Monitoring, Control, and Adjustment of Non-Homogeneous Healthcare and Patient Data

本研究的目的是开发新的统计过程控制(SPC)和有界反馈调节(BFA)方法来处理医疗不良事件和表现出非同质性或自相关性的患者生理数据。例如手术并发症，每个患者的生存可能性不同，糖尿病患者的血糖水平需要控制，以将与目标值的偏差降至最低。风险调整后的研究将包括新的休哈特、EWMA和CUSUM方法，该方法基于不同二分事件的混合风险模型。将开发数字代码来计算准确的游程长度属性，并调查可能的近似的影响。反馈控制研究将SPC与死区调整图表(而不是连续调整)相结合，这是一个具有重要医疗应用的领域，仍然相对未被探索。这将包括确定有界调整的统计性质，以及评估将BFA和SPC结合起来的参数和非参数方法的性能。将为这些综合方法开发成本模型，并用于确定它们的最优同时设计，以及次优设计的后果和对模型错误指定的稳健性。开发的方法还将通过与三家学术医院的合作进行经验验证。这项研究的更广泛影响包括改善医疗过程的安全性，更好地控制患者的健康状况，以及显著降低相关成本。开发的方法将通过准确地考虑许多医疗保健过程中固有的自然统计行为，提供更强的检测重要变化的能力。重要的医疗不良事件包括用药错误、手术部位感染、呼吸机相关肺炎和错误部位手术，据估计，这些事件总共导致77万至200万名患者受伤，4.5万至9.8万人死亡，全国每年损失88亿美元。重要的反馈控制应用包括口服抗凝剂自我剂量、激素自我调节以及ICU患者的血细胞计数和血氧饱和度水平，在这些应用中，连续控制是不现实的，需要平衡调整的竞争成本、与预期水平的偏差以及延迟的变化检测。该项目的集成方法将优化对这些过程的控制，并减少检测系统性变化的时间和成本。这项研究也将惠及其他行业的类似问题，并将培养专注于医疗保健的研究生。

This research project is to develop new statistical process control (SPC) and bounded feedback adjustment (BFA) methods for healthcare adverse event and patient physiologic data that exhibit non-homogeneity or autocorrelation. Examples include surgery complications where each patient has a different survival likelihood and diabetic glucose levels that are desirable to control to minimize deviations from a target value. The risk-adjusted research will include new Shewhart, EWMA, and Cusum methods based on a mixed-risk model for non-identical dichotomous events. Numeric code will be developed to compute exact run length properties and investigate the effect of possible approximations. The feedback control research will integrate SPC with deadband adjustment charts (as opposed to continual adjustment), an area with important healthcare applications that remains relatively unexplored. This will include determining statistical properties of the bounded adjustments and evaluating the performance of parametric and non-parametric approaches to integrating BFA and SPC. Cost models will be developed for these integrated methods and used to determine their optimal simultaneous design, as well as the consequence of sub-optimal designs and robustness to model misspecification. Developed methods will also be validated empirically working with three academic hospitals.Broader impacts of this research include improved healthcare process safety, better control of patients' health status, and significant reduction in associated costs. The developed methods will provide greater ability to detect important changes by accurately accounting for the natural statistical behavior inherent in many healthcare processes. Important medical adverse events include medication errors, surgical site infections, ventilator-associated pneumonia, and wrong-site surgery, together estimated to result in 770,000 to 2 million patient injuries, 45,000 to 98,000 deaths, and $8.8 billion annually nationwide. Important feedback control applications include oral anticoagulant self-dosing, hormone self-regulation, and ICU patient blood counts and oxygen saturation levels, where continuous control is not practical and competing costs of adjustments, deviations from desired levels, and delayed change detection need to be balanced. The project's integrated methods will optimize the control of these processes and reduce the time and costs in detecting systemic changes. This research also will be benefit similar problems in other industries and will develop graduate students with a healthcare focus.