Development and Validation of a Time‐Dependent Nomogram for Predicting Hematological Toxicity Risk During Linezolid Therapy in Drug‐Resistant Tuberculosis

ABSTRACT Background Linezolid is essential for drug‐resistant tuberculosis (DR‐TB) treatment but causes frequent hematological toxicity. We developed a time‐dependent nomogram to predict this risk. Methods A prospective cohort of 201 patients with DR‐TB receiving linezolid (600 mg/day) was enrolled. Blood trough concentration (C min ) and clinical variables were measured. Multivariable Cox regression identified predictors, with a nomogram constructed to estimate toxicity probability at 1, 3, and 6 months. Model performance was assessed via calibration curves, C‐index, and time‐dependent area under the curve (AUC). Results Ninety‐six patients (47.8%) developed hematological adverse events (anemia: 26.4%, leukopenia: 14.4%, thrombocytopenia: 7.0%). Five predictors were significant: C min > 2.08 mg/L [Hazard Ratio (HR) = 2.87, p < 0.001]; lower baseline white blood cells (HR = 0.84, p = 0.003), hemoglobin (HR = 0.99, p = 0.033), and creatinine clearance rate (HR = 0.99, p = 0.001); and initial treatment (HR = 0.56 vs. retreatment, p = 0.011). The nomogram showed good discrimination (C‐index = 0.73) and calibration. Time‐dependent AUCs were 0.74 (1‐month), 0.79 (3‐month), and 0.80 (6‐month). Internal validation via bootstrapping (1000×) confirmed robustness. Conclusions This nomogram, integrating C min with baseline clinical factors, enables early identification of patients with DR‐TB at high risk for hematotoxicity and could guide pre‐emptive interventions. However, external validation is required to confirm its generalizability before widespread clinical implementation.