Urinary metabolomics may improve prediction of overall survival beyond tumor stage in colorectal cancer: results from the ColoCare study.

Abstract

Background: Colorectal cancer (CRC) is a leading cause of cancer-related mortality. Prognosis is primarily guided by tumor stage despite substantial molecular heterogeneity. Urinary metabolomics may capture systemic and tumor-related biology beyond staging and could improve prognostic assessment. We hypothesized that incorporating urinary metabolomic profiles would improve overall survival (OS) prediction performance compared with a stage- and age-based reference model.

Method: A total of n = 76 stage I-IV CRC patients recruited as part of the ColoCare Study in Heidelberg Germany with pre-surgery urinary metabolomics were included (23 deaths; median follow-up 3.03 years). Four metabolomics-based penalized Cox models adjusted for tumor stage and age at diagnosis were developed using LASSO, adaptive LASSO, spike-and-slab LASSO, and iterative sure independence screening (iSIS)-LASSO. Model discrimination was assessed using Harrell's C-index and time-dependent AUC based on the nested cross-validation.

Results: Compared with the reference model (Cox model including only tumor stage and age at diagnosis), all metabolomics-based models provided better discrimination. The spike-and-slab LASSO Cox model demonstrated the best performance, achieving a C-index of 0.75 (vs. 0.68) and consistently higher time-dependent AUCs at 1-5 years of follow-up, with a peak AUC of 0.76 at year 3 (vs. 0.68). Three urinary metabolites were consistently selected across all metabolomics-based models: indolelactate, 2-hydroxyisobutyrate and a uridine-like metabolite.

Conclusions: Urinary metabolomics may improve CRC OS prediction beyond tumor stage and age at diagnosis, especially with the spike-and-slab LASSO Cox model. These results support urinary metabolomics as a promising noninvasive prognostic tool that merits external validation.