Key drug features.

Background
Rifampicin resistant tuberculosis remains a global health problem with almost half a million new cases annually. In high-income countries patients empirically start a standardized treatment regimen, followed by an individualized regimen guided by drug susceptibility test (DST) results. In most settings, DST information is not available or is limited to isoniazid and fluoroquinolones. Whole genome sequencing could more accurately guide individualized treatment as the full drug resistance profile is obtained with a single test. Whole genome sequencing has not reached its full potential for patient care, in part due to the complexity of translating a resistance profile into the most effective individualized regimen.
Methods
We developed a treatment recommender clinical decision support system (CDSS) and an accompanying web application for user-friendly recommendation of the optimal individualized treatment regimen to a clinician.
Results
Following expert stakeholder meetings and literature review, nine drug features and 14 treatment regimen features were identified and quantified. Using machine learning, a model was developed to predict the optimal treatment regimen based on a training set of 3895 treatment regimen-expert feedback pairs. The acceptability of the treatment recommender CDSS was assessed as part of a clinical trial and in a routine care setting. Within the clinical trial setting, all patients received the CDSS recommended treatment. In 8 of 20 cases, the initial recommendation was recomputed because of stock out, clinical contra-indication or toxicity. In routine care setting, physicians rejected the treatment recommendation in 7 out of 15 cases because it deviated from the national TB treatment guidelines. A survey indicated that the treatment recommender CDSS is easy to use and useful in clinical practice but requires digital infrastructure support and training.
Conclusions
Our findings suggest that global implementation of the novel treatment recommender CDSS holds the potential to improve treatment outcomes of patients with RR-TB, especially those with ‘difficult-to-treat’ forms of RR-TB.

研究背景：利福平耐药结核病（Rifampicin resistant tuberculosis, RR-TB）仍是全球性公共卫生难题，每年新增病例近50万。在高收入国家，患者通常先接受经验性标准化治疗方案，后续再根据药物敏感性试验（drug susceptibility test, DST）结果调整为个体化治疗方案。但在多数医疗场景中，DST检测结果难以获取，或仅能覆盖异烟肼（isoniazid）与氟喹诺酮类（fluoroquinolones）药物的敏感性检测。全基因组测序（whole genome sequencing, WGS）可通过单次检测获取完整的耐药谱，从而更精准地指导个体化治疗，但目前其在临床患者诊疗中的潜力尚未充分发挥，部分原因在于难以将耐药谱转化为最优的个体化治疗方案。
研究方法：本研究开发了一款治疗推荐临床决策支持系统（clinical decision support system, CDSS）及配套网页应用程序，旨在为临床医师提供便捷的最优个体化治疗方案推荐服务。
研究结果：经利益相关方专家研讨会及文献综述，研究团队确定并量化了9项药物特征与14项治疗方案特征。基于包含3895条“治疗方案-专家反馈”配对数据的训练集，通过机器学习方法构建了预测最优治疗方案的模型。本研究在临床试验与常规诊疗场景中评估了该治疗推荐CDSS的可接受性：在临床试验场景中，所有患者均接受了CDSS推荐的治疗方案；20例病例中有8例因药品缺货、临床禁忌证或药物毒性反应，对初始推荐方案进行了重新计算。在常规诊疗场景中，15例病例中有7例因推荐方案与国家结核病治疗指南不符，临床医师拒绝了该推荐。一项调研显示，该治疗推荐CDSS操作简便、临床实用性强，但需依托数字基础设施并开展相关培训。
研究结论：本研究结果表明，这款新型治疗推荐CDSS的全球推广应用，有望改善利福平耐药结核病患者的治疗结局，尤其针对那些“难治性”利福平耐药结核病亚型患者。