Gemcitabine and Platinum-Based Agents for the Prediction of Cancer-Associated Venous Thromboembolism: Results from the Vienna Cancer and Thrombosis Study

Simple Summary Certain chemotherapy agents (gemcitabine, platinum-based agents) have been suggested to increase the risk of venous thromboembolism in cancer patients. Our aim was to evaluate, whether treatment with these agents can be used to better predict the risk of cancer-associated venous thromboembolism. Within a prospective observational cohort study, including 1409 patients, we found that treatment with gemcitabine and/or platinum-based agents is only of limited value in predicting the risk of venous thromboembolism beyond known risk factors included in an established risk prediction model (tumor type, blood levels of D-dimer). These findings suggest that a large part of the observed rate of venous thromboembolism in patients treated with these agents might be related to the underlying thrombotic risk rather than the agent itself. Abstract Gemcitabine and platinum-based agents could increase the risk of venous thromboembolism (VTE) in patients with cancer. We evaluated the additive predictive utility of these agents towards cancer-associated VTE beyond a recently developed and externally validated clinical prediction model, which was based on tumor entity and continuous D-dimer levels. Analysis was performed in the derivation cohort of this model, obtained from the Vienna Cancer and Thrombosis Study (CATS), a prospective observational cohort study (n = 1409). Patients were followed for the occurrence of VTE for a maximum of two years. Competing-risk analysis was performed to obtain cumulative incidences and to conduct between-group comparisons of VTE risk. Cumulative two-year incidences of VTE were not elevated with gemcitabine treatment (10.2% vs. 7.5%, p = 0.148), whereas they were higher for platinum-based therapy (11.6% vs. 5.9%, p < 0.001). In a multivariable analysis, adjusting for tumor site category and D-dimer, gemcitabine was not associated with increased risk of VTE (subdistribution hazard ratio (SHR) 0.82, 95% confidence interval (CI) 0.53–1.28, p = 0.390), whereas platinum-based therapy predicted for a numerically increased VTE risk (SHR 1.44, 95% CI 0.96–2.17, p = 0.080). Similar results were obtained in a sensitivity analysis (updated cohort, n = 1870). Our findings suggest limited additional value of chemotherapy for the prediction of cancer-associated VTE, beyond a validated clinical prediction model.


Introduction
Various tumor-, patient-and treatment-related factors contribute to the high risk of venous thromboembolism (VTE) in patients with cancer [1]. Among antineoplastic treatments, especially platinum-based agents and gemcitabine have been particularly implicated in increasing risk of cancer-associated thromboembolism [2][3][4]. Consequently, treatment with either platinum-based agents or gemcitabine has previously been incorporated in the PROTECHT-score, a modified version of the Khorana score for the prediction of VTE in patients with cancer [5]. The performance of this score has been evaluated in two independent cohort studies that reported c-indices of 0.59 and 0.61, indicating only moderate discrimination [6,7].
Recently, we developed and externally validated the Vienna Cancer and Thrombosis model (CATS model), a simple clinical prediction model for VTE in patients with cancer based solely on tumor entity and continuous pretherapeutic levels of D-dimer [8]. The aim of the present analysis was to evaluate platinum-based or gemcitabine therapy for the prediction of cancer-associated VTE and to explore the additive predictive value of these variables beyond those incorporated in the validated CATS score.

Prediction of Cancer-Associated Thrombosis Beyond the CATS Score
Gemcitabine was not associated with risk of VTE in the univariable analysis (SHR 1.39, 95% CI 0.88-2.18, p = 0.156). This association remained similar upon multivariable adjustment for tumor type (SHR 0.83, 95% CI 0.53-1.31, p = 0.432), D-dimer (SHR 1.18, 95% CI 0.75-1.85, p = 0.481), and tumor type and D-dimer (SHR 0.82, 95% CI 0.53-1.28, p = 0.390, Model #1-3 in Table 2). These results suggest a lack of independent predictive utility of gemcitabine therapy towards cancer-associated VTE beyond tumor type or levels of D-dimer.  Table 2). These findings suggest a potential weak association between platinum-based therapy and risk of cancer-associated VTE beyond tumor type or levels of D-dimer.

Sensitivity Analysis
In order to further explore the putative association of therapy with platinum-based agents or gemcitabine with VTE, a sensitivity analysis was performed with an updated cohort of patients recruited within the Vienna Cancer and Thrombosis study (n = 1870). Consistent with the main analysis, therapy with gemcitabine was not associated with risk of VTE in the univariable analysis (SHR 1.17, 95% CI 0.76-1.80, p = 0.478). Therapy with platinum-based agents was associated with higher VTE risk in the univariable analysis (SHR 1.50, 95% CI 1.09-2.07, p = 0.014). This association was weakened upon adjustment for tumor site category and D-dimer levels in the multivariable analysis (SHR 1.21, 0.86-1.70, p = 0.280).

Discussion
The present findings suggest that platinum-based therapy and gemcitabine do not significantly improve prediction of VTE in patients with cancer beyond an established risk prediction model. Univariable analyses confirmed a relative increase in risk of VTE with platinum-based therapy and gemcitabine as reported previously [2,4]. This concept is further supported by data on vascular toxicity of these agents from preclinical work and clinical observations [11]. Platinum-based agents have been shown to induce endothelial cell dysfunction, including proinflammatory changes and an increased expression of cell adhesion molecules, whereas preclinical data on vascular and hemostatic effects of gemcitabine remain scarce and inconclusive [12][13][14][15].
Nonetheless, our findings suggest that the high burden of VTE in patients treated with platinum agents or gemcitabine are possibly not be fully causal, but to a large degree be mediated by underlying prothrombotic risk factors that are correlated with treatment with these agents. For example, gemcitabine is one mainstay of treatment for pancreatic cancer, and platinum-based agents are a treatment standard for gastric cancer, and both these tumor entities are associated with the highest known risk of VTE [16,17].
A limitation of our analysis is that treatment with platinum-based agents and gemcitabine was not analyzed as a time-dependent variable and applied cumulative doses were not considered. Although time-dependent and dose-specific analyses could have yielded a more refined appreciation of the association between these treatments and cancer-associated VTE risk, time-dependent variables are not useful for clinical prediction purposes as their information only becomes available over time, and thus cannot be used for clinical decision making.

Methods
This analysis was conducted within a prospective, observational cohort study, the Vienna Cancer and Thrombosis Study (CATS). The design, procedures, as well as inclusion and exclusion criteria have been previously reported in detail [18], and the study was approved by the Ethics Committee of the Medical University of Vienna (Project identification code: 126/2003, date of approval: 2 September 2003). Briefly, patients with newly diagnosed cancer or recurrent or progressive disease after remission were eligible for inclusion. Patients were followed prospectively for the occurrence of objectively confirmed, independently adjudicated VTE for a maximum of 2 years.
We used the same dataset as the CATS model derivation study (n = 1423) [8], with risk of VTE stratified according to tumor type, i.e., very high risk (pancreas and gastric), high risk (lung, colorectal, oesophagus, kidney, lymphoma, bladder/urothelial, uterus, cervical, ovarian, and others), and intermediate/low risk (breast and prostate). D-dimer levels were obtained from blood withdrawn at study inclusion and measured with the STA-Liatest assay (Diagnostica-Stago, Asnières, France). Data on chemotherapy during the 2-year observation period were obtained from the in-house pharmacy prescription records, with missing chemotherapy data in 14 patients (analysis cohort, n = 1409).

Statistical Analysis
Baseline differences between groups were analyzed by means of the rank sum test or Chi-square test, as appropriate. Between-group differences in risk of VTE were assessed with competing risk cumulative incidence estimators and Fine and Gray proportional subdistribution hazards regression models [19,20]. A sensitivity analysis was conducted with the most recent dataset available from CATS, consisting of 1870 patients.

Conclusions
Therapy with gemcitabine was not associated with an increased risk of VTE. Platinum-based therapy could be a weak independent risk factor for cancer-associated VTE, with only limited predictive value beyond tumor site category and D-dimer levels. These findings do not support the use of these variables in clinical prediction models for VTE in patients with cancer.