Machine learning-driven dynamic prognosis for primary colorectal lymphoma.

[UNLABELLED] Primary colorectal lymphoma (PCL) is a rare extranodal non-Hodgkin lymphoma that is often diagnosed at an advanced stage, resulting in poor prognosis and challenges in survival prediction. This study aimed to evaluate conditional survival (CS) outcomes and develop a dynamic CS-nomogram for individualized prognostic assessment. Data from the Surveillance, Epidemiology, and End Results (SEER) database (2004–2021) were used to identify 2,743 patients with PCL. CS analysis was integrated with a random survival forest (RSF) algorithm and multivariate Cox regression to construct a dynamic, individualized prognostic nomogram. Model performance was evaluated using calibration plots, time-dependent ROC curves, and decision curve analysis (DCA). Patients were stratified into risk groups according to nomogram-derived scores. Seven prognostic variables—age, histology, tumor stage, chemotherapy, tumor site, household income, and marital status—were identified through RSF. The nomogram demonstrated excellent calibration and consistent discriminatory performance, with AUC values remaining stable from 1 to 10 years in both the training and validation cohorts. Specifically, the 3-, 5-, and 10-year AUCs were 0.780, 0.789, and 0.812 in the training cohort, and 0.781, 0.780, and 0.808 in the validation cohort. CS analysis demonstrated progressively increasing survival probabilities with longer follow-up, reflecting the dynamic nature of prognosis. Risk stratification effectively distinguished groups with significantly different survival outcomes across a 10-year period. This RSF-based CS-nomogram provides accurate and dynamic survival predictions for PCL patients, outperforming traditional static models. Using a large population-based cohort, it supports personalized risk assessment and informed clinical decision-making. Future studies incorporating molecular profiling and prospective validation may further refine prognostic precision and guide individualized treatment strategies.

[SUPPLEMENTARY INFORMATION] The online version contains supplementary material available at 10.1038/s41598-026-36995-0.