Proposal
General Project Description
This project involves investigating the feasibility of enhancing the treatment of diabetes through a web-based case similarity and retrieval system that permits physicians to pool their knowledge and optimize treatment of their patients.
Type 2 (adult onset) diabetes is simple to diagnose, but has many complications in treatment. A doctor often relies on his or her experience with other patients who were "similar" in some relevant way–typically by having a combination of common factors such as age, family history, ethnic background, reported diet and exercise habits, and so on. If doctors had a greater experience pool to draw from, better treatment recommendations might result.
The goal of this project is to create a web-based application that contains anonymous patient data including background information, treatments used, and treatment outcomes. The application would allow doctors or medical staff to enter new patient data and use it to call up similar cases. The project will investigate the applicability of alternative methods for determining which cases in the database should be considered similar to the new data. Such alternatives range from simple statistical methods to artificial intelligence techniques including probabilistic and case-based reasoning.
Ideally, the application would be capable of adaptation by considering the results most often selected or most favored by doctors. The model could be adjusted by the program itself in response to it having determined which factors doctors actually pay most attention to.
The experimental system will consist of a back-end database, a similarity and reasoning component, and an end-user web interface. Fictitious, but representative, data will be used for proof-of-concept. In actual practice, anonymity of patient data is a critical prerequisite of the database.
Professional Collaboration
If such a system proves feasible and is subsequently developed and fielded, physicians who have registered with the system may contribute data and access the database. Physicians (or their staff) would enter patient information and treatment details. They would then be able to compare those treatments and patient successes against similar entries in the database. The system must include a component that utilizes the physician's own experience and clinical judgment, allowing physicians to pool their knowledge and thereby optimize treatment of their patients. Further, the system must be usable as well as useful, so ergonomic human-computer interface factors need to be addresses.
Specific Questions/Hypotheses
We seek to determine whether the treatment of Type II diabetes could be enhanced through the use of a database-centric similarity-recognition system that is accessed via a web interface.
Our principle objective is a feasibility assessment that includes investigations of (1) the relevant attributes of patient case data and their embodiment in a database, (2) potential similarity metrics and methods, and (3) web-based human-computer interaction appropriate to the intended segment of the medical community.
If feasibility is established, the next step would be development of demonstration prototypes that permit physicians to experience how they could pool their knowledge and optimize treatment of their patients.
For such a system to be successfully deployed, it must engage the interest and participation of the medical community. We hypothesize that the level of interest will be sufficient to encourage participation and contribution of knowledge. We have established the following criteria for the service provided by such a system to achieve the requisite interest and participation:
- The service must be useful in supplementing doctors' own knowledge.
- The service must be easy to access and to use.
- The service must have high availability and reliability.
- The service must guarantee patient anonymity.
- The underlying processes must be subject to review by a board of physicians.
- The underlying system must be easily maintained.
In addition, the system itself must be scalable and extensible. Key concerns regarding scalability include how readily additional data can be integrated, implications for explosive growth of the database, and how quickly users can be added. Extensibility of the system must be considered along multiple dimensions, such as, access via other modalities (e.g., physician-carried mobile devices with both audio and video input/output); access by other users (e.g., health clinics); use of the same data for other purposes (e.g., medical research); and adaptation of the system for other diseases.
Methods
Initially, we will research the following:
- Identify current "best practices" in diabetes treatment and management, including factors likely to influence treatment recommendations.
- Identify methods and techniques for determining similarity given complex and uncertain data and retrieving best-match results (e.g., statistical methods, Bayesian inferencing, case-based reasoning, rule-based expert systems). We will pay particular attention to applications of such techniques to data mining.
- Which architectures best support a web-based application for remote subscribers that facilitates entering and accessing data.
Armed with this initial research, and if feasibility is still adequate, we will build a prototype to test the concept.
The prototype, with mock data and examples of retrieval and data mining capabilities, will then be used as a presentation to the medical community in order to gauge the level of interest in such a system.
References
Burkow, T.M. & Nilsen, L.L. "Success and Failure in Web-Based Medical Collaboration." Journal of Telemedicine and Telecare, 2005 December; 11(8): 11-3.
Durso, Samuel C. "Using Clinical Guidelines Designed for Diabetes Mellitus and Complex Health Status." Journal of the American Medical Association, 2006 April; 295(16): 1935-1940.
Inzucchi, Silvio E. & Sobel, Burton E. "A Case-Based Approach to Type 2 Diabetes and the Metabolic Syndrome." National Diabetes Education Initiative. January 2005.
Lehmann, E.D., Deutsch, T. et al. "Combining Rule-Based Reasoning and Mathematical Modeling in Diabetes Care." Artificial Intelligence in Medicine, 1994 Apr; 6(2): 137-60.
Pearl, J. Causality: Models, Reasoning and Inference. Cambridge University Press, 2000.
Impact on the Goal of CREU
This project would support the goals of CREU by allowing the student participants to engage in a combination of research and project design and execution. Working with a faculty mentor, the students will gain understanding of what research in computer science entails. The experience will help prepare them to consider graduate studies in computer science, and by completing the project, they will also become better candidates for graduate school.
Student Activity and Responsibility
- Background research and literature survey
- Hypothesis construction
- Decisions about programming language & architecture for building prototype system
- Presentation of prototype system to medical professionals
- Analysis of results of experiment, including hypothesis testing and assessment of professional interest
- Documentation and reporting
Faculty Activity and Responsibility
Support for students' activities, including but not limited to
- Source of domain expertise
- Research protocol and reporting guidance
- Reviews of interim and final products
Students and faculty will initially meet at least twice per week until the project is well underway, at which time project meetings may be reduced to a minimum of once per week. At each meeting, progress will be monitored and short-term goals will be developed to facilitate the meeting of long-term objectives. Student participants will keep a weekly log of activities, results, and findings relative to the project.
Last updated: October 1, 2006
Created September 29, 2006