This article is more than 5 years old. Information may no longer be current.
‘Simple’ model predicts risk for cancer-related venous thromboembolism
Click Here to Manage Email Alerts
A clinical prediction model that used tumor-site category and D-dimer as the only variables accurately predicted risk for venous thromboembolism among patients with solid cancers seen outside the hospital setting, according to findings published in The Lancet Haematology.
“This simple model is a considerable improvement on previous models for predicting cancer-associated venous thromboembolism, and could aid physicians in selection of patients who will likely benefit from thromboprophylaxis,” Ingrid Pabinger, MD, professor in the clinical division of hematology and haemostaseology, of the department of internal medicine at Medical University of Vienna in Austria, and colleagues wrote.
Pharmacological thromboprophylaxis reduces the relative risk for cancer-associated VTE; however, it is not often used in clinical practice because the absolute risk reduction has appeared low.
A personalized approach to risk assessment for VTE via clinical prediction models may classify which patients are at high risk and appropriate candidates for pharmacological thromboprophylaxis, according to the researchers.
Studies have shown clinical parameters and biomarkers of hemostatic activation — including D-dimer, thrombin and soluble P-selectin — are associated with VTE risk among patients with cancer.
Risk scores for cancer-associated VTE include the Khorana score, the Vienna update of the Khorana score and the CONKO score.
“However, the performance of these scores have been limited by deficient predictive power and poor handling,” the researchers wrote.
Further, tests for thrombin generation and soluble P-selectin are seldom available in routine clinical practice and have not been included in clinical prediction models.
Therefore, Pabinger and colleagues sought to develop and externally validate a clinical prediction model for cancer-associated VTE.
The development cohort included 1,423 patients enrolled in the prospective Vienna Cancer and Thrombosis Study, whom researchers evaluated to select prognostic variables to include in the model. Researchers then validated the model in a validation cohort comprised of 832 patients enrolled in the prospective Multinational Cohort Study to Identify Cancer Patients at High Risk of Venous Thromboembolism.
In the final clinical prediction model, researchers included tumor site risk category — an important element of the Khorana score — and continuous D-dimer concentrations as variables.
The multivariable subdistribution HR was 1.96 (95% CI, 1.41-2.72) for high or very high tumor site risk vs. low or intermediate tumor site risk, and 1.32 (95% CI, 1.12-1.56) per doubling of D-dimer concentration.
The model showed a 6-month risk for VTE of 5.7% (range, 2.2-36) in the development cohort and 6.4% (range, 2.3-23) in the validation cohort.
Researchers observed a cross-validated c-indice of 0.66 (95% CI, 0.63-0.67) in the development cohort and 0.68 (95% CI, 0.62-0.74) in the validation cohort, which showed the predicted risk for VTE agreed with observed incidence.
Researchers set a cutoff for predicted cumulative 6-month risk for VTE in the development cohort at 10%, which indicated a model sensitivity of 33% (95% CI, 23-47), specificity of 84% (95% CI, 83-87), positive predictive value of 12% (95% CI, 8-16) and negative predictive value of 95% (95% CI, 94-96).
At a cutoff of 15%, researchers observed a model sensitivity of 15% (95% CI, 8-24), specificity of 96% (95% CI, 95-97), positive predictive value of 18% (95% CI, 9-29) and negative predictive value of 95% (95% CI, 94-96).
At a 10% cutoff for predicted 6-month risk for VTE in the validation cohort, researchers observed a sensitivity of 21% (95% CI, 10-35), specificity of 87% (95% CI, 85-90), positive predictive value of 9% (95% CI, 4-16) and negative predictive value of 95% (95% CI, 93-96).
At a cutoff of 15%, researchers observed a sensitivity of 8% (95% CI, 2-20), specificity of 99% (95% CI, 98-99), positive predictive value of 29% (95% CI, 98-58) and negative predictive value of 95% (95% CI, 93-96).
“We showed that our clinical prediction model could outperform previous clinical prediction scores in predicting those patients at high risk of developing VTE,” researchers wrote. “The model is available for clinical use as a printed nomogram and as an online prediction tool.”
The study by Pabinger and colleagues highlights what may viewed as an “existential issue” as risk tools expand and become simpler, favoring clinical applicability, or become more complex, improving prediction, Alok A. Khorana, MD, co-director of Cleveland Clinic Cancer Center’s Upper Gastrointestinal and Colorectal Cancer programs, wrote in a related editorial.
For risk assessment tools to improve, new variables or biomarkers need to be added or populations need to be narrowed, he added.
“However, focusing on narrower populations could result in the proliferation of risk scores for every single type or subtype of cancer, which might overwhelm clinicians and lead to these tools never being used,” he wrote. “Future iterations of risk tools might incorporate -omics approaches or microRNA profiling, but how and when these would be applied clinically is uncertain.”
The question remains on how to reduce the cancer-associated VTE burden.
“My own bias is that applicability is key and risk stratification should be as automated as possible for optimal clinical use,” Khorana wrote. “If risk tools must be made more complex, the implementation bar should be set at a much higher positive predictive value (minimum 70%).
“Perhaps we can heed the advice of the aphorism attributed to many sources, including Albert Einstein: everything should be made as simple as possible, but not any simpler,” he added. “This, too, is more complicated than it seems.” – by Melinda Stevens
Disclosures: The authors report no relevant financial disclosures. Khorana reports the Khorana score was sponsored by a grant from NCI, and he reports research support from the Sondra and Stephen Hardis Chair in Oncology Research; the NHLBI; and the Scott Hamilton Cancer Alliance for Research, Education and Survivorship Initiative.
Click Here to Manage Email Alerts