Abstract: Background Medication-taking behavior is crucial for the treatment of drug-resistant tuberculosis (DRTB),while loss to follow-up (LTFU)increases the risk of further transmission of DR-TB.Previous studies have lacked a systematic classification based on patient medication-taking behavior characteristics and an exploration of their association with LTFU. Objective To understand the medication-taking behaviors of patients with drug-resistant tuberculosis (DR-TB)in Guizhou Province,categorize them to construct patient profiles and explore the associations between these behavioral patterns and LTFU,in order to provide a reference basis for the development of DR-TB prevention interventions.Methods Data collection for this study was conducted from January to February 2024.The data were sourced from the National Tuberculosis Information Management System and the electronic medical record system.A retrospective analysis was performed on DR-TB patients with treatment outcomes of either success or LTFU from three designated hospitals in Guizhou Province between December 2019 and December2023.Two-step cluster (TSC)algorithm was performed on 12 medication-taking behavior characteristics to construct patient profiles.Univariate analysis and multiple binary Logistic regressions were applied to identify the associations between medicationtaking behavior clusters and LTFU behaviors.Results A total of 565 DR-TB patients were included，and 130 (23.01%)patients were LTFU.The characteristics were classified into four clusters:109 (19.29%)positive benefit-seeking type,151 (26.73%)local treatment type，152 (26.90%)high endurance type,and 153 (27.08%)low adherence type.Multiple binary Logistic regression analysis revealed that compared to the positive benefit-seeking type,high endurance type (OR=3.597,95%CI=1.536-8.426),low adherence type (OR=11.149,95%CI=5.146-24.156)，and local treatment type (OR=2.342,95%CI=1.011-5.427)clusters were significantly associated with an increased risk of LTFU.Furthermore,advanced age (OR=1.043,95%CI=1.023-1.063)，ethnic minority status (OR=2.063,95%CI=1 .281-3.323)，BMI 2 (OR=1.806,95%CI=1 .085-3.006),and unwillingness to report marital status (OR=11.030,95%CI=4.176-29.135)were also risk factors for LTFU.Conclusion Based on medicationtaking behavior characteristics,DR-TB patients can be categorized into four types:positive benefit-seeking type,local treatment type,high endurance type,and low adherence type.Compared with the positive benefit-seeking type,the other three types were associated with a higher risk of being LTFU.Tailored intervention strategies should be formulated based on distinct patient types to optimize DR-TB control programs.Prioritized management of high-risk subgroups，targeted reduction of LTFU，and reinforcement of treatment adherence protocols are critical measures to enhance treatment success rates.

Key words: Drug-resistant tuberculosis, Medication-taking behavior characteristics, Two-step cluster analysis, Patient profile, Loss to follow-up behavior

摘要： 背景 服药行为是耐药结核病（DR-TB）治疗的关键环节，而患者失访会增加DR-TB的传播风险。既往研究尚未结合患者服药行为特征进行系统分类，也缺乏此类分类与失访结局关联的探讨。目的 了解贵州省DR-TB患者的服药行为特征，对其进行分类画像，并分析各类服药行为与失访行为的关联，为制订DR-TB精准干预措施提供参考依据。方法 于2024年1—2月开展数据收集工作，数据来源于中国疾病预防控制信息系统结核病信息管理系统及电子病历系统。本研究回顾性分析2019年12月—2023年12月贵州省3家定点医院中，治疗结局为成功和失访的全部DR-TB 患者信息。针对12 项服药行为特征进行二阶聚类分析，并据此构建患者画像。采用单因素分析及多因素Logistic 回归分析，探究服药行为特征分类与失访行为的关联。结果 共收集DR-TB患者565例，发生失访行为的患者为130 例（23.01%），服药行为特征分为积极获益型109 例（19.29%）、本地就医型151 例（26.73%）、高忍耐型152 例（26.90%）和低依从型153 例（27.08%）4 类。多因素Logistic 回归分析结果显示，以积极获益型患者为对照组，高忍耐型（OR=3.597，95%CI=1 .536~8 .426）、低依从型（OR=11.149，95%CI=5 .146~24 .156）和本地就医型患者（OR=2.342，95%CI=1 .011~5 .427）发生失访行为的风险更高（P<0.05）；年龄较高（OR=1.043,95%CI=1 .023~1 .063）、少数民族（OR=2.063,95%CI=1 .281~3 .323）、BMI2（OR=1.806，95%CI=1 .085~3 .006）、不愿意报告婚姻状况（OR=11.030，95%CI=4 .176~29 .135），均为发生失访行为的危险因素（P<0.05）。结论 根据服药行为特征可将DR-TB 患者分为积极获益型、本地就医型、高忍耐型和低依从型4 类，与积极获益型相比，其他3 类患者具有更高的失访风险。今后的结核病防治工作应针对不同类型患者的特征，制订个性化的患者干预措施，加强对高风险患者的管理，减少失访行为的发生，提高患者治愈率。

关键词: 耐药结核病, 服药行为特征, 二阶聚类分析, 患者画像, 失访行为