Research
Extracting Semantic Knowledge from Clinical Data Sources
The investigators for IU subcontracting to Logical Semantics, have undertaken the following:
Dr. Josette Jones has:
(1) Evaluated the structure and consistency of the large scale knowledge base with relationship to domain of coverage. (2) Constructed a language-independent ontology to represent the concepts, processes, qualities and relationship represented in radiology and other medical reports by devising a clustering algorithm for semantically equivalent sentences and phrases, and (3) Developed a coding system to uniquely capture semantically equivalent sentences and reject semantically incorrect sentences. The goal was to develop a clustered model of semantical knowledge by representing language specific distinctions (denotational, attitudinal, and stylistic) inside the cluster of semantically and near-semantically equivalent sentences.
Dr. Malika Mahoui has:
(1) Enhanced the text mining approaches currently adopted to perform the mapping process between report sentences (resp. SnowMed terms) and propositions, and (2) Made necessary alterations to the knowledge base in order to ensure scalability when aiming for maximum coverage in a medical domain.
Jeff Friedlin (Regenstrief Research Scientist) has:
(1) Investigated strategies and methods in which to automate the proposition creation process, (2) Tested the Bayesian and IR methods for automatically mapping text lines to propositions with the goal of determining which of the two processes is most likely to yield the most accurate result, (3) Oversaw the installation of SnomedCoder service, provide technical support to the users, and (4) Elicited feedback regarding the user satisfaction with the service.
Knowledge Engineering for My Health Care Manager, LLC
Project Summary
My Health Care Manager assists older adults and their families in managing health treatment plans and interactions with the health care system and related providers. The company employs social workers and nurses as Health Care Managers to provide the services using information technology-enabled capabilities developed and administered by the company. In this joint proposal between My Health Care Manager and Indiana University Purdue University Indianapolis (IUPUI), we present a method for developing knowledge engineering tools for effectively building knowledge models of Geriatric Care Management and effective utilization of this data by Health Care Managers.
Subcontract Summary
The IUPUI team which involves Drs. Mathew Palakal, Yuni Xia and Josette Jones will develop the knowledge representation model using data collected from Health Care Managers and best practice data from literature. Further, the IUPUI team will develop effective traversal and decision support using this knowledge for My Health Care Manager.
Subcontract Deliverables
Dr. Palakal, Dr. Xia, and Dr. Josette and their team will fully develop, implement and test the knowledge engineering model components based on data provided by My Health Care Manager and available data from the literature. The IUPUI team will assist in the search and loading of best practices from field experience, direct research, and literature review and analysis. The IUPUI team and My Health Care Manager team will mutually determine the specific deliverables of the research and development activities. The IUPUI team will provide My Health Care Manager with a quarterly report describing the outcome from project-related activities and an assessment of progress status. The quarterly reporting will be augmented with monthly progress meetings (at locations and dates to be determined) attended by the entire research team.
The Indiana Cancer TRAIN
Translating Research – the Answers and Information Network): A Web Portal Linking IUCC Patients, Clinicians and Service Providers
Cancer is one of the most prevalent chronic illnesses in this country. Cancer patients and their families have significant needs for information about diagnosis and treatment. Research has demonstrated that increased knowledge of cancer enhances coping in cancer patients. Provision of cancer-related information helps patients and their families to accept their diagnosis, improve symptom management and compliance, and reduce anxiety and uncertainty, thus increasing overall quality of life.
Over 2 million people living with cancer use the Internet to obtain health information. Although countless Web sites exist that address patients’ questions about cancer prevention and treatment, the information is often without a strong research base and can be difficult to interpret. At best, many Web sites deliver information that is not individualized to a specific patient concern and offer no opportunity for interaction with health care professionals. At worst, they deliver information that may be detrimental. The purpose of this project is to establish a secure web portal for Indiana University Cancer Center (IUCC) patients and families to serve as a platform for translation of evidence-based cancer control behavioral interventions and information dissemination.
The specific aims of this project are: (1) to develop and test a customized web portal for the delivery of personalized cancer information to IUCC patients and their families; (2) to adapt a previously tested symptom monitoring and management system that links patients and their providers for web-based delivery; (3) to evaluate the feasibility, acceptability, satisfaction and usability of the Indiana Cancer TRAIN web portal for delivery of an evidence-based symptom monitoring and management system.
This overall research program will unfold in several phases. In the first phase, we will develop and test a web portal using existing information technology. The portal will provide two levels of interaction. The public level will allow any user to obtain quality-filtered, evidence based cancer information through the web site. A second, secure layer will require authorized access to specific applications that deliver customized cancer control interventions. We will pilot test a previously tested symptom management system that has been modified for delivery via the web. The outcome of this project will be a functional web portal that can be used by interdisciplinary teams of investigators in subsequent years.
The IUCC web portal will serve as a platform for accelerating the translation of efficacious interventions directly to patients and their families, who experience complex knowledge management challenges. By leveraging existing resources on the Indianapolis campus, this project will advance research efficiently and cost-effectively. Furthermore, the portal will increase outreach, education, and cancer information to our larger community, an expectation of NCI comprehensive cancer centers.
Watch Anna McDaniel and Renee' Stratton talk about the Cancer Portal.
Value of New Drug Labeling Knowledge for e-Prescribing
Study the value that the upcoming FDA mandated electronic drug labeling HL7 standard will bring to existing and emerging Computerized Provider Order Entry (CPOE) systems and e-prescribing tools on the example of the Regenstrief Medical Gopher CPOE system and on a newly developed completely HL7 standards-based open-source e-prescribing tool used for delivering decision support enabled safe prescribing assistance to practitioners in diverse healthcare settings (incl. small and rural).
THRUST 1: Map the Gopher drug knowledge structures to and from the HL7 standard drug knowledge structures, studying the feasibility of automatically importing the public drug knowledge to populate indication-based dosage, SIG defaults, contraindications, interactions, and consequent orders data structures. Study the minimum necessary need for customization, the gain in knowledge coverage and the resulting cost-savings for knowledge maintenance. This experience will be contributed into the standardization of the labeling content.
THRUST 2: Based on existing open-source HL7 v3 tools development, create a light-weight open-source prescribing tool that delivers decision support functions (incl. default dosing and dose checks, contraindication and allergy checking, consequent orders) based solely on standards and open terminology and knowledge structures. A focus-group of providers from rural West-Wisconsin and North-Carolina negotiates a minimal set of common requirements to keep the tool light-weight and generalizable. Implement and pilot the tool and evaluate its benefits and costs with (1) a time-series controlled trial demonstrating efficacy to reduce prescribing errors (benefits), (2) time-motion study evaluating usability and cost (in terms of providers' time.) User and patient satisfaction survey will evaluate perceived value. After the pilot, grow the user community to assess the cost of implementation at new sites, demonstrating that standards-based open-source light-weight design can lead to generalizable tools allowing providers to realize the value of CPOE.
Computerized Education to Prevent Hypoglycemia When Driving
Our long-term goal is to decrease the risk for driving accidents associated with severe hypoglycemia among adolescents/young adults with Type 1 Diabetes Mellitus. The specific aims are to: develop a brief computer-assisted blood glucose awareness training program (BGAT) for adolescents with Type 1 Diabetes Mellitus and their parents to increase their knowledge about preventing hypoglycemia when driving; demonstrate patient and parent satisfaction, increased knowledge changes in perceived susceptibility, perceived seriousness/threat, perceived benefits to taking action, and barriers to taking action; collect pilot data to support a subsequent application for external funding to create a more intensive web-base version to be used by adolescents throughout the U.S. Type 1 diabetes is a common chronic disease affecting millions of US children and adolescents. Although maintaing blood sugar levels near normal will decrease the risk for long-term complications of blindness and kidney failure, tight control carries the risk of severe hypoglycemia (low levels of blood sugar) which may result in impaired judgment, loss of consciousness and seizures. If this occurs while driving, it may result in severe injury and death.
Intensive group-based BGAT has been shown to decrease the risk for hypoglycemia when driving for adults but the existing program not been designed or tested with adoelscents. This interdisciplinary project will develop and test a unique computer-based delivery system to provide the initial education about driving and hypoglycemia to adolescents and their parents using combined audio/visual formats provided on a lap top computer. Multi-media computer-based health education is effective with adolescents and can increase the capacity of the health team to provide health education with minimal cost. In Phase 1 an existing BGAT program for adults will be substantially modified and adapted for adolescents, computer programs written, attitudinal/behavioral measures develop; these will then be piloted and modified as needed. In Phase 2 we will recruit at least 150 adolescents and a parent to take the educational module and evaluate its impact on increasing knowledge, changing attitudes related to hypoglycemia and assisting adolescents and parents develop personal guidelines for preventing hypoglycemia while driving. The final task in Phase 2 is to prepare an application for extramural support to expand the intervention to a web-base format for use in a multi-center collaborative study.
Preparing Residents for the Next Revolution in Medicine – Genetics and the Promise of Personalized Medicine
The objective of this project is to improve the quality of education provided to residents in the area of genomic medicine at Clarian and Indiana University School of Medicine sites and thereby improving the care provided to patients at Clarian facilities. Recent literature has identified grave deficiencies in the knowledge of genetics in practicing physicians and in the care provided to patients with hereditary conditions. To address this need, innovative methods including computer-aided instruction will be used to develop efficient and easily accessible ways of teaching relevant information and skills. These modules would eventually be made available online through the Indiana University School of Medicine Office of Continuing Medical Education for Clarian and community physicians.
The specific aims of this project are (1) to increase the physician’s knowledge of specific diseases for which genetic testing and/or personalized therapy is available and indicated on a clinical basis; (2) to increase the physician’s knowledge of technologies used in genetic testing including the capabilities and limitations of each; (3) to increase the physician’s knowledge of how ethical, legal, and social issues (ELSI) are interwoven with genetic testing; and (4) to improve the physician’s skills in communicating complex medical information to patients. A needs assessment will be conducted by polling the residency program directors of the target training programs (Internal Medicine, Medicine/Pediatrics and Pediatrics) and local clinical geneticists. Computer-based modules in a case-based format will be developed to address these goals and objectives. Each module will be designed to be completed in 15 to 20 minutes and will consist of a pre- and post- test; a list of learning objectives; patient cases incorporating video clips highlighting important issues; and references to primary literature sources. Positive communication skills as well as common mistakes will be demonstrated through video clips. Ethical, legal and social issues involved in genetic testing will be highlighted in each module. To assess the effectiveness of the modules, pre- and post-test questions will be used to gauge improvement in physicians’ knowledge. Feedback from evaluation data for each module will be assessed continually and used to frequently modify the modules. These modules would eventually be packaged and made available online to Clarian and community physicians through the Office of Continuing Medical Education, Indiana University School of Medicine.
Onco-Miner: a Personalized Health Information Management System
Cancer is one of the most prevalent chronic illnesses in this country. Cancer patients and their families have significant needs for information about diagnosis and treatment. Over 2 million people living with cancer use the Internet to obtain health information. Although countless Web sites exist that address patients’ questions about cancer prevention and treatment, the information is often without a strong research base and can be difficult to interpret. At best, many Web sites deliver information that is not individualized to a specific patient concern and offer no opportunity for interaction with health care professionals, and at worst, deliver information that may be inaccurate. The purpose of this project is to develop and build an intelligent health-medical literature information management system that would provide patient-centric information and knowledge as needed in real-time as part of the IU Simon Cancer Center Burdette Web Portal.
Large amounts of biomedical and health related literature data are currently available today with varying degrees of accessibility over the internet. Health and medical related information can be searched on general purpose search engines, such as Google, or using special purpose health and medical websites. Regardless of which search tools are used, the challenge is to overcome the overload of irrelevant information and the sheer volume of data. Further, the current practices of manually specifying search patterns, coordinating search activities from multiple sources, and analyzing the retrieved data for information value, are tedious and time-consuming even for a highly motivated individual.
To alleviate the problem of information overload, significant advances has been made in the areas of Information Retrieval (IR), Information Filtering (IF), and Information Mining (IM). Information filtering (IF) deals with providing customized information to the users to stay abreast with the ever evolving information repositories without drowning them in an ocean of irrelevant or unwanted data. The retrieved documents are rank ordered and presented to the user based on a customized profile.
Information mining (IM), on the other hand, is primarily concerned with discovering associations among interesting entities from the filtered document collection using association rules. An association is an implication of form A®B where ‘A’ is a set of antecedent items and ‘B’ is the consequent item. [Wong et al., 1999]. The intuitive meaning is that the documents that contain ‘A’ tend to contain ‘B’. In text mining the associations are of the form WordA®WordB and/or vice versa (e.g. Smoking®Lung Cancer, Loss of apoptosis®Cancer).
Efficient information management systems can be developed by effectively combining both information filtering and information mining techniques. Such information management systems will be able to deliver personalized and highly relevant information to the user if and when needed. Such systems will also span across time searching for pertinent information thereby providing the user with historical as well as most recent information.
In this research, we propose to develop a tool called OncoMiner, for providing customized information to cancer patients. This tool will have two basic functionalities: (a) provide newsworthy articles in a rank ordered fashion based on relevance to the user, and (b) discover pertinent knowledge from the newsworthy articles and deliver this knowledge to the user. Essentially, the former deals with the thematic aspects whereas, the latter deals with semantic content. The overall goal is to provide relevant information for improving the quality of life through supportive care at diagnosis, during treatment, and throughout cancer survivorship. Our focus is to design the system such a way that the final system is adaptable, scalable, and most importantly, it is “user-centric”. Mapping and implementation of the algorithms for data-to-information transformation will utilize automated and efficient methods to provide patients with an active, personalized, and integrated information delivery with minimal user interaction. The proposed OncoMiner will be able to effectively handle text-based health documents and present them to the user based on the user's specific interest profile. The system will utilize documents only from authorized and reliable sources such as Medline Plus, the National Comprehensive Cancer Network, and the National Cancer Institute, in consultation with the project collaborators.
