Causal Pattern Mining in Highly Heterogeneous and Temporal EHRs Data
2017-03
Loading...
View/Download File
Persistent link to this item
Statistics
View StatisticsJournal Title
Journal ISSN
Volume Title
Title
Causal Pattern Mining in Highly Heterogeneous and Temporal EHRs Data
Authors
Published Date
2017-03
Publisher
Type
Thesis or Dissertation
Abstract
The World Health Organization (WHO) estimates that the total healthcare spending in the U.S. is around 18\% of its GDP for the year 2011. Even with such a high per-capita expenditure, the quality of healthcare in U.S. lags behind as compared to the healthcare in other industrialized countries. This inefficient state of the U.S. healthcare system is attributed to the current Fee-for-service (FFS) model. Under the FFS model, healthcare providers (doctors, hospitals) receive payments for every hospital visit or service rendered. The lack of coordination between the service providers and patient outcomes, leads to an increase in the costs associated with the healthcare management, as healthcare providers often recommend expensive treatments. Several legislations have been approved in the recent past to improve the overall U.S. healthcare management while simultaneously reducing the associated costs. The HITECH Act, proposes to spend close to \$30 billion dollars on creating a nationwide repository of electronic Health Records (EHRs). Such a repository would consist of patient attributes such as demographics, laboratories test results, vital information and diagnosis codes. It is hoped that this EHR repository will be a platform to improve care coordination between service providers and patients healthcare outcomes, reduce health disparities thereby improving the overall healthcare management system. Data collected and stored in the EHR (HITECH) and the need to improve care efficiency and outcome (ACT) would help to improve the current state of U.S. healthcare system. Data mining techniques in conjunction with EHRs can be used to develop novel clinical decision making tools, to analyze the prevalence and incidence of diseases and to evaluate the efficacy of existing clinical and surgical interventions. In this thesis we focus on two key aspects of EHR data, i.e. temporality and causation. This becomes more important considering that the temporal nature of EHRs data has not been fully exploited. Further, increasing amounts of clinical evidence suggest that temporal nature is important for the development of clinical decision making tools and techniques. Secondly, several research articles hint at the the presence of antiquated clinical guidelines which are still in practice. In this dissertation, we first describe EHR along with the following terminologies : temporality, causation and heterogeneity. Building on this, we then describe methodologies for extracting non-causal patterns in the absence of longitudinal data. Further, we describe methods to extract non-causal patterns in the presence of longitudinal data. We describe such methodologies in the context of Type-2 Diabetes Mellitus (T2DM). Furthermore, we describe techniques to extract simple and complex causal patterns from longitudinal data in the context of sepsis and T2DM. Finally, we conclude this dissertation, by providing a summary of our work along with future directions.
Keywords
Description
University of Minnesota Ph.D. dissertation. March 2017. Major: Computer Science. Advisor: Vipin Kumar. 1 computer file (PDF); ix, 112 pages.
Related to
Replaces
License
Collections
Series/Report Number
Funding information
Isbn identifier
Doi identifier
Previously Published Citation
Other identifiers
Suggested citation
Yadav, Pranjul. (2017). Causal Pattern Mining in Highly Heterogeneous and Temporal EHRs Data. Retrieved from the University Digital Conservancy, https://hdl.handle.net/11299/188895.
Content distributed via the University Digital Conservancy may be subject to additional license and use restrictions applied by the depositor. By using these files, users agree to the Terms of Use. Materials in the UDC may contain content that is disturbing and/or harmful. For more information, please see our statement on harmful content in digital repositories.