Background: The purpose of this study was to explore the prognostic value of Immunoscore in patients with early-onset colorectal cancer.
Methods: We retrospectively analyzed 708 colorectal adenocarcinoma patients (2017-2020), ultimately including 36 early-onset colorectal cancer cases after exclusions. CD3+/CD8 + lymphocytes were quantified using immunohistochemistry and a self-trained neural network model from Cellpose 2.0. Immunoscore was calculated based on T cell densities in tumor cores and invasive margins, stratified as high or low. Prognostic associations were assessed via Kaplan-Meier analysis, Cox regression, and restricted cubic spline models. Results of Cox regression was validated by post-hoc analysis.
Results: Of all early-onset colorectal cancer patients, 23(63.9%) patients were graded as Immunoscore-high, 13(36.1%) were graded as Immunoscore-low. The self trained model achieved high consistency with manual counting. High Immunoscore correlated with earlier clinical stages (stage I/II: P = 0.011), reduced metastasis risk (N0, P = 0.042; M0, P = 0.009), and lower mortality (P = 0.009). Univariate Cox regression analysis identified Immunoscore as a possible predictor for overall survival (Hazard Ratio = 5.82, P = 0.030) and progression free survival (Hazard Ratio = 3.68, P = 0.014). The Post-hoc power analysis showed the type II error probability (β) of univariate Cox analysis for overall survival with a hazard ratio of 5.82 was 0.282 (28.2%), while for progression free survival with a hazard ratio of 3.68, β was 0.006 (0.6%). Restricted cubic spline showed that the influence of CD3+/CD8 + cells in different region on prognosis was not simply linear. Although Immunoscore didn't remain statistically significant as an independent predictor of OS (Hazard Ratio = 4.76; P = 0.138) and PFS (Hazard Ratio = 1.83; P = 0.360) in multivariate Cox regression analysis, stratified Kapan-Meier curves by MMR status and clinical stage showed well separation.
Conclusion: Immunoscore can serve as a possible indicator in predicting prognosis of patients with early-onset colorectal cancer, but still need large sample research validation.
Keywords: Cellpose 2.0; Early onset colorectal cancer; Immunoscore; Machine learning.
© 2025. The Author(s).