Development and validation of a multi-feature immune scoring prognostic model for stage II and III rectal cancer patients.

Abstract

e15636 Background: Stage II and III rectal cancer patients undergoing direct surgery after clinical diagnosis remain at risk of postoperative recurrence, with significant variability in their prognoses. Immune scoring, based on the assessment of immune cells within the tumor microenvironment, offers a promising approach to predict outcomes and guide clinical decision-making. Methods: Multiplex immunohistochemistry (mIHC) was performed to simultaneously stain CD3, CD8, and PANCK on a single slide. Besides, an algorithm was developed for image segmentation, region identification, and cell recognition, resulting in the extraction of 429 features. The Lasso-Cox regression model was applied for feature selection and training, leading to the development of a multi-feature immune scoring prognostic model. Results: The training set for modeling included 384 samples, while the single-center validation set comprised 165 samples, and the multi-center test set contained 157 samples. For the disease-free survival (DFS) model, the Lasso-Cox regression model with 10-fold cross-validation was applied to the training set for feature selection, using partial likelihood deviance as the evaluation function. Features corresponding to the lambda value with the minimum partial likelihood deviance were selected, resulting in 18 features being incorporated. The multi-feature ensemble model’s prediction signatures were identified as significant prognostic factors for DFS in rectal cancer patients who underwent surgery (p < 0.001, HR 0.42, 95% CI 0.29–0.60). For the overall survival (OS) model, the Lasso-Cox regression model with 10-fold cross-validation was used for feature selection in the training set, with the C-index as the evaluation metric. Features corresponding to the lambda value with the maximum C-index were selected, incorporating 19 features. The prediction signatures of the multi-feature ensemble model were significant prognostic factors for OS in rectal cancer patients who underwent surgery (p < 0.001, HR 0.37, 95% CI 0.24–0.58). Notably, the multi-feature model outperformed patient stratification based on the Lancet Immunoscore. Specifically, the hazard ratio (HR) for the 360-μm margin dichotomy was 0.49 for DFS and 0.53 for OS, while the HR for the 500-μm margin dichotomy was 0.49 for DFS and 0.52 for OS. Additionally, the C-index for the multi-feature model was 0.67 for DFS and 0.74 for OS, demonstrating its superior prognostic performance. Conclusions: The DFS-based model enables the assessment of postoperative recurrence risk in patients, while the OS-based model provides predictions of patient prognosis. Together, the multi-feature immune scoring prognostic model demonstrates significant potential for clinical application.