Health care operational research under uncertainty and data variability

几十年来，与医疗保健服务相关的复杂性和挑战已被证明是实践运筹学（OR）和工业工程的富有成效的领域。研究项目和产出推动了这些学科的科学发展，并改善了医疗保健业务。随着医疗保健提供商在数据分析（基础设施和专业人力资源）方面投入更多资金，并继续寻求大学帮助解决最复杂的挑战，这种合作关系的重要性将不断增加。随着数据的爆炸式增长，许多以前不确定的问题特征现在可以被量化。如何充分利用这一点，面对这些新数据中经常发现的高变异性，如何做出正确的决策，是本项目研究的重点。具体来说，这个研究项目的研究目标是：******提高我们从新的、丰富的、和不同的医疗保健数据源***将这些数据和信息用于推进医疗保健分析***将这些数据和信息用于减少不确定性和改进的OR模型***将从分析中获得的解决方案与OR模型进行比较和对比******一旦了解了底层流程，就可以开发OR方法以做出更好的决策。排队理论和排队网络模型将用于战略容量规划，并能够很好地确定哪种投资（例如，更多的资源、更快的服务时间、病人的改道等）将带来最大的改善。当有大量数据并且我们不限于仅由理论概率分布描述的数据（许多排队网络模型中的常见问题）时，将使用仿真和仿真优化。马尔可夫模型也非常适合于不确定环境下的优化，并将用于当潜在的系统动力学与泊松过程很好地建模时。******分析模型能够根据经验数据进行描述、预测和规定，而不需要全面了解底层结构。随机OR模型同样可以描述、预测和规定行为，但这样做需要对底层系统有更全面的了解（通常需要更少的数据）。因此，目标是相似的，但方法和所需的信息是不同的。在研究每一种方法的差异、相似之处、挑战和结果时，我的目标是描述每种方法最适合的问题类型，并推导出一个连接OR方法和分析方法的统一框架，以便它们如何相互补充可以被理解并用于推进这两个领域。利用新的数据以及OR和Analytics方法的共同优势将改善加拿大人的医疗保健服务。**

The complexity and challenges associated with healthcare delivery has proved a fruitful area to practice operational research (OR) and industrial engineering for several decades. Research projects and outputs have advanced the science of these disciplines and improved healthcare operations. The importance of this partnership is poised to grow as healthcare providers invest more in data analytics (infrastructure and professional human resources) and continue to look to universities to help with their most complex challenges. With the explosion of data, many previously uncertain problem characteristics can now be quantified. How to take full advantage of this and how to make good decisions in the face of the high variability often found in this new data is the focus of this program of research. Specifically, the research objectives of this program of research are:****** To enhance our ability to extract information related to operational processes and patient populations from new, abundant, and disparate healthcare data sources*** To use this data and information for advancing healthcare analytics*** To use this data and information for less uncertain and improved OR models*** To compare and contrast solutions obtained from Analytics with OR models ******Once the underlying processes are understood, OR methods can be developed to make better decisions. Queueing theory and queueing network models will be used for strategic capacity planning and are well equipped to determine which investment (e.g. more resources, faster service times, rerouting of patients, etc.) will result in the most improvement. Simulation and simulation optimization will be used when there are large amounts of data and we are not limited to data described only by theoretical probability distributions (a common problem in many queueing network models). Markov models are also well suited for optimizing in uncertain settings and will be used when the underlying system dynamics are well modelled with Poisson processes. ******Analytics models are able to describe, predict and prescribe based on empirical data and without a need for a comprehensive understanding of the underlying structure. Stochastic OR models can likewise describe, predict and prescribe actions but to do so requires a more complete understanding of the underlying systems (and typically less data). As such, the objectives are similar, but the approach and the information needed is different. In studying the differences, similarities, challenges, and outputs of each, I aim to characterize problem types where each is most appropriate and to derive a unifying framework connecting OR methods and Analytics methods such that how they complement each other can be understood and used to advance both fields. Capitalizing on new data and the joint benefits of OR and Analytics methods will lead to improved healthcare delivery for Canadians.**