Background
Colorectal cancer (CRC) is the third most commonly diagnosed cancer and the second leading cause of cancer-related mortality worldwide, with an estimated 1.9 million new cases and nearly 1 million deaths in 2022. Despite advances in diagnosis and treatment, CRC prognosis remains poor, with only approximately 64% of patients surviving five years or longer after diagnosis. Prognostic assessment in CRC patients predominantly relies on demographic and tumour-related characteristics; however, these factors do not fully capture underlying biological heterogeneity relevant to disease progression. Identifying biomarkers, such as circulating proteins, associated with survival among CRC patients may enhance understanding on CRC progression and aid prognostic assessment.

Objectives
(1) To assess associations between circulating proteins and both all-cause and CRC-specific mortality and (2) develop and validate prediction models for these outcomes, assessing the added predictive value of proteomic markers beyond established prognostic factors.

Methods
We conducted a prospective analysis among 675 CRC from seven countries (Argentina, Brazil, Czech Republic, Poland, Russia, Serbia, and Thailand) within the DISCERN study. After quality control, data on 1,455 circulating proteins (O-link cardiometabolic, inflammation, neurology, and oncology panels) were available. Multivariable Cox proportional hazards models were used to estimate hazard ratios (HRs) and 95% confidence intervals (CIs) for associations between individual proteins and all-cause and CRC-specific mortality. Three prognostic prediction models were developed: (1) a basic model including sex, age, lifestyle factors (alcohol intake, body mass index, and tobacco smoking), tumour grade (low, high), and tumour stage (local, regional, distant); (2) a model including variables selected through least absolute shrinkage and selection operator (LASSO) regression; and (3) an extended model combining variables in (1) and proteins selected by LASSO. Model discrimination was assessed using cross-validated C-indices.

Results
At an FDR <0.05, a total of 211 proteins were associated with all-cause mortality (208 positively; e.g., ANGPTL4, BAMBI, RELT; HR range 1.23–1.80, and 3 negatively; e.g., NAAA, TNFSF11, GFOD2; HR range 0.69–0.73), while 13 proteins (e.g., ANGPTL4, KRT19, SPP1) were positively associated with CRC-specific mortality (HR range 1.53–1.97). These proteins were predominantly enriched in inflammatory pathways and are highly expressed in foetal and placental tissues. The basic prediction model achieved a C-index of 0.73 (95% CI: 0.68–0.78) for all-cause mortality and 0.77 (95% CI: 0.69–0.84) for CRC-specific mortality. LASSO-based models selected up to 14 proteins in addition to distant tumour stage and demonstrated improved discrimination, with C-indices of 0.85 (95% CI: 0.81–0.88) for all-cause mortality and 0.90 (95% CI: 0.86–0.93) for CRC-specific mortality. Similar or slightly higher predictive performance was observed for the extended model, with C-indices of 0.86 (95% CI: 0.83–0.89) and 0.92 (95% CI: 0.88–0.95), respectively.

Conclusion
In this international analysis, multiple circulating proteins were associated with mortality among CRC patients, and the inclusion of selected proteomic markers substantially improved mortality risk prediction beyond established demographic, lifestyle, and tumour-related factors. These findings highlight the potential value of proteomic profiling for enhanced prognostic stratification in CRC and support further validation and investigation of the underlying biological pathways in CRC.