MICA: Radio+ TB trial development: Determining the optimal regimen for bacteriologically negative, radiographically apparent TB diagnosed usingnewtest

结核病（TB）是全球传染病死亡的主要原因，大多数病例发生在非洲和亚洲。现在，国际社会对在未来15年内结束结核病流行作出了强有力的政治承诺。目前，在世界上估计的1000万结核病例中，每年只有70%（700万）被发现并开始接受治疗。终结结核病战略在未来几年的一个主要目标是“找到这些失踪的数百万人”，使至少90%的结核病病例开始接受治疗。要做到这一点，人们认识到，我们需要采取更积极主动的方法，并需要拥抱技术进步。数字胸部X光（图像出现在计算机屏幕上而不是电影）越来越多地被推荐为一种重要的筛查工具，因为它可以在患者出现任何症状之前发现肺部疾病的最早迹象，从而确定需要进行结核病测试的人（咳嗽痰样本）。每年，非洲和亚洲有数百万人正在接受数字胸部X光检查（仅在巴基斯坦的一个城市就有超过40万人）。虽然这已经被证明是非常有效的，但这种方法的一个新问题是，它发现很多人在胸部X光片上看起来像是结核病，但他们的痰液测试是阴性的。我们知道其中一些人在随后的几年中患结核病的风险很高，但目前还没有商定的简化方法来治疗这些环境中的这一群体，他们往往没有得到随访，成为结核病病例丢失的一部分。部分问题是超过35年来没有对这一群体进行临床试验。我们希望在现代试验中直接解决这个问题，以找出需要多少药物治疗来治疗胸部X光检查异常但痰液结核病检测阴性的人，以阻止他们在未来发展为结核病。我们将使用两种新的测试来确定那些最有可能患有结核病并从治疗中受益的人，如果发现效果良好，这些测试已经准备好使用。首先，我们将使用人工智能软件来分析胸部X光图像，看看模式是否表明结核病。这个软件以低成本在几分钟内提供答案，现在是如此复杂，它比人类读者表现得更好，更一致。如果软件认为胸部X光片看起来像结核病，我们将对少量血液进行第二次测试（最终可能只是刺破手指）。这种血液样本被直接放入一个药筒中，并装载到一台机器上，分析表明身体正在与结核病作斗争的遗传模式，并在2小时内得到答案。这种新的血液检测是由同一家公司（Cepheid）生产的，与最广泛使用的结核病痰液检测（Xpert）在同一台机器上运行，因此该机器甚至可以在移动的X射线车上找到，这些X射线车前往难以到达的社区以发现结核病。为了进行这项试验，我们需要解决我们建议在这项开发资助中解决的一些关键问题，即：1）有几个人工智能软件包已经以不同的方式进行了训练。什么是最好的人工智能软件包，为我们的试验，以检测和预测结核病在我们感兴趣的设置？2)增加新的结核病血液检测将在多大程度上改善这种检测/预测？3)如果一千人接受X光筛查，人工智能软件会发现多少人异常，其中有多少人的结核病血液检测呈阳性？4)患者和他们的医生对我们的方法有什么看法？他们在治疗中最看重什么？持续时间短，成功率高还是诊断确定？5)我们应该在哪里进行审判？哪一种环境能够正确地结合足够的患者和专业知识来进行试验？

Tuberculosis (TB) is the leading cause of death from an infectious disease globally, with most of the cases in Africa and Asia. There is now a strong, international political commitment to ending the TB epidemic over the next 15 years. At the moment, each year only 70% (7 million) of the world's estimated 10 million TB cases are picked up and started on treatment. A major goal of the End-TB strategy over the next few years is to "find these missing millions" starting at least 90% of TB cases on treatment. To do this it is recognised that we need a much more proactive approach and need to embrace technological advances. Digital chest X-Ray (where image appears on a computer screen rather than a film) is increasingly being recommended as an important screening tool to do this as it can pick up the earliest signs of disease on the lungs before the patient has any symptoms and therefore identify people who need testing for TB (on a coughed up sample of sputum). Each year millions of people in Africa and Asia are now being screened for TB with digital chest X-Ray (>400,000 in one city in Pakistan alone). Although this has proved very effective, an emerging issue with this approach is that it finds lots of people who look like they have TB on the chest X-ray but their sputum test is negative. We know some of these people are at very high risk to developing TB in subsequent years, but there is currently no agreed simplified approach on how to treat this group in these settings and they often are not followed up and become part of the lost TB cases. Part of the problem is than no clinical trials have been done for this group for over 35 years. We want to directly address this problem in a modern trial to find out, how much of what drug treatment is needed to treat people who have abnormal chest X-Ray but negative sputum tests for TB to stop them from developing TB in the future. We will use two new tests to identify those most likely to have TB and benefit from treatment, these tests are ready to be used if found to work well. Firstly we will use software which uses artificial intelligence to analyse the chest x-ray image to see if the pattern suggests TB. This software provides an answer in minutes at low cost and now is so sophisticated that it performs better and more consistently than human readers. If the software thinks the chest X-Ray looks like TB, we will conduct a second test on a small volume of blood (ultimately may only be a finger prick). This blood sample is placed directly into a cartridge and loaded onto a machine and analysed for a genetic pattern which suggests the body is fighting TB, with an answer available in 2 hours. This new blood test is made by the same company (Cepheid) as the most widely used sputum test for TB (Xpert) and run on exactly the same machine, therefore the machine is even found on mobile X-ray vans that travel into hard to reach communities to find TB.In order to undertake this trial we need to address some key questions which we propose to do in this development grant namely:1) There are several artificial intelligence software packages that have been trained in different ways. What is the best artificial intelligence software package for our trial, to detect and predict TB in the settings we are interested in?2) By how much will adding the new TB blood test improve this detection/prediction?3) If a thousand people were screened with X-Ray how many will the artificial intelligence software call abnormal and how many of those will have a positive TB blood test? 4) What do patients and their doctors think of our approach and what would they value the most in treatment? Short duration, high success rate or certainty of diagnosis?5) Where should we conduct our trial? Which setting has the right combination of seeing enough patients and expertise to conduct a trial?