Optimizing the Prediction of Venous Thromboembolis by a Risk Assessment Model in Patients with Glioblastoma

Purpose To nd an optimal model to predict venous thromboembolism specially for glioblastoma by a prospective research. Methods Patients with newly histologically conrmed glioblastoma multiform (GBM) were recruited for this study. Status of IDH1, PTEN, P53, BRAF, MGMT and TERT were determined using genetic sequencing through polymerase chain reaction (PCR). Amplication of EGFR was established through uorescence in situ hybridization (FISH). Competing-risks regression model was performed to calculate the risk of VTE. Clinical and laboratory parameters that were independently predicted risk of VTE were used to develop a risk assessment model (RAM). 145 patients with GBM were included in the present analysis. A total of 48/145 patients (33.1%) developed VTE, all 48 patients were found to be IDH1wt . D-dimer, ECOG score and EGFR staus were sugggested to be signicantly associtaed with the VTE risk in multivariable analysis. High ECOG score (>2), high D-dimer (>1.6 μg/mL) and EGFR amplication were used as the strongest independent predictors of increased risk of VTE. The cumulative incidence of VTE was 17.2% for patients with score 0 (n =29), 23.6% for patients with score 1 (n =55 ) and 63.8% for patients with score 2 (n = 35) or score 3 (n = 12) by application of a RAM . IDH1 were very low In IDHwt patients, by a VTE risk assessment model, very high low


Introduction
Venous thromboembolism (VTE) is common in cancer patients, and its incidence rates differ widely in different tumor types [1][2]. Compared with cancers in other sites, brain tumor patients are reported to be at relatively higher risk group of developing VTE, especially in malignant glioma patients, with the incidence rates between 10% and 30% [3][4]. Besides cancer type, the incidence of VTE is also closely related with the malignant degree of the tumor. Glioblastoma, the most common and malignant primary brain tumor, is suggested to be at higher risk of VTE than other types of gliomas [5][6]. These suggest that VTE is closely associated with speci c tumor type and there may be cancer type-speci c pathways of VTE.
Glioblastoma multiform (GBM) is a highly malignant and heterogeneous tumor. According to the 2016 WHO classi cation of CNS tumors, gliomas can be divided into different groups by oncogenetic mutations and tumor grade [7]. But currently, no consensus has been made regarding the optimal molecular classi cation for clinical practice of treatment for GBM. By large-scale pro ling studies of GBM, Verhaak et al described a molecular classi cation of GBM into Proneural, Neural, Classical, and Mesenchymal subtypes [8]. Labussière et al have proposed another new molecular classi cation of GBM based on TERT promoter (TERTp) mutation, EGFR ampli cation, and IDH1 mutation by analysis of prognosis [9]. However, it has already formed a consensus that molecular classi cation of glioma is better than histological grading in guiding treatment and judging the prognosis of glioma.
However, only little data exist to predictive of venous thromboembolism by combined oncogenetic mutations and blood biomarkers. In this study, we have combined analysis of clinical markers, oncogenetic mutations and plasma biomarkers in routine clinical practice to try to nd an optimal model to predict venous thromboembolism in GBM by a prospective research.

Patient selection
From May 2017 to August 2019, patients with glioblastoma were recruited for this study based on the following eligibility criteria: 18 years or older, newly histologically con rmed glioblastoma, received total or subtotal tumor resection, information regarding status of IDH, BRAF, TERTp, PTEN, P53 mutations, EGFR ampli cation and MGMT promoter methylation was available. Patients with a prior history of VTE or continuous anticoagulation therapy were excluded. The pathological grading criteria were based on the latest standards of World Health Organization (WHO) classi cation system [7] and pathology was determined by a senior neuropathologist in all cases. All patients signed the information consent to review and use their medical records. 18 months follow up were performed for each patient until the occurrence of VTE or death or loss of follow-up.

Molecular data
Mutational status of IDH1, PTEN, P53, BRAF and TERTp were determined using genetic sequencing through polymerase chain reaction (PCR). MGMT promoter methylation was determined by methylation speci c PCR. Ampli cation of EGFR was established through uorescence in situ hybridization (FISH).

Outcome measures
The baseline clinical characteristics of the patients regarding age, sex, Eastern Cooperative Oncology Group (ECOG) score, body mass index (BMI), hemiparesis, duration of follow-up and occurrence of VTE was collected. The main outcome measure was the occurrence of VTE within 18 months after inclusion of the study. Either deep venous thrombosis (DVT), pulmonary embolus (PE), or cerebral venous sinus thrombosis (CVST) con rmed by imaging was considered to be VTE in this study. Venous blood sample was collected on the day of enrollment, D-dimer and platelet count were measured within 2 hours after collection.

Statistical analysis
The primary outcome was time to rst VTE after study inclusion,with the patient's death as a competing event. The data were censored after 18 months follow-up or loss to follow-up. Competing-risks regression model according to Fine and Gray [16] was performed to calculate the risk of VTE. Univariate and multivariable competing risk analysis were used to compare differences in time to rst VTE between age, sex, ECOG score, BMI, hemiparesis, IDH, PTEN, P53, BRAF, TERTp mutation status, MGMT promoter methylation status and EGFR ampli cation status. Parameters that were independently predicted risk of VTE were used to develop a risk assessment model (RAM). Statistical signi cance was set at p < = 0.05. Analyses were performed using Stata 15.

Baseline characteristic of the study population
In this study, 145 patients with GBM were included in the present analysis. The median follow-up of patients included in the analysis was 15.8 months (range, 1.6 to 18 months). during the 18 months follow up period, 21(14.4%) patients died without VTE. The detailed clinical characteristics of the study population, extent of surgical resection are summarized in Table 1, baseline laboratory parameters and molecular information are summarized in Table 2.

Competing-risk regression
The competing-risk regression was used to analyze the hazard ratios (HRs) of developing VTE for each parameter in univariate and multivariable proportional subdistribution models. The detailed information are listed in Table 5. ECOG score, D-dimer, PTEN status and EGFR status were sugggested to be signi cantly associated with the VTE risk in univariate analysis. In multivariable analysis, all the variables in the univariate analysis were included. ECOG score, D-dimer and EGFR status remained signi cantly associtaed with the VTE risk. We used a statistical stepwise forward-selection process to design a VTE risk assessment model in 131 IDHwt patients. High ECOG score (> 2), high D-dimer (> 1.6 µg/mL), and EGFR ampli cation were used as the strongest independent predictors of increased risk of VTE. One point was assigned, respectively. The cumulative incidence of VTE was 17.2% for patients with score 0 (n = 29), 23.6% for patients with score 1 (n = 55 ) and 63.8% for patients with score 2 (n = 35) or score 3 (n = 12) by application of a RAM ; (P < 0.001; Fig. 1).

Discussion
In this study, we have designed a prospective research to investigate the relationship between tumorspeci c biomarkers and VTE in GBM patients. We have examined 7 routinely tested molecular factors in IDHwt patients, shows that the status of EGFR and PTEN are signi cantly correlated with VTE by univariate analysis. Other prognostic factors such as TERTp, P53, BRAF and MGMT has no in uence in the risk of VTE. PTEN is no longer signi cantly correlated with VTE events by multivariable analysis.We nally found 1 clinical marker, 1 molecular and 1 plasma biomarker predicting substantially increased risk of VTE in IDHwt GBM patients: namely, ECOG score, EGFR status and D-dimer. We tested the biomarkers by a VTE risk assessment model and found that patients with low ECOG score (≤ 2), low Ddimer (< 1.6µg/mL) and no EGFR ampli cation have the lowest risk of developing VTE (17.2%, score = 0), the risk of VTE was 63.8% for patients assigned to the high-risk group (score = 2 or 3).
Our ndings that combined analysis of ECOG score, EGFR and D-dimer to built a GBM speci c VTE risk assessment model is completely novel. Most of previous studies have tested different biomarkers in pooled patient populations with different types of cancer, however, the numbers of samples for speci c cancer are small [17][18][19]. A multinational cohort study evaluated four clinical prediction scores for VTE in patients with advanced cancer and found that the prediction scores were poor at predicting VTE in cancer patients [20]. The sensitivity and speci city of current prediction models for VTE in clinical practice is not satis ed in cancer patients, explore new models based on biomarkers and pathways of cancerassociated thrombosis in individual cancers are thought to be the best way to solve this problem [21].
With the publication of the 2016 WHO Classi cation of Tumors of the Central Nervous system, the o cial diagnostic and grading system of glioma has changed from solely on morphologic features to both on molecular and morphologic features.
Some molecular biomarkers like IDH, TERTp, EGFR, P53, PTEN, BRAF, MGMT are reported to be closely related with prognosis in GBM and recommened to test in clinical practice to assist diagnosis and treatment. Some preclinical and clinical research revealed that these biomarkers are also closely related with the risk of VTE. A study by Unruh et al. reported that mutant IDH1 has potent antithrombotic activity by the production of D-2-hydroxyglutarate (D-2-HG) in glioma patients [22]. NAZARI et al analyzed the different risk of VTE in brain tumor patients and found that the status of IDH1 was closely linked to the risk of VTE, patients with IDH1 mutation were at very low risk of VTE [23]. In line with the present study, we also found that IDH1 mutation was highly linked with low risk of VTE, none of 14 IDH1 mutant patients in this study developed VTE during follow-up. Rong et al have evaluated the relationship between the expression of EGFR and PTEN and tissue factor expression in glioblastoma and found that they are also highly correlated [24]. Recently, a retrospective study showing no difference in VTE risk according to EGFR ampli cation status in Grade II-IV Glioma patients. However, only 73 GBM patients had information on EGFR expression and they were not analyzed separately [25]. In contrast with this study, our data shows that the status of EGFR amplication signi cantly related with VTE risk in IDHwt GBM patients. Our large number of available EGFR information patients and separate analysis may partially explain this discrepancy.
On the other hand, we also evaluated the predictive value of clinical and plasma markers. Previous study have shown that glioma patients with age > 44 years linked with high risk of VTE [26]. However, our result shows no difference between patients with different ages. The median age of diagnosis of GBM is reported to be 64 years old and more than 80% is over 50 years old [27]. In our study, the median age of diagnosis is 57 years old and less than 10% with age < 50 years old. The difference in age distribution may explain this discrepancy. Besides, by univariate and multivariable competing risk analysis, we found that ECOG score is the strongest independent predictors of increased risk of VTE among age, gender, ECOG score, BMI and hemiparesis. Platelet count and D-dimer have been used widely as useful indicators of intravascular thrombosis, consist with previous study [4], our result shows that patients with VTE have signi cant higher D-dimer than patients without VTE, D-dimer is an independent predictors of VTE. However, no signi cant difference in Platelet count has been noted between two groups. Some limitations of our study need to be addressed. This study is a single-center study, the number of IDHmt patients and VTE events is relatively low. Moreover, based on the results of previous studies and routine clinical practice, although we have analyzed 7 oncogenetic mutations and 2 plasma biomarkers and developed a RAM in IDHwt GBM patients, some information of other oncogenetic mutations is not available; for example, CDKN2A and NFKB1A deletion mut In conclusion, we have combined analysis of clinical and laboratory markers in routine clinical practice of GBM and found that patients with IDH1 mutation were at very low risk of VTE. In IDHwt GBM patients, 1clinical marker,1 molecular mutation and 1 plasma biomarkers predict substantially increased risk of VTE : namely, ECOG score, EGFR status and D-dimer. by applying a VTE risk assessment model, we could identify patients with a very high and low risk of VTE in clinical practice.

Declarations
Founding: This work was partially supported by the National Natural Science Foundation of China (Grant No. 81641116).