Development of novel plasma proteomic biomarkers for cancer-associated thrombosis in an advanced cancer cohort.

BACKGROUND: Existing risk models for cancer-associated thrombosis (CAT) show suboptimal performance in selective high-risk populations with cancer. Affinity-based plasma proteomics offer a novel approach for detecting CAT risk. OBJECTIVES: To identify plasma biomarkers for CAT using proximity extension assays in an advanced cancer cohort. PATIENTS/METHODS: We performed a nested case-control study using the Olink Explore HT panel. The final cohort included 57 patients with CAT and 113 matched control patients from five selected cancer types who had samples collected between cancer diagnosis and chemotherapy initiation. Random survival forest model was used to assess non-linear associations with CAT in 5,416 normalized protein expressions and 8 clinical variables. Evaluation metrics averaged across bootstrapped out-of-bag test sets included time-dependent receiver operation characteristic curve (TD-ROC), calibration plot, and cumulative incidence in high- versus low-risk predicted groups. We used SHapley Additive exPlanation (SHAP) for feature interpretability. We performed overrepresentation analysis (ORA) and gene set enrichment analysis (GSEA) to assess biological pathway plausibility. RESULTS: Our internally validated model predicted early thrombotic events well (TD-ROC 0.83 at 30 days and 0.73 at 90 days), but the discrimination waned with follow-up time (0.67 at 180 days). Calibration followed a similar pattern. In ORA and GSEA, important proteins were observed in hemostatic pathways including platelet activation, fibrin clot formation, and complement cascade regulation. CONCLUSION: Affinity-based plasma proteomics can be used as a novel strategy to identify biomarkers of CAT. External validation with larger sample size in a cohort setting is required for risk prediction models.