复杂医疗保健数据的统计推断和过程控制

项目名称： 复杂医疗保健数据的统计推断和过程控制

项目编号： No.11301364

项目类型： 青年科学基金项目

立项/批准年度： 2014

项目学科： 数理科学和化学

项目作者： 刘浏

作者单位： 四川师范大学

项目金额： 22万元

中文摘要： 医疗保健数据的质量监控问题在统计过程控制（SPC）领域日益受到重视。传统SPC技术一般是基于简单低维数据及工业背景假设来进行设计。本项目研究的目标是针对一些高维、复杂、过度散布的医疗保健数据，发展相应的SPC新方法及软件包。研究内容将集中在如下三个方面：（1）高维医疗保健数据的SPC监控及诊断方案：针对这类数据，提出某些具有稀疏性质的充分降维方法，并与传统的多元SPC方法及现代参数、非参数统计方法相结合，进一步提出一些能够自动识别超高维数据飘移方向并快速有效实施的序贯检验诊断方法，提高传统多元SPC方法的检验和诊断效率；（2）复杂医疗保健数据的风险调整建模及其监控和诊断方法：基于目标的Parsonnet score，提出某些带有风险调整因子的混合效应模型，并利用拉普拉斯逼近及Score检验构造统计量，对此模型进行有效的监控和诊断；（3）偏大离差医疗保健数据的建模及其SPC监控和诊断方案。

中文关键词： 医疗保健数据；质量控制；手术质量；CUSUM；EWMA

英文摘要： Statistical process control (SPC) for effective detection of the medical and health-care data has increasingly attracted SPC researchers' attention. Most conventional SPC technologies are designed based on simple, low-dimension data and the industrial background. The central idea of the proposed research is to develop some new methodologies and software packages for high-dimension, complex and overdispersed data in medical and health-care area. This proposed research includes the following three research contents: (1) SPC monitoring and diagnostic methods for high-dimension medical and health-care data. Some sparse sufficient dimensional reduction (SSDR) method would be proposed for these data. Then, some automatic and efficient sequential diagnostic methods would be proposed for the data shifts via the combination of modern parametric or nonparametric statistical methods, traditional multivariate SPC methods and the above SSDR method. This would develop the monitoring and diagnostic efficiency of the conventional multivariate SPC methods; (2) Modeling, monitoring and diagnosing for risk-adjusted complex data. Some risk-adjusted mixed models would be proposed based on Parsonnet sore and some statistics would be constructed by score test and Laplace approximation to efficiently monitor and diagnore this model; (

英文关键词： Healthcare Data；Quality Control；Surgical Outcomes；CUSUM；EWMA