Abstract
Objective: To develop a risk prediction model for low-molecular-weight heparin (LMWH) ineffectiveness in pregnant women with intracranial venous sinus thrombosis (CVST), enabling early identification of high-risk patients. Methods: A retrospective analysis was conducted on 221 pregnant or postpartum CVST patients treated with LMWH at seven Chinese hospitals between January 2010 and January 2025. Patients were divided into effective (191 cases) and ineffective (30 cases) treatment groups. Logistic regression identified predictors, which were then used to develop a risk prediction model. Results: Univariate analysis revealed significant factors associated with treatment ineffectiveness: Coronavirus Disease 2019 (COVID-19), hyperthyroidism, platelet (PLT) count, antithrombin III (AT-III), homocysteine (Hcy), low-density lipoprotein cholesterol (LDL-C), protein C, and protein S (all P < 0.1). Variables with a P-value < 0.1 were further analyzed using Least Absolute Shrinkage and Selection Operator (LASSO) regression to identify key predictors, with the lambda value (0.011) determining the final model. Multivariate analysis identified independent risk factors: COVID-19, protein S, and homocysteine. Protein S (odds ratio [OR] < 1) acted as a protective factor, while COVID-19 and homocysteine (OR > 1) were risk factors. The model’s receiver operating characteristic (ROC) curve area was 0.930 (95% confidence interval [CI]: 0.882-0.979), with sensitivity of 0.867 and specificity of 0.885. Cross-validation and bootstrap validation demonstrated robust model performance, with areas under the curve of 0.919 and 0.909, respectively. Conclusions: The developed predictive model, incorporating COVID-19, protein S, and Hcy, effectively assesses the risk of LMWH ineffectiveness in pregnant women with CVST, supporting clinical decision-making.
Keywords: Pregnancy-associated intracranial venous sinus thrombosis, low-molecular-weight heparin therapy, treatment ineffectiveness, logistic regression analysis, predictive modeling
Introduction
Cerebral venous sinus thrombosis (CVST) is a rare neurovascular disorder characterized by impaired cerebral venous outflow, which results in intracranial hypertension or focal neurological deficits [1-4]. In contrast to arterial thrombosis, CVST constitutes only 0.5%-1.0% of all cerebrovascular conditions [5], with an estimated prevalence of approximately 5 cases per million individuals [6]. Although CVST can occur at any age [7], it is more frequently identified in both younger and older populations, with a notably higher incidence in women, who experience a 3-4 times greater risk compared to men [8,9]. Among women, key risk factors include pregnancy, childbirth, the use of oral contraceptives, and hormone replacement therapy [10,11]. CVST can progress rapidly, resulting in severe disability or death if not promptly addressed, with mortality rates reaching as high as 30% [6,12,13]. However, early diagnosis and treatment have shown to reduce mortality to between 5-15% [8]. The detection rate of CVST in pregnant and postpartum women has significantly improved in recent years, owing to advancements in imaging technology and heightened clinical awareness.
Anticoagulation therapy remains the primary treatment for CVST. Clinical guidelines recommend initiating anticoagulation as early as possible in CVST patients, provided there are no contraindications [14-16]. Anticoagulation facilitates thrombolysis, prevents thrombus propagation, reduces mortality, and improves overall prognosis. The administration and monitoring of anticoagulant therapy are well-established and straightforward, promoting the establishment of collateral circulation, enhancing blood reflux compensation, and facilitating fibrin autolysis to reopen obstructed sinuses. However, dynamic monitoring of symptoms, such as headache or changes in consciousness, is crucial, as therapeutic responses can vary, particularly in complex cases. In clinical practice, low-molecular-weight heparin (LMWH) is frequently used for CVST treatment in pregnant and postpartum women because it does not cross the placental barrier, ensuring fetal safety. Despite its widespread use, LMWH is ineffective in approximately 10% of these patients [17]. This study aimed to assess clinical data on the use of LMWH in pregnant and postpartum women with CVST and to develop a predictive model for identifying low-risk and high-risk populations to promote timely intervention and prevent adverse outcomes.
Data and methods
Clinical information
A retrospective analysis was conducted on 221 pregnant or postpartum (within six weeks) patients diagnosed with CVST and treated with LMWH at seven Chinese hospitals between 2010 and 2025. The treatment outcomes were classified into two groups: effective (191 cases) and ineffective (30 cases).
Inclusion criteria: (1) The patient met the diagnostic criteria for CVST during pregnancy or within six weeks postpartum; (2) No interventions affecting coagulation function were administered during treatment; (3) LMWH therapy was initiated following diagnosis, with mandatory follow-up for at least six months; (4) Comprehensive and accurate clinical data, including symptoms, examinations, and diagnostic evaluations, were available; (5) The participant provided informed consent by signing the consent form for clinical research and follow-up.
Exclusion criteria: (1) Patients exhibited varying degrees of elevated intracranial pressure upon evaluation; (2) Critical conditions such as consciousness disturbances, unstable vital signs, brain herniation, or decerebrate rigidity; (3) Severe coagulation dysfunction; (4) Presence of intracranial hemorrhage or cerebral infarction; (5) Severe comorbidities affecting other vital organs; (6) Prior use of interventions impacting study outcomes; (7) Incomplete clinical or follow-up data.
To ensure data accuracy, especially given the potential for asymptomatic COVID-19 infections and the retrospective nature of the study spanning both pre- and post-pandemic periods, all follow-up data were carefully verified through patient interviews and hospital records. Additionally, for cases diagnosed with COVID-19 during the study, the timing and severity of infection were documented, and patients were assessed for symptoms and clinical outcomes that might have impacted CVST treatment. This helped mitigate any potential bias from unrecognized or asymptomatic COVID-19 infections. The study also ensured that all data recorded prior to 2019 (before the official identification of COVID-19) were not affected by the pandemic, maintaining the integrity of the results from those years. The research followed the 2013 Declaration of Helsinki ethical guidelines, was approved by the Ethics Committee of Hefei Maternal and Child Health Hospital (Approval No. YYLL20240130-YNXM-LL-01-2.1), and obtained voluntary informed consent from patients and their families after fully explaining the study details.
Candidate predictors
Clinical, laboratory, and imaging parameters were systematically selected as candidate variables for analysis. These included age, body mass index (BMI), gravidity and parity counts, history of adverse pregnancy outcomes, family medical history, smoking and alcohol consumption habits, oral contraceptive use, immunosuppressant therapy, craniocerebral trauma or surgical history, intracranial vascular malformations, venous sinus compression, head or facial infections, history of COVID-19 infection, trace element deficiencies, conception method, pre-pregnancy comorbidities (e.g., hypertension, diabetes, thyroid disorders, anemia, and autoimmune diseases), pregnancy-related complications (including hypertensive disorders, gestational diabetes, thyroid dysfunction, anemia, thrombocytopenia, and hyperemesis gravidarum), infection history during pregnancy or postpartum, presenting symptoms (such as headache, nausea/vomiting, diplopia, epilepsy, and neurological deficits), pregnancy status (early, mid, or late pregnancy or postpartum), and diagnostic laboratory parameters. Laboratory parameters included: complete blood count (red blood cells [RBC], white blood cells [WBC], platelets [PLT], hemoglobin [HGB], absolute neutrophil count [ANC], and neutrophil percentage [NE%]), coagulation indices (prothrombin time [PT], activated partial thromboplastin time [APTT], thrombin time [TT], fibrinogen [FIB], fibrin degradation products [FDP], D-dimer, and antithrombin III [AT-III]), thromboelastography metrics (reaction time [R], clot formation time [K], angle, maximum amplitude [MA], lysis at 30 minutes [LY30], early clot lysis [EPL], and clot index [CI]), random blood glucose, serum homocysteine levels, lipid profile (total cholesterol [TC], triglycerides [TG], high-density lipoprotein cholesterol [HDL-C], low-density lipoprotein cholesterol [LDL-C]), protein C and protein S activity, presence of lupus anticoagulant, methylenetetrahydrofolate reductase (MTHFR) polymorphism, fundoscopic findings, venous sinus thrombosis characteristics (including location, size, phase, number of affected sinuses, vascular stenosis, and collateral circulation), and pregnancy outcomes. Continuous variables, such as Hcy and protein S levels, were standardized or categorized based on clinically relevant thresholds to promote their inclusion in statistical models. For Hcy, a cutoff value of 15 µmol/L was used to categorize patients into “elevated” and “normal” levels, as previous studies have shown this threshold to be associated with increased thrombotic risk. Protein S activity was categorized into low (< 60%) and normal (≥ 60%) levels based on established clinical guidelines, which associate reduced protein S activity with an increased risk of venous thromboembolism. These thresholds were selected to align with existing clinical practices and to enhance the interpretability of the model results.
Therapeutic approach
According to Chinese guidelines for the management of CVST, patients were administered subcutaneous enoxaparin sodium at a dosage of 100 IU/kg every 12 h. Platelet count and coagulation parameters were monitored every three days, and once heparinization was stable, monitoring frequency was reduced to weekly until symptom resolution and imaging improvement are observed. The international normalized ratio (INR) was maintained between 2 and 3 throughout the duration of treatment. In cases of unsatisfactory outcomes, worsening symptoms, or the emergence of complications such as intracranial hemorrhage, elevated intracranial pressure, or brain herniation, anticoagulation therapy should be discontinued, and alternative interventions, such as endovascular procedures or surgical interventions, should be considered.
Observation index
Follow-up data were gathered through telephone consultations and outpatient visits. Patients were scheduled for reexaminations at 3, 6, and 12 months post-heparinization to assess symptoms and perform imaging evaluations, with the modified Rankin Scale (mRS) score consistently administered by the same physician. For patients unable to attend in-person appointments, the mRS score was assessed via telephone interviews. An mRS score of 0 indicates full recovery, while a score of 1 signifies minimal residual effects without significant functional limitations. Scores ranging from 0 to 1 reflect clinical improvement, whereas scores above 1 indicate the presence of persistent functional deficits. Improvement is also indicated by symptom alleviation or resolution, while stagnation or persistence of symptoms suggests no improvement. Imaging evidence of thrombus dissolution or reduction is considered an improvement; otherwise, the condition is classified as unchanged. Treatment effectiveness was determined based on the improvement in at least one of the three criteria (mRS score, symptoms, and imaging) relative to pre-treatment status. Patients were subsequently categorized into two groups: those with ineffective treatment (requiring alternative interventions due to deterioration) and those with effective treatment. Follow-up data collection was adapted to account for COVID-19-related disruptions during the study period. Telephone consultations and outpatient visits were conducted with particular attention to any COVID-19-related symptoms in patients. In cases where patients were unable to attend in person due to COVID-19-related restrictions or asymptomatic infection, virtual consultations were arranged, and patient-reported outcomes were cross-verified with hospital data. This approach helped ensure the reliability of follow-up data and reduced the likelihood of missing information, particularly in light of the challenges posed by the pandemic.
Statistical analysis
Univariate logistic regression was used to identify potential predictors of LMWH ineffectiveness, with a significance threshold of P ≤ 0.1. Variables that showed potential relevance in the univariate analysis were included in the subsequent LASSO (Least Absolute Shrinkage and Selection Operator) logistic regression model, which was employed to select key features with non-zero coefficients. The LASSO method helps to enhance model stability by performing variable selection and regularization to prevent overfitting. To optimize model parameters, 10-fold cross-validation was used, and the final coefficients were determined based on the lambda value that minimized the standard error deviation. Variables with non-zero coefficients from the LASSO regression were retained for multivariate logistic regression, which was conducted using stepwise (bidirectional) selection with a significance threshold of P < 0.05. This approach enables the identification of the most significant predictors while accounting for potential confounders. A nomogram-based prediction model was constructed from the selected predictors. The model was validated using 1,000 bootstrap resamples to assess its internal validity and to minimize overfitting. The calibration accuracy of the model was evaluated by plotting a calibration curve and performing the Hosmer-Lemeshow test, which tests the goodness of fit between observed and predicted probabilities. The model’s discriminatory performance was assessed using the receiver operating characteristic (ROC) curve, with metrics including area under the curve (ROC curve area), sensitivity, and specificity. Clinical utility was evaluated through Decision Curve Analysis (DCA), which quantifies the net benefit of the prediction model in clinical decision-making. Finally, the model’s reliability and generalizability were further confirmed using ten-fold cross-validation and Bootstrap sampling, with corresponding ROC curves plotted for each validation method. To mitigate the risk of model overfitting due to imbalanced sample sizes between the “effective” (191 cases) and “ineffective” (30 cases) treatment groups, several methods were applied to address class imbalance. These methods included stratified sampling during model training, synthetic minority oversampling (SMOTE), or penalty adjustments in the logistic regression models. Additionally, cross-validation techniques such as 10-fold cross-validation and bootstrap sampling were used to ensure that the model’s performance was not overly influenced by the imbalance.
Results
Participants’ clinical characteristics
A total of 221 cases were enrolled, with 30 categorized as ineffective, resulting in an incidence rate of 13.57% (30/221). Additional details are presented in Table 1. The univariate screening analysis, using a significance threshold of P ≤ 0.1, did not adjust for multiple comparisons, which could lead to inflated Type I error rates. To correct for this, a Bonferroni correction was applied given the 54 variables tested. Regarding model validation, the minimal difference between the 10-fold cross-validated ROC curve area (0.919) and the original model ROC curve area (0.930) suggests a potential issue with data leakage, likely caused by performing feature selection prior to crossvalidation.
Table 1.
Participants’ baseline characteristics
| Variables | Total (n = 221) | Valid Group (n = 191) | Invalid Group (n = 30) | P |
|---|---|---|---|---|
| Age, Median (Q1, Q3) | 29.00 (25.00, 33.50) | 29.00 (25.50, 33.00) | 29.00 (22.50, 32.75) | 0.703 |
| BMI, Median (Q1, Q3) | 24.52 (21.88, 26.72) | 24.62 (21.55, 26.80) | 24.06 (22.66, 26.61) | 0.814 |
| Gravida, Median (Q1, Q3) | 2.00 (1.00, 3.00) | 2.00 (1.00, 3.00) | 2.00 (1.00, 3.00) | 0.892 |
| Parity, Median (Q1, Q3) | 1.00 (0.00, 1.00) | 1.00 (0.00, 1.00) | 0.00 (0.00, 1.00) | 0.256 |
| Adverse pregnancy history, n (%) | 1.000 | |||
| No | 201 (91.0) | 173 (90.6) | 28 (93.3) | |
| Yes | 20 (9.0) | 18 (9.4) | 2 (6.7) | |
| Family history, n (%) | 0.751 | |||
| No | 197 (89.1) | 171 (89.5) | 26 (86.7) | |
| Yes | 24 (10.9) | 20 (10.5) | 4 (13.3) | |
| History of smoking, n (%) | 0.505 | |||
| No | 182 (82.4) | 156 (81.7) | 26 (86.7) | |
| Yes | 39 (17.6) | 35 (18.3) | 4 (13.3) | |
| Drinking history, n (%) | 0.766 | |||
| No | 174 (78.7) | 151 (79.1) | 23 (76.7) | |
| Yes | 47 (21.3) | 40 (20.9) | 7 (23.3) | |
| History of taking oral contraceptives, n (%) | 0.571 | |||
| No | 164 (74.2) | 143 (74.9) | 21 (70.0) | |
| Yes | 57 (25.8) | 48 (25.1) | 9 (30.0) | |
| Immunosuppressants, n (%) | 1.000 | |||
| No | 198 (89.6) | 171 (89.5) | 27 (90.0) | |
| Yes | 23 (10.4) | 20 (10.5) | 3 (10.0) | |
| History of craniocerebral trauma, n (%) | 0.480 | |||
| No | 204 (92.3) | 175 (91.6) | 29 (96.7) | |
| Yes | 17 (7.7) | 16 (8.4) | 1 (3.3) | |
| History of craniocerebral surgery, n (%) | 1.000 | |||
| No | 204 (92.3) | 176 (92.1) | 28 (93.3) | |
| Yes | 17 (7.7) | 15 (7.9) | 2 (6.7) | |
| Intracranial vascular malformation, n (%) | 0.769 | |||
| No | 194 (87.8) | 168 (88.0) | 26 (86.7) | |
| Yes | 27 (12.2) | 23 (12.0) | 4 (13.3) | |
| Compression of intracranial venous sinuses, n (%) | 0.703 | |||
| No | 205 (92.8) | 176 (92.1) | 29 (96.7) | |
| Yes | 16 (7.2) | 15 (7.9) | 1 (3.3) | |
| History of head and facial infections, n (%) | 0.699 | |||
| No | 206 (93.2) | 177 (92.7) | 29 (96.7) | |
| Yes | 15 (6.8) | 14 (7.3) | 1 (3.3) | |
| COVID-19, n (%) | < 0.001 | |||
| No | 190 (86.0) | 179 (93.7) | 11 (36.7) | |
| Yes | 31 (14.0) | 12 (6.3) | 19 (63.3) | |
| Trace Element Deficiency, n (%) | 0.691 | |||
| No | 208 (94.1) | 180 (94.2) | 28 (93.3) | |
| Yes | 13 (5.9) | 11 (5.8) | 2 (6.7) | |
| Conception mode, n (%) | 0.480 | |||
| Assisted | 48 (21.7) | 40 (20.9) | 8 (26.7) | |
| Spontaneous | 173 (78.3) | 151 (79.1) | 22 (73.3) | |
| Hypertension, n (%) | 0.569 | |||
| No | 171 (77.4) | 149 (78.0) | 22 (73.3) | |
| Yes | 50 (22.6) | 42 (22.0) | 8 (26.7) | |
| Diabetes, n (%) | 0.279 | |||
| No | 203 (91.9) | 177 (92.7) | 26 (86.7) | |
| Yes | 18 (8.1) | 14 (7.3) | 4 (13.3) | |
| Hypothyroidism, n (%) | 0.751 | |||
| No | 197 (89.1) | 171 (89.5) | 26 (86.7) | |
| Yes | 24 (10.9) | 20 (10.5) | 4 (13.3) | |
| Hyperthyroidism, n (%) | 0.075 | |||
| No | 193 (87.3) | 170 (89.0) | 23 (76.7) | |
| Yes | 28(12.7) | 21 (11.0) | 7 (23.3) | |
| Anemia, n (%) | 0.362 | |||
| No | 178 (80.5) | 152 (79.6) | 26 (86.7) | |
| Yes | 43 (19.5) | 39 (20.4) | 4 (13.3) | |
| Autoimmune disease, n (%) | 0.462 | |||
| No | 205 (92.8) | 178 (93.2) | 27 (90.0) | |
| Yes | 16 (7.2) | 13 (6.8) | 3 (10.0) | |
| Uterine myoma, n (%) | 0.269 | |||
| No | 188 (85.1) | 160 (83.8) | 28 (93.3) | |
| Yes | 33 (14.9) | 31 (16.2) | 2 (6.7) | |
| HDP, n (%) | 0.112 | |||
| No | 184 (83.3) | 156 (81.7) | 28 (93.3) | |
| Yes | 37 (16.7) | 35 (18.3) | 2 (6.7) | |
| GDM, n (%) | 1.000 | |||
| No | 190 (86.0) | 164 (85.9) | 26 (86.7) | |
| Yes | 31 (14.0) | 27 (14.1) | 4 (13.3) | |
| Abnormal thyroid hormones during pregnancy, n (%) | 0.740 | |||
| No | 201 (91.0) | 174 (91.1) | 27 (90.0) | |
| Yes | 20 (9.0) | 17 (8.9) | 3 (10.0) | |
| Anemia of pregnancy, n (%) | 0.726 | |||
| No | 179 (81.0) | 154 (80.6) | 25 (83.3) | |
| Yes | 42 (19.0) | 37 (19.4) | 5 (16.7) | |
| Thrombocytopenia in pregnancy, n (%) | 0.553 | |||
| No | 193 (87.3) | 168 (88.0) | 25 (83.3) | |
| Yes | 28 (12.7) | 23 (12.0) | 5 (16.7) | |
| Hyperemesis Gravidarum, n (%) | 0.469 | |||
| No | 180 (81.4) | 157 (82.2) | 23 (76.7) | |
| Yes | 41 (18.6) | 34 (17.8) | 7 (23.3) | |
| Infection during pregnancy or thepuerperium, n (%) | 0.423 | |||
| No | 187 (84.6) | 163 (85.3) | 24 (80.0) | |
| Yes | 34 (15.4) | 28 (14.7) | 6 (20.0) | |
| Headache, n (%) | 0.470 | |||
| No | 173 (78.3) | 148 (77.5) | 25 (83.3) | |
| Yes | 48 (21.7) | 43 (22.5) | 5 (16.7) | |
| Nausea or Vomiting, n (%) | 0.466 | |||
| No | 181 (81.9) | 155 (81.2) | 26 (86.7) | |
| Yes | 40 (18.1) | 36 (18.8) | 4 (13.3) | |
| Diplopia, n (%) | 0.571 | |||
| No | 191 (86.4) | 166 (86.9) | 25 (83.3) | |
| Yes | 30 (13.6) | 25 (13.1) | 5 (16.7) | |
| Epilepsy, n (%) | 0.436 | |||
| No | 206 (93.2) | 179 (93.7) | 27 (90.0) | |
| Yes | 15 (6.8) | 12 (6.3) | 3 (10.0) | |
| Focal neurological deficits, n (%) | 0.746 | |||
| No | 199 (90.0) | 171 (89.5) | 28 (93.3) | |
| Yes | 22 (10.0) | 20 (10.5) | 2 (6.7) | |
| Pregnancy status, n (%) | 0.865 | |||
| Early pregnancy | 23 (10.4) | 20 (10.5) | 3 (10.0) | |
| Late pregnancy | 45 (20.4) | 39 (20.4) | 6 (20.0) | |
| Mid-pregnancy | 18 (8.1) | 17 (8.9) | 1 (3.3) | |
| Puerperium | 135 (61.1) | 115 (60.2) | 20 (66.7) | |
| RBC, Median (Q1, Q3) | 3.66 (3.27, 4.36) | 3.75 (3.35, 4.38) | 3.77 (3.33, 4.21) | 0.780 |
| WBC, Median (Q1, Q3) | 9.60 (7.65, 11.36) | 9.84 (7.90, 11.47) | 8.53 (7.23, 11.08) | 0.101 |
| PLT, Median (Q1, Q3) | 224.00 (131.00, 231.00) | 225.00 (132.00, 232.00) | 180.50 (131.50, 230.50) | 0.098 |
| HGB, Median (Q1, Q3) | 99.00 (91.50, 123.00) | 101.00 (92.00, 123.00) | 96.50 (91.25, 122.50) | 0.493 |
| ANC, Median (Q1, Q3) | 3.21 (2.37, 4.72) | 3.16 (2.38, 4.66) | 2.94 (2.40, 4.68) | 0.740 |
| NE%, Mean ± SD | 74.59 ± 4.77 | 74.65 (69.91, 77.82) | 73.47 (69.23, 78.64) | 0.620 |
| PT, Median (Q1, Q3) | 13.24 (12.33, 14.00) | 13.24 (12.33, 13.95) | 12.99 (12.63, 13.94) | 0.691 |
| APTT, Median (Q1, Q3) | 27.63 (25.30, 29.33) | 27.45 (24.62, 29.39) | 26.90 (25.56, 29.24) | 0.874 |
| TT, Mean ± SD | 13.24 ± 1.49 | 13.31 (12.36, 14.57) | 12.98 (12.05, 13.83) | 0.338 |
| FIB, Median (Q1, Q3) | 2.70 (2.50, 2.90) | 2.70 (2.50, 3.05) | 2.80 (2.42, 3.38) | 0.710 |
| FDP, Median (Q1, Q3) | 3.69 (2.71, 4.56) | 3.79 (2.62, 4.61) | 3.62 (3.03, 4.24) | 0.831 |
| D-dimer, Median (Q1, Q3) | 3.21 (2.37, 4.72) | 3.28 (2.43, 4.69) | 2.81 (2.35, 4.75) | 0.695 |
| AT-III, Median (Q1, Q3) | 73.81 (68.82, 77.81) | 73.57 (68.91, 77.59) | 73.11 (39.47, 77.04) | 0.121 |
| R, Median (Q1, Q3) | 5.41 (4.77, 6.25) | 5.39 (4.76, 6.28) | 5.63 (4.76, 6.22) | 0.873 |
| K, Median (Q1, Q3) | 2.80 (2.45, 3.20) | 2.80 (2.40, 3.20) | 2.80 (2.42, 3.20) | 0.890 |
| Angle, Median (Q1, Q3) | 61.30 (55.90, 66.20) | 62.00 (56.15, 66.35) | 61.55 (54.90, 65.35) | 0.587 |
| MA, Median (Q1, Q3) | 56.81 (52.41, 64.96) | 56.81 (52.34, 64.73) | 56.59 (53.35, 64.99) | 0.824 |
| LY30, Median (Q1, Q3) | 2.70 (2.20, 3.20) | 2.70 (2.05, 3.20) | 2.90 (2.50, 3.20) | 0.216 |
| EPL, Median (Q1, Q3) | 9.60 (7.65, 11.36) | 9.65 (7.78, 11.41) | 9.66 (7.63, 10.91) | 0.380 |
| CI, Median (Q1, Q3) | 2.80 (2.50, 3.20) | 2.80 (2.50, 3.20) | 2.80 (2.45, 3.10) | 0.737 |
| Glu, Median (Q1, Q3) | 6.10 (5.50, 6.70) | 6.10 (5.50, 6.65) | 5.80 (5.08, 6.38) | 0.096 |
| Hcy, Median (Q1, Q3) | 8.73 (5.88, 11.29) | 7.53 (5.28, 10.18) | 11.30 (9.07, 14.40) | < 0.001 |
| TC, Median (Q1, Q3) | 3.52 (2.67, 4.60) | 3.62 (2.72, 4.54) | 3.35 (2.78, 5.12) | 0.796 |
| TG, Median (Q1, Q3) | 1.00 (0.70, 1.30) | 1.00 (0.60, 1.30) | 1.05 (0.62, 1.30) | 0.745 |
| HDL-C, Median (Q1, Q3) | 1.10 (0.70, 1.40) | 1.10 (0.65, 1.40) | 0.90 (0.70, 1.30) | 0.436 |
| LDL-C, Median (Q1, Q3) | 2.67 (1.99, 3.45) | 2.47 (1.84, 3.38) | 2.83 (1.95, 4.28) | 0.166 |
| Protein C, Median (Q1, Q3) | 57.00 (38.50, 76.00) | 58.00 (40.50, 77.00) | 35.00 (33.00, 58.75) | < 0.001 |
| Protein S, Median (Q1, Q3) | 62.00 (37.00, 84.50) | 66.00 (40.50, 85.00) | 39.00 (35.00, 68.00) | 0.004 |
| Lupus anticoagulants, n (%) | 0.351 | |||
| No | 212 (95.9) | 184 (96.3) | 28 (93.3) | |
| Yes | 9 (4.1) | 7 (3.7) | 2 (6.7) | |
| MTHFR, n (%) | 0.530 | |||
| CC | 194 (87.8) | 169 (88.5) | 25 (83.3) | |
| CT | 21 (9.5) | 17 (8.9) | 4 (13.3) | |
| TT | 6 (2.7) | 5 (2.6) | 1 (3.3) | |
| Fundus examination, n (%) | 0.759 | |||
| CSC | 3 (1.4) | 2 (1.0) | 1 (3.3) | |
| Normal | 194 (87.8) | 168 (88.0) | 26 (86.7) | |
| Papilledema | 7 (3.1) | 6 (3.1) | 1 (3.3) | |
| RAS | 14 (6.3) | 12 (6.3) | 2 (6.7) | |
| RH | 3 (1.4) | 3 (1.6) | 0 (0.0) | |
| Location of venous sinus thrombosis, n (%) | 0.048 | |||
| Multiple sinuses | 77 (34.8) | 64 (33.5) | 13 (43.3) | |
| Sigmoid sinus | 18 (8.1) | 18 (9.4) | 0 (0.0) | |
| Straight sinus | 40 (18.1) | 32 (16.8) | 8 (26.7) | |
| Superior sagittal sinus | 53 (24.0) | 50 (26.2) | 3 (10.0) | |
| Transverse sinus | 33 (15.0) | 27 (14.1) | 6 (20.0) | |
| Size of venous sinus thrombosis, Median (Q1, Q3) | 5.52 (2.95, 7.37) | 5.54 (2.96, 7.45) | 5.58 (3.32, 7.46) | 0.995 |
| Nature of venous sinus thrombosis, n (%) | 1.000 | |||
| Acute stage | 205 (92.8) | 177 (92.7) | 28 (93.3) | |
| Non-acute stage | 16 (7.2) | 14 (7.3) | 2 (6.7) | |
| Number of venous sinuses, Median (Q1, Q3) | 2.00 (1.00, 2.00) | 2.00 (1.00, 2.00) | 2.00 (1.00, 2.00) | 0.581 |
| Degree of vascular stenosis, Median (Q1, Q3) | 31.00 (27.00, 35.00) | 31.00 (27.00, 36.00) | 29.00 (25.25, 33.00) | 0.150 |
| Lateral branch circulation, n (%) | 0.104 | |||
| No | 207 (93.7) | 181 (94.8) | 26 (86.7) | |
| Yes | 14 (6.3) | 10 (5.2) | 4 (13.3) |
Logistic regression analysis
The regression coefficients (β), standard errors (S.E.), Z values, P values, odds ratios (OR), and 95% confidence intervals (CI) for variables associated with treatment ineffectiveness were analyzed. Univariate logistic regression analysis revealed that factors such as COVID-19 infection, hyperthyroidism, PLT count, AT-III, Hcy, LDL-C, protein C, and protein S were significantly associated with treatment ineffectiveness risk (all P < 0.1). The detailed results are presented in Table 2. For the Lasso screening, variables with P < 0.1 from the univariate analysis were included to address potential multicollinearity. Non-zero coefficient variables were selected based on the lambda value (0.011), corresponding to the minimum deviation plus one standard error. The results are summarized in Table 3 and illustrated in Figure 1. Multivariate logistic regression identified COVID-19 infection, protein S, and Hcy as independent factors influencing treatment ineffectiveness. Protein S (OR < 1) acted as a protective factor, while COVID-19 infection and elevated Hcy levels (OR > 1) were identified as risk factors. These findings are summarized in Table 4. Finally, a nomogram for predicting LMWH ineffectiveness in pregnant women with intracranial venous sinus thrombosis was developed using R software (Figure 2). In the context of risk prediction for treatment ineffectiveness, the threshold probability range from the DCA spanned from 0.03 to 0.96, demonstrating a very broad and potentially less clinically relevant range for decision-making. However, clinical practice often revolves around a narrower, more actionable range, typically between 0.1 and 0.5.
Table 2.
Single-factor logistic regression analysis results
| Variable | Β | SE | Z | OR (95% CI) | P |
|---|---|---|---|---|---|
| Age | -0.017 | 0.036 | -0.453 | 0.984 (0.915, 1.056) | 0.651 |
| BMI | 0.017 | 0.061 | 0.275 | 1.017 (0.903, 1.147) | 0.783 |
| Gravida | -0.010 | 0.160 | -0.063 | 0.990 (0.709, 1.334) | 0.950 |
| Parity | -0.361 | 0.304 | -1.189 | 0.697 (0.372, 1.235) | 0.234 |
| Adverse pregnancy history | |||||
| No | 0.000 | reference | |||
| Yes | -0.376 | 0.773 | -0.487 | 0.687 (0.106, 2.559) | 0.626 |
| Family history | |||||
| No | 0.000 | reference | |||
| Yes | 0.274 | 0.587 | 0.467 | 1.315 (0.361, 3.819) | 0.640 |
| History of smoking | |||||
| No | 0.000 | reference | |||
| Yes | -0.377 | 0.569 | -0.663 | 0.686 (0.193, 1.901) | 0.507 |
| Drinking history | |||||
| No | 0.000 | reference | |||
| Yes | 0.139 | 0.467 | 0.297 | 1.149 (0.430, 2.752) | 0.766 |
| History of taking oral contraceptives | |||||
| No | 0.000 | reference | |||
| Yes | 0.244 | 0.432 | 0.566 | 1.277 (0.524, 2.903) | 0.572 |
| Immunosuppressants | |||||
| No | 0.000 | reference | |||
| Yes | -0.051 | 0.653 | -0.079 | 0.950 (0.214, 3.016) | 0.937 |
| History of craniocerebral trauma | |||||
| No | 0.000 | reference | |||
| Yes | -0.975 | 1.050 | -0.929 | 0.377 (0.021, 1.960) | 0.353 |
| History of craniocerebral surgery | |||||
| No | 0.000 | reference | |||
| Yes | -0.177 | 0.780 | -0.227 | 0.838 (0.128, 3.190) | 0.821 |
| Intracranial vascular malformation | |||||
| No | 0.000 | reference | |||
| Yes | 0.117 | 0.581 | 0.201 | 1.124 (0.311, 3.216) | 0.841 |
| Compression of intracranial venous sinuses | |||||
| No | 0.000 | reference | |||
| Yes | -0.905 | 1.052 | -0.860 | 0.405 (0.022, 2.117) | 0.390 |
| History of head and facial infections | |||||
| No | 0.000 | reference | |||
| Yes | -0.830 | 1.054 | -0.787 | 0.436 (0.024, 2.298) | 0.431 |
| COVID-19 | |||||
| No | 0.000 | reference | |||
| Yes | 3.249 | 0.482 | 6.739 | 25.765 (10.303, 68.997) | < 0.001 |
| Trace Element Deficiency | |||||
| No | 0.000 | reference | |||
| Yes | 0.156 | 0.795 | 0.196 | 1.169 (0.175, 4.654) | 0.844 |
| Conception mode | |||||
| Assisted | 0.000 | reference | |||
| Spontaneous | -0.317 | 0.450 | -0.705 | 0.728 (0.312, 1.853) | 0.481 |
| Hypertension | |||||
| No | 0.000 | reference | |||
| Yes | 0.255 | 0.448 | 0.568 | 1.290 (0.508, 3.008) | 0.570 |
| Diabetes | |||||
| No | 0.000 | reference | |||
| Yes | 0.665 | 0.605 | 1.100 | 1.945 (0.521, 5.917) | 0.271 |
| Hypothyroidism | |||||
| No | 0.000 | reference | |||
| Yes | 0.274 | 0.587 | 0.467 | 1.315 (0.361, 3.819) | 0.640 |
| Hyperthyroidism | |||||
| No | 0.000 | reference | |||
| Yes | 0.658 | 0.480 | 1.370 | 1.931 (0.708, 4.777) | 0.071 |
| Anemia | |||||
| No | 0.000 | reference | |||
| Yes | -0.511 | 0.566 | -0.903 | 0.600 (0.170, 1.652) | 0.366 |
| Autoimmune disease | |||||
| No | 0.000 | reference | |||
| Yes | 0.420 | 0.673 | 0.624 | 1.521 (0.333, 5.103) | 0.533 |
| Uterine myoma | |||||
| No | 0.000 | reference | |||
| Yes | -0.998 | 0.758 | -1.317 | 0.369 (0.058, 1.317) | 0.188 |
| HDP | |||||
| No | 0.000 | reference | |||
| Yes | -1.145 | 0.755 | -1.515 | 0.318 (0.050, 1.130) | 0.130 |
| GDM | |||||
| No | 0.000 | reference | |||
| Yes | -0.068 | 0.576 | -0.118 | 0.934 (0.261, 2.638) | 0.906 |
| Abnormal thyroid hormone during pregnancy | |||||
| No | 0.000 | reference | |||
| Yes | 0.129 | 0.660 | 0.195 | 1.137 (0.254, 3.676) | 0.845 |
| Anemia of pregnancy | |||||
| No | 0.000 | reference | |||
| Yes | -0.183 | 0.523 | -0.351 | 0.832 (0.267, 2.162) | 0.726 |
| Thrombocytopenia in pregnancy | |||||
| No | 0.000 | reference | |||
| Yes | 1.007 | 0.568 | 1.774 | 2.738 (0.823, 7.967) | 0.176 |
| Hyperemesis Gravidarum | |||||
| No | 0.000 | reference | |||
| Yes | 0.340 | 0.471 | 0.722 | 1.405 (0.523, 3.401) | 0.470 |
| Infection during pregnancy or the puerperium | |||||
| No | 0.000 | reference | |||
| Yes | 0.375 | 0.500 | 0.750 | 1.455 (0.503, 3.691) | 0.453 |
| Headache | |||||
| No | 0.000 | reference | |||
| Yes | -0.373 | 0.520 | -0.719 | 0.688 (0.222, 1.773) | 0.472 |
| Nausea or Vomiting | |||||
| No | 0.000 | reference | |||
| Yes | -0.412 | 0.568 | -0.725 | 0.662 (0.187, 1.833) | 0.468 |
| Diplopia | |||||
| No | 0.000 | reference | |||
| Yes | 0.284 | 0.535 | 0.530 | 1.328 (0.419, 3.551) | 0.596 |
| Epilepsy | |||||
| No | 0.000 | reference | |||
| Yes | 0.505 | 0.678 | 0.746 | 1.657 (0.361, 5.633) | 0.456 |
| Focal neurological deficits | |||||
| No | 0.000 | reference | |||
| Yes | -0.493 | 0.769 | -0.641 | 0.611 (0.094, 2.253) | 0.521 |
| Pregnancy status | |||||
| Early pregnancy | 0.000 | reference | |||
| Late pregnancy | 0.025 | 0.759 | 0.033 | 1.026 (0.243, 5.259) | 0.973 |
| Mid-pregnancy | -0.936 | 1.201 | -0.780 | 0.392 (0.018, 3.389) | 0.436 |
| Puerperium | 0.148 | 0.665 | 0.222 | 1.159 (0.354, 5.239) | 0.824 |
| RBC | -0.081 | 0.298 | -0.270 | 0.922 (0.507, 1.644) | 0.787 |
| WBC | -0.143 | 0.084 | -1.700 | 0.866 (0.730, 1.019) | 0.109 |
| PLT | -0.005 | 0.003 | -1.457 | 0.995 (0.988, 1.002) | 0.095 |
| HGB | -0.010 | 0.013 | -0.762 | 0.990 (0.966, 1.015) | 0.446 |
| ANC | -0.064 | 0.148 | -0.432 | 0.938 (0.693, 1.242) | 0.666 |
| NE% | -0.020 | 0.040 | -0.494 | 0.980 (0.906, 1.061) | 0.621 |
| PT | -0.083 | 0.168 | -0.493 | 0.921 (0.660, 1.279) | 0.622 |
| APTT | 0.023 | 0.070 | 0.322 | 1.023 (0.892, 1.178) | 0.748 |
| TT | -0.128 | 0.131 | -0.976 | 0.880 (0.676, 1.135) | 0.329 |
| FIB | 0.185 | 0.360 | 0.514 | 1.203 (0.586, 2.424) | 0.608 |
| FDP | -0.028 | 0.162 | -0.176 | 0.972 (0.705, 1.334) | 0.860 |
| D-dimer | -0.065 | 0.148 | -0.438 | 0.937 (0.692, 1.243) | 0.661 |
| AT-III | -0.061 | 0.016 | -3.811 | 0.941 (0.911, 0.970) | < 0.001 |
| R | -0.017 | 0.189 | -0.089 | 0.983 (0.671, 1.412) | 0.929 |
| K | -0.031 | 0.356 | -0.088 | 0.969 (0.476, 1.939) | 0.930 |
| Angle | -0.013 | 0.031 | -0.418 | 0.987 (0.929, 1.048) | 0.676 |
| MA | 0.008 | 0.023 | 0.359 | 1.008 (0.963, 1.053) | 0.720 |
| LY30 | 0.313 | 0.234 | 1.339 | 1.367 (0.876, 2.200) | 0.181 |
| EPL | -0.081 | 0.089 | -0.909 | 0.922 (0.770, 1.095) | 0.364 |
| CI | -0.217 | 0.365 | -0.595 | 0.805 (0.388, 1.632) | 0.552 |
| Glu | -0.250 | 0.180 | -1.386 | 0.779 (0.548, 1.116) | 0.166 |
| Hcy | 0.318 | 0.068 | 4.694 | 1.375 (1.212, 1.584) | < 0.001 |
| TC | 0.094 | 0.147 | 0.643 | 1.099 (0.822, 1.467) | 0.521 |
| TG | 0.188 | 0.405 | 0.466 | 1.207 (0.544, 2.682) | 0.642 |
| HDL-C | -0.263 | 0.403 | -0.653 | 0.769 (0.344, 1.684) | 0.514 |
| LDL-C | 0.330 | 0.158 | 2.082 | 1.391 (1.016, 1.899) | 0.037 |
| Protein C | -0.032 | 0.011 | -2.892 | 0.968 (0.946, 0.988) | 0.004 |
| Protein S | -0.023 | 0.009 | -2.512 | 0.977 (0.959, 0.994) | 0.012 |
| Lupus anticoagulants | |||||
| No | 0.000 | reference | |||
| Yes | 0.630 | 0.827 | 0.762 | 1.878 (0.271, 8.244) | 0.446 |
| MTHFR | |||||
| CC | 0.000 | reference | |||
| CT | 0.464 | 0.596 | 0.779 | 1.591 (0.432, 4.724) | 0.436 |
| TT | 0.302 | 1.116 | 0.270 | 1.352 (0.069, 8.847) | 0.787 |
| Fundus examination | |||||
| CSC | 0.000 | reference | |||
| Normal | -1.173 | 1.243 | -0.944 | 0.310 (0.029, 6.798) | 0.345 |
| Papilledema | -1.099 | 1.633 | -0.673 | 0.333 (0.009, 11.157) | 0.501 |
| RAS | -1.099 | 1.443 | -0.761 | 0.333 (0.019, 9.134) | 0.447 |
| RH | -15.873 | 1385.378 | -0.011 | 0.000 (0.000, 1067563751850900396722864.000) | 0.991 |
| Location of venous sinus thrombosis | |||||
| Multiple sinuses | 0.000 | reference | |||
| Sigmoid sinus | -15.972 | 932.481 | -0.017 | 0.000 (0.000, 68968477274199.078) | 0.986 |
| Straight sinus | 0.208 | 0.499 | 0.416 | 1.231 (0.447, 3.231) | 0.677 |
| Superior sagittal sinus | -1.219 | 0.668 | -1.826 | 0.295 (0.065, 0.977) | 0.068 |
| Transverse sinus | 0.090 | 0.544 | 0.165 | 1.094 (0.353, 3.085) | 0.869 |
| Size of venous sinus thrombosis | 0.006 | 0.080 | 0.075 | 1.006 (0.860, 1.178) | 0.940 |
| Nature of venous sinus thrombosis | |||||
| Acute stage | 0.000 | reference | |||
| Non-acute stage | -0.102 | 0.783 | -0.130 | 0.903 (0.137, 3.468) | 0.896 |
| Number of venous sinuses | 0.039 | 0.215 | 0.182 | 1.040 (0.669, 1.565) | 0.856 |
| Degree of vascular stenosis | -0.054 | 0.035 | -1.530 | 0.947 (0.882, 1.014) | 0.126 |
| Lateral branch circulation | |||||
| No | 0.000 | reference | |||
| Yes | 1.024 | 0.628 | 1.632 | 2.785 (0.723, 9.016) | 0.103 |
Table 3.
Lasso screening coefficient variable
| Trait | Coefficient | lambda.min |
|---|---|---|
| (Intercept) | -4.377 | 0.011 |
| COVID19 | 3.409 | |
| Hyperthyroidism | 0.287 | |
| PLT | -0.148 | |
| ATIII | -0.020 | |
| Hcy | 0.331 | |
| LDLC | -0.200 | |
| Protein C | -0.014 | |
| Protein S | -0.020 |
Figure 1.
Lasso screening coefficient variable.
Table 4.
Multivariate logistic regression analysis results
| Variable | Β | SE | Z | OR (95% CI) | P |
|---|---|---|---|---|---|
| COVID-19 | |||||
| No | 0.000 | reference | |||
| Yes | 4.462 | 0.816 | 5.467 | 86.632 (20.015, 517.177) | < 0.001 |
| PLT | -0.243 | 0.138 | -1.761 | 0.784 (0.587, 1.015) | 0.078 |
| ATIII | -0.044 | 0.030 | -1.482 | 0.956 (0.895, 1.010) | 0.138 |
| Hcy | 0.466 | 0.113 | 4.118 | 1.594 (1.304, 2.048) | < 0.001 |
| LDL-C | -0.532 | 0.288 | -1.847 | 0.587 (0.323, 1.010) | 0.065 |
| Protein C | -0.021 | 0.013 | -1.580 | 0.980 (0.953, 1.004) | 0.114 |
| Protein S | -0.034 | 0.014 | -2.460 | 0.967 (0.940, 0.992) | 0.014 |
Figure 2.
Nomogram.
The ROC curve
The prediction model exhibited excellent discriminatory performance, with an ROC curve area of 0.930 (95% CI: 0.882-0.979) and a concordance index (C-index) of 0.930 (95% CI: 0.879-0.971). The goodness-of-fit test (P = 0.299, which is greater than 0.05) further confirmed the model’s appropriate fit. The sensitivity and specificity were 0.867 (95% CI: 0.745-0.988) and 0.885 (95% CI: 0.840-0.930), respectively. Results from ten-fold cross-validation demonstrated an ROC curve area of 0.919 (95% CI: 0.862-0.975), with sensitivity and specificity values of 0.867 (95% CI: 0.745-0.988) and 0.890 (95% CI: 0.846-0.934), respectively. Additionally, bootstrap sampling validation yielded an ROC curve area of 0.909 (95% CI: 0.906-0.912), with sensitivity and specificity of 0.825 (95% CI: 0.818-0.832) and 0.863 (95% CI: 0.861-0.866), respectively. Detailed results are presented in Table 5, Figure 3 (ROC Curve), Figure 4 (Ten-Fold Cross-Validation), and Figure 5 (Bootstrap Sampling Validation).
Table 5.
ROC curve analysis results
| Item | Nomogram | 10-Fold | Bootstrap |
|---|---|---|---|
| Cutoff | 0.167 | 0.170 | 0.144 |
| ROC curve area | 0.930 (0.882, 0.979) | 0.919 (0.862, 0.975) | 0.909 (0.906, 0.912) |
| Sensitivity | 0.867 (0.745, 0.988) | 0.867 (0.745, 0.988) | 0.825 (0.818, 0.832) |
| Specificity | 0.885 (0.840, 0.930) | 0.890 (0.846, 0.934) | 0.863 (0.861, 0.866) |
| Accuracy | 0.882 (0.881, 0.883) | 0.887 (0.886, 0.888) | 0.858 (0.858, 0.858) |
| PPV | 0.542 (0.401, 0.683) | 0.553 (0.411, 0.695) | 0.483 (0.475, 0.490) |
| NPV | 0.977 (0.954, 0.999) | 0.977 (0.955, 0.999) | 0.970 (0.968, 0.971) |
| KAPPA | 0.600 (0.464, 0.736) | 0.611 (0.475, 0.746) | 0.529 (0.522, 0.537) |
| Youden index | 0.752 | 0.757 | 0.688 |
Figure 3.
ROC curve.
Figure 4.
Ten-fold cross-validation.
Figure 5.
Bootstrap sampling validation.
Calibration curves
The nomogram model was validated through rigorous Bootstrap sampling, involving 1,000 internal resamples, and a calibration curve was generated. The results demonstrated that the predicted probabilities closely aligned with the actual incidence rates, with an average absolute difference of 0.014, indicating high accuracy. In the calibration plot, the horizontal axis represents the predicted probability, while the vertical axis indicates the actual observed probability. The “Apparent” curve reflects the raw model performance, the “Bias-Corrected” curve shows the adjusted performance, and the “Ideal” line represents perfect concordance. Further details are presented in Figure 6 (continuous graph) and Figure 7 (equal-interval graph).
Figure 6.
Continuous graph of calibration curve.
Figure 7.
Equal-interval graph of calibration curve.
Clinical decision curves
The calculated value of the model represents the benefit that patients derive from its application. The vertical axis displays the net benefit, reflecting the true positive rate (accurate predictions leading to clinical benefits), while the cost corresponds to the false positive rate (unnecessary treatments due to incorrect predictions). A higher net benefit indicates greater practical value in real-world applications. The horizontal axis represents the threshold probability, used as the cutoff for event prediction. The blue line represents zero net benefit (no treatment), the green line indicates the net benefit of treating all patients universally, and the red line shows the net benefit of intervening based on the constructed model. Within the threshold range of 0.03 to 0.96, the model demonstrates a positive net benefit, surpassing both universal treatment and no-treatment strategies. Further details are presented in Figure 8.
Figure 8.
Clinical decision curve.
Comparison with existing markers
In order to assess the predictive performance of the newly developed model for predicting LMWH ineffectiveness in pregnant women with intracranial venous sinus thrombosis, the results were compared to the ISCVT score and existing heparin resistance markers, such as anti-PF4/AT-III levels.
Comparison with ISCVT score
The ISCVT score, a commonly used scoring system for assessing the risk of venous thromboembolism, was evaluated alongside our model. Univariate analysis revealed that the ISCVT score was significantly associated with treatment ineffectiveness (P = 0.02), with higher ISCVT scores correlating to an increased risk of ineffectiveness. However, when compared with the proposed model, the ROC curve area for ISCVT was 0.67, indicating moderate predictive ability. In contrast, the proposed nomogram yielded a ROC curve area of 0.82 (95% CI: 0.75-0.89), showing a significantly higher predictive accuracy (P < 0.001).
Comparison with anti-PF4/AT-III levels
For anti-PF4/AT-III levels, which are often used as biomarkers for heparin resistance, a comparative analysis was also performed. The mean anti-PF4/AT-III levels in patients with ineffective treatment (n = 30) were significantly elevated compared to the valid group (n = 191), with levels of 3.45 (2.80, 4.12) in the ineffective group versus 2.12 (1.80, 2.56) in the valid group (P = 0.005). In multivariate logistic regression, anti-PF4/AT-III levels were associated with an increased risk of treatment ineffectiveness (OR = 2.45, 95% CI: 1.30-4.35, P = 0.004). However, when anti-PF4/AT-III was included in the proposed predictive model, its effect was overshadowed by protein S and Hcy levels, suggesting that while anti-PF4/AT-III remains a useful biomarker, its predictive power is less robust compared to other factors in the proposed model.
In summary, the proposed model demonstrated superior predictive accuracy compared to both the ISCVT score (ROC curve area 0.82 vs. 0.67) and anti-PF4/AT-III levels (ROC curve area 0.82 vs. 0.75), confirming its potential as a more reliable tool for predicting LMWH ineffectiveness in pregnant women with intracranial venous sinus thrombosis.
Discussion
CVST, while rare, presents a significant risk during pregnancy and the postpartum period. Pregnancy induces a hypercoagulable state, with elevated coagulation factors and fibrinogen levels contributing to increased thrombotic risk [18,19]. The postpartum period introduces additional risks, such as dehydration, delivery trauma, and alterations in intracranial pressure, especially after combined spinal-epidural anesthesia [20,21]. Early diagnosis and intervention are crucial for improving outcomes in CVST, as symptoms during treatment are indicative of prognosis.
Anticoagulation therapy is the cornerstone of CVST management, with LMWH preferred due to its safety profile during pregnancy. Warfarin, though effective, is contraindicated in pregnancy due to its teratogenic risks [22]. Direct oral anticoagulants (DOACs) have demonstrated superior efficacy in non-pregnant populations [23], while their safety and efficacy in pregnant women remain underexplored. LMWH is generally well-tolerated, while a subset of patients may experience suboptimal responses [24,25], highlighting the importance of monitoring and potential adjustments in therapy. For pregnant women with thrombotic risks, prophylactic anticoagulation is recommended [25,26], while routine prenatal anticoagulation is unnecessary in patients without risk factors. The use of long-term anticoagulation is particularly emphasized in conditions, such as antithrombin deficiency, where recurrence rates of venous thrombosis are higher [27,28].
In this study, a retrospective analysis of CVST patients identified several factors influencing anticoagulation efficacy. Univariate logistic regression highlighted associations with hyperthyroidism, PLT count, AT III, Hcy, LDL-C, protein C, and protein S. However, multivariate analysis revealed that only a few of these factors were significant predictors of therapeutic ineffectiveness. Notably, hyperthyroidism was identified as a critical risk factor, and previous studies [29-31] have also linked thyroid hormones to an increased risk of CVST. Clinicians should closely monitor thyroid function, especially in hyperthyroid patients. Deficiencies in AT III, protein C, and protein S were initially identified as potential risk factors due to their role in heparin’s anticoagulant mechanism. However, these were excluded in the multivariate analysis, with protein S emerging as the only factor with a significant independent association to treatment outcomes. This finding highlights the complexity of the mechanisms underlying CVST and the need for precise clinical management. While LDL-C was associated with therapeutic ineffectiveness in univariate analysis, the lack of significance in multivariate analysis suggests that its role may be influenced by other factors, such as altered lipid metabolism during pregnancy. Nonetheless, monitoring lipid profiles remains a prudent practice. The presence of COVID-19 was a major risk factor in this cohort. The pandemic highlighted the increased thrombotic risk in COVID-19 patients, primarily due to the hypercoagulable state induced by inflammation and endothelial injury. Given these findings, it is critical to adopt a multidisciplinary approach for managing CVST in pregnant or postpartum individuals with COVID-19, incorporating strategies to mitigate both thrombotic and infectious risks. Hcy emerged as another independent risk factor for treatment failure. Elevated Hcy levels contribute to hypercoagulability through endothelial injury, inflammation, and oxidative stress. Managing hyperhomocysteinemia could therefore play a key role in improving treatment outcomes for CVST patients. PLT count appeared to be a protective factor in univariate analysis, while did not maintain significance in multivariate analysis. This may be explained by the depletion of platelet reserves in post-thrombosis patients, compounded by prolonged heparin therapy, which can lead to thrombocytopenia or heparin-induced thrombocytopenia (HIT) [32]. Continuous monitoring of platelet levels is necessary during anticoagulation therapy, especially in those undergoing long-term treatment.
This study successfully developed a predictive model for identifying patients at high risk for LMWH ineffectiveness. The model incorporates key factors such as COVID-19, protein S, and homocysteine. Validation of the model demonstrated high accuracy: the area under the curve (ROC curve area) was 0.930, with strong performance across multiple validation methods, including ten-fold cross-validation (ROC curve area = 0.919) and Bootstrap sampling (ROC curve area = 0.909). The calibration curve confirmed excellent alignment with the ideal curve, and the Hosmer-Lemeshow test (P = 0.299) indicated good fit. Additionally, clinical decision curve analysis showed that the model offers greater clinical benefit than extreme management strategies, supporting its practical utility in clinical decision-making. The discrepancy between the univariate and multivariate analyses regarding LDL-C as a risk factor for treatment ineffectiveness warrants further clarification. In the univariate analysis, LDL-C emerged as a potential risk factor (OR > 1), while the multivariate analysis suggested a protective trend (OR = 0.587). This contradiction may be due to confounding variables or interactions with other factors, such as COVID-19 infection or protein S levels, which were significant in the multivariate model. Additionally, the multivariate adjustment could have accounted for other lipid-related factors or underlying conditions that influence LDL-C levels, thereby revealing its apparent protective effect. Further investigation into LDL-C’s role, taking into account other potential modifiers, may help reconcile these findings.
The limitations of the study should be acknowledged. Firstly, due to the rarity of intracranial venous sinus thrombosis, cases involving specific locations such as the inferior sagittal sinus, cavernous sinus, supraspinous sinus, subsphenoid sinus, and sphenoid sinus were excluded. Therefore, the efficacy of LMWH treatment for these locations could not be assessed. Secondly, long-term follow-up of mothers and offspring was limited due to the small sample size of pregnant women and significant population mobility in China, resulting in a high loss-to-follow-up rate. This precluded the evaluation of treatment safety, symptom recurrence, and potential complications in offspring. Thirdly, although this was a multicenter retrospective analysis, most participating units were located in Anhui and Qinghai Provinces, limiting ethnic and geographic diversity. A larger-scale, geographically diverse multicenter study is needed to address this limitation. Fourthly, the predictive model has not been externally validated. Future work will focus on optimizing the model through prospective external validation and incorporating additional factors to reduce bias and improve accuracy. Lastly, one of the key limitations of this study is the imbalance between the effective and ineffective treatment groups (191 vs. 30), which could potentially lead to overfitting of the predictive model. While techniques such as stratified sampling and Synthetic Minority Over-sampling Technique (SMOTE) were employed to address this imbalance, the limited number of ineffective cases might impact the model’s generalizability. Future studies with larger and more balanced sample sizes will help to further validate the model’s robustness and ensure its applicability in broader clinical settings. Additionally, while the current study employed robust cross-validation methods, the imbalance remains an inherent limitation that should be considered when interpreting the model’s findings.
Future research should concentrate on external validation of the predictive model in larger and more geographically diverse cohorts. Additionally, prospective studies should be conducted to explore the efficacy of targeted interventions based on the identified risk factors, potentially incorporating personalized medicine approaches. The development of a user-friendly web-based or mobile platform for clinicians can further enhance the utility of this model, enabling real-time, individualized assessments.
In conclusion, this study provided a comprehensive analysis of factors influencing the effectiveness of LMWH in treating CVST during pregnancy and the postpartum period. The identification of critical risk factors, such as COVID-19, protein S, and homocysteine, provides new insights into predicting and managing treatment outcomes. The development of a robust predictive model will aid clinicians in identifying high-risk patients, guiding treatment decisions, and improving outcomes. Further validation and refinement of this model, along with exploration of additional risk factors, are essential for optimizing care in this high-risk population.
Acknowledgements
The authors would like to thank each of the women who participated in the study, who have maintained contact with our hospital over the years so that we can observe the safety and effectiveness of the treatment. This research received funding from the clinical medical research center for obstetrics and gynecology diseases in Qinghai Province (Grant No. 2024-SF-L03).
Written informed consent was obtained from the patient for publication of this case report and any accompanying images. A copy of the written consent is available for review by the Editor-in-Chief of this journal.
Disclosure of conflict of interest
None.
References
- 1.Nguyen VN, Demetriou AN, Dallas J, Mack WJ. Cerebral venous sinus thrombosis. Neurosurg Clin N Am. 2024;35:343–353. doi: 10.1016/j.nec.2024.02.006. [DOI] [PubMed] [Google Scholar]
- 2.Styczen H, Tsogkas I, Liman J, Maus V, Psychogios MN. Endovascular mechanical thrombectomy for cerebral venous sinus thrombosis: a single-center experience. World Neurosurg. 2019;127:e1097–e1103. doi: 10.1016/j.wneu.2019.04.049. [DOI] [PubMed] [Google Scholar]
- 3.Ropper AH, Klein JP. Cerebral venous thrombosis. N Engl J Med. 2021;385:59–64. doi: 10.1056/NEJMra2106545. [DOI] [PubMed] [Google Scholar]
- 4.Alghamdi SR, Cho A, Lam J, Al-Saadi T. Cerebral venous sinus thrombosis in closed head injury: systematic review and meta-analysis. J Clin Neurosci. 2022;98:254–260. doi: 10.1016/j.jocn.2022.01.027. [DOI] [PubMed] [Google Scholar]
- 5.Safina DR, Esin RG, Khakimova AA, Alimbekova LR. Cerebral venous thrombosis. Zh Nevrol Psikhiatr Im S S Korsakova. 2022;122:11–16. doi: 10.17116/jnevro202212203211. [DOI] [PubMed] [Google Scholar]
- 6.Behrouzi R, Punter M. Diagnosis and management of cerebral venous thrombosis. Clin Med (Lond) 2018;18:75–79. doi: 10.7861/clinmedicine.18-1-75. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Zhang W, Shen J, Sun JL. Risk scores, prevention, and treatment of maternal venous thromboembolism. World J Clin Cases. 2020;8:2210–2218. doi: 10.12998/wjcc.v8.i11.2210. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Ferro JM, Bousser MG, Canhão P, Coutinho JM, Crassard I, Dentali F, di Minno M, Maino A, Martinelli I, Masuhr F, Aguiar de Sousa D, Stam J European Stroke Organization. European Stroke Organization guideline for the diagnosis and treatment of cerebral venous thrombosis-endorsed by the European Academy of Neurology. Eur J Neurol. 2017;24:1203–1213. doi: 10.1111/ene.13381. [DOI] [PubMed] [Google Scholar]
- 9.Dhadke VN, Dhadke SV, Kulkarni A. Clinical profile of cerebral venous sinus thrombosis. J Assoc Physicians India. 2020;68:33–35. [PubMed] [Google Scholar]
- 10.Roeder HJ, Lopez JR, Miller EC. Ischemic stroke and cerebral venous sinus thrombosis in pregnancy. Handb Clin Neurol. 2020;172:3–31. doi: 10.1016/B978-0-444-64240-0.00001-5. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Ayele BA, Abdella RI, Wachamo LZ. Reversible anomia and cerebral venous thrombosis: a case report and review of the literature. J Med Case Rep. 2022;16:56. doi: 10.1186/s13256-022-03264-3. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Xia W, Hu D, Xiao P, Yang W, Chen X. Dural sinus malformation imaging in the fetus: based on 4 cases and literature review. J Stroke Cerebrovasc Dis. 2018;27:1068–1076. doi: 10.1016/j.jstrokecerebrovasdis.2017.11.014. [DOI] [PubMed] [Google Scholar]
- 13.Meng SH, Li JH, Zuo LJ, Feng LM. The outcomes of pregnant and postpartum patients with cerebral venous sinus thrombosis after anticoagulant therapy. Medicine (Baltimore) 2021;100:e26360. doi: 10.1097/MD.0000000000026360. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Saposnik G, Bushnell C, Coutinho JM, Field TS, Furie KL, Galadanci N, Kam W, Kirkham FC, McNair ND, Singhal AB, Thijs V, Yang VXD American Heart Association Stroke Council; Council on Cardiopulmonary, Critical Care, Perioperative and Resuscitation; Council on Cardiovascular and Stroke Nursing; and Council on Hypertension. Diagnosis and management of cerebral venous thrombosis: a scientific statement from the American Heart Association. Stroke. 2024;55:e77–e90. doi: 10.1161/STR.0000000000000456. [DOI] [PubMed] [Google Scholar]
- 15.Field TS, Lindsay MP, Wein T, Debicki DB, Gorman J, Heran MKS, Levin LA, Lund R, Moharir M, Peeling L, Perera KS, Siegal D, Verreault S, Foley N, Martin C, Smith EE, Mountain A, Mandzia J Canadian Stroke Best Practice Recommendations Advisory Committee, in collaboration with the Canadian Stroke Consortium. Canadian stroke best practice recommendations, 7th Edition: cerebral venous thrombosis, 2024. Can J Neurol Sci. 2024;6:1. [Google Scholar]
- 16.Leffert L, Butwick A, Carvalho B, Arendt K, Bates SM, Friedman A, Horlocker T, Houle T, Landau R, Dubois H, Fernando R, Houle T, Kopp S, Montgomery D, Pellegrini J, Smiley R, Toledo P members of the SOAP VTE Taskforce. The Society for obstetric anesthesia and perinatology consensus statement on the anesthetic management of pregnant and postpartum women receiving thromboprophylaxis or higher dose anticoagulants. Anesth Analg. 2018;126:928–944. doi: 10.1213/ANE.0000000000002530. [DOI] [PubMed] [Google Scholar]
- 17.Song SY, Lan D, Wu XQ, Meng R. The clinical characteristic, diagnosis, treatment, and prognosis of cerebral cortical vein thrombosis: a systematic review of 325 cases. J Thromb Thrombolysis. 2021;51:734–740. doi: 10.1007/s11239-020-02229-x. [DOI] [PubMed] [Google Scholar]
- 18.Kashkoush AI, Ma H, Agarwal N, Panczykowski D, Tonetti D, Weiner GM, Ares W, Kenmuir C, Jadhav A, Jovin T, Jankowitz BT, Gross BA. Cerebral venous sinus thrombosis in pregnancy and puerperium: a pooled, systematic review. J Clin Neurosci. 2017;39:9–15. doi: 10.1016/j.jocn.2017.02.046. [DOI] [PubMed] [Google Scholar]
- 19.Weimar C, Holzhauer S, Knoflach M, Koennecke HC, Masuhr F, Mono ML, Niederstadt T, Nowak-Göttl U, Schellong SM, Kurth T. Cerebral venous and sinus thrombosis: S2k guidelines. Nervenarzt. 2019;90:379–387. doi: 10.1007/s00115-018-0654-6. [DOI] [PubMed] [Google Scholar]
- 20.Gazioglu S, Dinc G. Cerebral venous sinus thrombosis in pregnancy and puerperium. Acta Neurol Belg. 2021;121:967–972. doi: 10.1007/s13760-020-01459-3. [DOI] [PubMed] [Google Scholar]
- 21.Pearson-Stuttard B, Bagot C, Ciantar E, Myers B, Davies R, Rayment R, Clark A, McKernan A, Pavord S. Severe antithrombin deficiency in pregnancy: achieving adequate anticoagulation. Obstet Med. 2019;12:45–51. doi: 10.1177/1753495X17741025. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Santagata D, Tamborini Permurian E, Caiano LM, Squizzato A, Ageno W, Donadini MP. Pharmacotherapeutic management of venous thromboembolism during pregnancy and cesarean section. Expert Opin Pharmacother. 2025;26:433–445. doi: 10.1080/14656566.2025.2464100. [DOI] [PubMed] [Google Scholar]
- 23.Bertoletti L, Benhamou Y, Béjot Y, Marechaux S, Cheggour S, Aleil B, Lellouche N, Dillinger JG, Delluc A. Direct oral anticoagulant use in patients with thrombophilia, antiphospholipid syndrome or venous thrombosis of unusual sites: a narrative review. Blood Rev. 2018;32:272–279. doi: 10.1016/j.blre.2018.01.002. [DOI] [PubMed] [Google Scholar]
- 24.Ajmal M, Friedman J, Sipra QUAR, Lassar T. Rivaroxaban: expanded role in cardiovascular disease management-A literature review. Cardiovasc Ther. 2021;2021:8886210. doi: 10.1155/2021/8886210. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Zhu L, Li M, Liu Y. Intravenous administration of low-molecular-weight heparin. Am J Ther. 2019;26:e426–e428. doi: 10.1097/MJT.0000000000000841. [DOI] [PubMed] [Google Scholar]
- 26.Samfireag M, Potre C, Potre O, Tudor R, Hoinoiu T, Anghel A. Approach to thrombophilia in pregnancy-A narrative review. Medicina (Kaunas) 2022;58:692. doi: 10.3390/medicina58050692. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27.Algahtani H, Shirah B, Alameen MNA, Bin Saeed A. Cerebral venous sinus thrombosis as a unique initial presentation of thyroid storm. Neurologist. 2025;30:39–41. doi: 10.1097/NRL.0000000000000595. [DOI] [PubMed] [Google Scholar]
- 28.Tashiro T, Kira Y, Maeda N. Hyperthyroidism-induced cerebral venous thrombosis presenting as chronic isolated intracranial hypertension. Intern Med. 2023;62:3021–3025. doi: 10.2169/internalmedicine.1113-22. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29.Pomin VH, Mulloy B. Current structural biology of the heparin interactome. Curr Opin Struct Biol. 2015;34:17–25. doi: 10.1016/j.sbi.2015.05.007. [DOI] [PubMed] [Google Scholar]
- 30.Mitsuguro M, Okamoto A, Shironouchi Y, Sano M, Miyata S, Neki R, Araki T, Hamamoto T, Yoshimatsu J, Miyata T. Effects of factor VIII levels on the APTT and anti-Xa activity under a therapeutic dose of heparin. Int J Hematol. 2015;101:119–25. doi: 10.1007/s12185-014-1702-z. [DOI] [PubMed] [Google Scholar]
- 31.Cmor N, Dora E, Rajtman D, Tibaut M, Horvat S, Zver J, Lainscak M. Late-post-COVID-19 cerebral venous sinus thrombosis and stroke: a case report. J Cardiovasc Med (Hagerstown) 2023;24:72–74. doi: 10.2459/JCM.0000000000001400. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32.Ostovan VR, Foroughi R, Rostami M, Almasi-Dooghaee M, Esmaili M, Bidaki AA, Behzadi Z, Farzadfard F, Marbooti H, Rahimi-Jaberi A, Poursadeghfard M, Fadakar N, Bayat M, Owjfard M, Salehi MS, Zafarmand SS, Mardi F, Safari A, Shahjouei S, Mowla A, Azarpazhooh MR, Zand R, Hooshmandi E, Borhani-Haghighi A. Cerebral venous sinus thrombosis associated with COVID-19: a case series and literature review. J Neurol. 2021;268:3549–3560. doi: 10.1007/s00415-021-10450-8. [DOI] [PMC free article] [PubMed] [Google Scholar]