Statistics-based Optimization Methods for Adaptive Interdisciplinary Pain Management

慢性疼痛是一个多方面的问题，影响着近40%的美国人口，据估计，每年会产生5600亿至6350亿美元的增量医疗保健成本，以及与康复和生产力损失相关的成本。随着越来越多的治疗选择和新的药物，制定一个基于证据的、个性化的治疗计划变得越来越复杂。本研究项目为适应性跨学科疼痛管理开发了新的基于统计的优化方法，该方法使用患者数据来推荐跨学科治疗方案以控制疼痛结果。这项研究涉及来自阿灵顿德克萨斯大学（UT）的一个跨学科研究小组，并使用了达拉斯德克萨斯大学西南医学中心的数据。由于UT西南使用流行和标准化的数据集，该研究将适用于全国约100个疼痛中心，并将影响跨学科疼痛管理的实施方式。该研究涉及统计优化中的四个主要主题。一是研究如何协调统计元模型和优化算法的发展。具体来说，将开发几个元模型，它们是底层系统的准确表示，并且可以进行全局优化。第二个主要主题涉及使用一种学习方法来推算缺失值，其中缺失位置位于块结构中。第三，研究了一种复杂数据的逆处理概率加权估计方法。最后，利用仿真方法对两种预测模型优化后的治理方案进行了评价。虽然这项研究是关于疼痛管理的，但该项目开发的方法可能被推广到创建复杂的适应性治疗方案，以治疗疼痛以外的疾病。

Chronic pain is a multidimensional problem that affects nearly 40 percent of the United States population and is estimated to incur $560-$635 billion in incremental healthcare costs and costs related to rehabilitation and lost productivity annually. With a growing number of treatment options and new medications, formulating an evidence-based, individually-tailored treatment plan has become increasingly complex. This research project develops new statistics-based optimization methods for adaptive interdisciplinary pain management that use patient data to recommend an interdisciplinary treatment regime for controlling pain outcomes. The research involves an interdisciplinary research team from the University of Texas (UT) at Arlington and uses data from the UT Southwestern Medical Center at Dallas. Because UT Southwestern uses a prevalent and standardized dataset, the research will be applicable to approximately 100 pain centers across the nation and will influence how interdisciplinary pain management is implemented. The research addresses four major topics in statistical optimization. One, the research studies how to coordinate the development of statistical meta-models and optimization algorithms. Specifically, several meta-models will be developed, which are accurate representations of the underlying system and can be globally optimized. The second major topic involves using a learning method to impute missing values in which the missing locations are in a block-wise structure. Three, the research develops an inverse-probability-of-treatment weighted estimators method for complex data. Finally, the research evaluates treatment solutions found from optimizing one prediction model with the other prediction models using simulation. Although this research is on pain management, the developed methodologies of this project can potentially be generalized to create complex adaptive treatment regimes for ailments other than pain.