A mathematical modeling framework for tuberculosis burden estimation and economic evaluation of pharmaceutical interventions

结核病（TB）是全球疾病和死亡的主要原因。2015年，世卫组织估计有960万结核病例和1.5例结核死亡。其中近50万例对两种或两种以上用于治疗结核病的主要药物具有耐药性。目前正在开发治疗结核病的新药和药物组合，以及可以预防成人疾病的新疫苗。量化结核病负担对于了解其全球流行病学和做出适当的资源分配决策至关重要。对每年新结核病例数的大多数估计主要依赖于各国当年向世卫组织报告的病例数。不幸的是，三分之一的结核病病例被认为未被发现或未报告，因此报告的病例数低估了新病例数。虽然人们可以纠正这一点，但很难确切知道要调整多少报告的数字。一些国家有良好的死因记录系统，可用于估计结核病造成的死亡人数。越来越多地使用大型和昂贵的流行率调查来估计人口中患有活动性疾病的人数。这些估计值受偏差的影响较小，但衡量的是不同的数量。目前还没有研究如何将这三种数据源结合起来，本文的主要目标是使用数学传输模型进行负荷估计，并为所有数据提供一个统一的框架。这些模型产生了新病例、死亡人数以及疾病流行率。它们明确表示疾病传播，因此在不同年份的新病例数之间引入了依赖性。这些模型涉及从以前的流行病学工作中得到证明的参数，但必须进行校准，以从结核病报告、死亡和流行率的数据中学习。校准意味着调整不完全已知的模型参数，以便将观察到的模型输出与数据相匹配。这个过程提供了一个模型，可以用来预测负担，但也可以教我们一些关于潜在过程的东西。有关结核病流行病学和病程的许多参数都相当不确定，这种不确定性很少在需要校准的模型中得到充分体现，但在本项目中将使用统计技术来完成，这些技术也允许比较不同模型的性能。结核病负担估计和传播模型的校准几乎总是在逐个国家的基础上进行。描述疾病进展的许多参数在不同国家可能相似，即使其确切值因未知原因而不同。分层建模技术使这些参数在各国之间相互关联。这可以提高准确性，特别是对于数据很少的国家，因为估计数可以参考邻国的数据。我将探讨这些技术的传播模型，并在统计建模，旨在占观察到的耐药模式。随着新的治疗方法和疫苗的出现，负责公共卫生的人将希望了解这些新技术在改善健康和改变支出方面的潜在影响。提供成本效益和预算影响证据需要一个包括传播的模型，以说明通过避免二次病例而产生的间接效益。我们将使用我们的模型为决策者提供指导，寻求以有限的资源最大限度地提高健康收益。我们还将分析模型中的不确定性来源，以确定未来的研究，这将在提高负担估计的准确性和减少决策的不确定性周围引入新的干预措施最有价值。

Tuberculosis (TB) is a major cause of disease and death globally. In 2015, WHO estimated there were 9.6 million TB cases and 1.5 TB deaths. Nearly 500,000 of these cases were resistant to two or more of the main drugs used to treat TB. New drugs, and combinations of drugs, are being developed to treat tuberculosis, as are new vaccines that may protect against disease in adults.Quantifying the burden of TB is fundamental to understanding its global epidemiology and for making appropriate resource allocation decisions. Most estimates of new TB case numbers each year rely strongly on the number of cases reported by countries in that year to WHO. Unfortunately, one in three TB cases are thought to go either undetected or unreported, so the number of cases reported underestimates the number of new cases. While one can correct for this, it is hard to know exactly how much to adjust the reported numbers. Some countries have good systems for recording causes of deaths, which can be used to estimate the number of deaths caused by TB. Increasingly, large and expensive prevalence surveys are being used to estimate the number of people with active disease in a population. These estimates are less subject to bias, but measure a different quantity. Little work has explored the best way of combining these three data sources.A major goal of this work is to use mathematical transmission models for burden estimation and provide a unified framework for all data. These models yield the number of new cases, deaths, and also the prevalence of disease. They explicitly represent disease transmission and so introduce a dependence between the number of new cases in different years. These models involve parameters evidenced from previous epidemiological work, but must be calibrated to learn from data on TB reports, deaths and prevalence. Calibration means adjusting imperfectly known model parameters in order to match observed model outputs to the data. This process provides a model that may be used to make predictions about burden, but may also teach us something about the underlying processes. Many of the parameters concerning the epidemiology and disease course of TB are quite uncertain, and this uncertainty is rarely represented fully in models needing calibration, but will be done in this project using statistical techniques that also allow comparison of different models' performance. TB burden estimation and calibration of transmission models are almost always carried out on a country-by-country basis. Many parameters describing disease progression are likely to be similar in different countries, even if their exact values differ for unknown reasons. Hierarchical modelling techniques allow such parameters to be correlated between countries. This can improve precision, particularly for countries with little data, as estimates can be informed by data from neighbouring countries. I will explore these techniques for the transmission model, and also in statistical modelling aiming to account for the observed patterns of drug-resistance. The transmission model will ultimately be extended to include different types of drug resistance.As new treatments and vaccines emerge, those with responsibility for public health will want to understand the potential impact these new technologies can have in terms of gains in health, and changes in spending. Producing cost-effectiveness and budget impact evidence requires a model that includes transmission, in order to account for indirect benefits accrued by avoiding secondary cases. We will use our model to provide guidance to decision-makers seeking to maximise health gain with limited resources. We will also analyse sources of uncertainty in the model to identify future research that would have most value in increasing the precision of burden estimates and in reducing decision uncertainty around the introduction of new interventions.