Diagnostic Algorithms for Tuberculosis

2019年，1000万人患病，140万人死于结核病，2020年，结核病是成年人死于传染性病原体的第二大死因，仅次于COVID-19。2015年，世界卫生组织（世卫组织）启动了终结结核病战略，旨在到2035年将结核病病例减少90%。虽然这一战略为国家结核病控制方案提供了动力，但实现其雄心勃勃的目标的主要障碍之一是结核病诊断的机会很少。世卫组织建议，所有疑似结核病患者都应接受世卫组织推荐的分子诊断（WRD）检测。这些测试是敏感和具体的，但需要良好的实验室基础设施，通常位于遥远的参考实验室。然而，结核病是一种贫困疾病，大多数人在具有基本实验室设施的卫生中心寻求医疗建议，限制了获得WRD的机会。人们认识到，非常需要适合于资源有限的保健中心的诊断方法，这些方法简单，在当地能迅速提供结果。理想情况下，这些测试应该识别尽可能多的结核病患者（称为纳入测试），排除那些不太可能患有结核病的患者（称为排除测试），并识别需要在参考实验室进行测试的个人。虽然这些想法诊断是不可用的，一些新的诊断可以部分满足这些要求。因此，我们的目标是使用能够“修补”彼此缺陷的诊断组合，以开发在需要时有效的诊断方法。例如，Fujifilm的SILVAMP TB LAM（FujiLAM），一种基于尿液的测试，在尼日利亚发现了66%和70%的艾滋病毒感染者和非艾滋病毒感染者患有结核病，并在几分钟内得到诊断。我们还使用C-反应蛋白（CRP），一种炎症标志物，作为排除试验。通过10分钟的CRP测试，我们证明了CRP低的患者患结核病的可能性小于1%，允许快速排除约50%的患者的诊断。此外，尽管目前每个样本都用一个WRD检测盒进行检测，但可以将来自多名患者的样本合并在一起进行检测。如果合并检测结果为阴性，则认为所有样本均为阴性，而阳性样本将重新检测以识别阳性样本。我们还评估了池化方法，并表明它减少了工作负载和成本，使WDR的使用更有效。显然有潜力开发更有效的诊断算法。我们建议开发诊断算法，使用适合于设施有限的卫生设施的测试，这些测试是快速的，以患者为中心的，具有成本效益的，并与世卫组织目前基于WRD的诊断算法相结合。我们将从开发数学模型开始，该模型测试适合于低资源环境中的护理点的诊断组合的性能。我们将通过我们的研究或公共领域生成的数据来告知模型，并包括其性能的不确定性水平。我们将估计其敏感性、特异性和在当地和参考水平上实现诊断的患者比例。我们将模拟不同结核病和艾滋病流行率的性能表，以记录其预测价值以及它们在不同情况下的工作方式。我们还将与患者、卫生服务人员和利益攸关方一起评估这些模式的效率和可接受性，并在尼日利亚进行试点实施研究，以估计其成本效益。一旦诊断算法得到优化和参数化，我们将寻求资金在低收入环境中进行大规模临床试验。

In 2019, 10 million people fell ill and 1.4 million died from TB and in 2020 TB was the second cause of death in adults from an infectious agent, second only to COVID-19. In 2015, the World Health Organization (WHO) launched the End TB Strategy to reduce the number of TB cases by 90% by 2035. Although this strategy generated impetus for National TB control Programs, one of the major barriers to achieve its ambitious targets is the poor access to diagnostics for TB. WHO recommends that all people with presumptive TB should be tested with WHO-recommended molecular diagnostics (WRDs). These tests are sensitive and specific, but require a good laboratory infrastructure and are usually located in distant reference laboratories. TB however is a disease of poverty and most people seek medical advice at health centres with basic laboratory facilities, limiting access to WRDs. It is recognised that there is a major need for diagnostics suitable for health centres with limited resources, which are simple and provide rapid results locally. The tests should ideally identify as many patients with TB as possible (called a rule-in test), exclude those unlikely to have TB (called a rule out test) and identify individuals who need tests at the reference laboratories. Although these idea diagnostics are not available, a few new diagnostics can partially meet these requirements. We therefore aim to use combinations of diagnostics that are able to 'patch up' the deficiencies of each other's' to develop diagnostic approaches that are efficient at the point of need. For example, Fujifilm's SILVAMP TB LAM (FujiLAM), a urine-based test, identified 66% and 70% of people with and without HIV who had TB in Nigeria, reaching a diagnosis within minutes. We have also used C-Reactive Protein (CRP), a marker of inflammation, as a test of exclusion. With a 10-minute CRP test we demonstrated that patients with low CRP have less than 1% chance of having TB, allowing the rapid exclusion of the diagnosis in about 50% of patients. Moreover, although currently each sample is tested with one WRDs cartridge, samples from several patients can be pooled together for testing. If a pooled test is negative, all samples are considered negative, while positive samples are re-tested to identify the positive sample. We have also assessed the pooling method and shown it reduces workloads and costs, making the use of WDRs more efficient. There is clearly potential to develop more efficient diagnostic algorithms. We propose to develop diagnostic algorithms that use tests suitable for health facilities with limited facilities, which are rapid, patient-centred, cost-effective and articulated with the WHO current diagnostic algorithms based on WRDs. We will start by developing mathematical models that test the performance of combinations of diagnostics that are suitable for the point of care in low resource settings. We will inform the models with data generated by our studies, or in the public domain, and include levels of uncertainty of their performance. We will estimate their sensitivity, specificity and proportions of patients achieving a diagnosis at the local and reference level. We will simulate tables of performance with varying TB and HIV prevalence, to document their predictive value and how they would work in different scenarios. We will also evaluate the efficiency and acceptability of the models with patients, health service staff and stakeholders and conduct a pilot implementation study in Nigeria to estimate their cost-effectiveness. Once the diagnostic algorithms are optimised and parametrised, we will seek funding to conduct large scale clinical trials in low income settings.