Beyond estimation: broadening the dynamic treatment regime literature

动态治疗机制是决定做出什么决定的一种统计方法。它们起源于医学，在医学中，针对个别患者进行量身定做的治疗--而不是给每个患有特定疾病的人开同样的药物--可以改善健康。这是因为两个不同患者的最佳治疗方法可能不同，例如使用某种药物的老年患者比年轻患者的情况更好，或者患者对副作用的耐受程度不同。通过分析数据，我们可以为每个患者找到最佳的动态治疗方案：一套规则，告诉我们在任何给定的时间根据单个患者的特征开出最佳的治疗方案。更重要的是，动态治疗机制的潜力不仅限于医学。在做出一系列决策的任何情况下(例如工程过程中的步骤，或科学实验中随时间推移的决策)，我们都可以分析数据以确定要采取的最佳行动。 在统计数据中，动态治疗制度在过去十年中得到了相当多的研究，看到了许多不同方法的发展，以在任何给定的环境中找到最佳决策序列。它的重点是解决在这个相对年轻的领域中虽然重要但在很大程度上仍未解决的问题。例如，在分析数据时，在统计分析“填补空白”之前，专家们会对不同的决定可能如何影响结果进行广泛的猜测。另一个问题是，在现实世界中，数据的测量往往存在误差(想想你测量一块木头的精确度有多高，或者你的汽车燃油表的精确度有多高)；这可能会严重破坏我们的结果，所以需要做一些工作来考虑这个问题。 这项研究将在统计和更广泛的研究界产生影响。对于前者，它将解决重要的理论问题，同时在不同的统计领域之间形成新的联系(这反过来又将推动整个领域的发展)。除了统计之外，这项工作虽然看似专门，但在决策过程中具有广泛的潜力。通过迅速扩大可以应用动态治疗制度的背景，来自不同专业的研究人员将找到更多使用这些方法并从中受益的机会。最后，该计划将培训许多高素质的个人，他们将继续为统计、更广泛的科学界、行业或整个加拿大做出自己的重要贡献。

Dynamic treatment regimes are a statistical way of deciding what decisions to make. They originate in medicine, where tailoring treatments to individual patients - rather than prescribing the same drug to everyone with a particular disease - can lead to improved health. This is because the best treatment for two different patients may not be the same, such as older patients faring better than younger ones using a certain medication, or patients having varying degrees of tolerance to side-effects. By analyzing data we can find the best dynamic treatment regime for each patient: a set of rules that tell us the best treatment to prescribe at any given time based on an individual patient's characteristics. What's more, the potential of dynamic treatment regimes is not limited to medicine. In any situation where a sequence of decisions is made (such as steps in an engineering process, or decisions over time in a scientific experiment), we can analyze data to work out the best actions to take. Within statistics, dynamic treatment regimes have received considerable study over the last decade, seeing the development of many different approaches to finding the best sequence of decisions in any given setting. This research program will expand this area of statistics into new, exciting directions. It focuses on solving problems that, though important, remain largely unaddressed in this relatively young field. For example, when analyzing data, specialists take a broad guess at how different decisions might affect an outcome, before statistical analyses 'fill in the blanks'. So far, relatively little work has been done in checking how good these broad guesses are. Another problem is that in the real world data are often measured with error (think how precisely you might measure a piece of wood, or the accuracy of your car's fuel gauge); this can badly undermine our results, so work needs to be done to take this problem into account. This research will have impact within statistics as well as the wider research community. For the former, it will address important theoretical problems while forming new links between different areas of statistics (which in turn will advance the field as a whole). Beyond statistics, though seemingly specialized, this work has wide-reaching potential in decision-making processes. By rapidly expanding the contexts where dynamic treatment regimes can be applied, researchers from different specialties will find more opportunities to use - and benefit from - these methods. Finally, this program will see the training of numerous highly qualified individuals, who will go on to provide important contributions of their own, be it to statistics, the wider scientific community, industry, or Canada as a whole.