Statistical Quality Methods to Monitor and Transform Healthcare Data

Open Access
Kim, Min-Jung
Graduate Program:
Industrial Engineering
Doctor of Philosophy
Document Type:
Date of Defense:
July 02, 2012
Committee Members:
  • Harriet Black Nembhard, Dissertation Advisor
  • Harriet Black Nembhard, Committee Chair
  • Dr Paul Griffin, Committee Member
  • Ling Rothrock, Committee Member
  • Rhonda Belue, Committee Member
  • Li Wang, Special Member
  • Heatlh Monitoring
  • Statistical Quality Methods
The last decade has presented critical health challenges, ranging from the threat of new infections, to the emerging burdens of chronic disease. As experienced in the H5N1 avian influenza epidemic in 2007 and the H1N1 influenza pandemic in 2009, infectious diseases have emerged as a significant threat to global health. Furthermore, the rise in the prevalence of chronic conditions across all socioeconomic classes has made these conditions the leading causes of death and disability. These and other similar challenges have led to a growing interest in leveraging healthcare data better to facilitate disease prevention and control and better health care. In this dissertation, three statistical quality control (SQC) based methods are proposed to deal with some of the distinguishing characteristics in healthcare data. The first method integrates SQC with forecasting methods to support syndromic surveillance in health monitoring. Health data are likely to be non-stationary, non-normal, or autocorrelated, which calls for caution in the application of traditional SQC approaches. Furthermore, in order to address the need to track movements of the disease progression/dispersion as well as obtain a timely and sensitive signal for disease change, the method proposes pretreatment steps using forecasting based methodologies (regression and time-series models) that strengthen in modeling baseline patterns of disease. The pretreatment steps are applied to health data in such a way that results in independently and normally distributed observations so that SQC procedures can be appropriately applied. The proposed health monitoring framework provides a customized procedure for dealing with various statistical characteristics of health data which lead to use satisfy general assumptions of SQC. As an example of influenza syndromic surveillance, we monitor the weekly influenza-like-illness (ILI) incidence data and weekly over-the-counter and prescribed medication sales related to the ILI symptoms. Statistical analysis of two sources of data shows that they are non-stationary processes with a high level of autocorrelation which is a barrier to the direct use of traditional SQC. The application of our monitoring framework to the multiple streams of ILI and drug sales results in earlier warning than the official announcement of outbreak. The second method combines two multivariate SQC approaches – the multivariate self-starting exponentially weighted likelihood ratio (MSS-EWLR) chart and the multivariate change-point (MCP) chart – for the complexity found in health monitoring applications where there is interest in two or more quality characteristics that may be correlated and where in-control process data is limited. As an example of its application in monitoring chronic disease, systolic blood pressure, diastolic blood pressure and mean arterial pressure are monitored for assessing hypertension and its related complications. Our approach allows the monitoring of both mean and variability of healthcare data in a single chart. The combination of the two SQC approaches provides better performance enabling the detection of subtle changes that can be masked by unsuspected change in the correlation among quality characteristics by monitoring both mean and variability. Also, the combined approach has the added benefit of working without a priori parameter information of a baseline. The third method combines time-series analysis and regression to transform cross-sectional data into repeated measurement data describing individual patient profiles that can be monitored with SQC tools. The majority of accessible national surveys are based on cross-sectional data that depict a snapshot of health conditions in the population. Although national surveys are performed periodically with same individuals, identifiers are not provided to protect participant privacy and confidentiality. This profile data estimation (PDE) method will allow the performance of proposed health monitoring methods to be demonstrated with real clinical data that is publicly available. This may ease the burden of validating new methodologies. Moreover, it may result in a better, more rapid understanding of the relationship among multiple data characteristics that explain disease progression and dispersion over a complex disease process within and between patients. The methods developed here link quality engineering research in the field of industrial engineering with health monitoring systems research in the field of healthcare. Enhanced healthcare data monitoring will contribute to the goal of ensuring that care teams and relevant organizations receive accurate, timely, and up-to-date information about given health situations so that they can determine the appropriate intervention to undertake. In this sense, the work is a part of the broader partnership between healthcare and its associated clinical discoveries, and the engineering systems that help translate the discoveries into action.