Latent class analysis-derived classification improves the cancer-specific death stratification of lymphomas: A large retrospective cohort study.

Abstract

Lymphomas have diverse etiologies, treatment approaches, and prognoses. Accurate survival estimation is challenging for lymphoma patients due to their heightened susceptibility to non-lymphoma-related mortality. To overcome this challenge, we propose a novel lymphoma classification system that utilizes latent class analysis (LCA) and incorporates demographic and clinicopathological factors as indicators. We conducted LCA using data from 221,812 primary lymphoma patients in the Surveillance, Epidemiology, and End Results (SEER) database and identified four distinct LCA-derived classes. The LCA-derived classification efficiently stratified patients, thereby adjusting the bias induced by competing risk events such as non-lymphoma-related death. This remains effective even in cases of limited availability of cause-of-death information, leading to an enhancement in the accuracy of lymphoma prognosis assessment. Additionally, we validated the LCA-derived classification model in an external cohort and observed its improved prognostic stratification of molecular subtypes. We further explored the molecular characteristics of the LCA subgroups and identified potential driver genes specific to each subgroup. In conclusion, our study introduces a novel LCA-based lymphoma classification system that provides improved prognostic prediction by accounting for competing risk events. The proposed classification system enhances the clinical relevance of molecular subtypes and offers insights into potential therapeutic targets.