Developing and evaluating a framework for the rational design of antibiotic prescribing policies in resource-constrained hospital settings

当住院患者有严重的细菌感染时，他们通常会被给予抗生素。在富裕国家，三分之一的时间抗生素是无效的。这通常是因为细菌获得了一种使他们对这种抗生素产生抗药性的基因。虽然检测这种耐药性是可能的，但需要三四天才能得到结果。对于严重感染的患者来说，这种延迟可能是生死攸关的。在低收入国家，情况可能也是如此，但获得的数据很少。然而，我们确实知道，在贫穷国家的医院里的患者更容易感染，当他们感染时，他们更有可能死亡。由抗药性细菌引起的感染也是一个主要问题。除了延误有效的治疗外，这种耐药性还可能在低收入环境中夺走生命，因为无法获得剩余的有效抗生素。即使是这样，对许多患者来说也可能太贵了。这项研究旨在解决这样一个问题，即我们如何在患者需要的时候更多地给他们有效的抗生素，而在资源有限的医院里，当他们不需要的时候，我们如何更少地给他们有效的抗生素。我们还想找出不同的抗生素是如何影响最危险的耐药细菌的传播的，我们想看看通过改变抗生素处方的模式，我们是否可以减少感染耐药细菌的数量。拟议工作的一部分将使用患者数据(年龄、住院时间、最后一次住院日期等)，并寻找有助于预测感染不同类型细菌的可能性的模式。例如，我们知道，长期住院的患者更有可能感染耐药细菌。我们可以利用这些信息来帮助选择哪种抗生素最有可能有效。我们的直觉是，通过使用计算机模型，我们可以优化利用这些信息，并比目前更频繁地选择有效的抗生素。医生在开抗生素时必须考虑的第二个因素是，这将如何影响其他患者。原因是，抗生素的使用越多，就越能创造出有利于抗药性细菌的环境。一般说来，抗生素使用量的增加与该抗生素耐药性的增加有关。然而，细节很复杂：一些抗生素比其他抗生素更容易产生耐药性，有时使用一种抗生素会导致对另一种完全不同的抗生素的耐药性增加。为了帮助设计好的抗生素政策，我们需要更好地理解这些复杂的机制。因此，这项工作的另一部分将使用计算机建模方法和新的统计技术来开发和应用更好的方法，以了解耐药性水平如何随着抗生素使用的变化而变化。下一阶段的研究将结合这些计算机模型，并广泛利用传染病医生的专业知识，为两家医院设计最佳的抗生素政策。我们将通过两种方式评估新政策：首先，我们将运行计算机模拟，使用来自医院的真实数据来预测新政策的执行情况。如果它的表现比目前的做法差，我们将重新设计政策，直到它表现得更好。然后，在其中一家医院，我们将进行一项干预研究，引入新政策，并评估它是否真的像预测的那样改善抗生素处方和减少耐药性。最后，使用新的快速测试来帮助确定导致感染的细菌类型，可能意味着患者在需要时有更多机会获得有效的抗生素治疗，而不必要的治疗机会更少。我们将使用之前开发的计算机模型来估计患者将从此类测试中受益多少，并评估哪些测试具有良好的性价比。

When hospitalised patients have a serious bacterial infection, they are usually given antibiotics. In rich countries, one third of the time the antibiotics are ineffective. This is often because the bacteria have acquired a gene that makes them resistant to that antibiotic. While it is possible to test for such resistance, it takes three or four days to get a result. For patients with serious infections this delay can be the difference between life and death.In lower income countries the same is likely to be true, but little data are available. We do, however, know that patients in hospitals in poorer countries get infections more often and when they do they are more likely to die. Infections caused by antibiotic-resistant bacteria are also a major problem. As well as delaying effective treatment, such resistance can claim lives in low-income settings because remaining effective antibiotics are not available. Even if they are, they may be too expensive for many patients. The research aims to address the question of how we can more often give patients effective antibiotics when they need them and less often when they don't in hospitals with limited resources. We also want to find out how different antibiotics affect the spread of the most dangerous antibiotic-resistant bacteria, and we want to see if by changing patterns of antibiotic prescribing we can reduce the number of infections with resistant bacteria.One part of the proposed work will use patient data (age, time in hospital, date last hospitalised, etc) and look for patterns that help to predict how likely infections with different types of bacteria are. For example, we know that patients who have been in hospital a long time are more likely to have infections with resistant bacteria. We can use this information to help choose which antibiotic is most likely to be effective. Our hunch is that by using computer models we can make optimal use of the information and choose an effective antibiotic more often than currently happens.The second consideration doctors have to take into account when prescribing antibiotics is how this will affect other patients. The reason is that the more an antibiotic is used the more it creates an environment favourable to antibiotic-resistant bacteria. In general, increasing antibiotic use is associated with increased resistance to that antibiotic. The specifics, however, are complicated: some antibiotics promote resistance much more than others, and sometimes use of one antibiotic can cause an increase in resistance against a completely different antibiotic. To help design good antibiotic policies we need to understand these complex mechanisms better. Another part of the work will therefore use a computer modelling approach and new statistical techniques to develop and apply better methods to understand how levels of resistance change in response to changing antibiotic use. The next stage of the research will combine these computer models and make extensive use of expertise from infectious disease doctors to design the best antibiotic policy we can for two hospitals. We will evaluate new policies in two ways: first we will run computer simulations, using real data from the hospitals to predict how well the new policy performs. If it performs worse than current practice we will redesign the policy until it performs better. Then, in one of the hospitals, we will perform an intervention study where we introduce the new policy and evaluate whether it really does improve antibiotic prescribing and reduce resistance as predicted.Finally, use of new rapid tests that help determine what type of bugs are causing an infection can mean a patient has more chance of getting effective antibiotic treatment when it is needed and less chance of unnecessary treatment. We will use the previously-developed computer models to estimate how much patients would benefit from such tests, and evaluate which would represent good value for money.