Analytics for Managing Health Care Wait Lists: Predictive and Prescriptive Approaches

等待护理的时间过长是加拿大的一个紧迫问题。2004年，政府投资了55亿美元用于“减少等待时间基金”。然而，几份报告明确指出，仍需进一步大幅减少等待时间。该DG的目标是开发和应用预测性和规范性分析，以改善等待列表管理。作为我们工作的应用，我们将与BC儿童医院（BCCH）的外科医生和管理人员合作，改善等待择期手术的儿科患者的等待时间。在预测方面，我们有： 目标1：开发等待时间预测工具，为患者提供更准确的手术预期等待时间估计。 由于当预期和实际等待时间不匹配时，等待的挫折感往往会发生，目标1旨在改善个人的等待时间体验。此外，准确的等待时间估计可以帮助患者做出更好的“等待或离开”的决策，特别是当他们有外部护理选择时。 我们将应用排队理论，模拟和机器学习的工具来开发等待时间预测工具，用于加入手术等待名单的患者。我们已经开发了一个离散事件模拟（DES）模型的等待列表动态BCCH，我们将使用估计等待时间。然而，每次外科医生想要为患者提供等待时间估计时运行模型在计算上是昂贵的。相反，我们将离线运行DES，以生成从各种状态加入等待列表的患者的等待时间样本，从中我们将拟合回归和机器学习模型。最终的预测模型将在准确性、透明度和易用性之间取得平衡。 虽然有出版物描述了急诊科的等待时间预测模型，但我们不知道用于估计手术等待时间的预测模型。 在规范方面，我们有： 目的2：确定有效和实用的策略，汇集手术等待名单。 传统上，与患者进行初步咨询的外科医生也会亲自进行手术。虽然从持续护理的角度来看这可能是好的，但有些手术可能由具有相同专业知识的其他外科医生合理执行。这就提出了一个有趣的规范性问题，即如何最好地设计一个系统，让一些外科医生集中他们的等待名单。 我们将与外科医生密切合作，首先确定潜在的患者类型和可以合并的手术，然后我们将在模拟优化框架中使用我们的DES模型来确定有希望的系统设计变更。我们的目标将是减少手术等待时间，同时也仔细考虑实施的便利性。总干事的这一目标也将有助于工艺灵活性文献，通过将其扩展到医疗保健环境。

Excessive wait times for care are a pressing problem in Canada. In 2004, the government invested $5.5 billion dollars in a “Wait Times Reduction Fund.” However, several reports make it clear that further significant reductions in wait times are still needed. The goal of this DG is to develop and apply predictive and prescriptive analytics to improve wait list management. As an application of our work, we will work with surgeons and administrators at BC Children's Hospital (BCCH) to improve the wait time experience for pediatric patients awaiting elective surgery. On the predictive front, we have: Aim 1: Development of wait time prediction tools to provide patients with more accurate estimates of their anticipated wait times for surgery. Since frustration with waiting often occurs when there is a mismatch between anticipated and actual wait times, Aim 1 intends to improve individuals' wait time experiences. Furthermore, accurate wait time estimation may help patients make better “wait or leave” queueing decisions, especially when they have outside options for care. We will apply the tools of queueing theory, simulation, and machine learning to develop wait time prediction tools for patients joining a surgical wait list. We have already developed a discrete-event simulation (DES) model of wait list dynamics at BCCH, which we will use for estimating wait times. However, it is computationally expensive to run the model every time a surgeon wants to provide a wait time estimate for a patient. Instead, we will run the DES offline to generate wait time samples of patients joining the wait list from various states, from which we will fit regression and machine learning models. The final predictive model will strike a balance between accuracy, transparency, and ease-of-use as a bedside tool. While there are publications describing wait time prediction models for emergency departments, we are unaware of predictive models for estimating wait times for surgery. On the prescriptive front, we have: Aim 2: Identification of effective and practical strategies for pooling surgical wait lists. Traditionally, surgeons who conduct the initial consultation with a patient also perform the surgery itself. While this may be good from a continuity-of-care point of view, there are some procedures which may be reasonably performed by other surgeons with the same expertise. This raises the interesting prescriptive question of how best to design a system in which some surgeons pool their wait lists. We will work closely with surgeons to first identify potential patient types and surgeries that can be pooled, and then we will use our DES model in a simulation-optimization framework to identify promising system design changes. Our objective will be to reduce surgery wait times while also giving careful consideration to ease of implementation. This aim of the DG will also contribute to the process flexibility literature, by extending it to health care settings.