childbearing age: age-period-cohort analysis 2021. J Infect. 2024;89(4): 106250. [DOI] [PubMed] [Google Scholar]

[CR33] 33.Liu H, Zhang L, Ainiwan D, Chi P, Cheng H, Alifu X, Qiu Y, Huang Y, Zhuang Y, Zhou H, et al. Associations of maternal liver biomarkers in the first trimester with the risk of hypertensive disorders of pregnancy. Hypertens Res. 2024;47(8):2183–94. [DOI] [PubMed] [Google Scholar]

[CR34] 34.Alemayehu E, Mohammed O, Belete MA, Mulatie Z, Debash H, Gedefie A, Weldehanna DG, Eshetu B, Shibabaw A, Tekele SG, et al. Association of prothrombin time, thrombin time and activated partial thromboplastin time levels with preeclampsia: a systematic review and meta-analysis. BMC Pregnancy Childbirth. 2024;24(1):354. [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR35] 35.Mendez-Barbero N, Oller J, Sanz AB, Ramos AM, Ortiz A, Ruiz-Ortega M, Rayego-Mateos S. Mitochondrial dysfunction in the cardio-renal axis. Int J Mol Sci. 2023. 10.3390/ijms24098209. [DOI] [PMC free article] [PubMed] [Google Scholar]

PERMALINK

Organ-system-based subclassification of preeclampsia using machine learning predicts pregnancy outcomes

Yanhong Xu

Yizheng Zu

Xiaosi Lu

Yiping Wang

Jiaying Zheng

Xia Xu

Jianying Yan

Abstract

Background

Methods

Results

Conclusions

Supplementary Information

Introduction

Methods

Study design and participants

Data collection

K-means cluster analysis

Statistical analysis

Results

K-means cluster result

Fig. 1.

Function characteristics

Fig. 2.

Pregnancy outcomes

Table 1.

Table 2.

Table 3.

Table 4.

Organ function interactions

Fig. 3.

Fig. 4.

Discussion

Conclusions

Supplementary Information

Acknowledgements

Abbreviations

Authors’ contributions

Funding

Data availability

Declarations

Ethics approval and consent to participate

Consent for publication

Competing interests

Footnotes

Contributor Information

References

Associated Data

Supplementary Materials

Data Availability Statement

ACTIONS

PERMALINK

RESOURCES

Similar articles

Cited by other articles

Links to NCBI Databases