Prognostic Performance of the FLAMB Model in Primary Gastric Diffuse Large B-Cell Lymphoma

Aims/Background Primary gastric diffuse large B-cell lymphoma (PG-DLBCL) is a common gastrointestinal malignancy. While rituximab plus cyclophosphamide, doxorubicin, vincristine, and prednisone (R-CHOP)-based regimens have improved survival, reliable prognostic tools remain scarce. The International Prognostic Index (IPI), though validated for nodal DLBCL, shows limited accuracy in PG-DLBCL. To address this, we developed an intelligent prognostic model integrating key clinical variables to optimize individualized risk stratification, particularly for resource-limited clinical settings. Methods A retrospective cohort study was conducted at Nanfang Hospital, Southern Medical University, enrolling patients diagnosed with PG-DLBCL between January 2007 and July 2022. Clinical data and survival outcomes were systematically collected. Optimal cut-off values were systematically determined for continuous variables using receiver operating characteristic (ROC) curve analysis. Survival rates were estimated using the Kaplan-Meier method, with survival curves plotted and univariate survival associations assessed using the log-rank test. Multivariable analyses were performed through Cox proportional hazards regression and random forest algorithms. A novel (ferritin, lactate dehydrogenase (LDH), age, monocyte count (mono), β2-microglobulin (β2-MG)) FLAMB prognostic model was constructed by integrating the Cox regression model with random forest classification. Model performance was evaluated by comparing its discriminative accuracy with that of the IPI scoring system. Results Statistically significant differences in 5-year survival among PG-DLBCL patients were observed for ferritin, age, mono, LDH, β2-MG, B symptoms, and cell of origin (COO) in univariate survival analysis (p < 0.05). We developed the FLAMB model using five routinely available variables (ferritin, LDH, age, mono, and β2-MG) to enhance risk stratification in PG-DLBCL. Compared to the IPI, FLAMB demonstrated superior discriminative power (C-index: 0.653 vs. 0.637, Δ = 1.6%) and more effectively identified high-risk patients requiring treatment intensification. This enhanced risk stratification was confirmed by a statistically significant log-rank test (p < 0.05). Survival analysis in subgroups of non-germinal center B-cell like (GCB) and B symptoms-negative patients yielded consistent results. Conclusion The newly developed FLAMB prognostic model offers more precise prognostic stratification than the IPI for patients with PG-DLBCL. FLAMB comprises five key variables derived from routine laboratory tests and general clinical characteristics, making it readily accessible. This model enables clinicians, particularly in primary care or community hospitals, to efficiently stratify patient risk, assess underlying disease severity, and inform timely treatment planning.