Recent Changes - Search:

Brigham and Women's Hospital
Harvard Medical School



Seminars

Weekly seminars are held on Fridays, 1-2 PM,
at the DSG Conference Room. Everyone is welcome to attend!

5/29/2009
Ronilda Lacson MD PhD
Future of the Decision Systems Group

Recent Seminars (2009)

Previous Seminars (2008)

Previous Seminars (2007)

Previous Seminars (2006)




09/28/06 Lucila Ohno-Machado MD PhD
Title: SMART: Scalable Medical Alert Response Technology
Abstract:
Brigham & Women's Hospital proposes to combine existing and new technologies to develop SMART: Scalable Medical Alert and Response Technology, a system for patient tracking and monitoring that begins at the emergency site and continues through transport, triage, stabilization, and transfer between external sites and health care facilities as well as within a health care facility. The system is based on a scalable location-aware monitoring architecture, with remote transmission from medical sensors and display of information on personal digital assistants, detection logic for recognizing events requiring action, and logistic support for optimal response. Patients and providers, as well as critical medical equipment will be located by SMART on demand, and remote alerting from the medical sensors can trigger responses from the nearest available providers. The emergency department at the Brigham and Womens Hospital in Boston will serve as the testbed for initial deployment, refinement, and evaluation of SMART. This project will involve a collaboration of researchers at the Brigham and Womens Hospital, Harvard Medical School, and the Massachusetts Institute of Technology.


10/5/2006 Ronilda Covar Lacson MD PhD
Title: Automatic Analysis of Medical Dialogue in the Home Hemodialysis Domain: Structure Induction and Summarization
Abstract:
Spoken medical dialogue is a valuable source of information for patients and caregivers. This work presents a first step towards automatic analysis and summarization of spoken medical dialogue. A dialogue is abstracted into a sequence of semantic categories using linguistic and contextual features integrated in a supervised machine-learning framework. This model has a classification accuracy of 73%, compared to 33% achieved by a majority baseline (p<0.01). A summarizer is then described and implemented utilizing this automatically induced structure. The evaluation results indicate that automatically generated summaries exhibit high resemblance to summaries written by humans. This work demonstrates the feasibility of automatically structuring and summarizing spoken medical dialogue.


10/12/2006 Jose Trevejo MD PhD
Title: Medical Applications of the Breath Analyzer


10/19/2006 Jerry Chen MD
Title: The use of machine learning techniques for classifying microcalcifications in digital mammograms
Abstract:
Dr. Chin-Yu Chen will be leading the discussion on the use of machine learning techniques for classifying microcalcifications in digital mammograms. He will begin by reviewing the attached journal article and then proceed with some thoughtful questions and ideas regarding his current research study.


10/26/2006 Michael Matheny MD
Title: Methodologies and Automated Applications for Post-Marketing Outcomes Surveillance of Medical Devices and Medications


11/2/2006 Margarita Sordo PhD
Title: On Sample Size and Classification Accuracy: A Performance Comparison
Abstract:
We investigate the dependency between sample size and classification accuracy of three classification techniques: Naïve Bayes, Support Vector Machines and Decision Trees over a set of ~8500 text excerpts extracted automatically from narrative reports from the Brigham & Women’s Hospital, Boston, USA. Each excerpt refers to the smoking status of a patient as: current, past, never a smoker or, denies smoking. Our empirical results confirm that size of the training set and the classification rate are indeed correlated. Even though these algorithms perform reasonably well with small datasets, as the number of cases increases, both SMV and Decision Trees show a substantial improvement in performance, suggesting a more consistent learning process. Unlike the majority of evaluations, ours were carried out specifically in a medical domain where the limited amount of data is a common occurrence. This study is part of the I2B2 project, Core 2.


11/9/2006 Samuli Niiranen PhD
Title: Types of Unintended Consequences Related to Computerized Provider Order Entry Authors: Emily Campbell, Dean Sittig, Joan Ash, Kenneth Guappone, Richard Dykstra
Abstract:
Dr. Samuli Niiranen will begin by reviewing the attached journal article and then proceed to talk about how the paper relates to his current research.


11/30/2006 Pankaj Sarin MD
Title: Encouraging changes in anesthesiology practice through electronic feedback
Abstract:
This research focuses on providing electronic feedback to physicians regarding their practices and patient outcomes. After introducing the methods, there will be a discussion about the design and implementation of a study on electronic feedback to attending anesthesiologists and whether this feedback induced practice change in the group.


12/7/2006 Staal Vinterbo PhD
Title: Approximating the Minimal Cover Problem
Abstract:
The Minimal Cover problem is the problem of selecting m out of n finite sets such that the union of these m sets is minimal. This problem is interesting because an algorithm for Minimal Cover can be used to find a solution to the k-ambiguity by cell suppression problem. This latter problem is related to privacy in disseminated data. The speaker will present a polynomial time algorithm for Minimal Cover that delivers a solution that is guaranteed not to be worse than m times the optimal. This algorithm allows us to find a solution to the k-ambiguity by cell suppression problem that is guaranteed not to be worse than (k-1) times the optimal. If time allows, he will show that there probably is no polynomial time algorithm that can guarantee solution quality within a constant factor of the optimal for both problems. These results are, as far as the speaker knows, the first of their kind for these problems.


12/14/2006 Robert Greenes MD PhD
Title: funding opportunities and possibilities at AHRQ and TATRC


12/21/2006 Esther Shilcrat PhD
Title: National and Local Databases to Support ePrescribing
Abstract:
DSG has been part of an ongoing effort to analyze the use of HL7 GELLO expression language in support of electronically mediated prior authorization within the context of ePrescribing. In particular, these efforts have also been directed towards the use of GELLO to create a nationally mandated ePrescribing prior authorization database (National).

The existence of a National will afford institutions the opportunity to create their own Local "version" DB. The Local can reflect rules from the National, as well as institution specific medical and business rules. Thus, an immediate issue is how to 'meld' local formularies and other rules with those presented in the National.

In addition, the Local may be further structured to support a variety of tasks beyond prior authorization. For one example, the Local could reflect local business rules. As another example, the Local could be designed to facilitate the execution of electronic guidelines. This would require information to be stored to facilitate quick identification of drugs in "categories of interest." We need to discover what these categories are by studying prescribing needs, such as are expressed in guidelines (ex., "is patient using inhaled steroid?" suggests the need to classify a drug as a steroid for quick retrieval.)

These two issues have considerable overlap, as adapting rules from the National to the Local would seem to require similar use of categories. For example, ideally, with the right categories, for any rule in the National, we can answer: does the local formulary already contain a 'close enough' drug? This information would then be used to decide on the contents of the Local.

There are many open questions, such as maintaining 'consistency' between the National and its Local variant. We also seek to create an 'updatable' ontology which will change as we learn more about the problem domain. This work is very preliminary and suggestions are most welcome!


1/11/2007 Chris Tsai MD MPH
Title: VIR-POX: An Agent-Based Analysis of Smallpox Preparedness and Response Policy
Authors: Benjamin M. Eidelson and Ian Lustick
Link: http://jasss.soc.surrey.ac.uk/7/3/6.html
Abstract:
He will be reviewing a journal article entitled "VIR-POX: An Agent-Based Analysis of Smallpox Preparedness and Response Policy". The link to the journal article is shown below. He will then discuss Agent-Based Modeling in general and its use in mass vaccinations.


1/18/2007 Qing Zeng PhD
Title: Text Characteristics of Clinical Reports and Their Implications for the Readability of Personal Health Records
Abstract:
Through personal health record applications (PHR), consumers are gaining access to their electronic health records (EHR). A new challenge is to make the content of these re-cords comprehensible to consumers. To address this challenge, we analyzed the lexical, syntactic and semantic characteristics of three sets of health texts: clinical reports from EHR, known difficult materials and easy-to-read materials. Our findings suggest that EHR texts are more different from easy texts and more similar to difficult texts in terms of syntactic and semantic characteristics; and EHR texts are more similar to easy texts and different from difficult texts in regard to lexical features. Since commonly used readability formulas focus more on lexical characteristics, this study points to the need to tackle syntactic and semantic issues in the effort to measure and improve PHR readability.


1/25/2007 Lola Ogunyemi PhD
Title: Probabilistic Reasoning for Trauma Decision Support
Abstract:
This talk will briefly examine different approaches adopted in the biomedical informatics domain for assessing the effects of penetrating trauma. The majority of the talk will be devoted to TraumaSCAN-Web, a web-based diagnostic decision support tool for penetrating trauma developed at DSG. Results of an assessment on Brigham & Women's Hospital Emergency Department residents, and an evaluation of a Bayesian network-only approach for penetrating trauma will be presented.


2/1/2007 Lucila Ohno-Machado MD PhD
Title: Uses and Misuses of the Coefficient of Variation: Implications for Studies of Ecosystem Stability
Abstract:
The ongoing controversy about the effects of biodiversity on ecosystem stability contributes to the challenge of establishing evidence-based conservation and restoration policies. The controversy may be partially explained by differences in the definition of stability and measurement indices. Some authors define stability as the lack of variability in biomass, often measured by the coefficient of variation (CV). Some studies conclude that increased biodiversity is associated with increased stability, and that stability necessarily increases from the species to the functional group to the community levels; others report either negative or no association. Even disregarding the effects of statistical averaging (the portfolio effect), the decline in CV may not necessarily indicate that systems are less variant, but be simply an artifact derived from misuse of the CV. Here we show that, in the context of observational studies in which variability for a single species is assessed in a multispecies environment, (1) the CV is a misleading index that can give rise to erroneous interpretations and (2) comparisons using an alternative index are more appropriate for the assessment of ecosystem variability. The alternative index requires few but realistic assumptions about the relationship between the quantities for which variabilities are being compared.


2/8/2007 Mark Finlayson
Title: The Difficulties of (Medical) Text Annotation
Abstract:
The digitization of medical records presents exciting opportunities for improving medical care. For example, digitally-accessible medical records could conceivably enable a doctor to search his patient's records like he might search Google, allow a researcher to analyze large sets of records to study trends or the effectiveness of novel treatments, and make possible the training of exceptionally accurate, domain-specific diagnosis programs. But, despite the great efforts made toward these applications, they are foundering on what seems, on the face of it, a simple problem, but in truth is an extremely difficult technological shoal: the ability to translate free text into computer-understandable form. This is a general knowledge capture problem that has plagued many areas of artificial intelligence and computer science, and is not specific to medical text understanding. In this talk I will review the general problem of natural language understanding with a view toward knowledge-capture systems, drawing my examples from medical record understanding, and present some recent progress that has been made in semi-automatic annotation tools that are the only feasible near-term solution to this problem. Short Bio: Mark A. Finlayson is a doctoral student under Patrick H. Winston at MIT's Computer Science and Artificial Intelligence Laboratory. His research is at the interface between Artificial Intelligence and Cognitive Science, focusing primarily on fielding models of psychological experiments regarding the effects of knowledge, culture, and expertise on cognition.


2/15/2007 Ronilda Covar Lacson MD PhD
Title: Predicting Mortality in Hemodialysis Patients Using Wavelets
Abstract:
The role of blood pressure in cardiovascular morbidity and mortality in hemodialysis patients has not been fully elucidated. High blood pressures as well as low blood pressures are poor prognostic indicators for mortality in this large population of patients. Hypertension is a known risk factor for mortality and cardiovascular disease in the general population. Recent studies, however, indicate an inverse association between blood pressure values and mortality in hemodialysis patients. Further, the range of blood pressure which would be considered hypertension is not clear for this subset of patients. Blood pressure is clearly a useful predictor to monitor because it is both easily measurable and it is modifiable. This talk will review various methods that have been used to predict mortality using blood pressure measurements. The speaker will then proceed to discuss a method that predicts mortality using multiple blood pressure values measured over time at varying intervals that takes the direction, magnitude and rate of change into consideration.


2/22/2007 Jerry Chin-Yu Chen MD
Title: Medical audit of mammography - what the data tell us.
Abstract:
The presentation will be based on a recently finished manuscript (attached file), which is not yet submitted for publication. The talk will begin with an introduction of mammography medical audit, and then proceed to discuss the results of medical audit when applied to the Taiwanese population. A comparison of these results will be done with those from North America. Finally, differences between the different subgroups of the study cohort will be discussed.


3/1/2007 Ilmo Parvinen, MD, Director, Sitra (The Finnish Innovation Fund, www.sitra.fi) Health Care Program
Title: Sustainable Health Care Systems : Impact of Value Based Improvements
Abstract:
The current state of the Finnish health care system is analyzed as a starting point for presenting Sitra?s view on how sustainable developments in health care can be attained. The key drivers and agents of change are - Deployment of electronic services - Structural changes in the health care delivery system - Changes in the stature, actions and responsibilities of the citizen The utility of these drivers and agents is also considered in an international context with a focus on the US. The key message is that although national situations vary, co-operation is vital and that discussion is not enough, concrete actions are needed.


3/8/2007 Lucila Ohno-Machado MD PhD
Title: Clustering medical informatics: a didactic presentation
Abstract:
In this talk, I will give a technical overview of unsupervised learning techniques, primarily hierarchical clustering and multidimensional scaling. I will then briefly discuss two examples that illustrate these techniques and relate to the analysis of co-citations, specifically for biomedical informatics literature. If time permits, I will take the opportunity to discuss the primary venues for biomedical informatics publication, emphasize the importance of participating in article and grant reviews, and some practical hints (particularly for those who want to pursue an academic career).


3/15/2007 Margarita Sordo PhD
Title: "Tutorial on Agent-Based Modeling and Simulation Part 2: How to Model with Agents" by Charles M. Macal & Michael J. North
Abstract:
The paper is an introduction to Agent Based Modeling and Simulation (ABMS). It is organized in two parts: 1) How to think about ABMS. The background, motivating principles and main concepts are described. Second part is on how to do ABMS. It presents practical applications, toolkits and development approaches. Examples are presented throughout the paper.


3/22/2007 Samuli Niiranen PhD
Title: Information Management in the Emergency Department
Abstract:
I will discuss two specific examples in this domain from my own work: computational understanding of chief complaint information and, briefly, visualization of emergency department patient flow.


4/5/2007 Ramin Khorasani, MD
Title: Information Management Systems Group, BWH
Abstract:
He will share the various research and initiatives of his group at the BWH Department of Radiology.


4/12/2007 Staal Vinterbo PhD
Title: Discernibility Properties of Bernoulli Process Data
Abstract:
Finding patterns that discern data-points in multidimensional data is NP-hard in general, forcing the use of approximations. The use of such patterns can be in classification or disclosure control. The talk will present properties of a discernibility matrix computed from binary data stemming from Bernoulli trials, and show that these can be used to gain insight into the efficacy of pattern recognition algorithms.


4/19/2007 Bonnie Kaplan PhD
Title: Same System - Different Views: Dilemmas and Research Avenues for IT
Abstract:
Many exciting uses of IT for healthcare were developed over the past 50 years. This experience informs us both of barriers and facilitators to IT adoption and use. My research focuses on how to design, deploy, and study IT in healthcare by investigating how people respond to IT applications. I will discuss research findings and suggest new research avenues involving management, organizational, ethical, social, and policy issues that may arise from new technologies in health care.

Bonnie Kaplan, Ph.D., is a Lecturer at the Yale Center for Medical Informatics; Adjunct Clinical Professor, Department of Biomedical and Health Information Sciences, University of Illinois at Chicago; and President of Kaplan Associates. She chairs the International Medical Informatics Association's Working Group on Organizational and Social Issues, and founded the Yale Interdisciplinary Bioethics Center Working Group on Technology and Ethics as well as the Yale Whitney Humanities Center Working Group on Science, Technology, and Utopian Visions. She is author of more than 70 papers and book chapters on people's reactions to new technologies and computer systems in health care. Dr. Kaplan is a recipient of the American Medical Informatics Association President's Award and a Fellow of the American College of Medical Informatics.

She is an authority on people's reactions to new technologies in health care and on evaluating applications of computer information systems. She specializes in change management, benefits realization, and identifying and addressing clinician and patient concerns. Her clients include the US Department of Veterans Affairs (VA) and the US National Institute of Standards and Technology (NIST); the Universities of Pittsburgh, Chicago, Washington, and Cincinnati; Yale, Boston, Oregon Health Sciences, and Johns Hopkins Universities; and Massachusetts General Hospital.

Dr. Kaplan is the author of more than 70 refereed papers and book chapters, as well as numerous other articles and publications. She wrote invited chapters in the most important books on evaluating information systems in health care. Her most recent publications concern patients' reactions to using an automated telephone advisory system; a usability evaluation of a web based case for in medical education; issues in evaluating clinical decision support systems and medical informatics applications; a review of people, organizational, and social issues in medical informatics; and an volume of edited papers.


5/3/2007 Michael Matheny MD
Title: Impact on Patient Satisfaction with Physician Use of an Automated Test Results Management System
Abstract:
He will be discussing the strengths and weaknesses of using a pre-post randomized controlled design relative to a straight-forward randomized intervenion study. This statistical introduction will be followed by an application of this methodology for the evaluation of patient satisfaction with test results communication after the implementation of automated test results management system (Results Manager).


5/10/2007 Hyeoneui Kim PhD
Title: Beyond Surface Characteristics: A New Health Text-Specific Readability Measurement
Abstract:
Accurate readability assessment of health related materials is a critical first step in producing easily understandable consumer health information resources and personal health records. Existing general readability formulas may not always be appropriate for the medical/consumer health domain. We developed a new health-specific readability pilot measure, based on the differences in semantic and syntactic features as well as text unit length. The tool was tested with 4 types of materials: consumer health texts, electronic health records, health news articles, and scientific biomedical journals. The results were compared with those produced by three commonly used general readability formulas. While the general formulas underestimated the difficulty of health records by placing them at the same grade levels as consumer health texts, our method rated health records as the most difficult type of documents. Our ratings, however, were highly correlated with general formulas ratings of consumer health, news, and journal articles (r=0.81~0.85, p<.0001).


5/17/2007 Pankaj Sarin MD
Title: Improving healthcare for patients with cardiovascular disease in Tanzania:Technological limitations and solutions
Abstract:
Cardiovascular disease is 8% of disease burden in Tanzania and one of the top causes of morbidity and mortality amongst non-communicable diseases. Although focus has been on communicable diseases, it is projected that NCD will match CD in developing countries in disability adjusted life years (DALY) by 2030. I will provide an overview of the current healthcare system and needs, limitations of infrastructure affecting technology choices, and examine how technological solutions can be used within these constraints to provide improved healthcare to patients.


5/31/2007 Lucila Ohno-Machado MD, PhD
Title: Intermediate Phenotypes for Genetic Markers


6/7/2007 Chris Tsai MD MPH
Review
Title: Informatics Systems to Promote Improved Care for Chronic Illness: A Literature Review DAVID DORR, MD, MS, LAURA M. BONNER, PHD, AMY N. COHEN, PHD, REBECCA S. SHOAI, MPH, MSW, RUTH PERRIN, MA, EDMUND CHANEY, PHD, ALEXANDER S. YOUNG, MD, MSHS
Abstract:
Objective: To understand information systems components important in supporting team-based care of chronic illness through a literature search. Design: Systematic search of literature from 1996-2005 for evaluations of information systems used in the care of chronic illness. Measurements: The relationship of design, quality, information systems components, setting, and other factors with process, quality outcomes, and health care costs was evaluated. Results: In all, 109 articles were reviewed involving 112 information system descriptions. Chronic diseases targeted included diabetes (42.9% of reviewed articles), heart disease (36.6%), and mental illness (23.2%), among others. System users were primarily physicians, nurses, and patients. Sixty-seven percent of reviewed experiments had positive outcomes; 94% of uncontrolled, observational studies claimed positive results. Components closely correlated with positive experimental results were connection to an electronic medical record, computerized prompts, population management (including reports and feedback), specialized decision support, electronic scheduling, and personal health records. Barriers identified included costs, data privacy and security concerns, and failure to consider workflow. Conclusion: The majority of published studies revealed a positive impact of specific health information technology components on chronic illness care. Implications for future research and system designs are discussed. J Am Med Inform Assoc. 2007;14:156-163.


6/14/2007 Qing Zeng PhD
Title: Multifaceted Adaptive Testing Approach for Literacy in Older Adults
Abstract:
The overall objective of this study is to develop and validate a flexible health literacy instrument for older adults employing the state-of-art computational technology. A comprehensive health literacy framework as described in this proposal can be used to support research on health literacy, develop screening tools for clinicians and tailor health education materials for older adult patients and consumers at varying levels of literacy. This can contribute to empowering patients and health consumers as well as serving to narrow the digital divide.


6/21/2007 Esther Shilcrat PhD
Title: Toward a More Comprehensive Taxonomy of Consumer Health Information Needs
Abstract:
As consumers become savvier about health related resources, and more desirous of understanding their own, or family members’ health conditions, it is increasingly important to analyze and represent consumers’ health related information needs. On the basis of prior efforts, we developed a more comprehensive and detailed taxonomy of such needs. The taxonomy is also designed to capture consumer needs related to information resource characteristics. We present a description of our initial taxonomy and validation study. We also describe our on-going efforts to uncover and classify more consumer health needs.


7/12/2007 Aziz Boxwala, MD, PhD
Title: Obtaining medical history from the patient by computer-based interviewing
Abstract:
Medical history obtained by a physician during a clinic visit has many important deficiencies. Obtaining the history by having the patients enter it in a computer has several advantages. While computer-based patient interviewing programs have been experimented with for over 50 years, clinical adoption is limited due to challenges including those related to workflow and funding for deployment of such tools. A confluence of factors, such as the availability of computers and high-speed network connections, use of electronic medical records, and reimbursement predicated on the comprehensiveness of documentation make this a more favorable time for use of computer-based patient history taking. A recent pilot study at the Brigham and Women's Hospital indicated that patients responded favorably to computer-based interviews. Important lessons were learned on factors affecting physician adoption.


8/9/2007 James Signorovitch PhD
Title: Practical Optimality Theory for Large-Scale Multiple Hypothesis Testing
Abstract:
Very large data sets are often generated with the goal of discovering a few interesting patterns for further study. Genomewide association mapping, health surveillance, text mining and gene expression microarray studies, to name some current examples, all aim to discover the most promising signals against vast backgrounds of random noise. The potential for chance findings in such efforts is well-appreciated, and statistical tools for limiting false discoveries are widely used. However there has been little statistical theory to help us make as many true discoveries as possible while limiting chance findings. Progress has been made, with new ways of using information providing impressive gains in power -- but these improvements have not been guided by a practical optimality theory that tells us how to use all relevant information in the data. Further improvements could be possible.

This talk will begin by briefly surveying the history of multiple hypothesis testing as a scientific and a statistical endeavor. I'll then describe a concise and practical optimality theory for multiple hypothesis testing that tells us how to construct the best possible procedure, exploiting all available information, for a wide range of problems. The proposed procedures are evaluated through simulations and applications to gene expression microarray data. Surprisingly, the new theory tells us that commonly used testing procedures are far from optimal. Guided by the optimality theory, we propose new testing procedures that can substantially outperform existing methods by exploiting strong patterns in the data that have generally been ignored.


8/23/2007 Joseph Kvedar MD
Title: The Connected Health Imperative: using patient-centric technologies to improve quality, access and efficiency.


8/30/2007 David Kaelber MD PhD
Title: The Value of Information Technology-Enabled Diabetes Management.
Abstract:
Objective: To determine the financial and clinical costs and benefits of implementing information technology enabled diabetes management systems. Research Design and Methods: A computer model was created to project the impact of information technology enabled disease management on care processes, clinical outcomes and medical costs for type 2 diabetic patients over the age of 25 in the United States. Several information technologies were modeled: diabetes registries, computerized decision support, remote monitoring, patient self-management systems and payer based systems. Estimates of care processes improvements were derived from published literature, trade publications, the general press, and experts. Simulations projected outcomes for both payer and provider organizations, scaled to the national level. Results: All forms of information technology enabled disease management improved the health of diabetic patients and reduced health care expenditures. Over ten years, diabetic registries saved $14.5 billion, computerized decision support saved $10.7 billion, payer-centered technologies saved $7.10 billion, remote monitoring saved $326 million and self-management saved $285 million. Integrated provider-patient systems yielded $16.9 billion in savings. Provider-sponsored diabetes registries are projected to be the least expensive approach for small and medium sized practices. For large practices with electronic health record systems, it is most economical to modify such systems with diabetes-specific clinical decision support capabilities. Conclusions: Information technology enabled diabetes management has the potential to improve care processes, delay diabetic complications and save healthcare dollars. Of existing systems, provider-centered technologies such as diabetes registries show the most current potential for benefit. Fully integrated provider-patient systems would have even greater potential for benefit. Provider registries are generally the most economical approach.


9/14/2007 Chris Hinske MD
Title: Developing a Software for the Human Simulation Center
Abstract:
The Human Simulation Center is one of the recent projects of the Department of Emergency Medicine and Medical Management of the Ludwig-Maximilians-Universität München (http://www.inm-online.de). It i s a training center designed to train physicians and medical staff involved in emergency care in the setting of "the complete emergency transportation chain" (from accident to OR or ICU). It is also designed to investigate common error types that occur under stressful conditions and in the process of passing on information to others. One of the challenges was to develop a software that offers an intuitive user-interface for tutors as well as provides the flexibility to be used in changing settings and different environments.


9/21/2007 Bill Long PhD
Title: Lessons Extracting Diseases from Discharge Summaries
Abstract:
We have developed a program using very limited natural language processing and the UMLS to extract diseases and procedures from discharge summaries and have applied this program to 96 cases annotated by physicians. To evaluate the effectiveness of the program, we compared the concepts it extracted to those extracted by the annotators. The program extracts 93% of the desired concepts including some more specific than those the annotators used. This talk will discuss the reasons concepts were missed, including ambiguous phrases, phrases missing words or were separated or needing deduction, among other reasons. The false positives included insignificant findings, ambiguous phrases, or concepts that did not apply to the patient now. The analysis shows that extraction of medical concepts from discharge summaries with limited natural language processing and no domain inference is effective with still more potential.


9/28/2007 Leo Celi MD
Title: The Problem with Medicine: Re-Designing the ICU Workflow
Abstract:
Health care delivery is riddled with error-prone inefficiencies due to variation in task design. The management of the same patient admitted with pneumonia, for example, will vary from hospital to hospital, ward to ward, and doctor to doctor, with respect to type and frequency of laboratory testing, frequency of monitoring, antibiotics and other adjunctive treatments. This variation in practice is worse in the ICU, which is characterized by information overload. ICU workflow is a good reflection of how a system has grown by accretion rather than evolution. The information overload is driven by the false belief that outcomes are positively influenced by the amount of information and minute-to-minute manipulation of physiologic variables. The challenge for informatics in the ICU is not just to translate tasks and processes; an outcomes-driven transformation is necessary. The possible role of AI to this end is presented. Finally, barriers to the development and implementation of workflow re-design and other informatics applications in the ICU are discussed.


10/5/2007 Angela Haslbeck
Title: Remote Patient-Monitoring, Diagnostics and Therapy - An Attempt to Predict Innovation and Industry Evolution Using the Delphi Method
Abstract:
Advances in wireless networking technologies and the miniaturization of vital parameter sensors have, in the past half-decade and especially in the U.S., induced entrepreneurial activity and investment from venture capital firms into an emerging industry for mobile telemedicine solutions. The impending cost collapse of health care systems and hopes for the telemedicine industry to become an important economic factor have also led policy makers in Europe to set up projects to develop and validate innovative systems and services for remote and mobile health care. The development of the Mobile Health Industry might accelerate in the next years due to further technological advances in the field of biomedical computing and trends towards integrated care and disease management. Angela' s research aims to forecast the future technological and market development of remote and wireless monitoring in the U.S. using the Delphi Method, a widely approved qualitative forecasting approach utilizing expert opinion. Findings from the study she then will use to test Christensen et al.'s hypothesis on disruptive change in the healthcare industry.


10/12/2007 Steven Seltzer MD
Dr. Steven Seltzer is the Chairman of the Department of Radiology at the Brigham and Women’s Hospital.\\ Title: New Research Initiatives at the Department of Radiology


10/19/2007 Meghan Dierks MD
Title: Technology to Enhance Institutional M&M and QA Processes
Abstract:
Morbidity and mortality (M&M) conferences and peer review activities constitute important forums for the exploration and discussion of significant adverse clinical events. These activities are an ingrained component of hospital quality assurance and provide opportunities to systematically review system-based and individual factors that may have contributed to a patient injury, an unexpectedly poor clinical outcome or an adverse event with some degree of transparency. However, recent published reports suggest that there is wide variation in the methods by which cases are identified, criteria for case review and formal presentation, depth of review, format for presentation, actions required in response to the event/case, and final disposition of cases. This variation is prominent at both the inter-institutional and intra-institutional level (i.e., between different departments or disciplines within an institution). To counter this trend, we have developed a data- and workflow- management system for the M&M/QA process. The system is a web-enabled and database-backed application that establishes a minimal structure, format and data set for investigation and analysis of an adverse event, provides a platform for collaborative cross-disciplinary review, commentary and follow-up on a case, enables comprehensive classification for subsequent analysis and trending, and provides opportunities to audit case status. While the system has improved workflow relating to case review, we are also evaluating the extent to which the system actually changes behavior around reporting morbidities/mortalities, enhances the institution's early recognition and understanding of important system-wide safety issues, and enables rapid dissemination of this knowledge throughout the institution. Specifically, we will determine the extent to which the system: 1. lowers the threshold for reporting/submitting cases for review 2. changes the demographics of who initially reports a case 3. increases collaborative, multidisciplinary review/critique of cases that cross disciplinary boundaries 4. improves tracking of the status of action plans/interventions emerging from a specific adverse event 5. standardizes the process and quality of case review across departments 6. enables identification of system-based or common cause failures/errors across different departmental boundaries (Provide capability for institution-wide aggregation/review of data)


10/26/2007 Jihoon Kim MS
Title: Difference-based clustering of short time-course microarray data with replicates
Abstract:
Background
There are some limitations associated with conventional clustering methods for short time-course gene expression data. The current algorithms require prior domain knowledge and do not incorporate information from replicates. Moreover, the results are not always easy to interpret biologically.
Results
We propose a novel algorithm for identifying a subset of genes sharing a significant temporal expression pattern when replicates are used. Our algorithm requires no prior knowledge, instead relying on an observed statistic which is based on the first and second order differences between adjacent time-points. Here, a pattern is predefined as the sequence of symbols indicating direction and the rate of change between time-points, and each gene is assigned to a cluster whose members share a similar pattern. We evaluated the performance of our algorithm to those of K-means, Self-Organizing Map and the Short Time-series Expression Miner methods.
Conclusions
Assessments using simulated and real data show that our method outperformed aforementioned algorithms. Our approach is an appropriate solution for clustering short time-course microarray data with replicates.


11/2/2007 Ivar Helgason MD
Title: The Theriak Medication Management (TMM)
Abstract:
TMM is a "closed loop" medication system, handling everything from prescription, pharmacy review and logistics to bedside verification. TMM is based on the unidose concept, and controls dispensing robots in the pharmacy and in some cases automated ward delivery systems. I will talk about our design philosophy and the challenges when implementing large clinical systems.


11/9/2007 Joaquin Blaya MS
Title: Implementing and evaluating laboratory information systems in resource-poor settings
Abstract:
Multi-drug resistant tuberculosis (MDR-TB) patients in resource-poor settings experience large delays in starting appropriate drug regimens and are often not monitored appropriately due to an overburdened health care system, communication delays, and missing or error-prone data. Information management and communication in medical care are especially critical in resource-poor settings where increased barriers to treatment include lack of coordination between national, regional and local health institutions, inefficient communication systems, and high turnover rates among personnel. Medical information systems can be used to alleviate these problems by increasing the timeliness and quality of laboratory information available. This thesis proposes to develop and implement such systems in the urban, resource-poor setting of Lima, Peru in both networked and non-networked institutions. Formal evaluation of these systems will then assess their effect on administrative and clinical metrics. The work presented will be divided into two parts. The first focuses on the electronic communication and reporting of tuberculosis laboratory information to health care personnel within a networked group of institutions. This should reduce the communication times of laboratory results, accelerate the start of treatment for high-risk patients, and decrease the frequency of errors. The second addresses the electronic collection of tuberculosis (TB) laboratory information from a distributed group of non-networked laboratories to reduce delays, errors and costs. These two parts will provide verified medical informatics tools for settings both with and without internet connectivity.


11/30/2007 Winston Kuo, DDS, MS, DMSc
Title: The Roles of miRNAs in Craniofacial Development


12/07/2007
Presenter: Melody In Chang PhD
Title: Regional Health Information Exchange- Experience in Taiwan
Abstract:
Health information technology has been identified as a key tool in addressing the major challenges that health care faces in efficiency, safety, and quality. With the growing healthcare expenditure, the government of Taiwan has shown its commitment to health information technology by initiating a five-year project, called National Health Information Project (NHIP), to establish a nationwide infrastructure for electronic medical record and health information exchange. A case of regional health information exchange in Taiwan, South Taiwan Medical Information Network (SMIN), will be presented. Some lessons learned in the SMIN project about data standards issues, policy and organizational support issues, information sharing issues, and privacy/security issues are discussed. These experiences, we expect, will be incorporated into the next generation of local health information infrastructure for building the nationwide network in the future.


12/14/07
Presenter: Hamish Fraser, MBChB, MRCP, MSc
Title: Clinical Information systems as tools to improve quality of care in developing countries: examples from HIV and TB treatment programs
Abstract:
The scale-up of HIV treatment in developing countries and the parallel creation of large scale treatment programs for multi-drug resistant tuberculosis (MDR-TB) require the creation of systems for chronic disease management in places where short-term care, or no care at all, is the tradition. Partners In Health (PIH), a health care non-profit based in Boston, provides care to some of the poorest communities in countries such as Haiti, Peru, Rwanda and Lesotho. PIH has pioneered the treatment of HIV and MDR-TB by focusing both on building capacity in local communities and bringing the best modern medications, investigations and training to these impoverished environments. One of the tools PIH has developed and deployed is a web-based electronic medical record system to track the treatment ofthese patients and their lab results and clinical progress. I will briefly describe the PIH-EMR and discuss how we use it to reduce medical errors, delays and oversights in treatment. I will also describe the evaluations we have performed on the system in use, and discuss the critical need for systems to reduce loss to follow-up in HIV treatment programs in Africa. Finally I will introduce the OpenMRS EMR architecture developed in collaboration with the Regenstrief Institute in Indiana and other groups in Africa.


12/21/07
Presenter: Blackford Middleton, MD, MPH, MSc
Director, Clinical Informatics Research and Development, Partners Healthcare
Title: Applied Clinical Informatics Research Highlights from Partners Clinical Informatics Research & Development (CIRD) and Center for Information Technology Leadership (CITL)
Abstract:
In this presentation, Dr. Middleton will provide an overview of the Clinical Informatics R&D group, and the Center for Information Technology Leadership, at Partners Healthcare. Selected research studies will be highlighted, and issues surrounding conducting applied research in an operational information systems environment will be discussed. An overview of a new award for advanced clinical decision support research will also be presented.


1/11/08
Presenter: Griffin Weber, M.D., Ph.D.
CTO, Harvard Medical School
Title: CTSA Data Sharing
Abstract:
The Clinical and Translational Science Award (CTSA) is an NIH-funded consortium of academic health centers located throughout the nation whose goal is to transform how clinical and translational research is conducted. Harvard submitted its proposal in November to become a CTSA institution. The 1,300 page proposal was the result of a two-year collaborative effort involving hundreds of individuals from Harvard and its affiliated hospitals and institutions. Novel tools for clinical research informatics are a core component of the proposal. One of these tools is the Shared Health Research Information Network (SHRINE), which will allow investigators to search the clinical data warehouses at each of the Harvard hospitals--a total population of 10 million patients. SHRINE will enable multi-institution queries to identify patient populations for clinical trials, and it will provide data analysis tools to help answer common questions investigators ask about their patient cohorts. SHRINE is an extension of software created for Informatics for Integrating Biology and the Bedside (i2b2), which is a National Center for Biomedical Computing based at Partners HealthCare System. This talk will present an overview of CTSA informatics at Harvard, show a demo of i2b2 software, and describe some of the challenges in creating SHRINE.


1/18/08
Presenter: Luke Sato, M.D.
CMO, Harvard Risk Management Foundation
Title: IT for Risk Management


1/25/08
Presenter: Kumiko Ohashi, RN, PhD
Title: Network and sensor technologies for medical fields - How can they ensure patient safety, and improve quality and efficiency of medical care?
Abstract:
In recent years, ubiquitous computing technologies have been applied in the field of medicine. Radio frequency identification (RFID) and small sensor networks provide information about medical practice and patient status that, in turn, improve the quality of medical care. We developed a new system named the "smart stretcher," which continuously monitors patients' vital signs and detects apnea during transfer within a hospital. We developed a bedside safety system for improving medical efficiency and patient safety at the bedside. I will talk about these systems and some other future projects.


2/01/08
Presenter: Jonathan Jackson MEng
Title: Health technology in resource constrained settings
Abstract:
Dimagi is a Cambridge based startup from the MIT Media Lab and Harvard that focuses on providing technology solutions for healthcare in resource constrained settings. Dimagi's CEO, Jonathan Jackson, will discuss current projects as well as preliminary results from field trials in South Africa and Tanzania using decision support on mobile devices. His talk will also focus on the challenges and lessons learned from his first-hand field experience.


2/08/08
Presenter: Ronilda Lacson MD PhD
Title: Predicting Hemodialysis Mortality Using Serial Blood Pressure Measurements
Abstract:
Blood pressure is a significant predictor of mortality. This finding applies to hemodialysis patients who have the advantage of having serial blood pressure measurements recorded before and after each hemodialysis treatment. Several studies have focused on the effect of both high blood pressure and low blood pressure on mortality risk. In most studies, however, blood pressure is utilized as a fixed, often single average value, which summarizes all available blood pressure measurement for each individual. This method provides an acceptable representation of the overall blood pressure as well as limiting the number of variables in the prediction model(s). However, this method does not provide a good representation for the blood pressure variability and the trends over time, both of which are readily available from patients' data. I will discuss the results of utilizing time-series modeling techniques with support vector machine to predict hemodialysis mortality that incorporates serial systolic blood pressure measurements over time.


2/15/08
Presenter: Staal Vinterbo PhD
Title: Practical tools and tricks to keep your information safe
Abstract:
I will discuss some information on security issues and point out some steps one can take to avoid some of these. Particular focus will be on behavior on the net, communication security, authentication and how to manage passwords.


2/29/08
Presenter: Winston Hide PhD
Title: A computational approach towards the understanding of Tumor Initiating Cell gene expression
Abstract:
Gene expression data is challenging in its heterogeneity and breadth. By combining data my group has accessed from Microarray, Massively Parallel Signature Sequencing (MPSS), Expressed Sequence Tag (EST), rt-PCR and Capped Analysis of Gene Expression(CAGE), it is possible to overcome artifacts and biases inherent in each of the platforms, and exploit the strengths of each. Together with our collaborators my group has developed an analysis platform for RNA from CD133hi cells that have been extracted from melanomas of patients. The CAGE analysis platform provides unique insight into internally consistent transcription start site (TSS) expression level and switching, together with direct context of publicly available FANTOM 3.0 CAGE TSS. We have discovered melanoma progenitor specific signals that will be discussed.


3/14/08
Presenter: Qing Zeng-Treitler PhD
Title: Developing Informatics Tools to Empower Patients
Abstract:
In the presentation, I will briefly review our work in the area of consumer health informatics including vocabulary, information retrieval and health literacy, and then highlight a few new research directions we are pursuing. I will also discuss a few areas that we are interested in exploring.


3/21/08
Presenter: Chris Tsai MD MPH
Title: Characterizing Calibration in Cardiovascular Risk Prediction Models
Abstract:
Risk prediction models have been employed in cardiovascular disease since the Framingham Study in 1950s. These global risk assessments are the basis for guidelines-based preventive interventions, including drug therapy. External validation has shown to be sometimes poor (while discrimination is usually adequate, calibration is not). Thus, absolute risk for an individual may not be reliably predicted. We are conducting a systematic review specifically addressing the calibration of cardiovascular models in external populations. I'll describe the results we have so far.


3/28/08
Presenter: Francis X. Campion MD FACP
Director, Complex Chronic Care Program, Harvard Vanguard Medical Associates
Clinical Informatics Research, HMS Department of Ambulatory Care and Prevention
Title: Future of Public Health: Automating disease reporting and vaccine safety monitoring


4/04/08
Steven Labkoff MD, FACP
Senior Director, Pfizer Inc.
Title: Life after Fellowship: Getting into and Working in the Business World
Abstract:
After Medical Informatics fellowship is completed, a significant number of fellows will pursue interests that lie in the business world. While fellowships get one ready for a wide array of activities in life, I have found that there is a significant opportunity to learn more about the business world before you venture in. I will discuss pathways into business, tools needed to learn about the business world, the role of recruiters, what areas in industry are looking for Informatics Fellows and more.


4/11/08
Carlos Nakamura PhD
Title:
The effects of specific support to hypothesis generation on the diagnostic performance of medical students Abstract:
Some researchers have argued that hypothesis generation is far more challenging to medical students than hypothesis evaluation because they are initially taught to reason from the diseases (hypotheses) to the symptoms (clues) rather than the other way around. In this presentation I report the results of a study on the effects of computer-based support to hypothesis generation on the diagnostic performance of 2nd and 4th year medical students.


4/18/08
Aziz Boxwala MD PhD
Title:
Reliable execution of important care processes
Abstract:
Errors in clinical care, especially those related to communications and handoffs in care, are a major cause of morbidity and mortality. For example, various studies have found that 10 to 30% of patients with positive mammograms do not have a timely follow-up. Current approaches, including informatics tools, to detect and prevent such failures are limited in their impact. I will describe a software system called ACE, that I am involved in the development of, for monitoring patient's clinical data. It can monitor the patient's status over long periods of time, detect important clinical events, notify a responsible provider, and then ensure that appropriate actions are taken.


4/25/08
Xiaole Shirley Liu PhD
Title:
ChIPing the Human Cistrome
Abstract:
Cistrome defines the set of cis-acting targets of a trans-acting factor on a genome scale. I will discuss our analysis of ChIP-chip on genome tiling microarrays and ChIP-seq on Solexa sequencing to discovery human cistromes. I will also discuss how to use the cistromes to understand transcriptional and epigenetic regulation of gene expression in cancer.


5/2/2008
Pedro Galante PhD
Title:
The complexity of the transcriptome: a large-scale analysis of intron retention and genes sense-antisense in the human and mouse genomes.
Abstract:
I will present a large-scale analysis of intron retention (a type of alternative splicing) and genes sense-antisense in the human and mouse genomes.


5/9/2008
Heimar Marin, RN, PhD and Eduardo Marques, MD, PhD
Federal University of Sao Paulo
Topic: Criteria for Certification of Hospital Information Systems software Abstract:
Education on Health Informatics - IMIA Recommendations: In this presentation I will share the updated recommendations that IMIA is preparing related to the curriculum content to enable health care professionals to efficiently and responsibly use information and knowledge processing methodology and information and communication technology. I will also introduce some activities of the Brazilian Health Informatics Society related to the Process of Certification for Software in Healthcare. In the second part, a methodology for strategic planning of health information systems will be presented with the following goals:

- to translate the strategic objectives into a set of coherent performance indicators;
- to associate strategic objectives with long term targets and annual budgets;
- to clarify and obtain consensus regarding the strategy;
- to communicate strategy to all the organization;
- to align institutional and personal targets to strategy;
- to identify and align all the strategic initiatives;
- to stimulate timely and systematized strategic revisions and
- to obtain feedback to increase the knowledge about the strategy and to improve it


5/16/2008
Wallace Leung, PhD
Hong Kong Polytechnic University
Title: Clinical Decision Support Systems
Abstract:
We are developing a Clinical Decision Support System (CDSS) in RIIPT integrating various info (protein & gene biomarkers, patient data, lab test data, imaging etc.). By using rule-based, case-based or model-based engines (intelligent algorithms such as support vector machine, neural network, Bayesean network, AI etc.), we can analyze the extracted features and correlate the data for analysis and forecast. It can be used for diagnosis, prognosis and therapy. We are currently applying this for lung and Nasal pharangeal cancer but have interest to extend to any diseases, especially chronic diseases (cardiovascular, kidney, liver, etc.). Also we are partnering on this project with hospitals in Hong Kong and our biology and computing departments at the university.


5/23/2008
Michael Matheny MD MS MPH
Vanderbilt University
Title: Towards Automated Medication Outcomes Surveillance Among Hospitalized Patients in the Veteran's Administration
Abstract:
Post-marketing medication surveillance has received strong attention from the public and media in the last few years as a result of high profile medication and medical devices recalls by the FDA. As a result of a series of public hearings and IOM solicited recommendations, the FDA has begun to focus on clinical registry and routine clinical data as complementary sources to its existing surveillance network. Effective use of each of these data sources for medical product surveillance faces a number of challenges at the intersection of medical informatics, observational cohort biostatistics, and pharmacoepidemiology.

This discussion will focus on aspects of the VA electronic health record and the current VA Health Services Research & Development research agenda that provide unique opportunities for medical informatics research, highlight particular opportunities and challenges for using natural language processing on VA free text clinical records to produce computable concepts, facts, and values, and underscore the importance of risk modeling and risk stratification using both traditional and machine learning approaches as the keystone for addressing confounding in observational cohorts as well as predicting the propensity for use of medications or devices.


5/30/2008
Pankaj Sarin MD MS
Title: THE INTEGRATED CLINICAL EDUCATION (ICE) SYSTEM: A PILOT OF A SERVICE SPECIFIC EDUCATION PORTAL EMBEDDED INTO CLINICAL WORKFLOW
Abstract:
Clerical tasks required to provide patient care take up significant physician time but often do not directly have significant educational value. Educational questions may arise during these tasks but are often lost to follow-up because of time constraints. In anesthesia, the process of finding and assembling information needed for a preoperative patient assessment is one such example. Resident physicians navigate through a number of unlinked electronic based and paper based systems to locate information, taking away from time that can be spent with patients or discussing cases with attending physicians. We aim to develop an integrated web based portal application that will significantly reduce the time the above clerical tasks take, integrate patient, case, and physician level of training specific education (literature and protocols/guidelines) into the workflow, and improve collaboration between attending and resident level physicians. This proposed Integrated Clinical Education (ICE) system will be created by enhancing previously created Partners IS and Dept of Anesthesiology technologies. Integration of technologies used by non-medical information web sites, such as mash-up sites that use Google services, Amazon.com "Recommendations for You", and del.icio.us social bookmarking, will also occur. The proposed ICE system will be covered during the presentation.


6/13/2008
Jeeyae Choi RN PhD
Title: Development and Evaluation of a Computerized Decision Support System for Genetic Counselors in Korea
Abstract:
Cancer rates for Korean women have been steadily increasing. The incidence of breast cancer among Korean women was 8.7% in 1985, and that rose to 11.4% in 1992 and to 12.5% in 1997. Other statistics showed that not only incidence but also the mortality rate of Korean women who are diagnosed with breast cancer have been continuously increasing since the 1980s.

Genetic counseling has been used for prevention and early diagnosis of breast cancer in the primary care settings. Korean clinicians have utilized genetic counseling as a part of their routine care since 1996. They use various breast cancer risk assessment models to assess the breast cancer risk during a counseling session. However, these models have been developed in western societies and are difficult to use within a limited time of genetic counseling.

The project proposes to validate four models, which are Gail, Claus, Couch and BRACAPRO, among Korean population and develop a suitable model for Koreans. Then, we would like to design and implement a computerized decision support system that can provide a breast risk assessment value to the clinician at the point-of-care.


6/20/2008
Robert A. Greenes, MD, PhD,
Ira A. Fulton Chair and Professor,
Department of Biomedical Informatics,
Arizona State University
Abstract:
DSG's founder will present an overview of BMI research and teaching activities at ASU.


8/25/2008
David Ahern PhD
Title: The Role of HIT/eHealth in Patient Engagement and Quality Improvement
Abstract:
HIT/eHealth is the use of technology platforms, e.g., EMRs and emerging interactive technologies to enable health improvement and health care services. Dr Ahern will discuss the role of eHealth in various initiatives. He will talk about how eHealth can help diverse consumers and patients to become:
1. more informed and knowledgeable about their health
2. activated in addressing their health concerns with their providers
3. skilled in self-management


9/5/2008
Lucila Ohno-Machado MD PhD
Title: State of the DSG
Abstract:
She will do a review of 2007-2008, highlighting achievements, and discussing DSG goals for 2008-2009.


9/12/2008
Frederic Resnic MD MS
Director, BWH Cardiac Catheterization Lab
Title: Automated Surveillance of Clinical Registries to Monitor Medical Device Safety
Abstract:
Post-approval surveillance of the safety of medical devices and drugs has become a topic of intense interest for healthcare providers, regulators, government agencies and patient advocates. With the continued expansion of routine clinical registries, an opportunity for automated surveillance of both expected and unexpected adverse events is becoming increasingly feasible. In conjunction with collaborators from the Decision Systems Group, Fred Resnic has helped to develop an automated platform for the continuous monitoring of medical device registries, with flexible statistical monitoring strategies and alerting algorithms. The system, called DELTA, has been evaluated in a single center environment and validated with prospective randomized trial data, and is now undergoing further validation in a multi-center implementation in the state of Massachusetts. This seminar will review the structure of the DELTA system, the initial system evaluations and describe the ongoing multi-center validation study.


9/19/2008
Aziz Boxwala, MD PhD
Title: Knowledge Representation in the CDS Consortium
Abstract:
The Clinical Decision Support Consortium is a multi-institutional collaborative funded by the Agency for Healthcare Research and Quality. The objective of the CDSC is to improve the speed with which evidence-based guidelines are integrated into practice using clinical decision-support (CDS) tools. One of the factors responsible for the lack of incorporation of guidelines into CDS is the effort required to translate the textual recommendations into computable knowledge. Therefore, an important part of the solution that the CDSC is designing is to disseminate knowledge in a format that is structured for implementing into CDS tools. The heterogeneity of the clinical decision-support systems, the clinical information systems, and the practice workflows make this endeavor challenging. The CDSC is proposing a multilayered representation format in which knowledge is increasingly structured in successive layers. This approach allows delivery of knowledge for a variety of implementation targets.


9/26/2008
Dorothy Curtis MS
MIT Networks and Mobile Systems Group
Computer Science and Artificial Intelligence Laboratory
Department of Electrical Engineering and Computer Science
Title: Physiological Signal Monitoring in the Waiting Area of an Emergency Room
Abstract:
The Scalable Medical Alert and Response Technology (SMART) System was developed to monitor physiological signals from patients in the waiting areas of an emergency department. The system monitors the SpO2 (oxygenation level in the blood), ECG (electrical activity of the heart) and the location of multiple patients wirelessly. It was deployed at the Brigham and Women’s Hospital in Boston, MA, between June, 2006, and December, 2007. This talk will describe the overall architecture, the sensors used, challenges in deploying this technology in a hospital and the degree of patient acceptance. Some potential future projects will be mentioned, time-permitting.

Dorothy Curtis is a Research Scientist at the Computer Science and Artificial Intelligence Laboratory (CSAIL) at MIT. In addition to physiological monitoring and location systems, her real-time systems work includes computer networking and shipboard collision detection and avoidance and navigation. Other computer science interests include natural language processing, compilers, and object-oriented databases. She has degrees from MIT, BU, and Simmons.


10/3/2008
Leonard D'Avolio, PhD
VA Boston Healthcare System
MAVERIC
Title: Using Information Extraction to Facilitate Surgical Outcomes Assessment
Abstract:
The most common surgical treatment for prostate cancer is prostatectomy, or surgical removal of the prostate. Despite the prevalence of this treatment, an estimated 28% of procedures result in tumor left at the margin of resection, increasing the risk of cancer recurrence in those patients by two to four times.

Like many unintended surgical outcomes, positive tumor margins are the consequence of surgical technique. Yet the lack of accessible information describing the surgical procedure has led outcomes assessment researchers to rely on indirect but quantifiable correlates such as hospital volume and surgical experience.

This research tests the hypothesis that information extraction techniques can facilitate more robust *process-based* surgical outcomes assessment by identifying and structuring patient conditions prior to surgery, key variations in procedures, and the result of the surgery from clinical records. The results of an attempt to automatically extract "before, during, and after" data from the reports of two hospitals will be presented along with discovered inconsistencies in urologists' approach to the prostatectomy.

This talk will be followed with an overview of the national VA resources available to DSG researchers through the VA Boston Healthcare System and the Massachusetts Veterans Epidemiology Research and Information Center (MAVERIC).


10/10/2008
Staal Vinterbo PhD
Title: Faking Congressional Voting Data
Abstract:
Low range categorical data offer particular problems in the context of privacy. This is particularly worrying as genetic data becomes increasingly available. I will present an argument that there it is possible to create artificial binary data that is 'safe' and 'useful'.


10/17/2008
Richard Lu MD
Title: Decision Support using American College of Radiology Appropriateness Criteria
Abstract:
Clinical practice guidelines are position statements published by medical professional societies that are designed to establish a uniform baseline for medical practice. In the radiology domain, the American College of Radiology's Appropriateness Criteria are designed to control costs and also to ensure that the correct test is ordered for each clinical scenario. Hundreds of radiologic tests are ordered at Brigham & Women's Hospital each day, but little is known about how appropriate these tests are according to commonly accepted standards of care. The talk will present a project proposal that aims to build a web-based, open source decision support tool that can lead to better understanding of radiologic test ordering patterns by healthcare providers.


10/24/2008
Adam Wright PhD
Clinical Informatics Research and Development, Partners Healthcare System
Title: Association Rule Mining for Clinical Decision Support
Abstract:
Substantial evidence exists which suggests that well-designed clinical decision support systems can be powerful tools for improving the quality and safety of healthcare. Developing clinical decision support systems, however, is generally a time-consuming and difficult manual process. In this talk I give an overview of two closely related and complementary data mining techniques: frequent itemset mining and association rule mining and discuss their application to the automated development of clinical decision support content based on electronic medical record data.


10/31/2008
Charles J. Weitz, M.D., Ph.D.
Robert Henry Pfeiffer Professor of Neurobiology
Harvard Medical School
Title: Distributed circadian clocks in mammals: what do they do?


11/7/2008
Kumiko Ohashi RN PhD
Sergey Goryachev MS
Sasikiran Kandula MS
AMIA Annual Meeting Presentations


11/21/2008
Angela Haslbeck
Title: Developing Measures for Clinical Decision Support
Abstract:
Measures play an important role in ensuring quality of care, and are increasingly used to determine reimbursement. While there have been evaluation studies for clinical decision support (CDS) systems in the past, CDS is still lacking comprehensive operational or run-time measures to identify and improve on problems, as well as to evaluate the implementation of a CDS from a quality of care perspective. The aim of the project presented is to develop a comprehensive framework of measures for guideline-based CDS, and apply it to the clinical decision support use cases developed by the Clinical Decision Support Consortium (CDSC).


12/05/2008
Chris Hinske MD
Title: Silencing the Host - The Role of Intronic miRNAs
Abstract:
Fifteen years ago lin-4 was reported to be the first endogenous small non-coding, but interfering RNA structure involved in developmental timing in C. elegans. First thought not, or only rarely, to occur in mammals, microRNAs are now among the major players in up-to-date genomic research. The mature molecules are ~22 nucleotides in length and, by targeting predominantly the 3¹ UTR of mRNAs, lead to translational repression or degradation of the target message, hence controlling important cellular mechanisms, including division, differentiation and death. This key role makes them excellent targets for cancer research. In fact they have been shown to have a major impact on cancer development in many cases. However, miRNAs are not a homogeneous class and can be subclassified into intragenic and intergenic, depending on their position on the genome. Whereas intergenic miRNAs are expected to be independent transcriptional units, intragenic miRNAs are commonly believed to be regulated through their host gene. Despite of the growing knowledge on how miRNA integrate into cellular regulatory networks, our current knowledge about the specific role of intronic miRNAs is rather limited. In this work we integrated current miRNA knowledge bases, ranging from miRNA sequence and genomic localization information to target prediction, with biochemical pathway information and publicly available expression data to investigate functional properties of intronic miRNAs and their relationship with their host genes. To the best of our knowledge, we are the first to show in a large-scale analysis that intronic miRNAs seem to act as negative feedback regulators on multiple levels. We furthermore investigate the impact of this model on the potential role of intronic miRNAs in cancer pathogenesis.


12/12/2008
Carlos Nakamura PhD
Title: Development of a pictographic grammar and guidelines for converting text into pictographs
Abstract:
This ongoing study is part of larger project intended to complement patient discharge instructions with pictographs to enhance comprehension and recall among low-liteacy patients. The system is based on automatic text-to-pictograph conversion using information extracting techniques. In the descriptive stage, health-related pictographs were collected and analyzed from a morphological, syntactic, and semantic perspectives. In the ongoing prescriptive stage, guidelines for converting text into pictographs and the creation of a pictographic grammar are being formulated. The theoretical and practical constrainsts in the creation of a pictographic communication system will be discussed.


1/09/2009
Rosa Figueroa Iturrieta
Title: Exploring Active Learning in Medical Text Classification
Abstract:
The task of classifying documents using machine learning and natural language processing techniques often requires a large labeled training set. However, creating a large labeled training set is time consuming and costly. It is thus important to minimize the size of the required training set.
Active learning is an interactive way of selecting the training data by finding the most informative instances from an unlabeled pool to request the true class label. The existing research in active learning shows that active learning methods can reach a better performance using fewer examples than the traditional random selection approach. We are exploring the use of active learning techniques to select samples for medical text classification. We experimented with three algorithms to interactively select informative and diverse training instances, the results of which will be discussed.


1/16/2009
Catherine Arnott Smith, PhD
Assistant Professor
School of Library and Information Studies
University of Wisconsin-Madison
Title: PatientsLikeMe: Consumer Health Vocabulary as Folksonomy
Abstract:
PatientsLikeMe is an online social networking community for patients. Subcommunities center on sixteen diagnoses: the oldest are oriented around Amyotrophic Lateral Sclerosis, Multiple Sclerosis and Parkinson’s Disease. Community members can describe their symptoms to others in natural language terms, resulting in folksonomic tags available for clinical analysis and for browsing by other users to find “patients like me”. Forty-three percent of PatientsLikeMe symptom terms from the original three diagnostic subcommunities are present as exact (24%) or synonymous (19%) terms in the Unified Medical Language System Metathesaurus (National Library of Medicine; 2007AC). Slightly more than half of the symptom terms either do not match the UMLS, or are unclassifiable. Analysis of the failed matches reveals challenges for online patient communication, not only with healthcare professionals, but with other patients. In a Web 2.0 environment with lowered barriers between consumers an d professionals, a deficiency in knowledge representation affects not only the professionals, but the consumers as well.


Jihoon Kim
Title: ExpressionCombiner: a web-based tool for cross-platform analysis of gene expression data
Abstract:
ExpressionCombiner is a tool that integrates multiple expression datasets from different technologies into a single file to enable meta-analysis. The input is a text file in the format of a probe by sample expression matrix. The output is a gene symbol by sample expression matrix in a text file. Users can remove lower quality probes by filtering based on their sequence matching onto gene, transcript and SNP. By combining datasets, the user can increase the statistical power and validate her own experimental data with published data obtained from public repositories. Appropriate filtering will ensure unambiguous interpretation of the results at the gene-level, provide expression measurements withstanding alternative-splicing and remove allele-specific gene expression. ExpressionCombiner is available at https://brca.partners.org/ExpressionCombiner.


Freddy Bafuka
Title: Image-Based Evaluation of Health-Related Document Readability
Abstract:
Many studies have shown that the readability of the health information provided to consumers does not match their reading levels. In this project, we explore an purely image-based approach to evaluate the readability level of documents, as a first step toward the end-goal of translating them from to reader-appropriate readability level.


Erik Pitzer MS
Title: Large Scale Microarray Data Re-Use
Abstract:
In the past, large repositories of microarray data have been created and accumulated a wealth of information. Unfortunately, these repositories seem more like data dumpsters because of their unstructured annotation. We have created a system to salvage the available information and open up a whole wealth of new possibilities.


Dominik Aronsky MD PhD
Assistant Professor of Biomedical Informatics
Director, Academic Programs for Fellowships, Infrastructure, and Interdisciplinary Training
Vanderbilt University Medical Center
Title: Computer-based Decision Support in the Emergency Department
Abstract:
The Emergency Department is a fast-paced, information intensive environment that can benefit from improved information management. The presentation will discuss how an integrated information system infrastructure can support providers to deliver high-quality patient care, optimize operational activities, and facilitate clinical and informatics research studies in the Emergency Department. Illustrative examples will include improvement of pneumonia-care processes, implementation of asthma guidelines, and forecasting Emergency Department overcrowding.


Alex Turchin MD MS
Senior Medical Informatician, CIRD, Partners HealthCare System
Department of Endocrinology, Diabetes and Hypertension, Brigham and Women's Hospital
Title: Using EMR Data in Clinical Research: Lessons Learned
Abstract:
Electronic medical record (EMR) systems have a great potential as a source of data for clinical research. They cover multiple aspects of medical care, include information on thousands of patients and allow for rapid data processing. However, using EMR data is not always a straight road to success. This presentation will describe common pitfalls in EMR data analysis with the emphasis on applications of natural language processing. Particular attention will be paid to the issues of data quality and validation of text analysis tools including stratified and prospective validation. This discussion will be illustrated on the example of a study of association of anti-hypertensive treatment intensification as a measure of quality of care and physician board certification.


Ronilda Lacson MD PhD
Title: Biomedical Data Annotation
Abstract:
This study describes a large-scale manual re-annotation of data samples in the Gene Expression Omnibus (GEO), using variables and values derived from the National Cancer Institute thesaurus. A framework is described for creating an annotation scheme for various diseases that is flexible, comprehensive, and scalable. Overall, we show that it is possible to perform manual re-annotation of a large repository in a reliable manner.


Robert El-Kareh MD
Title: Diagnosis Errors: Can Informatics Help?
Abstract:
Diagnosis errors are common and can have significant consequences for patients and clinicians. Despite their importance, these errors remain understudied. Highly computerized healthcare systems generate huge amounts of electronic data and provide an opportunity to expand our knowledge in this important field. This presentation will provide a brief overview of diagnosis errors, followed by the description of specific ways that electronic data can be used to identify and prevent diagnosis errors. Three ongoing projects will be discussed to illustrate these applications.


Guido Davidzon MD
Title: Improving Mortality Prediction Among Patients with Subarachnoid Hemorrhage
Abstract:
Current mortality prediction models in the intensive care unit discriminate and calibrate poorly when applied to specific disease-related subsets of patients. We hypothesize that mortality prediction can be optimized for patients with subarachnoid hemorrhage (SAH) by using a predictive model fitted with local institutional data. We developed an SAH mortality prediction model using heuristic-driven feature selection and compared its performance against the Simplified Acute Physiology Score (SAPS). We studied 223 ICU patients with an ICD-9 code for SAH seen at Beth Israel Deaconess Medical Center from 2003 to 2007. Best-fitted models were selected based on performance of tenfold cross-validation on a training set and subsequently evaluated on unseen data. Results: Using a locally re-weighted SAPS in this population resulted an AUC of 0.84 with a Hosmer-Lemeshow (HL) p-value of 0.2 . Our logistic regression model using age, leukocyte count, serum glucose, blood pressure variability and Glasgow Coma Score yielded an AUC of 0.95 with a non-significant HL p-value of 0.88. Mortality prediction was improved through the use of a model built with local institutional data from a homogeneous subset of patients.\\


Richard Lu MD
Title: Mortality Prediction in Patients with Septic Shock
Abstract:
Very few models exist that predict mortality from septic shock. The objectives of this research include (1) To utilize machine learning algorithms in predicting mortality among septic shock patients, and (2) To identify variables that are most highly associated with mortality. Methods: Data from patients seen at the Beth Israel Deaconess Medical Center from 2003-2007 on 1,372 patients with septic shock was obtained. Logistic Regression, Neural Network, Classification Tree and Bayesian Network models were developed and evaluated. Calibration of the logistic models was performed using Hosmer Lemeshow goodness of fit. Results: Lactic acid level is the best single predictor of mortality in septic shock (OR=1.33, p<0.000001) using logistic regression (AUC=0.71). Models using Neural Network, Classification Tree and Bayesian Network have AUCs of 0.71, 0.66 and 0.74, respectively. Conclusion: This research provides empirical evidence that machine learning algorithms can predict mortality in patients who develop septic shock.


Thomas Brox Roest
Title: Digging deep into primary care patient records: Some research results
Abstract:
My research is focused on finding ways of navigating information in primary care patient records. For most people, the primary care physician acts as the primary point of contact and as the gatekeeper for specialist care. As the amount of information about a patient grows, the time needed to get an overview of the patient's history increases. The application of data mining and information extraction techniques may aid the physician in finding information relevant at the point of care. This presentation outlines some of my results from applying classification and natural language processing techniques for the purposes of automated coding, part of speech tagging, and consultation note structuring.


Aziz Boxwala MD PhD
Title: Clinical Decision-Support Everywhere
Abstract:
Until recently, the use of computer-based clinical decision-support (CDS) has been limited largely to academic medical centers and commonly for medication ordering and prescribing. There are calls now to broaden the deployment of CDS, for example, as suggested by AMIA’s Roadmap for National Action on Clinical Decision Support. Even at the academic medical centers, there are efforts to broaden the use of CDS to a larger variety of clinical applications. There are many challenges ahead in the implementation of CDS. This presentation will discuss these challenges and will review various activities that aim to provide solutions.


Qing Zeng-Treitler PhD
Title: The Socio-Clinical Data Nexus: Comparing Data from Social Networks with Medical Records
Abstract:
Social network has emerged as a new information source for biomedical research, comparing with the traditional sources of clinical trial and medical record. By reflecting patients' perspective, social network could potentially complement the information from clinical trials and medical records. At the same time, the representativeness, completeness, and accuracy of data from social network remain to be studied. To assess the potential value of social network information, we conducted a study comparing the most frequent symptoms and treatments of multiple scoliosis (MS) patients obtained from a MS community on the PatientsLikeMe (a social network site) and from the Partners Healthcare System's Research Patient Data Registry (RPDR) (a medical record data repository). We found that order and prevalence of the top MS symptoms extracted from the two sources differ significantly. Further, the PatientsLikeMe and RPDR prevalences also appear to differ from what have been reported in the textbook/literature. The top treatments from the PatientsLikeMe and RPDR are fairly consistent, though the prevalences are also very different. These findings lead us to hypothesize that social network and medical record both have systematic biases in the recording of health data. When combined, they could capture a more comprehensive and accurate picture of a disease population's health conditions as well as health care actives. The findings also indicate gaps in patient-provider communication which need to be addressed in clinical practice and patient education.


Jens Huhn
Title: Some Recent Advances in Machine Learning Research
Abstract:
This presentation will cover three recent results from my Machine Learning research. I will (a) present how a crisp rule learning method was improved by fuzzification combined with a novel rule stretching technique in the fuzzy rule-based classification algorithm "FURIA". After that I will (b) highlight the connection between classification learning and decision making as implemented in the "FR3" algorithm and show how discerning different types of uncertainties - namely conflict and ignorance - between two alternatives in a classification decision might be exploited for better decision finding and understanding. Finally, I will (c) outline the problem of "Label Ranking", where every data instance is not assigned a single class label but a ranking over all labels, and explain how this problem can be solved by using a decision tree like approach.


Shobha Phansalkar RPh,PhD
Title: Exchanging Medication Information Across the Healthcare Continuum
Abstract:
Incomplete medication histories contribute to over a quarter of all hospital prescribing errors. Accurate and complete medication histories can prevent possible adverse drug events (ADEs)and improve patient safety.The fragmented nature of the healthcare system necessitates distribution of medication history information among providers, retail pharmacies, pharmacy benefit managers (PBMs), and patients. Automated extraction of information, across entities could allow a more complete and accurate estimation of the patient's medication history. This study describes the medication reconciliation application currently used at Partners Healthcare System and its enhancement to reconcile medications received from retail pharmacies and insurance claims. We will also describe the evaluation of a new user interface for this application, in order to improve clinician's access to medication information.


Joaquin Blaya PhD
Title: Developing, Implementing, and Evaluating Tuberculosis Laboratory Information Systems for Resource-Poor Settings
Abstract:
Multi-drug resistant tuberculosis (MDR-TB) patients in resource-poor settings experience large delays in starting appropriate drug regimens and are often not monitored appropriately due to an overburdened health care system, communication delays, and missing or error-prone data. Medical information systems can be used to alleviate these problems by increasing the timeliness and quality of laboratory information available. This research developed, implemented, and evaluated two such systems in the urban, resource-poor setting of Lima, Peru in institutions with and without internet.

The first part addresses the electronic collection of tuberculosis (TB) laboratory information from multiple institutions without internet. A handheld computer-based system was developed and implemented. A cluster randomized controlled trial and before-and-after comparison showed that this system had a significant effect in reducing processing times from 23 to 8 days, the proportion of cultures with delays >90 days from 9.2% to 0.1%, the number of errors by 57.1%, and the work-hours necessary to process results by 60%. A cost and timeline framework was developed to allow other organizations in resource-poor settings to implement this technology.

The second part addresses a web-based system, e-Chasqui, developed to provide electronic communication and reporting of TB laboratory information to health care personnel within institutions with internet. A cluster randomized controlled trial showed that access to e-Chasqui resulted in significantly less time to receipt of test results, a 56% reduction in tests taking over 60 days to arrive and a 98% reduction of results that never arrived, as well as a significantly faster time to culture conversion among patients in intervention versus control centers. These two parts describe verified medical informatics tools and an implementation methodology for settings both with and without internet connectivity.


Staal Vinterbo PhD
Title: Anonymizing Trusted Data Sources
Abstract:
An aspect of disclosure lies in the control of the information that can be learned from knowing which source the data comes from. For one, knowing the source implicitly adjoins each data point with the information available about the particular clinic or institution. Another concern is that knowing the source allows inferences to be made about the source using the data. I will present an approach allowing the querying of trusted data sources and the collection of the results in an anonymous way.

Lucila Ohno-Machado MD PhD
Title: Medical Informatics at UC San Diego This is a presentation of ongoing and future projects in Biomedical Informatics at the new Division of Biomedical Informatics at the University of California, San Diego.

Page last modified on June 03, 2009, at 11:04 AM