基于营养状态（CONUT）的列线图构建与验证：预测复发/难治性多发性骨髓瘤 CAR-T 细胞治疗后的持续性血液学毒性

This retrospective study included 302 consecutive patients with R/R MM who received CAR-T cell therapy. Patients were randomly allocated to training and validation cohorts (7:3 ratio). The primary endpoint was prolonged grade 3/4 neutropenia >28 days; predictors were identified using logistic regression. The model's performance was assessed by the area under the curve (AUC), calibration curves, and decision curve analysis (DCA).

Multivariable analysis confirmed four independent predictors for the primary endpoint (prolonged grade 3/4 neutropenia >28 days): high tumor burden ( p = 0.013), ferritin ( p = 0.002), interferon- (IFN- , p = 0.018), and CONUT score ( p = 0.011). The nomogram built on these factors demonstrated a bias-corrected AUC of 0.815 in the training cohort, which was superior to the CAR-HEMATOTOX model (AUC: 0.706, p < 0.001). The predictive performance remained robust in the internal validation cohort (AUC: 0.824). The calibration curves showed good agreement between prediction and observation, and DCA confirmed the clinical utility of the model. The nomogram also exhibited excellent discriminative ability for predicting a composite PHT endpoint (AUC: 0.821, p = 0.417).

We developed a validated nomogram that incorporates the baseline CONUT score and key clinical variables (e.g., tumor burden, ferritin, IFN- ) to effectively predict PHT risk in R/R MM patients after CAR-T cell therapy, thereby facilitating early risk stratification and guiding personalized management.