Plasma proteomics improves thrombosis prediction in patients with cancer and identifies targetable IL-17–driven endothelial activation

Existing risk models fail to reliably predict venous thromboembolism (VTE), underscoring the need for more accurate predictive models. In this study, we conducted a high-throughput proteomic analysis of 1,105 plasma proteins from newly diagnosed lung and gastric cancer patients prospectively monitored for VTE development.

Utilizing a Bayesian probabilistic machine learning approach, we developed a predictive model incorporating 11 protein biomarkers and five clinical parameters (age, sex, history of VTE, body mass index, and hemoglobin), which significantly outperformed the Khorana prediction model. Orthogonal validation in an external placebo cohort from a phase III trial confirmed the model’s predictive power. Further investigation into the mechanistic role of CD200R1, a checkpoint receptor limiting leukocyte inflammatory response that contributed strongly to the model, showed that reduced levels in plasma correlated with higher D-dimer and thrombosis risk.