Early identification of high-risk myelosuppression subtypes in DLBCL patients undergoing R-CHOP chemotherapy

Despite prophylactic interventions, chemotherapy-related myelosuppression remains an important clinical concern in patients with diffuse large B-cell lymphoma (DLBCL) receiving R-CHOP. This study characterized its occurrence patterns, baseline predictors, temporal risk, and dynamic heterogeneity.

We established an 8-year longitudinal cohort of patients with DLBCL who received standard-dose R-CHOP between 2016 and 2024. Endpoints included WHO-defined grade ≥1 and grade ≥3 myelosuppression and severe lineage-specific complications. Correlation analyses evaluated associations between baseline laboratory parameters and HGB-, NEUT-, PLT-, and WBC-related suppression. Multiple machine-learning models were developed for each endpoint and hematologic dimension. L2-regularized Cox proportional hazards models were used for time-to-event analyses, with supplementary time-dependent Cox analyses evaluating G-CSF exposure for NEUT-related endpoints. K-means clustering of longitudinal hematologic features was performed to identify dynamic subphenotypes.

Among 262 patients, advanced Lugano stage and the high-risk IPI subgroup were associated with several severe myelosuppression outcomes, while sex, infection status, and functional or anthropometric characteristics showed endpoint-specific associations. HCT, HGB, and RBC were negatively associated with HGB- and PLT-related myelosuppression and grade ≥3 NEUT-related myelosuppression, whereas RDW-CV and RDW-SD were positively associated with several severe outcomes. Models for HGB- and PLT-related endpoints achieved AUCs >0.80 for grade ≥1, grade ≥3, and corresponding severe-complication outcomes, whereas NEUT-related models showed lower discrimination. Baseline HCT was independently associated with lower hazards of grade ≥1 HGB-related myelosuppression (HR = 0.918, 95% CI 0.894-0.943), grade ≥3 HGB-related myelosuppression (HR = 0.910, 95% CI 0.882-0.940), and severe anemia (HR = 0.934, 95% CI 0.902-0.966). RDW-CV independently increased the hazards of grade ≥1 and grade ≥3 PLT-related myelosuppression, while baseline NEUT# reduced the hazard of grade ≥1 NEUT-related myelosuppression (HR = 0.835, 95% CI 0.760-0.918; FDR-adjusted p = 0.0026). G-CSF exposure was not significantly associated with NEUT-related endpoints after endpoint-level FDR correction. Clustering identified a high-severity, high-variability subtype and a comparatively mild, stable subtype; recovery patterns differed across hematologic dimensions.

R-CHOP-related myelosuppression is heterogeneous in occurrence, temporal evolution, and recovery. Baseline clinical and laboratory features may support early risk stratification, dynamic monitoring, and individualized supportive care.