En route to personalized prophylaxis

Predicting a patient's outcome reliably is perhaps the one of the most challenging aspects of the art of medicine. In this issue of Blood, Ay and colleagues introduce science to this art by validating the Khorana risk score (the first risk assessment model that predicts how likely a cancer patient is to develop symptomatic VTE) and confirming the value of 2 biomarkers, D-dimer and soluble P-selectin (sP-selectin), as predictors of thrombosis.¹ 

Thromboembolism is a dreaded complication in patients with cancer. Not only does it complicate cancer treatment, it also shortens survival, reduces quality of life, and consumes significant health care resources. Despite therapy with anticoagulants, up to one-third of oncology patients with venous thromboembolism (VTE) will experience recurrent thrombotic events or serious bleeding.²  Primary prevention is the most effective way to reduce disease burden and this is the recommended standard of practice in oncology patients who are hospitalized or having surgery.³  But given that the risk of VTE ranges from 1%-30% among ambulatory patients with cancer and that it varies over time even in an individual patient, a one-size-fits-all solution of providing thromboprophylaxis to all ambulatory patients is unlikely to be beneficial, practical, or cost-effective. In fact, it may cause more harm because of the higher risk of bleeding in patients with cancer. A more sensible and scientific approach would be to identify patients at particularly high risk who would benefit the most from thromboprophylaxis.

So how do we identify this population? Khorana and colleagues first took up this challenge and developed a VTE risk assessment model for predicting the likelihood of VTE in patients receiving outpatient chemotherapy.⁴  Using registry data collected to evaluate febrile neutropenia and other complications of chemotherapy, they found 5 independent risk factors that predicted for symptomatic VTE during the first 4 cycles of chemotherapy: (1) site of cancer; (2) prechemotherapy platelet count ≥ 350 × 10⁹/L; (3) hemoglobin level < 100 g/L or the use of erythropoiesis-stimulating agents; (4) prechemotherapy leukocyte count > 11 × 10⁹/L; and (5) body mass index > 35 kg/m². Patients with none of these risk factors had a very low risk of VTE at < 0.8%, while those with 3 risk factors or more, or a high-risk cancer type with 1 or more additional risk factors, had a VTE risk of 7%. In the Khorana study, some cancers are underrepresented and all patients were receiving chemotherapy; data on central venous catheters, use of anticoagulants, and previous history of VTE were not collected. In addition, some experts questioned the validity of the results because thrombotic events were not independently adjudicated and others wondered whether the model is generalizable to unselected patients.

Ay and colleagues provide the answers in this issue.¹  They confirm that the Khorana model is able to stratify a broad range of cancer patients into distinct risk groups for VTE using these 5 readily available clinical variables. In the well-characterized cohort of patients in the Vienna Cancer and Thrombosis Study (CATS) that was designed to follow the risk of symptomatic, objectively confirmed VTE, Ay et al report the 6-month cumulative probability of VTE is 17.7% in the highest risk group with a score of ≥ 3, 9.6% in those with a score of 2, 3.8% in those with a score of 1, and 1.5% in patients with a score of 0. In patients with a score of < 3, the likelihood of not having VTE is 94.9%; in the patients with a score of ≥ 3, the likelihood of having VTE is 22.1%.

Ay and colleagues also evaluated whether the addition of 2 biomarkers, D-dimer and sP-selectin, to the Khorana model provides greater accuracy in predicting the risk of VTE. The associations between these markers and VTE were previously demonstrated by the same and other investigators.^5-8  Not surprisingly, this expanded model teases out further risk groups. In the highest score group with ≥ 5 risk factors present, 35% of the patients had a VTE event over 6 months, while those without any factors had a VTE risk of only 1%. In patients with a score of < 5, the likelihood of not having VTE is 94.4%; in the patients with a score of ≥ 5, the likelihood of having VTE is 42.9%.

The results are robust yet there are outstanding issues. There are relatively few patients in the high-score (≥ 3) categories, so we are less confident in the estimated VTE risk and the accuracy indices in these groups. The cutoff values used for D-dimer and sP-selectin may not be applicable for different assays. Additional risk factors for VTE, such as distant metastasis, previous history of VTE, use of newer thrombogenic cancer treatments (eg, bevacizumab, thalidomide) were either not significant or not examined in the models. Finally, it is uncertain whether the Ay model offers a clinically meaningful improvement in accuracy that justifies the added complexity and cost.

So, what does this research mean to patients with cancer and the physicians caring for them? Do the models predict the risk of VTE? Yes, they do. Can we reliably use them to recommend thromboprophylaxis in those with a high score? No, not quite yet. Before we use these models to make management decisions about primary thromboprophylaxis, intervention trials must be conducted to prove such strategies are safe and effective. This is important because preventing thrombosis cannot come at the cost of excessive bleeding and the optimal dose of prophylaxis remains uncertain.⁹  The first clinical trial, funded by the National Heart, Lung, and Blood Institute (www.clinicaltrials.gov no. NCT00876915), is now ongoing to test the efficacy and safety of a low-molecular-weight heparin in preventing VTE in high-risk patients with a Khorana score of ≥ 3 who are starting chemotherapy.

The Ay and Khorana models also raise fascinating questions about the biology of thrombosis in cancer patients. Are the risk factors causally related or simply “bystanders” of hypercoagulability? With leukocytosis, anemia, and thrombocytosis acting as independent risk factors in these models, hematopoiesis or changes in bone marrow milieu may play an important mechanistic role. We still have a long way to go in understanding the mechanisms that govern the interactions between coagulation and tumor biology.

Khorana and colleagues took the first step toward personalizing primary thromboprophylaxis in ambulatory patients with cancer, and Ay and colleagues have propelled us further ahead. We are looking forward to the next leap.