A Knowledge and Data-Driven Computerized Physician Order Entry System

Abstract

This thesis proposes a knowledge and data-driven framework for a Computerized Physician Order Entry System (CPOE). The proposed framework includes the de- sign of a dynamic UI/UX that provides uniform healthcare delivery to patients. Uni- formity in patient care is a major issue facing the healthcare industry globally. This is because healthcare provision is highly dependent on the expertise and experience of doctors, resulting in unjustified clinical variation. Avoiding inappropriate variation in healthcare delivery is critical for both patient safety and quality of care. CPOE is an integral part of any healthcare information system that plays an essential role in patient care delivery because it provides automated provider order entry, archival, and dissemination that is integrated with electronic medical records of patients on more sophisticated platforms. In this context, the thesis proposed using data-driven approaches and takes advantage of the vast amount of patient-related medical data stored in hospital data archives. The thesis further augments the methodology by utilizing knowledge-driven approaches to incorporate evidence-based care. The pro- posed analytical framework integrates both of these methodologies. The first part of our proposed methodology develops a knowledge-driven framework that is based on doctors’ input and healthcare manuals. The thesis then considers incorporating data-driven methodology. To implement the data-driven approach, the thesis uses frequent pattern mining and association rule mining. The thesis then incorporates the use of published clinical pathways for the knowledge-driven approach into the en- hanced framework. The proposed framework is built on a knowledge base populated with disease quadruples. Each disease quadruple includes symptoms, tests, diagno- sis, and treatment for a specific disease. This framework acts as a backbone for the design of the CPOE system as it restricts the user interface to project diagnosis and treatment options supported by the knowledge base. Hence the user interface of the proposed CPOE system is dynamic and implicitly ensures consistency in health- care diagnosis and treatment by avoiding inappropriate clinical variance. The thesis describes the framework in terms of architecture, components, and algorithms. It demonstrates the use of data mining for avoiding clinical variance by ensuring uni- formity in healthcare provision. The framework presents a scalable methodology to incorporate any number of diseases and the flexibility of providing diagnosis and treatment options for multiple diseases based on the available data. To validate the proposed framework, we used both published and local data sets. The published data sets used include the Medical Information Mart for Intensive Care (MIMIC) dataset and the Disease-Symptom Knowledge Base of New York Presbyterian Hos- pital (NYPH). Data sets were gathered locally also, including those related to the specialty of Ophthalmology and the disease Aplastic Anemia. This broad range of published and local data sets enhances the credibility of the presented work while also providing local insights. The framework is validated from two aspects: the CPOE prototype application and the backend knowledge base. A diverse group of medical practitioners was selected for testing the framework on multiple use cases. A method for cross-validating the knowledge base using patient data has also been used, along with the matching score formula. Based on the medical practition- ers’ user experience with the platform and the cross-validation results, the thesis i concludes the effectiveness of the platform in assisting doctors in making informed decisions about patient diagnosis and improving accuracy in treating patients. The proposed framework shall optimize the healthcare delivery system of the hospital. As a result, the proposed research can aid clinical decision-making and thus be directly applied in medical practices.