List of Participants for
Electronic Health Record Informatics Workshop
28 ^th Human-Computer Interaction Lab Symposium
University of Maryland
Thursday May 26, 2011

All workshop participants, their talk titles and abstracts could be found below.
Click here to return to main Electronic Health Record Informatics Workshop page.

Speakers

• Sigfried Gold (M.A., M.F.A.), Colonel Trinka Coster (M.D., M.S.), Suji Xie (M.S.), Tamra Meyer (Ph.D., M.P.H.), Rose Thelus (Ph.D., M.P.H.) and Lockwood Taylor (Ph.D., M.P.H.) -- Oracle Corporation, U.S. Army Office of the Surgeon General Pharmacovigilance Center

  
  Visual Representation of Exposure Patterns in Drug Safety Research
  

  The U.S. Army Pharmacovigilance center has assembled a data warehouse of medical records and claims representing the clinical care of 12 million soldiers, dependents and retirees over a period of five years. They use that data to promote patient safety in the military and to serve as a federal partner for the FDA's Sentinel program studying drug safety signals across the U.S. population.

  This talk will offer a brief tour through the challenges encountered when trying to use large databases of medical records and claims for drug safety research, focusing on visualization techniques that help researchers understand the data. The goal of drug safety or pharmacovigilance research is to discover and confirm causal relationships between drug exposure and adverse events. We will show some graphical methods for exploring this relationship.

  However, these methods depend on sometimes shaky assumptions regarding the definition of drug exposure. Drug exposure "eras" are defined by stringing together overlapping or proximal drug prescriptions. Researchers often make the assumption that a gap of up to 14 days does not constitute a break in exposure. For some drugs this may be a reasonable assumption, but the effort to characterize patterns of drug exposure, both in terms of formulating rules to define eras and then describing the general length of those eras and the frequency and length of drug holidays is fraught with complexity. The complexity is considerable even when looking at a single drug, but often researchers need to understand patterns of exposure to more than one drug.

  We will look at various techniques for representing these patterns, including LifeFlow, an innovative visualization method for temporal summary developed by Krist Wongsuphasawat at the HCIL.

• A. Zach Hettinger (M.D., M.S.) -- National Center for Human Factors Engineering in Healthcare, MedStar Institute for Innovation & Department of Emergency Medicine, Washington Hospital Center

  
  Recognition Of Wrong Patient Errors In A Simulated Computerized Provider Order Entry System
  

  Background: Computerized provider order entry (CPOE) systems have the potential to improve patient care in emergency medicine, but it is unclear how interface design can impact error rates.

  Objectives: The objective of this study was to determine if the addition of contextual information to a CPOE interface improves provider recognition and correction of patient selection errors, also known as "wrong-patient" errors, for ED radiology orders.

  Methods: Forty-six emergency medicine providers were randomly assigned to three groups (standard design, bolded patient identifiers, and contextual information) using a between-subjects design. Each participant navigated through 3 scenarios ordering radiology studies in a simulation of an existing CPOE system. A "wrong patient" error was introduced by the simulation in the final scenario. The primary outcome measure was recognition and correction of the introduced error. Secondary measures were the correction interval (time from introduction to correction) and the step in the ordering process at which recognition occurred.

  Results: We found an overall 24% error recognition rate. The error recognition rate and correction interval were both non-significant between groups. 82% of error recognitions occurred during the data entry(brief history, indication, room location) phase of ordering (p=0.03).

  Conclusion: In this study 76% of providers did not recognize the occurrence of wrong-patient selection error, which suggests providers do not routinely confirm patient identity after patient selection from the menu. However, when recognized, recovery took place significantly more often in the data entry phase which suggests this is where efforts should be focused. In this preliminary study there was no significant impact from the addition of contextual information to the ordering screen. Further study is needed to determine the most effective intervention to reduce the wrong-patient order hazard.

• Dick Horst (Ph.D.), Dana Douglas (M.S.), and Mark Becker (M.A.) -- UserWorks, Inc.

  
  Usability Evaluation of Electronic Health Records -- Where We Are; Where We Need to Be
  

  Momentum seems to be building to include usability evaluation as a more formal part of the development and certification of electronic health record applications (EHRs). Yet there is pushback from some in industry and regulatory bodies, questioning whether the usability field is mature enough to provide reliable, valid, unbiased measures of EHRs. This talk will attempt to summarize the following:

       * Some of the unique challenges in evaluating EHRs

       * What has been done thus far in addressing these challenges

       * Ongoing efforts that have been publicized that will likely push the envelope in evaluating EHR usability

       * Lessons we can possibly learn from other usability certification efforts

       * What is needed in the way of additional research, demonstration projects, and validation of procedures

  In providing this "lay of the land" overview, we will draw upon published literature, ongoing projects that we are aware of, efforts that have been publicized by funding and regulatory agencies, and our own experience in usability evaluation and user experience design in the healthcare field. There appear to be unprecedented opportunities on the EHR horizon but the terrain is rocky and the path not entirely clear. We hope this presentation will be a "wayfinding" exercise.

• Eliz Markowitz (M.S.) & Todd R. Johnson (Ph.D.)

  
  A Framework for Developing Systematic Yet Flexible Systems
  

  The introduction of consistent, systematic processes has improved task efficiency, safety, and effectiveness in many fields. In the field of healthcare, the introduction of structure, in the form of evidence-based medicine, checklists, and structured documentation has demonstrated improved benefits, however, such methods have also been criticized for their extreme rigidity. While HIT must provide enough systematicity to ensure consistency, efficiency, and safety, they must also allow flexibility to accommodate patient variability. Health care information technology that is either too flexible or too structured can lead to an increase in adverse events, morbidity, and mortality. Systematic Yet Flexible (SYF) systems provide a balanced approach by encouraging the user to complete the necessary steps while allowing users enough flexibility to address unexpected situations.

  Accordingly, there needs to be a way to determine the optimal mix of systematicity and flexibility for a given task. We propose a framework for analyzing and designing SYF systems. The framework uses three problem spaces: the natural space, the idealized space and the system space. The natural space captures only the task constraints that are present in the environment. The idealized space captures how the task should be done under ideal conditions, following best practices. The system space captures how the task can be done in a designed system or user interface. The goal of an SYF system is to design the system space so that it supports graceful degradation from the idealized space to the natural space.

  We also propose an information theoretic measure of flexibility. Intuitively, an inflexible task is one in which only a single unique sequence of actions will lead to the goal, whereas a flexible task is one in which any sequence of actions will lead to the goal. We define flexibility for a task problem space as the average number of bits needed to select the next action for each non-terminal state in a problem space. We can convert bits, n, to percentage flexibility using the formula: 100[1-(1/(n+1))]. Consider the idealized space for the following three tasks:

       1) Only one action is necessary -- Table A has ten blocks and Table B is empty. The goal is to place any one block from Table A onto Table B. If there are n possible actions, there are n possible paths. There is one state with ten possible actions, all correct, for a flexibility of Log2(10)/1 = 3.32 bits, and 76.86% flexibility.

       2) Order does not matter -- There are ten blocks on Table A and the sequence in which you move the blocks to Table B does not matter as long as you move all ten blocks. Once you have accomplished one sub-goal, the remaining sub-goals are constrained. If there are n sub-goals, there are n! possible paths. The initial state has 10 possible actions, and each successive state has only 9, and so on, for a flexibility of Sum[10^i*Log[2, 10 - i], {i, 0, 9}]/Sum[10^i, {i, 0, 8}] = 1.06, and 51.5% flexibility.

       3) Order matters -- There are ten blocks numbered 1 through 10, which must be dropped into a deep chute in numerical order. Here, there is only one possible path. There are ten non-terminal states with 1 action per state, for a flexibility of 10*Log2(1)/10 = 0 bits, and 0% flexibility.

  While the task requiring only one action may seem to be equally flexible to the task where order does not matter, each sub-goal achieved in moving all ten blocks decreases the flexibility of the actions that follow.

• Anne Miller (Ph.D.), Janos Mathe, Andras Nadas, Michael Hooper, Lisa Weavind -- Center for Research and Innovation in Systems Safety; Department of Anesthesiology, Institute for Software Integrated Systems; School of Nursing, Vanderbilt University Medical Center

  
  Alerts and reminders: Is This All There is to Clinical Decision Support?
  

  Alerts and reminders can improve clinical decision-making and reduce errors, but their success has been variable. Habituation, clinician frustration and lack of contextual integration are reasons for failure. More significantly, alerts and reminders are limited to just-in-time process support. They do not support critical clinical decisions: what is wrong with this patient? Is this patient deteriorating or recovering? Was the last intervention successful? What other treatment options could we use and in what measures/combinations? The purpose of this presentation is to show the evolution, through three user interface design iterations, of an application designed to support complex clinical decisions. Iteration 1 presents a prototype that integrates patient data using physiological principles. Using this interface, nurses were better able to detect patient change and physicians were better able to agree about patient diagnoses. The usefulness of this prototype was limited by its 3-D complexity and its overall use of real-estate. Iteration 2 involved breaking the fully integrated display into parts based on physiological functions (circulation, respiration, renal etc). Nurses' ability to detect patient change was further enhanced, but physicians were less likely to agree about diagnoses. These findings suggest that nurses’ decisions are better supported by narrower information views, whereas physicians’ decisions may be better supported by more integrated representations that highlight disease-related gestalts. Increasingly, patient care protocols based on clinical evidence of efficacy are improving patient outcomes. The 24-hour protocol for managing severe blood stream infections is well established. We designed an interface based on this protocol and outcomes from iterations 1 and 2. Residents using the interface were better able to assess scenario patients, identify patients who were not septic and assume patient care process from a previous physician than they were using an interface designed using protocol rules alone. The integrated interface is being developed for clinical trial.

• Craig Nichols (M.D.) -- Director Testicular Cancer Consortium

  
  Complex, Personalized Cancer Clinical Decision Support Utilizing Aggregated Clinical and Biological Information
  

  Background: Clinical decisions in patients with cancer are the primary driver of outcomes and costs; more so than new technologies. Within our current medical system, most critical cancer decisions are made at a local level without the benefit experience available at centers of excellence for the condition. Despite the widespread availability of clinical guidelines, locally, there are substantial deviations from these guidelines and non-compliance can exceed 50%. There are emerging data in cancer and other specialties that models of centralized expert care can improve outcomes by up to 10% or more and that there is up to a 30% reduction in resource utilization (primarily accrued by elimination of unnecessary tests and treatments as well as improved safety profiles).

  We are seeking to utilize the ability to aggregate large clinical data sets to accomplish two primary goals:

       1.) Enhance value in cancer clinical decision making by integrating real time human and computable expertise into the cancer provider: cancer patient interface

       2.) Build biological and clinical research capacity by utilizing aggregated de-identified population-based clinical data sets.

  Our pilot explorations will be in the area of testicular cancer. This is a disease where we have clear cut demonstration of the value of population-based clinical decision support and navigation. If successful, we would anticipate this model could extend to other cancer disease sites and to other non-malignant diseases.

  Hypothesis: Centralized, real time expert review of actionable clinical data will improve outcomes in patients with testicular cancer and reduce resource utilization.

       Specific Aim 1.) To build and validate tools for actionable real time information flow between patients, providers and knowledge centers to enhance outcomes and value in testicular cancer.

       Specific Aim 2.) To build cancer research standardizing large volumes of aggregated clinical information from platforms by conserving and populations of testicular cancer patients

       Specific Aim 3.) To build and validate practical, value-added patient engagement tools in cancer

• Sanjay Patel (M.S.) and Adam Hammouda -- WebFirst, Inc.

  
  Open Source mHealth Solution based on Drupal 7
  

  Many data collection platforms exist in the world of mobile health. Examples of these include: EpiSurveyor, Ushahidi, MOTECH, RapidSMS, FrontlineSMS. While these solutions are low-cost and effective for many mHealth interventions, they are limited in their flexibility, scalability, and ability to publish and share data. In order to fully achieve the goals of the mHealth community (improving access to information, improve data reliability, improve the quality of care by providing better decision-support tools) it is important to consider platforms that are open and well-supported by a large developer community.

  By using platforms that have readily available modules, important functions can be easily added at relatively low cost. Some of these functions include: data quality review, workflow, data visualization/GIS and social media sharing, and ability to collect data from and publish to a variety of platforms (web, android, iOS,SMS).

  In this workshop, we will discuss the advantages of using Open Data Kit integrated with the Drupal 7 (open source) framework. Drupal 7 will be used as a model through which mobile health researchers can analyze data more effectively and communicate with each other in a more effective and efficient way, furthering mutual goals in public health.

  We will demonstrate and discuss: (i) A real-world use case - an immunization reminder system, (ii) Data collection via SMS and Android/j2me phones, (iii) Collect/store GPS location and visualize data with Google maps, (iv) Collect/store multimedia (pictures, audio, video), (v) Ability to define workflow rules and triggers (e.g. reminder settings, data quality), (vi) Ability to publish data to a website and share datasets via social media.

• Matt Quinn -- NIST

  
  NIST and AHRQ - EHR Usability Roadmaps
  

  A recent report from the HIMSS Usability Task Force identified usability as one of the major factors - possibly the most important factor -- hindering widespread adoption of EMRs. The National Institute of Standards and Technology (NIST) and the Agency for Healthcare Research and Quality (AHRQ) have each developed research agendas focused on improving the usability of electronic health records (EHRs). Matt Quinn, who led research in this area at AHRQ, and who has recently moved to NIST will provide an overview of the NIST's role and roadmap for improving EHR usability, including details of key AHRQ and NIST projects.

• Cui Tao (Ph.D.) -- Mayo Clinic College of Medicine

  
  Semantic-Web based Annotation and Reasoning for Temporal Information in Clinical Narratives
  

  The temporal dimension in medical data is essential in clinical research. Efficient and timely dissemination and analysis of the temporal aspects could boost an array of clinical translational research such as disease progression studies, disease support systems, and personalized medicine. Time aspects that are important for clinical findings, however, are often embedded in clinical narratives. This makes it challenging to retrieve temporal relations and reason about them, especially when it involves multiple patients across different sites. In order to facilitate clinical researchers to expose the temporal dimension in medical data analysis, scalable software platforms that allow users to ask time-oriented questions and retrieve temporal information automatically from large volumes of clinical records are highly desired. Toward this goal, we have designed a framework where temporal information in narratives can be annotated with respect to an OWL ontology, and annotated data can be queried and reasoned.

  The system is centered by a Semantic-Web ontology, CNTRO (Clinical Narrative Temporal Relation Ontology, cntro.org) for modeling temporal information and relations in the clinical domain. It models clinical events, different kinds of temporal expressions (such as time instants, time intervals, repeated time periods, and durations), different levels of time granularity, temporal relations, and time uncertainties. We have also conducted analyses for limitations and possible improvements of CNTRO, based on which, a newer version of CNTRO is under development with finer levels of classifications, more robust coverage, and more formal semantic definitions.

  With respect to CNTRO, temporal data in clinical narratives can be annotated and represented in RDF. A Protege plug-in is being implemented for annotating events and time-related information. This GUI is built upon the Knowtator, an annotation interface developed in Mayo Clinic for manual text annotation to XML files based on a pre-defined schema. The new interface makes it work with the Semantic Web, so that users can annotate information of interested with respect to CNTRO, the annotated information can be stored as RDF files, and queried by SPARQL.

  We have also developed a prototype framework for querying and inferring temporal information. This framework combines DL-based reasoning, SWRL-based reasoning, and SWRL Built-Ins library. It provides basic functions for answering a list of time-related questions. We evaluated the feasibility of using CNTRO with existing Semantic-Web technologies and discussed possible limitations and extensions that we found necessary or desirable to achieve the purposes of querying time-oriented data from real-world clinical narratives.

• Rosemary Tate -- Brighton and Sussex Medical School & School of Informatics, University of Sussex, Brighton

  
  Visualising Words: Uncovering Hidden Information in Large UK Primary Care Databases
  

  Almost all UK residents are registered with a primary care physician (General Practitioner or GP) who oversees healthcare and acts as gatekeeper to specialist care, usually within a group practice serving 5000-30000 individuals. All GP practices use electronic recording, partly or exclusively, and many have been doing so since the early 1980's. Primary care databases collate anonymised information from a large number of practices, enabling analyses of the real life care of large populations -- for example, the General Practice Research Database collects data from ~10% of UK practices. These datasets contain a mine of information for pharmacological and health services research and are used for these purposes worldwide.

  However, most research uses only the coded part of the record, ignoring the rich information contained in free text notes or letters. The cost of anonymisation, and lack of tools for efficiently processing this diverse information, discourages use of free text, but at the cost of poorly understood bias in our understanding of health and health care.

  Our team at Sussex (Patient Records Enhancement Project) includes statisticians, computational linguists, informatics specialists and clinicians. We are developing tools to extract and represent hidden free text information, alongside socio-technical exploration of decisions to record free text vs. coded. We are also developing tools for integrated display of coded and free text (and free text derived) data to help researchers visualize, process and interpret the huge amount of diverse information contained in primary care health records.

  First I shall outline briefly some EHR systems and interfaces currently used by UK GP practices. I shall then present results of our recent work investigating ovarian cancer symptoms (with a live demo using Lifelines 2), and discuss some of the challenges involved in designing visualisation tools for researchers using primary care databases.

• Lauren Wilcox (M.S.) & Steven Feiner (Ph.D.), Department of Computer Science, Columbia University, New York; David Vawdrey (Ph.D.), Department of Biomedical Informatics, Columbia University, New York

  
  An Application to Provide Views of the Electronic Health Record to Cardiology Patients
  

  As patients are encouraged to become active participants in their own care, recent research has begun to explore the direct sharing of electronic health information with patients during hospital visits. Through modified views into the electronic health record (EHR), hospital patients could access information on their care status and progress, the identities of their care team members, and expected upcoming events, such as scheduled radiology exams or pending physician consults. Such sharing of information can provide unprecedented opportunities to educate patients and their family members about treatment progress and post‐discharge care planning. This talk will discuss our research on the design, development and evaluation of patient-facing clinical views of the EHR, focusing on:

       1. Studies with patients, nurses, and physicians discussing candidate information types found in the medical record, such as the patient problem list, medication list, test results, etc., and how these structured information types can be visualized by patients to meet specific patient information needs at the point of care.

       2. Design guidelines for creating patient-facing views of the EHR for two hospital patient groups, resulting from two formative studies.

       3. A technology solution that we have created through a collaboration between Columbia University Departments of Computer Science and Biomedical Informatics, and NewYork-Presbyterian Hospital. We have built upon Personal Health Record (PHR) technology to provide inpatient views of care team information and active and discontinued medication lists.

This participant list is for Electronic Health Record Informatics Workshop.