多药耐药/利福平耐药结核病患者早期痰液细菌负荷动态变化可预测治疗疗效：一项多中心前瞻性队列研究

Abstract

Objectives Early assessment of treatment response is critical for optimizing the management of multidrug-resistant/rifampicin-resistant tuberculosis (MDR/RR-TB). We aimed to establish a molecular bacterial load assay (MBLA) and evaluate early bacterial load dynamics for predicting 6-month treatment response. Methods A multicenter prospective cohort study enrolled patients with MDR/RR-TB in China. MBLA was established and validated against conventional culture methods. Generalized linear mixed models (GLMM) and artificial neural network (ANN) models were constructed by integrating 14-day and 2-month bacterial load changes, and evaluated using area under the curve (AUC), accuracy, sensitivity, specificity, and F1 score. Results MBLA enabled quantification within 24   hours and strongly correlated with conventional culture (r = 0.792, -0.790; both P < 0.001). Patients with 6-month sputum culture conversion showed a rapid decline in sputum bacterial load, especially within the first 14 days after treatment initiation. Sputum bacterial load changes over the first 14 days of treatment yielded acceptable prognostic performance for predicting 6-month treatment response, albeit inferior to those obtained at 2 months. The ANN model outperformed the GLMM, with testing set AUCs of 0.91 (14-day) and 0.98 (2-month). Conclusions MBLA enables rapid quantification of viable sputum bacterial load. The ANN model incorporating 14-day bacterial load dynamics achieves a favorable AUC for predicting 6-month sputum culture conversion. It may only serve as an auxiliary early-warning indicator to supplement routine clinical assessment for MDR/RR-TB patients. Prospective interventional trials and independent external validation are required before any consideration of its use to inform personalised anti-tuberculosis treatment decisions.