Understanding and Predicting Loss to Follow-up from Multi-Drug Resistant Tuberculosis Treatment in the Setting of High-HIV Burden

项目摘要 在全球范围内，结核病（TB）是导致死亡的主要传染性原因之一，也是各国特别关注的问题 艾滋病病毒感染率很高。只有57%的病例得到成功治疗，耐多药结核病（MDR-TB） 已经成为结核病控制的一个实质性障碍。高失访率（LTFU）（即，缺少两个或更多 连续几个月的治疗）是造成耐多药结核病治疗成功率低的主要原因。LTFU 可能导致额外的抗生素耐药性、耐多药结核病治疗失败和死亡。世界卫生组织 建议优先关注有LTFU风险的患者，但目前尚无证据支持 来识别这些患者。为了解决这一差距，拟议的研究将开发一个预测模型 基于治疗开始时存在的特征，评估MDR-TB治疗的LTFU。如果准确的话，这个模型 将确定LTFU的高风险患者以及从干预中获益最大的患者 促进护理参与和保留。虽然LTFU的原因很复杂，但过去的研究表明， 产生了一些潜在的预测因素，将告知拟议的预测模型，包括男性性别，年龄， 住房不稳定性、酒精使用、物质使用、就业状况、教育水平、农村居住地和 结核病发作。除了治疗开始时存在的因素外，LTFU与以下因素之间的关系 将检查整个治疗过程中的变化，包括不良治疗事件和治疗方案 更广泛地了解耐多药结核病护理工作。拟议的研究将嵌套在 南非耐多药结核病患者随机分组试验的对照组（R 01 AI 104488）。具体目标 这项拟议的研究题为“了解和预测非洲国家耐多药结核病治疗后的失访情况”， 高HIV负担的设定”，在LTFU患者中进行一项巢式回顾性队列研究 或成功完成耐多药结核病治疗（即，治愈或完成治疗），以：（1a）制定预测 基于治疗开始时可用的患者特征的耐多药结核病治疗的LTFU模型， LASSO回归和k折交叉验证;（1b）将目标1a中开发的预测模型调整为工具 可由提供者在护理点用于估计患者的LTFU风险;（1c）确定是否存在 治疗方案是LTFU的一个风险因素，如果它改善了目标1a中开发的预测模型的拟合度; 以及（2）检查LTFU与不良治疗作用的时间和负担之间的关系。这 这项研究将是第一个采用预测建模方法来指导耐多药结核病提供者识别患者的研究 长期随访的高风险患者，并优先考虑他们接受支持服务，以最终改善耐多药结核病 在资源有限的情况下的治疗结果。通过拟议的学习和培训计划，申请人将 获得分析大型复杂纵向数据和应用机器学习优化患者的经验 参与和临床护理。

Project Summary Globally, Tuberculosis (TB) is one of the leading infectious causes of death and a particular concern in countries with a high HIV burden. With only 57% of cases being successfully treated, multi-drug resistant-TB (MDR-TB) has become a substantial barrier to TB control. High rates of loss to follow up (LTFU) (i.e., missing two or more consecutive months of treatment) are a major contributor to the low MDR-TB treatment success rates. LTFU may lead to additional antibiotic resistance, MDR-TB treatment failure, and death. The World Health Organization recommends that patients at-risk for LTFU be given priority attention, but there is currently no evidence-based way to identify these patients. In order to address this gap, the proposed study will develop a prediction model for LTFU from MDR-TB treatment based on characteristics present at treatment initiation. If accurate, this model will identify the patients who are at high-risk for LTFU and who will draw the greatest benefit from interventions that promote care engagement and retention. Although the reasons for LTFU are complex, past research has yielded a number of potential predictors that will inform the proposed prediction model, including male sex, age, housing instability, alcohol use, substance use, employment status, education level, rural residence, and prior episode(s) of TB. In addition to factors present at treatment initiation, the relationship between LTFU and factors that change throughout treatment, including adverse treatment events and treatment regimen, will be examined to develop a broader understanding of MDR-TB care engagement. The proposed study will be nested within the control arm of a cluster-randomized trial of MDR-TB patients in South Africa (R01 AI104488). The specific aims of the proposed study, titled “Understanding and Predicting Loss to Follow-up from MDR-TB Treatment in the Setting of High-HIV Burden”, are to conduct a nested, retrospective cohort study among patients who were LTFU or successfully completed MDR-TB treatment (i.e., cured or completed treatment) to: (1a) develop a prediction model for LTFU from MDR-TB care based on the patient characteristics available at treatment initiation utilizing LASSO regression and k-fold cross-validation; (1b) adapt the prediction model developed in Aim 1a into a tool that can be used by providers at the point of care to estimate a patient’s risk for LTFU; (1c) determine if type of treatment regimen is a risk factor for LTFU and if it improves the fit of the prediction model developed in Aim 1a; and (2) examine the relationship between LTFU and the timing and burden of adverse treatment effects. This study will be the first to take a predictive modeling approach to guide MDR-TB providers in identifying patients at high-risk for LTFU and prioritizing their receipt of support services in order to ultimately improve MDR-TB treatment outcomes in resource-limited settings. Through the proposed study and training plan, the applicant will gain experience analyzing large, complex longitudinal data and applying machine learning to optimize patient engagement and clinical care.