Abstract
Objective
To investigate the association between the change in D-dimer levels (ΔD-dimer) and the development of localized intravascular coagulation (LIC) following surgical intervention in patients with venous malformations.
Methods
This retrospective cohort study enrolled 267 patients with venous malformations who underwent surgical treatment at Fujian Maternity and Child Health Hospital between June 2021 and December 2025. The primary exposure variable was the preoperative-to-postoperative ΔD-dimer. The principal outcome was postoperative LIC. Multivariable logistic regression analysis was employed to estimate the effect size, adjusted for potential confounders.
Results
The incidence of LIC was 31.1% (83/267). Each unit increase in ΔD-dimer was significantly associated with a higher risk of LIC (adjusted odds ratio [aOR] = 2.51; 95% confidence interval [CI]: 1.93–3.26; p < 0.001). This association was consistent across stratified subgroups, including sex, age, number of lesions, tissue layer involvement, lesion volume, lesion location, surgical approach, and anticoagulation status (all P for interaction > 0.05), with a borderline interaction by head/neck involvement (P for interaction = 0.054). ΔD-dimer is independently associated with postoperative LIC in patients with venous malformations, with a monotonic dose–response relationship. A threshold near 4.8 mg/L FEU is suggested by exploratory segmented analysis but requires external validation before any clinical use.
Conclusion
ΔD-dimer is independently associated with postoperative LIC in patients with venous malformations, demonstrating a significant dose–response relationship. Collectively, these findings suggest that ΔD-dimer is independently associated with postoperative LIC and may serve as a risk marker for perioperative coagulation monitoring.
Keywords: D-dimer, venous malformations, localized intravascular coagulation, biomarker, postoperative complications
GRAPHICAL ABSTRACT
Introduction
Venous malformation (VM), the most prevalent slow-flow vascular malformation encountered in clinical practice, fundamentally arises from congenital structural abnormalities [1,2]. A critical pathophysiological feature of VM involves numerous dilated, tortuous venous channels that foster sluggish or stagnant blood flow. This abnormal hemodynamics initiates localized activation of the coagulation cascade, leading directly to localized intravascular coagulation (LIC) [3,4]. LIC, a consumptive coagulopathy, is consistently evidenced by reduced fibrinogen, elevated D-dimer, and diminished specific coagulation factors (e.g. FV, FVIII, FXIII), often accompanied by mild to moderate thrombocytopenia in severe presentations [5,6]. Despite its intrinsic confinement to the lesion site and distinction from systemic disseminated intravascular coagulation (DIC), LIC poses a substantial elevation in perioperative risk due to its potential for rapid progression to severe hemorrhagic or thrombotic events following surgical or interventional trauma [7–9].
D-dimer, a specific degradation product of cross-linked fibrin, serves as a pivotal biomarker reflecting both fibrinolytic activity and coagulation activation in vivo [10]. Indeed, in patients with VM, baseline D-dimer levels consistently correlate with lesion volume, multiplicity, deep tissue infiltration, and the presence of phleboliths, thereby establishing D-dimer as a critical laboratory indicator for assessing LIC [3,5,6]. However, current understanding primarily relies on static measurements. Recent investigations have increasingly focused on the dynamic changes in D-dimer. For instance, Leung et al. [11] observed that in children with VM undergoing ethanol sclerotherapy, postoperative D-dimer levels significantly increased beyond baseline, correlating with treatment response and long-term recurrence. This critical observation underscores a limitation of static preoperative D-dimer values for comprehensive coagulation risk assessment. Instead, we hypothesize that the perioperative D-dimer delta (i.e. the difference between postoperative and preoperative values) could more sensitively reflect the perturbation in local hemostatic–fibrinolytic equilibrium induced by surgical trauma, thus serving as a novel and superior predictor of postoperative LIC.
To date, evidence linking the D-dimer delta to postoperative localized LIC remains limited. Although established in other surgical contexts—such as colorectal cancer, bone tumor resection, and aortic dissection—as independent predictors of venous thromboembolism or adverse outcomes [12–14], the applicability of these preoperative or postoperative D-dimer levels to the distinct VM patient population is uncertain. This uncertainty stems from the unique pathophysiology of VM-associated LIC, which primarily arises from chronic, localized coagulation activation driven by abnormal hemodynamics within the lesion, as opposed to systemic inflammation or malignancy-induced hypercoagulability [3,15]. Consequently, prospective investigations in VM patients are needed to elucidate the utility of dynamic D-dimer changes in perioperative coagulation risk stratification and to inform evidence-based management strategies.
To address this need, we leveraged clinical data from Fujian Provincial Maternity and Children’s Hospital to conduct a retrospective cohort study. This investigation aimed to delineate the association between perioperative D-dimer delta and the occurrence of postoperative LIC in patients with venous malformations. To achieve this objective, we integrated patient demographics (sex, age), lesion characteristics, surgical approach, and multiple potential confounders into comprehensive multivariable analyses. Our goal was to provide clinicians with a simple, objective risk marker to inform perioperative coagulation monitoring.
Methods
Study population
This single-center retrospective cohort study was conducted at Fujian Maternity and Child Health Hospital, enrolling patients from June 2021 to December 2025, with follow-up concluding in February 2026. The study population consisted of 267 individuals diagnosed with venous malformations based on established clinical and imaging criteria. Diagnosis strictly adhered to the International Society for the Study of Vascular Anomalies (ISSVA) classification system [1], integrating comprehensive evaluation via medical history, physical examination, ultrasonography, and magnetic resonance imaging (MRI).
Patient enrollment was governed by rigorous inclusion and exclusion criteria. Inclusion criteria comprised: confirmed diagnosis of venous malformations, undergoing surgical intervention, availability of complete perioperative D-dimer measurements, and accessible postoperative follow-up data. Conversely, exclusion criteria encompassed: coexisting other vascular malformations (e.g. Klippel-Trenaunay syndrome) or malignancies, receipt of systemic anticoagulation or thrombolytic therapy for other diseases during the perioperative period, missing data on key variables (exposure or outcome), and pre-existing LIC. In this study, “surgical intervention” refers to any definitive procedural treatment for venous malformations performed by the vascular surgery team, including percutaneous ethanol sclerotherapy, surgical resection, or combined procedures.
For data acquisition, all data were extracted from the hospital’s electronic medical record system by two independently trained researchers using a standardized form. This form facilitated the systematic collection of demographic characteristics, clinical manifestations, treatment details, and laboratory parameters.
Variables
The primary exposure variable was the D-dimer delta(ΔD-dimer), defined as the change (in mg/L fibrinogen equivalent units [FEU]) between the peak postoperative D-dimer level, measured within 24 h postsurgery, and the preoperative baseline value. This D-dimer measurement was derived from peripheral blood, was both anatomically and temporally independent of the D-dimer used for outcome adjudication described below, and was not used for outcome adjudication. D-dimer levels were measured using an immunoturbidimetric assay (reported in mg/L fibrinogen equivalent units [FEU]; normal reference range: < 0.55 mg/L FEU, the analytical lower limit of detection is 0.10 mg/L FEU, and the upper limit of the reportable range is 1000 mg/L FEU), with sampling strictly confined to the 48-hour preoperative window (single measurement), and the initial 24 h postsurgery. When multiple measurements were available, the peak postoperative value was defined as the highest value. The assay platform remained unchanged throughout the study period.
The primary outcome was the occurrence of LIC after surgery. LIC diagnosis required the simultaneous fulfillment of at least one clinical sign and at least one laboratory finding, as follows. Clinical signs included: localized skin cyanosis, exacerbation of swelling, or acute worsening of pain, laboratory finding included: elevated D-dimer in local drainage or aspirate fluid (greater than 2 times the local laboratory upper limit of normal, i.e. >1.1 mg/L FEU). Imaging findings (ultrasonography or MRI demonstrating stasis or microthrombi within the lesion) were recorded as supportive evidence but were not a mandatory diagnostic requirement. Furthermore, to ensure diagnostic fidelity, two senior vascular surgeons independently adjudicated LIC status based on predefined criteria; adjudicators were blinded to the exposure (peripheral blood D-dimer values and calculated ΔD-dimer). Any discrepancies were resolved by a third expert.
A comprehensive set of covariates was also incorporated into the analysis, encompassing: sex, age (age was recorded in years; for infants aged < 1 year, age was expressed in decimal years, e.g. 0.87 years ≈10 months), lesion location (head/neck, trunk, extremities), number of lesions (solitary vs. multiple), involved tissue layers (subcutaneous, muscular), lesion volume, length of hospital stay, payment method, number of surgical procedures, presence of preoperative swelling or pain, surgical approach, intraoperative use of absolute ethanol, comorbid cardiovascular disease, family history of thrombosis or bleeding disorders, Puig classification (types I–IV), and receipt of perioperative anticoagulation. Surgical approach was recorded as a binary variable: sclerotherapy (percutaneous ethanol sclerotherapy, with or without concurrent embolization) versus resection (open surgical resection, with or without concurrent sclerotherapy). There is currently no established practice guideline for anticoagulant prophylaxis against LIC in VM patients. At our institution, perioperative anticoagulation follows a standardized protocol: low-molecular-weight heparin (LMWH, enoxaparin) is administered at prophylactic dose (40 mg subcutaneously once daily in adults, or 0.5 mg/kg in children weighing < 40 kg), initiated 6–12 h postoperatively and continued for 3–5 days, extended to 1 weeks for patients with extensive lesions (Puig types III–IV), multifocal disease, or baseline coagulopathy. The decision to initiate anticoagulation was made by the treating surgeon based on lesion characteristics (volume, location, Puig classification) and baseline coagulation profile (baseline D-dimer and fibrinogen), not on the postoperative ΔD-dimer that serves as the exposure variable. Anticoagulation was recorded as a binary variable (yes/no); agent, dose, and duration were not retained for analysis. Lesion volume was assessed preoperatively by MRI (primary modality) or ultrasonography (for superficial lesions). The three maximum orthogonal diameters were measured on axial and coronal T2-weighted fat-saturated MRI sequences, and lesion volume was estimated using the ellipsoid formula V = (L × W × D) × π/6; for multifocal lesions, individual nodule volumes were summed. Volume was then categorized as small (< 10 cm³ or maximum diameter < 3 cm), moderate (10–50 cm³ or maximum diameter 3–5 cm), or large (> 50 cm³ or maximum diameter > 5 cm), based on clinically meaningful cutoffs used in our institutional practice. All covariates were abstracted from medical records, having been selected based on prior literature suggesting their potential association with postoperative coagulopathy in venous malformations. Cases were excluded only if data on the primary exposure or outcome variable were missing; for other covariates, missingness was <5%, and complete-case analysis was performed without imputation.
Statistical analysis
Continuous variables were assessed for normality using the Shapiro–Wilk test. Normally distributed continuous variables were presented as mean ± standard deviation (SD) and compared using independent-samples t-tests (two-group comparisons) or one-way analysis of variance (ANOVA; comparisons across more than two groups). Non-normally distributed continuous variables were presented as median (interquartile range [IQR]) and compared using the Mann–Whitney U test (two-group comparisons) or the Kruskal–Wallis test (comparisons across more than two groups). Categorical variables were reported as frequency (percentage) and compared using the chi-square test. All hypothesis tests were two-sided, with statistical significance defined at α = 0.05.
Univariate logistic regression estimated the association between each covariate and LIC, reporting odds ratios (ORs) with 95% confidence intervals (CIs). Variables exhibiting complete or quasi-complete separation (preoperative swelling, preoperative pain, Social Insurance payment category) were excluded from standard univariate analysis; their Firth-corrected estimates are reported in the supplementary material.
For multivariable analysis, covariates were selected a priori based on biological plausibility and prior VM-LIC literature: sex, age, number of lesions, lesion volume, Puig classification (ordinal I–IV), surgical approach, absolute ethanol use, and anticoagulation. Administrative variables (payment method), collinear variables (number of procedures), and separated variables (swelling, pain) were excluded. The parsimonious model contained 10 covariates—ΔD-dimer plus 9 confounders (11 parameters including intercept)—yielding an events-per-variable ratio of 8.3, consistent with the recommended minimum of ∼10. As a sensitivity analysis, Firth bias-reduced penalised logistic regression was additionally fitted on both the parsimonious and full covariate sets (with lesion location collapsed into three binary involvement indicators and payment into a binary indicator) to obtain stable estimates under separation. Profile-likelihood CIs were used for separated variables.
Subgroup analyses used Firth-penalised logistic regression within each stratum, with interactions tested by likelihood-ratio tests. Lesion location was stratified by head/neck involvement (yes vs no).
Nonlinear dose-response was explored using a generalised additive model (GAM) with a cubic B-spline basis (df = 4), compared with a linear-logit model by likelihood-ratio test. A two-piece segmented logistic regression with grid-search breakpoint identification was also fitted. These analyses were prespecified as exploratory. Model discrimination was assessed by the area under the receiver operating characteristic curve (AUC), with internal validation via 500-iteration bootstrap (optimism-corrected AUC); calibration by the Hosmer–Lemeshow test; and cutoff performance by sensitivity, specificity, positive predictive value, and negative predictive value. External validation is required before any cutoff is recommended clinically.
Analyses were performed in Python 3.11 (pandas, statsmodels, scipy, scikit-learn).
Results
Patient selection and study flow
From an initial pool of 467 patients identified with venous malformations, 169 were excluded based on predefined criteria, specifically: non-surgical intervention (n = 46) and incomplete preoperative and postoperative D-dimer measurements (n = 123). This initial screening process thus yielded 298 potentially eligible participants. Subsequently, the application of further predefined exclusion criteria led to the removal of an additional 31 patients. These individuals included those with coexisting other types of vascular malformations (n = 17), those who received anticoagulation for other diseases during the perioperative period (n = 3), and those with missing data for key variables (either exposure or outcome) (n = 11). Consequently, a final cohort of 267 patients with complete datasets was included in the analytical cohort (Figure 1).
Figure 1.
The flowchart of patients’ selection.
Baseline characteristics and distribution of LIC incidence
The study cohort comprised 267 patients with venous malformations who underwent surgical intervention. Demographically, the age ranged from 0.014 to 68 years (median 8.0 years; IQR 4.0–11.0), including 20 infants (< 1 year), 223 pediatric patients (1–17 years), and 24 adults (≥ 18 years; 9.0% of the cohort). Gender distribution was near-even (130 males [48.7%] and 137 females [51.3%]). For analytical purposes, patients were stratified into three groups based on tertiles of D-dimer delta (ΔD-dimer): low (−2.10 to 1.51 mg/L FEU, n = 89), intermediate (1.54 to 3.12 mg/L FEU, n = 91), and high (3.20 to 6.89 mg/L FEU, n = 87). The overall incidence of postoperative LIC was 31.1% (83/267), with a significant and escalating gradient observed across the three ΔD-dimer groups: 6.7% (6/89) in the low group, 25.3% (23/91) in the intermediate group, and 62.1% (54/87) in the high group (χ2 = 65.041, p < 0.001) (Table 1). Of the 267 patients, 32.2% (86/267) had preoperative D-dimer levels exceeding the upper limit of normal. A positive ΔD-dimer was observed in 96.6% (258/267) of patients. The median time from surgery to LIC diagnosis was 4 days (IQR: 3–6 days; range: 2–8 days). All exposure (ΔD-dimer) measurements were completed within 24 h postoperatively, preceding the outcome ascertainment window.
Table 1.
Baseline characteristics.
| Variable | Overall | Low | Medium | High | Test | P Value |
|---|---|---|---|---|---|---|
| N | 267 | 89 (−2.10 to 1.51 mg/L FEU) | 91 (1.54 to 3.12 mg/L FEU) | 87 (3.20 to 6.89 mg/L FEU) | – | – |
| LIC | – | – | – | – | χ²=65.041 | <0.001 |
| No | 184 (68.9%) | 83 (93.3%) | 68 (74.7%) | 33 (37.9%) | – | – |
| Yes | 83 (31.1%) | 6 (6.7%) | 23 (25.3%) | 54 (62.1%) | – | – |
| Sex | – | – | – | – | χ²=0.476 | 0.788 |
| Male | 130 (48.7%) | 42 (47.2%) | 43 (47.3%) | 45 (51.7%) | – | – |
| Female | 137 (51.3%) | 47 (52.8%) | 48 (52.7%) | 42 (48.3%) | – | – |
| Age | 8.00 (4.00–11.00) | 8.00 (5.00–11.00) | 8.00 (4.00–11.00) | 9.00 (4.50–11.50) | F = 0.196 | 0.822 |
| Lesion location | – | – | – | – | χ²=56.739 | <0.001 |
| Head/neck | 36 (13.5%) | 22 (24.7%) | 9 (9.9%) | 5 (5.7%) | – | – |
| Trunk + extremities | 63(23.6%) | 12 (13.5%) | 21 (23.1%) | 30 (34.5%) | – | – |
| Extremities | 50 (18.7%) | 21 (23.6%) | 22 (24.2%) | 7 (8.0%) | – | – |
| Trunk | 30 (11.2%) | 8 (9.0%) | 9 (9.9%) | 13 (14.9%) | – | – |
| Head/neck + trunk | 42 (15.7%) | 11 (12.4%) | 17 (18.7%) | 14 (16.1%) | – | – |
| Head/neck + extremities | 36 (13.5%) | 14 (15.7%) | 11 (12.1%) | 11 (12.6%) | – | – |
| Head/neck + trunk + extremities | 10 (3.7%) | 1 (1.1%) | 2 (2.2%) | 7 (8.0%) | – | – |
| Number of lesions | – | – | – | – | χ²=14.628 | <0.001 |
| Solitary | 116 (43.4%) | 51 (57.3%) | 40 (44.0%) | 25 (28.7%) | – | – |
| Multiple | 151 (56.6%) | 38 (42.7%) | 51 (56.0%) | 62 (71.3%) | – | – |
| Involved tissue layers | – | – | – | – | χ²=3.257 | 0.196 |
| Muscular | 211 (79.0%) | 65 (73.0%) | 73 (80.2%) | 73 (83.9%) | – | – |
| Subcutaneous | 56 (21.0%) | 24 (27.0%) | 18 (19.8%) | 14 (16.1%) | – | – |
| Lesion volume | – | – | – | – | χ²=9.716 | 0.046 |
| Small | 104 (39.0%) | 43 (48.3%) | 35 (38.5%) | 26 (29.9%) | – | – |
| Moderate | 116 (43.4%) | 37 (41.6%) | 40 (44.0%) | 39 (44.8%) | – | – |
| Large | 47 (17.6%) | 9 (10.1%) | 16 (17.6%) | 22 (25.3%) | – | – |
| Length of hospital stay | 4.16 ± 1.72 | 4.18 ± 1.66 | 4.09 ± 1.74 | 4.22 ± 1.77 | 0.875 | |
| Payment method | – | – | – | – | χ²=8.917 | 0.349 |
| Employee Medical Insurance | 14 (5.2%) | 4 (4.5%) | 5 (5.5%) | 5 (5.7%) | – | – |
| Social Insurance | 3 (1.1%) | 3 (3.4%) | 0 (0.0%) | 0 (0.0%) | – | – |
| Rural Medical Insurance | 7 (2.6%) | 3 (3.4%) | 3 (3.3%) | 1 (1.1%) | – | – |
| Resident Medical Insurance | 228 (85.4%) | 74 (83.1%) | 76 (83.5%) | 78 (89.7%) | – | – |
| Self-pay | 15 (5.6%) | 5 (5.6%) | 7 (7.7%) | 3 (3.4%) | – | – |
| Number of surgical procedures | 2.139 ± 1.63 | 2.146 ± 1.67 | 2.143 ± 1.79 | 2.126 ± 1.46 | 0.990 | |
| Preoperative swelling | – | – | – | – | χ²=7.977 | 0.019 |
| No | 257 (96.3%) | 89 (100.0%) | 88 (96.7%) | 80 (92.0%) | – | – |
| Yes | 10 (3.7%) | 0 (0.0%) | 3 (3.3%) | 7 (8.0%) | – | – |
| Preoperative pain | – | – | – | – | χ²=9.327 | 0.009 |
| No | 258 (96.6%) | 89 (100.0%) | 89 (97.8%) | 80 (92.0%) | – | – |
| Yes | 9 (3.4%) | 0 (0.0%) | 2 (2.2%) | 7 (8.0%) | – | – |
| Surgical approach | – | – | – | – | χ²=5.543 | 0.063 |
| Sclerotherapy | 219 (82.0%) | 68 (76.4%) | 73 (80.2%) | 78 (89.7%) | – | – |
| Resection | 48 (18.0%) | 21 (23.6%) | 18 (19.8%) | 9 (10.3%) | – | – |
| Use of absolute ethanol | – | – | – | – | χ²=0.780 | 0.677 |
| No | 156 (58.4%) | 49 (55.1%) | 56 (61.5%) | 51 (58.6%) | – | – |
| Yes | 111 (41.6%) | 40 (44.9%) | 35 (38.5%) | 36 (41.4%) | – | – |
| Cardiovascular disease | – | – | – | – | χ²=2.116 | 0.347 |
| No | 264 (98.9%) | 88 (98.9%) | 91 (100.0%) | 85 (97.7%) | – | – |
| Yes | 3 (1.1%) | 1 (1.1%) | 0 (0.0%) | 2 (2.3%) | – | – |
| Family history of thrombosis or bleeding disorders | – | – | – | – | χ²=0.201 | 0.904 |
| No | 253 (94.8%) | 84 (94.4%) | 87 (95.6%) | 82 (94.3%) | – | – |
| Yes | 14 (5.2%) | 5 (5.6%) | 4 (4.4%) | 5 (5.7%) | – | – |
| Puig classification | – | – | – | – | χ²=9.936 | 0.127 |
| I | 48 (18.0%) | 21 (23.6%) | 18 (19.8%) | 9 (10.3%) | – | – |
| II | 32 (12.0%) | 13 (14.6%) | 10 (11.0%) | 9 (10.3%) | – | – |
| III | 134 (50.2%) | 43 (48.3%) | 46 (50.5%) | 45 (51.7%) | – | – |
| IV | 53 (19.9%) | 12 (13.5%) | 17 (18.7%) | 24 (27.6%) | – | – |
| Anticoagulation | – | – | – | – | χ²=2.650 | 0.266 |
| No | 150 (56.2%) | 52 (58.4%) | 45 (49.5%) | 53 (60.9%) | – | – |
| Yes | 117 (43.8%) | 37 (41.6%) | 46 (50.5%) | 34 (39.1%) | – | – |
Total sample size: 267. Continuous variables: mean ± SD (normally distributed, compared by t-test/ANOVA) or median (IQR) (non-normally distributed, compared by Mann–Whitney U/Kruskal–Wallis test); normality assessed by the Shapiro–Wilk test; Categorical variables: n (%), compared by chi-square test. P-values are from two-sided tests; P-values are from two-sided tests.LIC:localized intravascular coagulation.
3. Univariate and multivariable logistic regression analyses
In univariate logistic regression analysis, each 1-unit increase in ΔD-dimer was significantly associated with an elevated risk of LIC (odds ratio [OR] = 2.463, 95% confidence interval [CI]: 1.954–3.105, p < 0.001). Beyond ΔD-dimer, other variables exhibiting significant associations with LIC included: number of lesions (multiple vs. solitary: OR = 6.131, 95% CI: 3.220–11.672, p < 0.001); lesion volume (moderate vs. small: OR = 2.095, 95% CI: 1.113–3.943, p = 0.022; large vs. small: OR = 6.039, 95% CI: 2.817–12.949, p < 0.001); tissue layer involvement (subcutaneous vs. muscular: OR = 0.411, 95% CI: 0.196–0.862, p = 0.019); surgical approach (surgical resection vs. sclerotherapy: OR = 0.452, 95% CI: 0.208–0.983, p = 0.045); Puig classification (type 2 vs. type 3: OR = 3.113, 95% CI: 1.126–8.607, p = 0.029; type 4 vs. type 3: OR = 3.273, 95% CI: 1.085–9.872, p = 0.035); and anticoagulation use (yes vs. no: OR = 0.466, 95% CI: 0.270–0.805, p = 0.006). Conversely, sex, age, length of hospital stay, number of surgeries, absolute ethanol use, cardiovascular disease, and family history of thrombosis or bleeding disorders showed no statistically significant associations (Table 2). Preoperative swelling, preoperative pain, and the Social Insurance payment category exhibited complete or quasi-complete separation in standard maximum-likelihood logistic regression; their Firth-corrected estimates are reported in Supplementary Table S1.
Table 2.
Univariate regression analysis.
| Variable | OR | 95% CI | P value |
|---|---|---|---|
| ΔD-dimer | 2.463 | (1.954, 3.105) | <0.001 |
| Sex | 0.960 | (0.571, 1.612) | 0.876 |
| Age | 1.006 | (0.983, 1.030) | 0.628 |
| Lesion location | – | – | – |
| Number of lesions | 6.131 | (3.220, 11.672) | <0.001 |
| Involved tissue layers | 0.411 | (0.196, 0.862) | 0.019 |
| Lesion volume | – | – | – |
| Small | 1.00 | – | – |
| Moderate | 2.095 | (1.113, 3.943) | 0.022 |
| Large | 6.039 | (2.817, 12.949) | <0.001 |
| Length of hospital stay | 1.06 | (0.92, 1.23) | 0.428 |
| Payment method | – | – | – |
| Employee Medical Insurance | 1.00 | – | – |
| Rural Medical Insurance | 0.300 | (0.028, 3.250) | 0.322 |
| Resident Medical Insurance | 0.865 | (0.280, 2.672) | 0.801 |
| Self-pay | 0.450 | (0.085, 2.395) | 0.349 |
| Number of surgical procedures | 1.182 | (0.201, 6.932) | 0.853 |
| Surgical approach | 0.452 | (0.208, 0.983) | 0.045 |
| Use of absolute ethanol | 1.379 | (0.817, 2.326) | 0.229 |
| Cardiovascular disease | 4.519 | (0.404, 50.546) | 0.221 |
| Family history of thrombosis or bleeding disorders | 0.881 | (0.268, 2.895) | 0.835 |
| Puig classification | – | – | – |
| I | 1.00 | – | – |
| II | 3.113 | (1.126, 8.607) | 0.029 |
| III | 1.246 | (0.376, 4.130) | 0.719 |
| IV | 3.273 | (1.085, 9.872) | 0.035 |
| Anticoagulation | 0.466 | (0.270, 0.805) | 0.006 |
OR: Odds Ratio; CI: 95% Confidence Interval; ΔD-dimer: the change between the peak postoperative D-dimer level within 24 h after surgery and the preoperative baseline value.
For the multivariable analysis, a parsimonious model was constructed based on a priori clinical justification, retaining 10 covariates (the exposure ΔD-dimer plus 9 confounders: sex, age, number of lesions, lesion volume, Puig classification as ordinal, surgical approach, absolute ethanol use, and anticoagulation; EPV = 8.3). After adjustment, ΔD-dimer remained independently associated with LIC (adjusted OR = 2.51, 95% CI: 1.93–3.26, p < 0.001) (Table 3). A Firth-penalised sensitivity analysis (Supplementary Tables S2 and S3) yielded consistent estimates (ΔD-dimer OR 1.70 in the parsimonious set and 1.57 in the full-covariate set; both p < 0.001), confirming robustness across model specifications (OR range 1.57–2.51).
Table 3.
Multivariable logistic regression (parsimonious model).
| Variable | β (SE) | Adjusted OR | 95% CI | P value |
|---|---|---|---|---|
| ΔD-dimer (per 1 mg/L FEU) | 0.921 (0.133) | 2.51 | (1.93, 3.26) | <0.001 |
| Sex (female vs male) | 0.578 (0.374) | 1.78 | (0.86, 3.71) | 0.122 |
| Age (per year) | 0.014 (0.021) | 1.01 | (0.97, 1.06) | 0.502 |
| Number of lesions (multiple vs solitary) | 2.018 (0.600) | 7.52 | (2.32, 24.39) | 0.001 |
| Lesion volume—moderate vs small | 0.347 (0.614) | 1.41 | (0.42, 4.71) | 0.572 |
| Lesion volume—large vs small | 1.684 (0.811) | 5.39 | (1.10, 26.41) | 0.038 |
| Puig classification (ordinal, per higher type) | −1.047 (0.442) | 0.35 | (0.15, 0.84) | 0.018 |
| Surgical approach (resection vs sclerotherapy) | −0.784 (0.925) | 0.46 | (0.07, 2.80) | 0.397 |
| Absolute ethanol use (yes vs no) | 0.373 (0.383) | 1.45 | (0.68, 3.08) | 0.331 |
| Anticoagulation (yes vs no) | −1.036 (0.393) | 0.36 | (0.16, 0.77) | 0.008 |
10 covariates (11 parameters incl. intercept); EPV = 8.3; McFadden pseudo-R2 = 0.389; likelihood-ratio χ2 = 128.66 on 10 df, p < 0.001. Puig treated as ordinal (I = 1, II = 2, III = 3, IV = 4). OR: Odds Ratio; CI: 95% Confidence Interval; ΔD-dimer:the change between the peak postoperative D-dimer level within 24 h after surgery and the preoperative baseline value.
Nonlinear association and threshold effect analysis
A generalised additive model (GAM) with a cubic B-spline basis (df = 4) for ΔD-dimer was compared with a linear-logit model by likelihood-ratio test (LR = 11.07, 2 df, p = 0.0039), providing modest evidence of nonlinearity. The smoothed curve (Figure 2) showed a monotonically increasing relationship with a possible mid-range plateau and a steepening at high ΔD-dimer values; confidence bands were wide above ΔD-dimer ≈ 4 mg/L FEU. A two-piece segmented logistic regression identified an optimal breakpoint at K = 4.86 mg/L FEU (LR = 9.46, 1 df, p = 0.0021 vs linear model). Below the breakpoint (n = 242), the slope was positive and significant (OR = 2.07 per unit, 95% CI: 1.60–2.67, p < 0.001); the above-breakpoint segment (n = 25) was too sparse for stable estimation, and no above-breakpoint OR is reported. The threshold analysis is exploratory; no clinical recommendation is made on its basis alone.
Figure 2.
Non-linear associations between D-dimer change and LIC risk. (A) GAM-adjusted OR (reference = median ΔD-dimer 2.34 mg/L FEU) with 95% CI band; red dash-dot line indicates segmented breakpoint K = 4.86 mg/L FEU. (B) Observed LIC incidence by ΔD-dimer decile (marker size ∝ n) with GAM-predicted probability overlaid.
The parsimonious multivariable model showed good discrimination (AUC = 0.888 apparent; 0.878 optimism-corrected via 500-bootstrap; 95% CI of bootstrap AUC 0.855–0.936) and adequate calibration (Hosmer–Lemeshow χ2 = 8.52, p = 0.384). The univariate AUC of ΔD-dimer alone was 0.827 (Youden-optimal cutoff 3.23 mg/L FEU). Performance at candidate cutoffs is reported in Table 4; a cutoff of 4.86 mg/L FEU yielded high PPV (96.0%) and specificity (99.5%) but low sensitivity (28.9%). External validation in an independent cohort is required before any cutoff can be recommended for clinical use.
Table 4.
Threshold effect analysis and performance of ΔD-dimer at candidate cutoffs.
| Cutoff (mg/L FEU) | Sensitivity | Specificity | PPV | NPV | TP | FP | FN | TN |
|---|---|---|---|---|---|---|---|---|
| 3.00 | 0.675 | 0.788 | 0.589 | 0.843 | 56 | 39 | 27 | 145 |
| 3.23 (Youden-optimal) | 0.651 | 0.837 | 0.643 | 0.842 | 54 | 30 | 29 | 154 |
| 4.00 | 0.494 | 0.924 | 0.745 | 0.802 | 41 | 14 | 42 | 170 |
| 4.86 | 0.289 | 0.995 | 0.960 | 0.756 | 24 | 1 | 59 | 183 |
| 5.50 | 0.120 | 1.000 | 1.000 | 0.716 | 10 | 0 | 73 | 184 |
| 6.00 | 0.072 | 1.000 | 1.000 | 0.705 | 6 | 0 | 77 | 184 |
PPV: positive predictive value; NPV: negative predictive value. Univariate AUC = 0.827; multivariable AUC = 0.888 (optimism-corrected 0.878); Hosmer–Lemeshow p = 0.384.
Subgroup analyses
Subgroup analyses were performed using Firth-penalised logistic regression of ΔD-dimer on LIC within each stratum, with interactions tested by likelihood-ratio tests. The ΔD-dimer–LIC association was consistently positive in direction across all subgroups (all OR > 1), and statistically significant in the majority of strata (p ≤ 0.05), with a few small-sample subgroups showing a positive but non-significant association (e.g. Puig Type II, n = 32, OR = 1.53, p = 0.143), consistent with limited statistical power rather than effect attenuation. No significant interactions were observed (all P for interaction > 0.05). A borderline interaction was noted by head/neck involvement (P for interaction = 0.054; OR 1.76 with head/neck involvement vs 2.32 without). All other interactions were non-significant, including sex (p = 0.208), age (p = 0.765), number of lesions (p = 0.482), tissue layer (p = 0.565), lesion volume (p = 0.395), surgical approach (p = 0.637), ethanol use (p = 0.973), Puig classification (p = 0.397), and anticoagulation (p = 0.530) (Table 5). Lesion location was stratified by head/neck involvement (yes vs no) rather than the original 8-level nominal classification, because the head/neck-only and extremities-only strata contained zero LIC events, precluding stable estimation.
Table 5.
Subgroup analysis (firth-penalised logistic regression).
| SubgroupVariable | Samplesize | OR(95%CI) | P value | P for interaction |
|---|---|---|---|---|
| Sex | – | – | – | 0.208 |
| Male | 130 | 2.16 (1.62, 2.89) | <0.001 | – |
| Female | 137 | 1.85 (1.41, 2.43) | <0.001 | – |
| Age, years | – | – | – | 0.765 |
| ≤8 (0.01-8) | 141 | 2.09 (1.56, 2.79) | <0.001 | – |
| >8 (8–68) | 126 | 1.94 (1.48, 2.55) | <0.001 | – |
| Lesion location | – | – | – | 0.054 |
| Head/neck involvement | 144 | 1.76 (1.36, 2.28) | <0.001 | – |
| No head/neck involvement | 123 | 2.32 (1.68, 3.21) | <0.001 | – |
| Number of lesions | – | – | – | 0.482 |
| Solitary | 116 | 1.76 (1.24, 2.48) | 0.001 | – |
| Multiple | 151 | 1.97 (1.53, 2.53) | <0.001 | – |
| Involved tissue layers | – | – | – | 0.565 |
| Muscular | 211 | 2.07 (1.65, 2.60) | <0.001 | – |
| Subcutaneous | 56 | 1.69 (1.13, 2.51) | 0.010 | – |
| Lesion volume | – | – | – | 0.395 |
| Small | 104 | 1.88 (1.34, 2.63) | <0.001 | – |
| Moderate | 116 | 1.81 (1.37, 2.40) | <0.001 | – |
| Large | 47 | 1.95 (1.23, 3.09) | 0.005 | – |
| Surgical approach | – | – | – | 0.637 |
| Sclerotherapy | 219 | 2.08 (1.66, 2.61) | <0.001 | – |
| Resection | 48 | 1.71 (1.10, 2.66) | 0.018 | – |
| Use of absolute ethanol | – | – | – | 0.973 |
| No | 156 | 2.01 (1.53, 2.65) | <0.001 | – |
| Yes | 111 | 1.98 (1.48, 2.64) | <0.001 | – |
| Puig classification | – | – | – | 0.397 |
| I | 48 | 1.71 (1.10, 2.66) | 0.018 | – |
| II | 32 | 1.53 (0.87, 2.69) | 0.143 | – |
| III | 134 | 1.87 (1.44, 2.42) | <0.001 | – |
| IV | 53 | 2.03 (1.28, 3.24) | 0.003 | – |
| Anticoagulation | – | – | – | 0.530 |
| No | 150 | 2.16 (1.65, 2.83) | <0.001 | – |
| Yes | 117 | 1.83 (1.34, 2.48) | <0.001 | – |
OR: odds ratio for LIC per 1-unit increase in ΔD-dimer, estimated by Firth-penalised logistic regression within each subgroup (unadjusted). Adjusted estimates for the anticoagulation subgroup are reported in Supplementary Table S4.
Sensitivity analyses for perioperative anticoagulation confirmed the robustness of the ΔD-dimer–LIC association: stratified by anticoagulation status, the adjusted OR was 1.77 (95% CI 1.38–2.28) in non-anticoagulated patients and 1.58 (95% CI 1.18–2.13) in anticoagulated patients (P for interaction = 0.156); excluding anticoagulated patients entirely yielded a consistent adjusted OR of 1.77 (95% CI 1.38–2.28)(Supplementary Table S4).
Discussion
This retrospective cohort study investigated the association between perioperative ΔD-dimer changes and the risk of postoperative LIC in patients with venous malformations. Among 267 patients, each one-unit increase in ΔD-dimer was significantly associated with a higher risk of postoperative LIC (adjusted odds ratio [OR] = 2.51; 95% confidence interval [CI]: 1.93–3.26; p < 0.001). This association was consistent across all subgroups, including those stratified by sex, age, number of lesions, tissue layer involvement, lesion volume, lesion location, surgical approach, and anticoagulation status, with no significant interactions observed (all P for interaction > 0.05); a borderline interaction by head/neck involvement (p = 0.054) was noted and warrants confirmation in independent cohorts. These findings suggest that ΔD-dimer is independently associated with LIC and may serve as a perioperative risk marker.
Our findings are consistent with prior literature concerning coagulopathy associated with venous malformations, providing further corroboration and expanding on existing knowledge. For instance, Hung et al. [5] reported that approximately 33.3% of children with venous malformations exhibited LIC, wherein elevated D-dimer levels were significantly associated with larger lesion volume, multifocality, and the presence of phleboliths. Aronniemi et al. [6] further confirmed the commonality of LIC in pediatric venous malformations (with an elevated D-dimer rate of 39%) and its close link to deep, diffuse, or multifocal lesions. A multicenter observational study by Blaise et al. [16] underscored the imperative for systematic evaluation of coagulation profiles in all patients with venous malformations, irrespective of lesion size or symptomatology, given the frequent underestimation of LIC risk. Notably, Leung et al. [11] prospectively observed that dynamic changes in D-dimer levels following alcohol sclerotherapy could predict treatment response and long-term recurrence, thereby highlighting the clinical utility of perioperative D-dimer monitoring. Similarly, in other hypercoagulable states such as COVID-19 and sepsis, persistently rising or fluctuating D-dimer levels have been associated with adverse outcomes [17–19]. Mechanistically, the pathogenesis of LIC in venous malformations is driven by sluggish blood flow, aberrant endothelial activation, and a local imbalance between coagulation and fibrinolysis [4,20]. As the end product of fibrinolysis, elevated D-dimer directly reflects ongoing local coagulation activation and enhanced fibrin degradation, thereby serving as a reliable biomarker for this complex process.
The clinical utility of this study lies in providing a quantifiable risk marker for perioperative coagulation monitoring in venous malformations. While current guidelines recommend preoperative assessment of D-dimer and fibrinogen in extensive or deep venous malformations [16], they do not emphasize dynamic changes. Our data demonstrate that ΔD-dimer—rather than a single measurement—is a stronger predictor of LIC, indicating that routine monitoring of D-dimer levels both preoperatively and in the early postoperative period may facilitate the timely identification of high-risk patients. Compared with existing evidence, our study contributes by quantifying the dose–response relationship between ΔD-dimer and LIC in a relatively large cohort (n = 267) and validating its consistency across multiple surgical modalities (sclerotherapy or resection). From a public health perspective, our findings support ΔD-dimer as a risk marker for perioperative coagulation monitoring in patients with venous malformations. Recent studies have indicated that DOACs can improve coagulation parameters and alleviate pain in this population [8,21]. Although our observational data cannot establish causality, the observed association is biologically plausible and clinically actionable. Therefore, future research should prioritize prospective interventional trials to determine whether individualized anticoagulation guided by ΔD-dimer reduces LIC-related complications such as bleeding or thrombotic events.
Methodologically, this study has three principal strengths. First, its retrospective cohort design defined the exposure (ΔD-dimer) and outcome (postoperative LIC), providing a temporal ordering consistent with—but not sufficient to establish—causal inference [22]. As an observational study, residual confounding and reverse causality cannot be entirely ruled out, and the findings should be interpreted as associative rather than causal. Second, data quality was high: all venous malformation diagnoses were confirmed by imaging; LIC was determined using standardized laboratory thresholds; and comprehensive covariate information—encompassing lesion characteristics and surgical details—was complete, minimizing potential information bias. Third, tthe statistical analysis used a parsimonious multivariable model (10 covariates; EPV = 8.3) as the primary analysis, with Firth penalised regression as sensitivity analysis. The ΔD-dimer–LIC association was robust across all model specifications (OR 1.57–2.51, all p < 0.001). Subgroup analyses (Table 5) showed a consistently positive association across all strata (all OR > 1) with no significant interactions (all p > 0.05). Threshold effect analysis (Table 4), prespecified as exploratory, suggested modest nonlinearity (GAM LR = 11.07, p = 0.004) with a possible breakpoint near 4.86 mg/L FEU, though the above-breakpoint segment was too sparse (n = 25) for stable estimation. Model discrimination (AUC = 0.888; optimism-corrected 0.878) and calibration (Hosmer–Lemeshow p = 0.384) were adequate, with internal validation via bootstrap resampling.
Nevertheless, several limitations warrant acknowledgment. First, the exclusion of 123 patients (26.3%) due to incomplete perioperative D-dimer measurements is a plausible source of selection bias, as patients undergoing repeated coagulation testing may have larger or more symptomatic lesions, potentially biasing the effect estimate away from the null; we were unable to perform a formal included-versus-excluded comparison because key lesion-severity variables were not abstracted for excluded patients. Future prospective studies with protocol-mandated perioperative D-dimer measurement for all VM patients would eliminate this bias at the design stage. Second, inherent to its retrospective cohort design, this study could not fully account for unmeasured confounders—such as precise operative duration, intraoperative blood loss, or detailed postoperative mobility—nor can it definitively establish causality [23]. Despite adjusting for known covariates, the potential for residual confounding therefore persists. Third, while D-dimer testing is routinely integrated into clinical practice, the absence of standardization across assay platforms may impact the comparability of absolute values [10]. Moreover, our study did not concurrently assess other critical LIC-related markers, such as fibrinogen or platelet count, thus precluding a comprehensive evaluation of the overall coagulation profile. Future investigations should endeavor to address these limitations through prospective designs, multicenter collaboration, and standardized laboratory protocols.
Supplementary Material
Acknowledgments
Not Applicable.
Funding Statement
This work was supported by the grants from Fujian Provincial Health Technology Project [No. 2025GGA064]; the Joint Funds for the Innovation of Science and Technology,Fujian Province [No. 2025Y9639].
Ethical approval
This study received approval from the Institutional Review Board of Fujian Maternity and Child Health Hospital (Approval No. 2026KY108-01). The approval was granted retrospectively for the analysis of pre-existing anonymized clinical data; all data were collected as part of routine clinical care prior to study approval, and no prospective intervention was performed. The study was conducted in strict accordance with the principles of the Declaration of Helsinki. Given the study’s retrospective design, all patient data were de-identified and irreversibly anonymized, thereby precluding individual re-identification. Consequently, as no interventions were made to clinical care pathways, the ethics committee waived the requirement for informed consent. Throughout the study, data confidentiality and patient privacy were maintained, with access to the anonymized database restricted exclusively to authorized research personnel.
Disclosure statement
No potential conflict of interest was reported by the author(s).
Informed consent
Not applicable. As this study was a retrospective analysis of de-identified data and the requirement for informed consent was waived by the institutional ethics committee, specific consent for publication was not required.
Data availability statement
The datasets generated and/or analyzed during the current study are not publicly available due to patient privacy restrictions and institutional data policies. However, de-identified data may be made available from the corresponding author upon reasonable request and with appropriate institutional and ethical approvals.
References
- 1.Kunimoto K, Yamamoto Y, Jinnin M.. ISSVA classification of vascular anomalies and molecular biology. Int J Mol Sci. 2022;23(4):2358. doi: 10.3390/ijms23042358. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Acord M, Srinivasan A.. Management of venous malformations. Semin Intervent Radiol. 2021;38(2):215–225. doi: 10.1055/s-0041-1729743. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Han YY, Sun LM, Yuan SM.. Localized intravascular coagulation in venous malformations: a system review. Phlebology. 2021;36(1):38–42. doi: 10.1177/0268355520946211. [DOI] [PubMed] [Google Scholar]
- 4.Zhuo KY, Russell S, Wargon O, et al. Localised intravascular coagulation complicating venous malformations in children: associations and therapeutic options. J Paediatr Child Health. 2017;53(8):737–741. doi: 10.1111/jpc.13461. [DOI] [PubMed] [Google Scholar]
- 5.Hung JWS, Leung MWY, Liu CSW, et al. Venous malformation and localized intravascular coagulopathy in children. Eur J Pediatr Surg. 2017;27(2):181–184. doi: 10.1055/s-0036-1582241. [DOI] [PubMed] [Google Scholar]
- 6.Aronniemi J, Långström S, Mattila KA, et al. Venous malformations and blood coagulation in children. Children (Basel). 2021;8(4):312. doi: 10.3390/children8040312. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Swerdlin RF, Briones MA, Gill AE, et al. Coagulopathy and related complications following sclerotherapy of congenital venous malformations. Pediatr Blood Cancer. 2022;69(5):e29610. doi: 10.1002/pbc.29610. [DOI] [PubMed] [Google Scholar]
- 8.Liu H, Hu L, Yang X, et al. Dabigatran etexilate is efficacious in consumptive coagulopathy and pain associated with venous malformations. J Vasc Surg Venous Lymphat Disord. 2023;11(2):397–403.e1. doi: 10.1016/j.jvsv.2022.09.015. [DOI] [PubMed] [Google Scholar]
- 9.Nakano TA, Zeinati C.. Venous thromboembolism in pediatric vascular anomalies. Front Pediatr. 2017;5:158. doi: 10.3389/fped.2017.00158. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Johnson ED, Schell JC, Rodgers GM.. The D-dimer assay. Am J Hematol. 2019;94(7):833–839. doi: 10.1002/ajh.25482. [DOI] [PubMed] [Google Scholar]
- 11.Leung YC-L, Leung MW-Y, Yam S-D, et al. D-dimer level correlation with treatment response in children with venous malformations. J Pediatr Surg. 2018;53(2):289–292. doi: 10.1016/j.jpedsurg.2017.11.030. [DOI] [PubMed] [Google Scholar]
- 12.Imamura H, Yasui M, Nakai N, et al. Preoperative D-dimer value and lower limb venous ultrasound for deep venous thrombosis prevents postoperative symptomatic venous thromboembolism in patients undergoing colorectal surgery: a retrospective study. J Anus Rectum Colon. 2023;7(3):159–167. doi: 10.23922/jarc.2022-071. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Tsantes AG, Loukopoulou I, Papadopoulos DV, et al. Fibrinolysis shutdown and elevated D-dimer levels have high prognostic capacity for postoperative thromboembolic complications in patients with bone tumors. J Thromb Thrombolysis. 2023;55(3):536–544. doi: 10.1007/s11239-023-02787-w. [DOI] [PubMed] [Google Scholar]
- 14.Pei F, Wei J, Yao Y, et al. Predictive value of preoperative platelet count and D-dimer levels for spinal cord injury following acute type a aortic dissection. J Cardiothorac Surg. 2024;19(1):121. doi: 10.1186/s13019-024-02597-y. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Seront E, Hermans C, Boon LM, et al. Targeted treatments for vascular malformations: current state of the art. J Thromb Haemost. 2024;22(11):2961–2975. doi: 10.1016/j.jtha.2024.07.013. [DOI] [PubMed] [Google Scholar]
- 16.Blaise S, Soya E, Nou M, et al. Localised intravascular coagulation complicating venous malformations: clinical characterisation. Phlebology. 2024;39(3):169–173. doi: 10.1177/02683555231216349. [DOI] [PubMed] [Google Scholar]
- 17.Iba T, Levy JH, Levi M, et al. Coagulopathy in COVID-19. J Thromb Haemost. 2020;18(9):2103–2109. doi: 10.1111/jth.14975. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Fei Y, Tang N, Liu H, et al. Coagulation dysfunction. Arch Pathol Lab Med. 2020;144(10):1223–1229. doi: 10.5858/arpa.2020-0324-SA. [DOI] [PubMed] [Google Scholar]
- 19.Williams B, Zou L, Pittet JF, et al. Sepsis-induced coagulopathy: a comprehensive narrative review of pathophysiology, clinical presentation, diagnosis, and management strategies. Anesth Analg. 2024;138(4):696–711. doi: 10.1213/ANE.0000000000006888. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Ricci KW, Brandão LR.. Coagulation issues in vascular anomalies. Semin Pediatr Surg. 2020;29(5):150966. doi: 10.1016/j.sempedsurg.2020.150966. [DOI] [PubMed] [Google Scholar]
- 21.Lagneaux E, Boon LM, Revencu N, et al. Direct oral anticoagulants and venous malformations: literature review and retrospective study of 29 patients. Res Pract Thromb Haemost. 2024;8(3):102400. doi: 10.1016/j.rpth.2024.102400. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Andrade C. Observational studies. Respir Care. 2022;67(11):1585–1597. doi: 10.4187/respcare.11170. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Greenland S, Mansournia MA, Joffe M.. To curb research misreporting, remove rather than redefine statistical significance. Eur J Epidemiol. 2022;37(4):327–328. doi: 10.1007/s10654-022-00931-z. [DOI] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
The datasets generated and/or analyzed during the current study are not publicly available due to patient privacy restrictions and institutional data policies. However, de-identified data may be made available from the corresponding author upon reasonable request and with appropriate institutional and ethical approvals.