Association between D-dimer difference and risk of postoperative venous thromboembolism in endometrial cancer: a retrospective cohort study Authors: Weipeng Li, Hui Zheng, Lan Zhen, Qian Qiu, Leilei Zhu, Rong Zhang, Huan Yi Published: 2026 ABSTRACT ObjectiveTo investigate the association between D-dimer level changes and venous thromboembolism (VTE) risk in endometrial cancer patients undergoing surgery.MethodsThis retrospective cohort study included 307 endometrial cancer patients who underwent surgical treatment at Fujian Maternity and Child Health Hospital between January 2022 and December 2024, with follow-up to December 2025. D-dimer change was the exposure variable and VTE occurrence the outcome. Multivariable logistic regression assessed independent effects after adjusting for 17 covariates, including age, body mass index, FIGO stage, and comorbidities. Threshold effect and stratified analyses were conducted.ResultsEach one-unit increase in D-dimer change was significantly associated with higher VTE risk (adjusted OR = 1.202; 95% CI: 1.076–1.342; P = 0.001). Nonlinear testing revealed a significant threshold effect (P < 0.001) at 3.78: below this threshold, VTE risk increased sharply (OR = 7.761; 95% CI: 4.286–14.051); above it, risk declined (OR = 0.643; 95% CI: 0.526–0.785). Mediation analysis showed anticoagulant therapy, infection, and caprinin had negligible mediating effects (proportion < 0.2%; P > 0.86).ConclusionD-dimer level changes exhibit a nonlinear association with postoperative VTE risk, characterized by a distinct threshold effect. These hypothesis-generating findings suggest that monitoring D-dimer trajectories rather than absolute values may offer a novel dimension for perioperative VTE risk stratification, warranting prospective validation in multicenter cohorts before clinical application. Endometrial carcinoma is one of the most common malignancies of the female reproductive system, with a continuously rising global incidence [1, 2]. Although surgical intervention remains the primary treatment for early-stage disease and effectively controls localized lesions, patients remain at substantial risk for venous thromboembolism (VTE) during the perioperative period. VTE—encompassing deep vein thrombosis (DVT) and pulmonary embolism (PE)—represents a severe, potentially fatal postoperative complication that significantly impairs prognosis and quality of life [3, 4]. Evidence indicates that patients with gynecologic malignancies inherently exist in a hypercoagulable state, and this risk is further exacerbated by surgical trauma, prolonged immobilization, and lymphadenectomy [5, 6]. D-dimer, a fibrin degradation product, serves as a key biomarker reflecting both fibrinolytic activity and thrombus formation, and has been widely adopted for VTE screening and risk stratification [7, 8]. Within gynecologic oncology, multiple studies have demonstrated a strong association between elevated preoperative D-dimer levels and VTE occurrence. For instance, Habu et al. [9] reported a VTE detection rate of 11.0% among endometrial cancer patients undergoing contrast-enhanced computed tomography, identifying a D-dimer threshold of ≥ 1.2 µg/mL as a robust predictive indicator. Similarly, Tasaka et al. [6] documented a 11.5% prevalence of asymptomatic preoperative VTE in endometrial cancer patients and identified age ≥ 60 years and advanced disease stage as independent risk factors. Notably, Nakamura et al. [10] further revealed that elevated preoperative D-dimer levels not only predict VTE but are also independently associated with shorter overall survival, suggesting its potential reflection of tumor biological aggressiveness. Additionally, Beckmann et al. [11] illustrated through a case report that endometrial cancer cells can drive recurrent thromboembolic events by overexpressing tissue factor and releasing procoagulant microparticles, with D-dimer levels dynamically fluctuating in response to tumor burden. Although the clinical utility of preoperative D-dimer measurement is increasingly recognized, a single time-point assessment may inadequately capture the dynamic evolution of coagulation status throughout the perioperative period. Recently, attention has shifted toward the change in D-dimer levels—termed the “D-dimer delta” (defined as the difference between peak postoperative and baseline preoperative values)—as a more comprehensive indicator of the integrated impact of surgical stress, inflammatory response, and underlying thrombogenesis on the hemostatic–fibrinolytic system. In colorectal surgery, Imamura et al. [12] demonstrated that preoperative D-dimer > 1.0 µg/mL combined with lower-limb venous ultrasound screening significantly reduced symptomatic postoperative VTE incidence. In gynecologic oncology, Komatsu et al. [13] observed marked postoperative D-dimer elevation, with greater increases among patients receiving adjuvant chemotherapy, underscoring the importance of serial monitoring. Nevertheless, systematic investigations into the association between perioperative D-dimer delta and VTE specifically in endometrial cancer patients remain scarce, particularly within real-world clinical cohorts validating its predictive performance. Accordingly, we conducted a retrospective cohort study involving 307 patients with endometrial carcinoma who underwent surgical treatment at Fujian Maternity and Child Health Hospital (Wusi North Campus) between January 2022 and December 2024. The primary objective was to evaluate whether D-dimer delta serves as an independent predictor of postoperative VTE, after adjusting for multiple covariates including age, FIGO stage, operative duration, and comorbidities. This study aims to provide a novel quantitative tool for perioperative VTE risk stratification in endometrial cancer patients, thereby generating hypotheses for future prospective validation and informing the design of individualized thromboprophylaxis strategies. Methods Study population This study was a single-center retrospective cohort study conducted at the Fuzhou Maternity and Child Health Hospital (Wusi North Campus) from January 2022 to December 2024, with follow-up extending through December 2025. The study population comprised patients diagnosed with endometrial cancer who underwent surgical treatment during this period, resulting in a final cohort of 307 individuals. Inclusion criteria were as follows: histopathologic ally confirmed endometrial cancer based on postoperative specimens, age ≥ 18 years, receipt of radical or staging surgery, availability of complete preoperative and postoperative D-dimer measurements, and documented venous thromboembolism (VTE) outcomes during follow-up. Exclusion criteria included: concomitant malignancies, pregnancy-related endometrial lesions, perioperative mortality, substantial missing clinical data (particularly for exposure or outcome variables), and loss to follow-up precluding definitive VTE status determination. Endometrial cancer diagnosis and staging adhered to the International Federation of Gynecology and Obstetrics (FIGO) pathological and clinical staging criteria [14]. All data were extracted from the hospital’s electronic medical records and postoperative follow-up documentation by two independently trained researchers; key variables were cross-verified to ensure consistency. Variables The primary exposure variable was the D-dimer difference, defined as the change in D-dimer concentration (in mg/L fibrinogen equivalent units [FEU]) between postoperative day 3 and the preoperative measurement obtained within 24 h before surgery. D-dimer levels in this study were measured using a Sysmex CS-5100 fully automated coagulation analyzer with Sysmex COAG-DIMER-PLUS reagent, employing an immunoturbidimetric assay. Results were reported in mg/L fibrinogen equivalent units (FEU), with a normal reference range of < 0.5 mg/L FEU. Preoperative D-dimer samples were collected within 24 h before surgery. Postoperative D-dimer samples were collected on the morning of postoperative day 3 (with the day of surgery defined as day 0), between 06:00 and 08:00, by nursing staff who had undergone standardized training. All blood samples were venous blood and were transported to the clinical laboratory within 2 h of collection. This metric reflects the degree of perioperative coagulation activation and was derived from routine clinical coagulation assays. The primary outcome was the occurrence of venous thromboembolism (VTE), encompassing deep vein thrombosis (DVT) and/or pulmonary embolism (PE). VTE events were adjudicated based on lower extremity venous ultrasonography, computed tomography pulmonary angiography (CTPA), or a composite clinical assessment involving high clinical suspicion followed by significant symptom resolution after anticoagulant therapy. All patients underwent routine lower-extremity venous ultrasound screening on postoperative days 4–7, and during the 30-day follow-up period, additional evaluations were performed at any time if there was clinical suspicion. Final confirmation required consensus between two senior attending physicians who independently reviewed medical records and imaging reports. Covariates included age (years), educational attainment (Primary school and below, middle school, high school, college and above), body mass index (BMI, kg/m²), length of hospital stay (days), payment method (insurance-covered, self-pay, etc.), FIGO stage (I–IV), operative duration (minutes), intraoperative blood loss (mL), history of hypertension, diabetes mellitus, cardiovascular disease, pulmonary disease, cerebrovascular disease, Caprini score (treated as a continuous variable), use of intermittent pneumatic compression (IPC) devices, receipt of perioperative anticoagulation, and postoperative infection. Among patients who received anticoagulation, the standard regimen consisted of subcutaneous low-molecular-weight heparin (LMWH; enoxaparin 40 mg/day, or weight/renal function-adjusted dosing), initiated on postoperative day 1 and continued for a standard duration of 7 days, extendable as clinically indicated. These covariates were selected based on their established roles as potential confounders or clinically relevant factors for VTE in prior literature [9, 15–17]. Cases were excluded only if data on the exposure or outcome variables were missing. For other covariates with missingness below 5%, complete-case analysis was performed without imputation. Ethic statement This study was approved by the Ethics Committee of Fujian Maternity and Child Health Hospital (Approval No.: 2025KY115) and conducted in strict accordance with the ethical principles of the Declaration of Helsinki. Given its retrospective cohort design, all patient data were de-identified and irreversibly anonymized, rendering individual re-identification impossible. Furthermore, the study did not interfere with clinical care pathways. Consequently, the ethics committee granted a waiver of individual informed consent. Patient privacy and data security were rigorously maintained throughout the study, with access to the anonymized database restricted exclusively to authorized research personnel. Statistical analysis This study employed a retrospective cohort design, enrolling 307 patients with endometrial cancer who underwent surgical treatment at the Fujian Maternity and Child Health Hospital (Wusi North Campus) between January 2022 and December 2024, with follow-up censored in December 2025. All analyses were based on complete-case analysis; the proportion of missing data for all covariates was less than 5%, and no imputation was performed. Continuous variables were presented as mean ± standard deviation or median (interquartile range), depending on their distributional characteristics, while categorical variables were described as frequencies (percentages). Baseline characteristics between groups were compared using one-way analysis of variance (ANOVA) or independent-samples t-tests for continuous variables and chi-square tests for categorical variables. All P values were two-sided. The primary analysis used univariable and multivariable logistic regression models to assess the association between the change in D-dimer levels (ΔD-dimer) and the risk of venous thromboembolism (VTE). Results were reported as unadjusted and adjusted odds ratios (ORs) with corresponding 95% confidence intervals (CIs). The multivariable model simultaneously adjusted for all 17 prespecified covariates: age, education level, body mass index (BMI), length of hospital stays, payment method, FIGO stage, operative duration, estimated blood loss, hypertension, diabetes mellitus, cardiovascular disease, pulmonary disease, cerebrovascular disease, Caprini use, intermittent pneumatic compression, anticoagulation therapy, and infection status. To explore potential nonlinear relationships between ΔD-dimer and VTE risk, generalized additive models (GAMs) were constructed using cubic B-spline basis functions (with three knots) to fit smooth terms. Nonlinearity was evaluated using Wald tests and likelihood ratio tests. Furthermore, a two-piecewise linear logistic regression model was applied to identify possible threshold effects. The optimal cut-off point was automatically determined via maximum likelihood estimation, and model fit was compared between the piecewise and linear models using likelihood ratio tests. Pre-specified subgroup analyses were conducted by stratifying participants according to key covariates (e.g., age ≤ 60 vs. >60 years; BMI ≤ 24 vs. >24 kg/m²). Within each stratum, the association between ΔD-dimer and VTE was estimated, and interaction effects were tested using likelihood ratio tests. Additionally, mediation analyses were performed for three candidate mediators (anticoagulation therapy, infection status, caprine use) using a counterfactual framework-based Monte Carlo approach with 5,000 parametric bootstrap replications. A probit model was employed to handle the binary outcome, and results are reported as total effect, direct effect, indirect effect (average causal mediation effect, ACME), and the proportion mediated, with effect sizes expressed as risk differences. Results Patient selection and study flow Of the 1,977 patients initially identified with endometrial cancer, a total of 1,305 were excluded based on the following criteria: non-surgical management (n = 1,084), absence of VTE (n = 201), and explicit refusal or inability to complete follow-up (n = 20). This screening process identified 672 potentially eligible participants. Further application of predefined exclusion criteria resulted in the removal of an additional 365 patients, including those with coexisting malignancies (n = 21) and those with substantial missing clinical information (n = 344). Consequently, 307 patients with complete datasets were included in the final analytical cohort (Fig. 1). Fig. 1The flowchart of patients’ selection. VTE: venous thromboembolism The flowchart of patients’ selection. VTE: venous thromboembolism Baseline characteristics and distribution of VTE incidence Patients were stratified into three groups based on tertiles of ΔD-dimer: low (0.23–3.26, n = 104), medium (3.28–5.61, n = 101), and high (5.65–10.90, n = 102). The overall incidence of VTE was 47.9% (147/307). A significant difference in VTE incidence was observed across the three groups (χ²=48.034, P < 0.001): 25.0% (26/104) in the low ΔD-dimer group, 73.3% (74/101) in the medium group, and 46.1% (47/102) in the high group. Notably, the medium ΔD-dimer group exhibited substantially higher VTE risk compared to the other two groups, suggesting a potential nonlinear association. Except for age (F = 4.455, P = 0.012) and hypertension (χ²=9.636, P = 0.008), all other covariates were well balanced across groups (all P > 0.05) (Table 1). Table 1Baseline characteristicsVariableOverallLowMediumHighTestP valueN307104 (0.23–3.26)101 (3.28–5.61)102 (5.65–10.90)--Occurrence of VTE (n, %)----χ²=48.034< 0.001No160 (52.1%)78 (75.0%)27 (26.7%)55 (53.9%)--Yes147 (47.9%)26 (25.0%)74 (73.3%)47 (46.1%)--Age (years, mean ± SD)54.50 ± 8.7753.42 ± 7.9856.60 ± 8.5253.50 ± 9.46F = 4.4550.012Education level (n, %)----χ²=5.0920.532 Primary school and below99 (32.2%)35 (33.7%)30 (29.7%)34 (33.3%)-- Middle school142 (46.3%)44 (42.3%)54 (53.5%)44 (43.1%)-- High school51 (16.6%)19 (18.3%)15 (14.9%)17 (16.7%)-- College and above15 (4.9%)6 (5.8%)2 (2.0%)7 (6.9%)--BMI (kg/m², mean ± SD)24.28 ± 4.3023.93 ± 4.1724.22 ± 4.1924.70 ± 4.55F = 0.8440.431Length of hospital stays (days, mean ± SD)13.51 ± 4.4613.18 ± 3.8713.96 ± 5.1913.40 ± 4.26F = 0.8220.440Payment method (n, %)----χ²=7.1760.518 Urban Employee Basic Medical Insurance114 (37.1%)40 (38.5%)31 (30.7%)43 (42.2%)-- Urban Resident Basic Medical Insurance129 (42.0%)40 (38.5%)48 (47.5%)41 (40.2%)-- Rural Medical Insurance59 (19.2%)23 (22.1%)21 (20.8%)15 (14.7%)-- Self-paid4 (1.3%)1 (1.0%)1 (1.0%)2 (2.0%)-- Social Insurance1 (0.3%)0 (0.0%)0 (0.0%)1 (1.0%)--FIGO stage (n, %)----χ²=2.6910.847 I221 (72.5%)79 (76.0%)67 (67.0%)75 (74.3%)-- II51 (16.7%)16 (15.4%)19 (19.0%)16 (15.8%)-- III29 (9.5%)8 (7.7%)12 (12.0%)9 (8.9%)-- IV4 (1.3%)1 (1.0%)2 (2.0%)1 (1.0%)--Operative duration (minutes, mean ± SD)180.15 ± 51.20175.41 ± 47.23179.49 ± 51.54185.63 ± 54.63F = 1.0370.356Estimated blood loss (mL ², mean ± SD)39.00 (29.00-48.50)38.00 (29.00-47.25)39.00 (28.00–48.00)39.00 (29.00–50.00)F = 2.5180.082Hypertension (n, %)----χ²=9.6360.008 Yes102 (33.2%)33 (31.7%)24 (23.8%)45 (44.1%)-- No205 (66.8%)71 (68.3%)77 (76.2%)57 (55.9%)--Diabetes mellitus (n, %)----χ²=0.0210.990 No211 (68.7%)72 (69.2%)69 (68.3%)70 (68.6%)-- Yes96 (31.3%)32 (30.8%)32 (31.7%)32 (31.4%)--Cardiovascular disease (n, %)----χ²=3.7500.153 No285 (92.8%)97 (93.3%)90 (89.1%)98 (96.1%)-- Yes22 (7.2%)7 (6.7%)11 (10.9%)4 (3.9%)--Pulmonary disease (n, %)----χ²=0.6110.737 No284 (92.5%)95 (91.3%)93 (92.1%)96 (94.1%)-- Yes23 (7.5%)9 (8.7%)8 (7.9%)6 (5.9%)--Cerebrovascular disease (n, %)----χ²=1.6910.429 No302 (98.4%)103 (99.0%)98 (97.0%)101 (99.0%)-- Yes5 (1.6%)1 (1.0%)3 (3.0%)1 (1.0%)--Caprini Score (n, %)----χ²=14.3010.282 411 (3.6%)4 (3.8%)2 (2.0%)5 (4.9%)-- 5128 (41.7%)46 (44.2%)40 (39.6%)42 (41.2%)-- 6130 (42.3%)45 (43.3%)43 (42.6%)42 (41.2%)-- 726 (8.5%)8 (7.7%)11 (10.9%)7 (6.9%)-- 86 (2.0%)0 (0.0%)1 (1.0%)5 (4.9%)-- 94 (1.3%)1 (1.0%)3 (3.0%)0 (0.0%)-- 102 (0.7%)0 (0.0%)1 (1.0%)1 (1.0%)--Intermittent pneumatic compression (n, %)----χ²=2.9980.223 Yes274 (89.3%)89 (85.6%)94 (93.1%)91 (89.2%)-- No33 (10.7%)15 (14.4%)7 (6.9%)11 (10.8%)--Anticoagulation therapy (n, %)----χ²=0.9520.621 Yes192 (62.7%)68 (65.4%)59 (59.0%)65 (63.7%)-- No114 (37.3%)36 (34.6%)41 (41.0%)37 (36.3%)--Infection status (n, %)----χ²=4.9700.083 No298 (97.1%)103 (99.0%)95 (94.1%)100 (98.0%)-- Yes9 (2.9%)1 (1.0%)6 (5.9%)2 (2.0%)--Continuous variables: Mean ± SD when mean > 2 × SD, otherwise Median (IQR); Categorical variables: n (%); Statistics: For two groups - t-test (t value), Chi-square test (χ² value); For multiple groups - ANOVA (F value), Chi-square test (χ² value); P-values are from two-sided tests. VTE: venous thromboembolism; BMI: body mass index Table 1Baseline characteristicsVariableOverallLowMediumHighTestP valueN307104 (0.23–3.26)101 (3.28–5.61)102 (5.65–10.90)--Occurrence of VTE (n, %)----χ²=48.034< 0.001No160 (52.1%)78 (75.0%)27 (26.7%)55 (53.9%)--Yes147 (47.9%)26 (25.0%)74 (73.3%)47 (46.1%)--Age (years, mean ± SD)54.50 ± 8.7753.42 ± 7.9856.60 ± 8.5253.50 ± 9.46F = 4.4550.012Education level (n, %)----χ²=5.0920.532 Primary school and below99 (32.2%)35 (33.7%)30 (29.7%)34 (33.3%)-- Middle school142 (46.3%)44 (42.3%)54 (53.5%)44 (43.1%)-- High school51 (16.6%)19 (18.3%)15 (14.9%)17 (16.7%)-- College and above15 (4.9%)6 (5.8%)2 (2.0%)7 (6.9%)--BMI (kg/m², mean ± SD)24.28 ± 4.3023.93 ± 4.1724.22 ± 4.1924.70 ± 4.55F = 0.8440.431Length of hospital stays (days, mean ± SD)13.51 ± 4.4613.18 ± 3.8713.96 ± 5.1913.40 ± 4.26F = 0.8220.440Payment method (n, %)----χ²=7.1760.518 Urban Employee Basic Medical Insurance114 (37.1%)40 (38.5%)31 (30.7%)43 (42.2%)-- Urban Resident Basic Medical Insurance129 (42.0%)40 (38.5%)48 (47.5%)41 (40.2%)-- Rural Medical Insurance59 (19.2%)23 (22.1%)21 (20.8%)15 (14.7%)-- Self-paid4 (1.3%)1 (1.0%)1 (1.0%)2 (2.0%)-- Social Insurance1 (0.3%)0 (0.0%)0 (0.0%)1 (1.0%)--FIGO stage (n, %)----χ²=2.6910.847 I221 (72.5%)79 (76.0%)67 (67.0%)75 (74.3%)-- II51 (16.7%)16 (15.4%)19 (19.0%)16 (15.8%)-- III29 (9.5%)8 (7.7%)12 (12.0%)9 (8.9%)-- IV4 (1.3%)1 (1.0%)2 (2.0%)1 (1.0%)--Operative duration (minutes, mean ± SD)180.15 ± 51.20175.41 ± 47.23179.49 ± 51.54185.63 ± 54.63F = 1.0370.356Estimated blood loss (mL ², mean ± SD)39.00 (29.00-48.50)38.00 (29.00-47.25)39.00 (28.00–48.00)39.00 (29.00–50.00)F = 2.5180.082Hypertension (n, %)----χ²=9.6360.008 Yes102 (33.2%)33 (31.7%)24 (23.8%)45 (44.1%)-- No205 (66.8%)71 (68.3%)77 (76.2%)57 (55.9%)--Diabetes mellitus (n, %)----χ²=0.0210.990 No211 (68.7%)72 (69.2%)69 (68.3%)70 (68.6%)-- Yes96 (31.3%)32 (30.8%)32 (31.7%)32 (31.4%)--Cardiovascular disease (n, %)----χ²=3.7500.153 No285 (92.8%)97 (93.3%)90 (89.1%)98 (96.1%)-- Yes22 (7.2%)7 (6.7%)11 (10.9%)4 (3.9%)--Pulmonary disease (n, %)----χ²=0.6110.737 No284 (92.5%)95 (91.3%)93 (92.1%)96 (94.1%)-- Yes23 (7.5%)9 (8.7%)8 (7.9%)6 (5.9%)--Cerebrovascular disease (n, %)----χ²=1.6910.429 No302 (98.4%)103 (99.0%)98 (97.0%)101 (99.0%)-- Yes5 (1.6%)1 (1.0%)3 (3.0%)1 (1.0%)--Caprini Score (n, %)----χ²=14.3010.282 411 (3.6%)4 (3.8%)2 (2.0%)5 (4.9%)-- 5128 (41.7%)46 (44.2%)40 (39.6%)42 (41.2%)-- 6130 (42.3%)45 (43.3%)43 (42.6%)42 (41.2%)-- 726 (8.5%)8 (7.7%)11 (10.9%)7 (6.9%)-- 86 (2.0%)0 (0.0%)1 (1.0%)5 (4.9%)-- 94 (1.3%)1 (1.0%)3 (3.0%)0 (0.0%)-- 102 (0.7%)0 (0.0%)1 (1.0%)1 (1.0%)--Intermittent pneumatic compression (n, %)----χ²=2.9980.223 Yes274 (89.3%)89 (85.6%)94 (93.1%)91 (89.2%)-- No33 (10.7%)15 (14.4%)7 (6.9%)11 (10.8%)--Anticoagulation therapy (n, %)----χ²=0.9520.621 Yes192 (62.7%)68 (65.4%)59 (59.0%)65 (63.7%)-- No114 (37.3%)36 (34.6%)41 (41.0%)37 (36.3%)--Infection status (n, %)----χ²=4.9700.083 No298 (97.1%)103 (99.0%)95 (94.1%)100 (98.0%)-- Yes9 (2.9%)1 (1.0%)6 (5.9%)2 (2.0%)--Continuous variables: Mean ± SD when mean > 2 × SD, otherwise Median (IQR); Categorical variables: n (%); Statistics: For two groups - t-test (t value), Chi-square test (χ² value); For multiple groups - ANOVA (F value), Chi-square test (χ² value); P-values are from two-sided tests. VTE: venous thromboembolism; BMI: body mass index Baseline characteristics Continuous variables: Mean ± SD when mean > 2 × SD, otherwise Median (IQR); Categorical variables: n (%); Statistics: For two groups - t-test (t value), Chi-square test (χ² value); For multiple groups - ANOVA (F value), Chi-square test (χ² value); P-values are from two-sided tests. VTE: venous thromboembolism; BMI: body mass index Distribution of VTE types Among the 147 patients who developed venous thromboembolism (VTE) after surgery for endometrial cancer, the most common type was left calf muscle vein thrombosis (CMVT), accounting for 86 cases (58.5%). Bilateral CMVT and right CMVT were observed in 23 (15.6%) and 22 (15.0%) patients, respectively. Distal deep vein thrombosis (DVT) occurred in 8 patients (5.4%), while mixed DVT and central DVT were identified in 5 (3.4%) and 2 (1.4%) patients, respectively. Pulmonary embolism (PE) was observed in only one patient (0.7%)(Fig. 2). Among the 147 patients who developed postoperative VTE, the median time from surgery to VTE diagnosis was 10 days (IQR: 6–28 days), All occurred later than the ΔD-dimer measurement time point (postoperative day 3). Fig. 2Anatomical Distribution of Postoperative VTE in Endometrial Cancer Patients (N = 147). VTE: venous thromboembolism; PE: pulmonary embolism; DVT: distal deep vein thrombosis; CMVT: calf muscle vein thrombosis Anatomical Distribution of Postoperative VTE in Endometrial Cancer Patients (N = 147). VTE: venous thromboembolism; PE: pulmonary embolism; DVT: distal deep vein thrombosis; CMVT: calf muscle vein thrombosis Univariate and multivariable logistic regression analyses Univariate logistic regression revealed that each one-unit increase in ΔD-dimer was associated with 12.8% higher odds of VTE (OR = 1.128, 95% CI: 1.034–1.231, P = 0.007). Age was also a significant predictor (OR = 1.114, 95% CI: 1.076–1.152, P < 0.001) (Table 2). In the multivariable model adjusted for all 17 covariates, the association between ΔD-dimer and VTE was further strengthened: the adjusted OR was 1.202 (95% CI: 1.076–1.342, P = 0.001), indicating that each one-unit increase in ΔD-dimer was associated with a 20.2% higher risk of VTE after controlling for confounders (Table 3). Table 2Univariate regression analysisVariableOR95% CIP valueΔD-dimer1.128(1.034, 1.231)0.007Age1.114(1.076, 1.152)<0.001Education level-->0.05BMI1.019(0.967, 1.073)0.485Length of hospital stays0.961(0.912, 1.012)0.133Payment method-->0.05FIGO stage-->0.05Operative duration1.000(0.996, 1.004)0.997Estimated blood loss1.001(0.998, 1.004)0.596Hypertension--0.838Diabetes mellitus--0.216Cardiovascular disease--0.131Pulmonary disease--0.391Cerebrovascular disease--0.723Caprini Score-->0.05Intermittent pneumatic compression--0.507Anticoagulation therapy--0.742Infection status--0.264OR: Odds Ratio. CI: 95% Confidence Interval. VTE: venous thromboembolism; BMI: body mass index Table 2Univariate regression analysisVariableOR95% CIP valueΔD-dimer1.128(1.034, 1.231)0.007Age1.114(1.076, 1.152)<0.001Education level-->0.05BMI1.019(0.967, 1.073)0.485Length of hospital stays0.961(0.912, 1.012)0.133Payment method-->0.05FIGO stage-->0.05Operative duration1.000(0.996, 1.004)0.997Estimated blood loss1.001(0.998, 1.004)0.596Hypertension--0.838Diabetes mellitus--0.216Cardiovascular disease--0.131Pulmonary disease--0.391Cerebrovascular disease--0.723Caprini Score-->0.05Intermittent pneumatic compression--0.507Anticoagulation therapy--0.742Infection status--0.264OR: Odds Ratio. CI: 95% Confidence Interval. VTE: venous thromboembolism; BMI: body mass index Univariate regression analysis OR: Odds Ratio. CI: 95% Confidence Interval. VTE: venous thromboembolism; BMI: body mass index Table 3Multivariable regression analysisVariableUnadjusted ORUnadjusted 95% CIUnadjusted P valueAdjusted ORAdjusted 95% CIAdjusted P valueΔD-dimer1.137(1.041, 1.241)0.0041.202(1.076, 1.342)0.001OR: Odds Ratio. CI: 95% Confidence Interval Table 3Multivariable regression analysisVariableUnadjusted ORUnadjusted 95% CIUnadjusted P valueAdjusted ORAdjusted 95% CIAdjusted P valueΔD-dimer1.137(1.041, 1.241)0.0041.202(1.076, 1.342)0.001OR: Odds Ratio. CI: 95% Confidence Interval Multivariable regression analysis OR: Odds Ratio. CI: 95% Confidence Interval Nonlinear association and threshold effect analysis Generalized additive modeling (GAM) demonstrated a significant nonlinear dose–response relationship between ΔD-dimer and VTE risk (likelihood ratio test LR = 75.564, P < 0.001; Wald test P < 0.001). The smoothed curve (Fig. 3) revealed an inverted U-shaped pattern, with VTE risk rising sharply and then declining as ΔD-dimer increased. To quantify this nonlinearity, we performed piecewise linear regression with threshold effect analysis (Table 4). A two-segment model with a threshold significantly outperformed the linear model (LR = 73.173, P < 0.001). The optimal threshold was identified at 3.780 (95% CI: 3.035–4.264). Among the 129 patients with ΔD-dimer ≤ 3.780, each one-unit increase in ΔD-dimer was strongly associated with higher VTE risk (OR = 7.761, 95% CI: 4.286–14.051, P < 0.001). Conversely, among the 175 patients with ΔD-dimer > 3.780, each one-unit increase was associated with reduced VTE risk (OR = 0.643, 95% CI: 0.526–0.785, P < 0.001). The difference in effect estimates between the two segments was highly significant (OR = 0.083, 95% CI: 0.040–0.171, P < 0.001). Fig. 3Non-linear associations between D-dimer change and VTE risk. Generalized Additive Model solved via IRLS with cubic B-spline basis (2 basis functions, degree = 17), adjusted for 17 covariate(s). Predicted probability with 95% confidence intervals. Smooth-term Wald test P = < 0.001; Likelihood ratio test LR = 75.564, P = < 0.001.VTE: venous thromboembolism Non-linear associations between D-dimer change and VTE risk. Generalized Additive Model solved via IRLS with cubic B-spline basis (2 basis functions, degree = 17), adjusted for 17 covariate(s). Predicted probability with 95% confidence intervals. Smooth-term Wald test P = < 0.001; Likelihood ratio test LR = 75.564, P = < 0.001.VTE: venous thromboembolism Table 4Threshold effect analysisCategoryMetricEstimateP-valueModel Ifitting model using logistic modelOR = 1.202 (1.076, 1.342)0.001Model IIFitting model using two-piecewise linear model: Threshold (K)3.780-Model IIEffect 1 (< K)OR = 7.564 (4.215, 13.575)< 0.001Model IIEffect 2 (> K)OR = 0.638 (0.522, 0.781)< 0.001Model IIEffect difference (2 vs. 1)OR = 0.084 (0.041, 0.173)< 0.001Model comparisonLog-likelihood ratio test (df = 1)LR = 73.181< 0.001Sample distributionSample size ≤ K129-Sample distributionSample size > K175- Table 4Threshold effect analysisCategoryMetricEstimateP-valueModel Ifitting model using logistic modelOR = 1.202 (1.076, 1.342)0.001Model IIFitting model using two-piecewise linear model: Threshold (K)3.780-Model IIEffect 1 (< K)OR = 7.564 (4.215, 13.575)< 0.001Model IIEffect 2 (> K)OR = 0.638 (0.522, 0.781)< 0.001Model IIEffect difference (2 vs. 1)OR = 0.084 (0.041, 0.173)< 0.001Model comparisonLog-likelihood ratio test (df = 1)LR = 73.181< 0.001Sample distributionSample size ≤ K129-Sample distributionSample size > K175- Threshold effect analysis Subgroup analyses and interaction assessment To evaluate whether the association between ΔD-dimer and VTE was modified by other factors, subgroup analyses were conducted across key covariates (Table 4). Interaction testing revealed that only hypertension status exerted a statistically significant modifying effect (P for interaction = 0.034). Among patients with hypertension (n = 205), ΔD-dimer was strongly associated with VTE (OR = 1.224, 95% CI: 1.089–1.376, P < 0.001), whereas no significant association was observed in those without hypertension (n = 102; OR = 1.006, 95% CI: 0.876–1.156, P = 0.929). Additionally, the effect of ΔD-dimer was stronger in patients with FIGO stage III disease (n = 51; OR = 1.371, 95% CI: 1.058–1.778, P = 0.017) and remained significant in patients without infection (n = 298; OR = 1.118, 95% CI: 1.024–1.221, P = 0.013). No significant interactions were observed for other covariates—including age, BMI, education level, FIGO stage or anticoagulant use (all P for interaction > 0.05)—indicating robustness of the main effect across most subgroups (Table 5). Table 5Stratified analysisStratification VariableStratumSample sizeOR95% CIP valueP for interactionAge-0---0.289 ≤ 60 (30–60)≤ 60 (30–60)2391.159(1.048, 1.283)0.004- > 60 (61–75)> 60 (61–75)680.970(0.785, 1.199)0.780-Education level-0---0.303BMI-0---0.161 ≤ 24 (17.7–23.9)≤ 24 (17.7–23.9)1621.156(1.019, 1.311)0.024- > 24 (24.1–31.9)> 24 (24.1–31.9)1451.104(0.978, 1.247)0.109-FIGO stage-0---0.230Operative duration-0---0.836 ≤ 171 (120–171)≤ 171 (120–171)1571.127(0.989, 1.283)0.072- > 171 (172–327)> 171 (172–327)1501.121(0.996, 1.262)0.059-Estimated blood loss-0---0.076 ≤ 39 (21–39)≤ 39 (21–39)1631.187(1.053, 1.339)0.005- > 39 (40–485)> 39 (40–485)1441.059(0.931, 1.204)0.383-Hypertension-0---0.034Anticoagulation therapy-0---0.749Infection status-0----Customized Stratification Thresholds: Age: 60 years; BMI: 24; Operation Time: 171 min; Blood Loss: 39 ml. OR: Odds Ratio. CI: 95% Confidence Interval. BMI: body mass index Table 5Stratified analysisStratification VariableStratumSample sizeOR95% CIP valueP for interactionAge-0---0.289 ≤ 60 (30–60)≤ 60 (30–60)2391.159(1.048, 1.283)0.004- > 60 (61–75)> 60 (61–75)680.970(0.785, 1.199)0.780-Education level-0---0.303BMI-0---0.161 ≤ 24 (17.7–23.9)≤ 24 (17.7–23.9)1621.156(1.019, 1.311)0.024- > 24 (24.1–31.9)> 24 (24.1–31.9)1451.104(0.978, 1.247)0.109-FIGO stage-0---0.230Operative duration-0---0.836 ≤ 171 (120–171)≤ 171 (120–171)1571.127(0.989, 1.283)0.072- > 171 (172–327)> 171 (172–327)1501.121(0.996, 1.262)0.059-Estimated blood loss-0---0.076 ≤ 39 (21–39)≤ 39 (21–39)1631.187(1.053, 1.339)0.005- > 39 (40–485)> 39 (40–485)1441.059(0.931, 1.204)0.383-Hypertension-0---0.034Anticoagulation therapy-0---0.749Infection status-0----Customized Stratification Thresholds: Age: 60 years; BMI: 24; Operation Time: 171 min; Blood Loss: 39 ml. OR: Odds Ratio. CI: 95% Confidence Interval. BMI: body mass index Stratified analysis Customized Stratification Thresholds: Age: 60 years; BMI: 24; Operation Time: 171 min; Blood Loss: 39 ml. OR: Odds Ratio. CI: 95% Confidence Interval. BMI: body mass index Mediation analysis Based on clinical experience candidate mediators (anticoagulant use; infection; Caprini), mediation analysis was performed using Monte Carlo simulation (5,000 iterations). Results showed that none of the three mediators exhibited significant indirect effects: the proportion mediated by anticoagulation was only 0.1% (ACME = 0.0001, P = 0.912); by infection, 0.0% (ACME ≈ 0.0000, P = 0.992); and by Caprini, 0.2% (ACME = 0.0003, P = 0.860). The total effect (risk difference of approximately 0.125–0.129) was almost entirely explained by the direct effect (> 99.8%), suggesting that the impact of ΔD-dimer on VTE operates primarily through pathways other than these three proposed mediators (Table 6). Table 6T mediation analysis of potential mediators in the association between d-dimer difference and VTE riskEffect MetricAnticoagulation TherapyInfectionCapriniEstimate (P value)Estimate (P value)Estimate (P value)Total Effect (c)0.1287 (< 0.001)0.1286 (0.002)0.1255 (0.001)Indirect Effect (ACME)ᵃ0.0001 (0.912)0.0000 (0.992)0.0003 (0.860)Direct Effect (c’)0.1281 (< 0.001)0.1278 (< 0.001)0.1249 (0.001)Proportion Mediatedᵇ0.1% (0.912)0.0% (0.992)0.2% (0.860)Path a (X→M)ᶜ0.0012 (0.907)−0.0000 (0.996)0.0390 (0.215)Path b (M→Y|X)ᵈ0.1969 (0.248)1.7040 (0.011)0.0133 (0.822)ᵃ ACME, average causal mediation effect (indirect effect through the mediator)ᵇ Proportion of the total effect explained by the mediatorᶜ Path a: effect of D-dimer difference (X) on mediator (M)ᵈ Path b: effect of mediator (M) on VTE risk (Y), adjusted for D-dimer difference (X)VTE, venous thromboembolism; ACME, average causal mediation effect; X, exposure variable (D-dimer difference); M, mediator; Y, outcome variable (VTE occurrence) Table 6T mediation analysis of potential mediators in the association between d-dimer difference and VTE riskEffect MetricAnticoagulation TherapyInfectionCapriniEstimate (P value)Estimate (P value)Estimate (P value)Total Effect (c)0.1287 (< 0.001)0.1286 (0.002)0.1255 (0.001)Indirect Effect (ACME)ᵃ0.0001 (0.912)0.0000 (0.992)0.0003 (0.860)Direct Effect (c’)0.1281 (< 0.001)0.1278 (< 0.001)0.1249 (0.001)Proportion Mediatedᵇ0.1% (0.912)0.0% (0.992)0.2% (0.860)Path a (X→M)ᶜ0.0012 (0.907)−0.0000 (0.996)0.0390 (0.215)Path b (M→Y|X)ᵈ0.1969 (0.248)1.7040 (0.011)0.0133 (0.822)ᵃ ACME, average causal mediation effect (indirect effect through the mediator)ᵇ Proportion of the total effect explained by the mediatorᶜ Path a: effect of D-dimer difference (X) on mediator (M)ᵈ Path b: effect of mediator (M) on VTE risk (Y), adjusted for D-dimer difference (X)VTE, venous thromboembolism; ACME, average causal mediation effect; X, exposure variable (D-dimer difference); M, mediator; Y, outcome variable (VTE occurrence) T mediation analysis of potential mediators in the association between d-dimer difference and VTE risk ᵃ ACME, average causal mediation effect (indirect effect through the mediator) ᵇ Proportion of the total effect explained by the mediator ᶜ Path a: effect of D-dimer difference (X) on mediator (M) ᵈ Path b: effect of mediator (M) on VTE risk (Y), adjusted for D-dimer difference (X) VTE, venous thromboembolism; ACME, average causal mediation effect; X, exposure variable (D-dimer difference); M, mediator; Y, outcome variable (VTE occurrence) Discussion This study was a retrospective cohort investigation aimed at evaluating the association between the D-dimer delta (i.e., the difference in D-dimer levels) and the risk of venous thromboembolism (VTE) in patients following endometrial cancer surgery. After comprehensive adjustment for 17 covariates, each one-unit increase in D-dimer delta was associated with a statistically significant 20.2% higher risk of VTE (adjusted odds ratio [OR] = 1.202; 95% confidence interval [CI]: 1.076–1.342; P = 0.001). More importantly, nonlinear analyses revealed a distinct threshold effect: when the D-dimer delta was ≤ 3.78 µg/mL, VTE risk increased sharply (OR = 7.761; 95% CI: 4.286–14.051); beyond this threshold, however, the risk significantly decreased (OR = 0.643; 95% CI: 0.526–0.785), indicating a complex non-monotonic relationship. Furthermore, mediation analysis demonstrated that potential mediators—including anticoagulant therapy and infection status—contributed minimally to the total effect (mediation proportion < 0.2%; P > 0.86), suggesting that the influence of D-dimer delta on VTE occurs predominantly through a direct pathway. Our findings are generally consistent with prior reports in gynecologic oncology regarding the relationship between D-dimer and VTE, yet they also highlight important distinctions. Tasaka et al. [6] identified preoperative D-dimer elevation as an independent predictor of VTE in endometrial cancer patients among a cohort of 2,086 individuals with gynecologic malignancies, with age ≥ 60 years and advanced-stage disease as major risk factors. Similarly, Zhou et al. [18] demonstrated that preoperative D-dimer > 4.215 µg/mL predicted perioperative VTE events in ovarian cancer patients. The underlying mechanism may involve moderate D-dimer elevation reflecting physiological postoperative coagulation activation, whereas extreme elevations could signal the onset of consumptive coagulopathies such as disseminated intravascular coagulation (DIC), paradoxically reducing thrombotic potential [17, 19]. However, both studies focused exclusively on static preoperative D-dimer levels. In contrast, our study innovatively employed a dynamic metric—the difference between preoperative and postoperative D-dimer levels—to more precisely capture the degree of coagulation activation triggered by surgical trauma. This observation aligns mechanistically with the case report by Beckmann et al. [11], which described high tissue factor (TF) expression by endometrial cancer cells and the release of TF-bearing microvesicles that drive the coagulation cascade, leading to a sharp rise in D-dimer. Following effective surgical removal of the tumor burden, the procoagulant stimulus diminishes, potentially causing D-dimer levels to decline—thereby explaining the reduced VTE risk observed beyond the high-threshold point. Although Komatsu et al. [13] noted a general postoperative elevation of D-dimer in gynecologic cancer patients, they did not explore the nonlinear relationship between its dynamic changes and VTE; our study provides critical evidence addressing this gap. These findings have potential clinical relevance, though they should be interpreted as hypothesis-generating given the retrospective, single-center design. First, ΔD-dimer may serve as a candidate biomarker for identifying endometrial cancer patients at elevated risk of postoperative VTE. In particular, the identified threshold of 3.78 mg/L FEU—below which VTE risk increased steeply—could potentially inform risk stratification algorithms, if prospectively validated. Clinicians should be cautious, however, in directly applying this threshold in practice pending external validation, as the threshold was derived from the same dataset used to identify it and may be subject to overfitting. Second, in contrast to reliance solely on static preoperative D-dimer values [20], monitoring the trajectory of D-dimer change may offer additional prognostic information regarding perioperative coagulation dynamics. This hypothesis is mechanistically supported by prior work on tissue factor expression in endometrial cancer cells [11] and warrants investigation in prospective multicenter studies. From a health system perspective, if ΔD-dimer-guided stratification is ultimately validated, it could potentially support more targeted thromboprophylaxis, reducing both undertreatment in high-risk patients and unnecessary anticoagulation in low-risk patients. However, we emphasize that these translational implications remain speculative at this stage and should not be acted upon clinically without further confirmatory evidence. Methodologically, this study possesses several strengths. First, the retrospective cohort design allows for the examination of temporal precedence between exposure (ΔD-dimer) and outcome (VTE). While this design cannot establish causality—as residual confounding and unmeasured variables remain inherent limitations—it provides stronger observational evidence for hypothesis generation than cross-sectional designs [21, 22]. Second, data quality was high, with missingness for all covariates below 5%; complete-case analysis was employed to handle missing data, minimizing potential bias. Third, our statistical approach was rigorous and comprehensive: beyond conventional multivariable logistic regression to control for confounding, we applied generalized additive models (GAMs) and piecewise linear regression to explicitly test and quantify nonlinear relationships and threshold effects, thereby avoiding erroneous conclusions that might arise from assuming linearity. Finally, mediation analysis systematically evaluated the potential mediating roles of key variables such as anticoagulation and infection, deepening our mechanistic understanding and confirming that the effect of D-dimer delta on VTE is primarily direct. Despite these strengths, several limitations warrant acknowledgment. First, this is a single-center study, with patients drawn from a single region in Fujian Province and possessing specific demographic and clinical characteristics. Caution is therefore warranted when generalizing the findings to other regions, other ethnic groups, or medical institutions using different D-dimer assay platforms. Second, as an observational study, although we adjusted for a wide range of known confounders, residual confounding (such as unmeasured inherited thrombophilia or detailed tumor molecular subtypes) cannot be fully excluded; thus, causality cannot be established (Hess, 2023). Third, the degree of standardization in D-dimer testing methods and sampling time points (e.g., exact postoperative hours) may affect the accuracy of the measured differences, representing an inherent measurement limitation of retrospective studies. Fourth, the threshold of 3.78 mg/L FEU identified in this study was determined automatically via maximum likelihood estimation, and the cut-off selection was based on the same dataset, posing a risk of statistical overfitting. This threshold has not been validated in an independent external data.