Workshops take place on Saturday, November 16, and Sunday, November 17.
- Half-day workshops = 3 CME/CE credits
- Full-day workshops = 6 CME/CE credits
For sessions offering MOC-II credit, check the Self-assessment Booklet of Multiple Choice Questions.
Saturday, November 16, 2019
8:30 a.m. – 12:00 p.m.
N. Shimpi, Marshfield Clinic Research Institute; K. Williams, Indiana University, Regenstrief Institute, Inc.; A. Mahnke, Marshfield Clinic Research Institute; S. Kumar, University of Tennessee Health Science Center
According to the National Academy of Medicine, previously known as the Institute of Medicine, patient-centered care within health care has been identified as a crucial element of the care delivery process. Recognizing strategies to reduce incidence of chronic diseases, including oral-systemic association, and to improve effectiveness and efficiency of care through the application of informatics, is of paramount importance.2 The emergence of electronic health records has necessitated the capture and utilization of knowledge to support patient-centric care. At this digital frontier, technologic innovations in the field of dentistry and medicine have also been advancing. In addition to technologic innovations, knowledge surrounding oral and systemic association have created opportunities for improved cross-disciplinary care delivery, access, efficiency, and quality of oral and overall health care. Evidence suggest that application of clinical and research data through informatics platforms can facilitate this collaborative care.
The objective of this workshop is to equip attendees with practical competencies surrounding emerging concepts of clinical and research informatics that can be applied to various models of care across dental and medical practice. This introductory, interactive workshop brings together informatics researchers who leverage clinical and research data at point-of-care settings.
The workshop covers the following broad areas:
a. Holistic care delivery across medical and dental domains: facilitating care delivery for oral-systemic associations
b. Dental public health: System-level health informatics use and applicability to integrated care delivery
c. Usability: User-friendly interfaces to facilitate dentistry and cross-disciplinary care delivery
d. Teledentistry: The importance of teledentistry in public health and dental informatics
C. Pollack, Johns Hopkins Bloomberg School of Public Health; E. DuGoff, University of Maryland School of Public Health; Y. Park, P. Karampourniotis, IBM
Social network methods offer a novel way to investigate complex patterns of healthcare delivery. By explicitly modeling relationships between patients, providers, and organizations, social network methods enable researchers to study healthcare as a complex system. Increasingly, these approaches have been applied to insurance claims and electronic medical records data to study care coordination, healthcare costs, and quality. However, there are tremendous amount of research and application opportunities with network analytics in healthcare that are yet to be explored, including how insurance networks, hospital-based networks, and provider networks impact a range of outcomes.
This instructional workshop aims to provide researchers and professionals with an introduction to the conceptual issues and network analytic tools needed to analyze healthcare data, including both didactic and interactive components. The first part of the workshop will be devoted to an overview of the network analytics applied in healthcare research, presenting the landscape of state-of-the-art research papers, their methodologies and findings. The second section will focus on basic concepts of network analytics, including the concepts of bipartite networks and network projection, as well as fundamental network properties.
In the third and final section, attendees will have the opportunity to acquire hands-on experience (using python and/or R) on how to process electronic health data to network representation, how to conduct basic analysis on networks, and how to visualize networks. This workshop is intended to cover the needs and interests of individuals with novice to intermediate levels of experience in the field.
S. Craig, The Children's Hospital of Philadelphia; U. Huebner, University AS Osnabrück; E. Orenstein, Emory University; M. Gong, Chinese Academy of Medical Sciences; T. Cullen, Regenstrief Institute, Inc.; H. Fraser, Brown University
Global health initiatives are designed to improve human health and delivery of health care services in all communities and healthcare settings world-wide. High-level global health informatics (GHI) agendas have focused on developing long-term goals, driven by evidence-based recommendations. Despite this guidance, considerable gaps remain on how to successfully apply these recommendations. Over the last decade, GHI initiatives have been successful in beginning to translate this knowledge into practice but have had limited impact beyond the local or regional level. This workshop will use case-studies to demonstrate GHI principles, and how they lead to success or failure on the ground. Experts will lead discussions in five GHI domains: 1. Initiative planning; 2. International partnerships; 3. Research; 4. Education; and 5. Early Career Engagement. Panelists, experts and attendees will compare and contrast the key aspects to creating successful and sustainable GHI partnerships. Ultimately, this will allow the creation of a repository of standards, knowledge, as well as best practices around collaborative principles in GHI.
R. Jenders, Charles Drew University/UCLA; G. Del Fiol, University of Utah; P. Haug, Intermountain Healthcare; K. Kawamoto, University of Utah; B. Rhodes, Dynamic Content Group LLC
Clinical decision support (CDS) can help improve clinical practice and health behaviors. The use of health information technology (HIT) standards for encoding data, representing knowledge and delivering knowledge-based interventions in turn can help facilitate implementation of CDS. However, many standards from numerous standards development organizations (SDOs) exist that are variously incorporated into vendor software, and consensus on the use of these standards is lacking. Moreover, new work in the past year on CDS standards and in related areas such as clinical quality measurement has increased the complexity of this domain. The emergence of the Health Level Seven International (HL7) Fast Healthcare Interoperability Resources (FHIR) standard for patient data representation is particularly pertinent in light of work to leverage FHIR for CDS implementation.
Accordingly, the leaders of this instructional workshop, who are co-chairs of the HL7 CDS and Arden Syntax Work Groups, will address two overarching educational objectives. First, attendees will learn key details of and latest developments in extant and proposed HIT standards that are applicable to CDS and which they may need to use in their work. Second, the attendees will recognize how these standards relate to one another and can be used to implement CDS that improves outcomes. Through interaction with workshop leaders, attendees will be able to contribute ideas regarding refinement and use of HIT standards to facilitate CDS.
X. Bofil-De Ros, National Cancer Institute-Frederick; T. Davidsen, National Cancer Institute; A. Deslattes Mays, Jackson Laboratory for Genomic Medicine; I. Fore, National Cancer Institute; J. Pevsner, Kennedy Krieger Institute
Basic and clinical research in cancer is increasingly focused on generation of rich datasets to identify the molecular basis for disease and to match targeted therapies that factor in each patient’s unique biology. To progress towards this goal, the cancer research community will need to access, integrate, and analyze many different types of data, including genomics, proteomics, microbiomics, metabolomics, cancer models, clinical treatment and outcomes, multi-resolution, multi-modality imaging data, population-based data, and data contributed by health care providers and patients themselves. Investment in the informatics infrastructure to fully leverage these diverse data types is imperative and was called out as a priority by the Cancer MoonshotSM Blue Ribbon Panel. To this end, NCI has initiated development of a Cancer Research Data Commons (CRDC) to provide access to interoperable data repositories, analysis tools, and workspaces. The vision for the NCI CRDC is a virtual, expandable infrastructure that provides secure access to diverse data types, allowing users to analyze, share, and store results, leveraging the storage and elastic compute of the cloud. In this workshop, we will provide a brief overview of the vision and status of the CRDC and facilitate an interactive discussion with participants on its future direction. The discussion will focus on key topics to gain insights from the informatics community on how they would envision optimal use of the platform and what content and functionality will be of most value to them.
E. Austin, University of Washington; C. LeRouge, Florida International University, University of Washington; A. Hartzler, D. Lavallee , University of Washington; A. Chung, University of North Carolina School of Medicine; P. Hsueh, IBM; C. Petersen, Mayo Clinic
As one form of patient-generated health data, patient reported outcomes (PROs) are increasingly utilized, critical data sources for learning health system efforts to continually assess and improve healthcare delivery, quality of care, and population health services. A growing number of health systems prioritize the electronic capture and presentation of PROs (i.e., ePRO systems), necessitating a comprehensive and integrated approach to diverse uses of ePRO systems within complex healthcare organizations. Workshop leaders experienced in the research and application of ePRO systems will share their perspectives on: 1) planning for ePRO systems at the health system level, 2) developing tools for ePRO data reporting and visualization, 3) deploying ePROs systems across diverse clinical environments, and 4) evaluating sustained and ongoing use of ePRO systems. Through attendance at this workshop participants will learn to formulate strategies for user engagement, interface design, data flow and visualization, and workflow modeling for ePROs tools leveraging PGHD in the learning health system.
B. Rawat, A. Jagannatha, UMass Amherst; F. Li, H. Yu, UMass Lowell
This instructional workshop provides an introduction to techniques in multi-task modeling for deep learning techniques to perform automated clinical judgement studies (Naranjo Question Answering). We have organized successful workshops in MedInfo 2015, AMIA 2017, 2018 and have developed deep learning models for clinical judgement studies and for state-of-art natural language processing (including clinical event detection in electronic health record notes document, sentence classification, semantic entailment and question answering). In this workshop, we would cover the fundamentals of multi-task learning and deep learning. We will demonstrate the methodologies on the application of the widely accepted Naranjo questionnaire for clinical judgement studies. For deep learning, we will describe the state-of-the-art Recurrent Neural Networks and attention modeling. The focus would rest on using widely used Python programming language and its deep learning packages, such as PyTorch, to quickly implement a prototype and test different multi-task deep learning models. These deep learning techniques can be extended to other clinical natural language classification tasks other than predicting answers to Naranjo questionnaire.
K. Zheng, University of California, Irvine; T. Kannampallil, Washington University in St. Louis; J. Horsky, Northwell Health; V. Patel, The New York Academy of Medicine
Most clinical environments resemble a paradigmatic complex system with its dynamic and interactive collaborative work, non-linear and interdependent activities, and uncertainty. Addition of new organizational and systemic interventions, such as health IT, can cause considerable cascading effects in the clinical processes, workflow, and consequently, on throughput and efficiency. A 2011 IOM report called for a sociotechnical approach for designing and incorporating health IT in clinical settings. One of the critical aspects of a sociotechnical approach is to understand the progression and evolution of human interactions within a sociotechnical context. In this instructional workshop, we will discuss a set of convergent methodologies for analyzing human interactive behavior both with technology and with other humans or artifacts. These methodologies can help in capturing underlying patterns of human interactive behavior, and provide a mechanism to develop integrative, longitudinal metrics (e.g., metrics related to performance, or errors) for clinical activities for sustained interactive episodes that evolve over time. In addition, such analysis of interactive behavior can also provide significant input for patient safety outcomes through the design of safer and more efficient health IT. In this workshop, we will (a) identify challenges to studying human interactive behavior in complex clinical contexts; (b) discuss new approaches for capturing and analyzing sequences of human interaction using sequential analysis and network-theoretic, time-series based methods; (c) utilize one or more of these techniques to demonstrate their effectiveness as a viable mechanism for developing insights on clinical work activities through hands-on sessions; (d) provide participants hands-on experience in using data collection and data analysis tools; (e) discuss how research on human interaction informs the cognitive design of clinical information systems to improve their usability and support for collaborative team work; and (f) discuss the implications of these techniques for new care delivery models and patient safety initiatives.
C. Jaffe, Health Level 7 International; J. Mandel, Microsoft; S. Sartin, S. Posnack, Department of Health & Human Services; J. Overhage, Cerner Corporation; M. Tripathi, Massachusetts eHealth Collaborative
The path to health data interoperability has been a tortuous one. When HL7 first introduced FHIR® (Fast Healthcare Interoperability) nearly 9 years ago, a disruptive wave was set in motion. The changes that FHIR envisioned would embrace long-established internet technology and the precept that the implantation community was at the forefront. The report of the JASON Task Force provided a clear and achievable path to that goal, beginning with the enablement of Application Programming Interfaces (APIs). From that Task Force, the Argonaut Project, a private-sector initiative, emerged to drive FHIR implementation. In early 2018, Apple announced that it had leveraged the Argonaut implementation guide and embedded it within its operating system. To date, over 500 health systems have partnered to enable patients to access to their health data. FHIR has enabled the exchange of biomedical data over the broad continuum including patient care, population health, accountable care and personalized medicine. In March of this year, both the Center for Medicare and Medicaid Services (CMS) and the Office of the National Coordinated for Health IT (ONC) released two complementary Notices of Proposed Rulemaking (NPRM) to improve the Interoperability of health data. This panel will explore the emergence of FHIR, the milestones in its development, and the value brought to the FHIR implementers and the health of the worldwide community we serve.
8:30 a.m. – 4:30 p.m.
L. Heermann Langford, Intermountain Healthcare; R. Leftwich, Intersystems; J. McClay, UNMC
Beginning in 2010, HL7 created Fast Healthcare Interoperability Resources (FHIR) as a next generation standard to address clinical data interoperability. Clinicians on FHIR evolved in 2014 as an event held at each HL7 Working Group Meeting (3 times yearly) to educate clinicians about HL7 FHIR and provide feedback to the HL7 FHIR team regarding the clinical viability and usability of the FHIR standard. This AMIA pre-symposium will bring the Clinicians on FHIR activity to a broader clinical audience attending the AMIA Fall Symposium. The workshop is designed to educate attendees about HL7 FHIR and tools available to access, review, and provide feedback to the HL7 FHIR team regarding the evolving HL7 FHIR standard. It is also intended to make the audience aware of the potential of FHIR for innovation in their organizations. The faculty will provide lectures describing HL7 FHIR history, background, and fundamental principles. Examples of applications using the SMART on FHIR platform will also be discussed. After this initial overview of FHIR, the attendees will be guided through using online tools to examine HL7 FHIR Resources (the basic building blocks of FHIR) and build FHIR Profiles (implementation guides for specific use cases) for care plan and care coordination use cases.
O. Uzuner, George Mason University; M. Yetisgen, University of Washington; H. Liu, Mayo Clinic
Natural Language Processing (NLP) for clinical and biomedical narratives has received growing attention over the past decade. As these efforts grew, so has the need for broader community engagement and resource sharing. As a result, many NLP-ready data sources and software have been developed and shared. These efforts have provided significant learning opportunities for new comers to the field and have supported advancement of the state of the art. In its support of these goals, the AMIA NLP working group pre-symposium continues the tradition since its inception in 2012 to provide a unique platform for close interactions among students, scholars, and industry professionals who are interested in clinical NLP for data and resource sharing. This year’s proposed event will consist of three sections: 1) a graduate student consortium, where students can present their work and get feedback from faculty; 2) a NLP community challenges and workshops session, that will invite organizers and participants of NLP shared task challenges and workshops in the clinical and biomedical domain to present the major lessons learned from the most recent shared tasks and workshops, highlighting the newly available corpora for NLP research as well as the state of the art solutions to various NLP tasks; and 3) System demos session, where researchers will demo their existing systems and disseminate their software.
1:00 p.m. – 4:30 p.m.
D. Vreeman, Indiana University and Regenstrief Institute, Inc; S. Abhyankar, Regenstrief Institute, Inc.
LOINC® (Logical Observation Identifiers Names and Codes) is a freely available international standard for identifying laboratory and other clinical observations. Presently, LOINC is used by more than 76,000 users from 175 countries and has been adopted as a national standard in 35 countries. LOINC has been translated into 18 variants of 12 languages and is used by healthcare organizations, reference laboratories, ministries of health and other federal agencies, professional societies, healthcare information exchange networks, insurance companies, healthcare IT vendors, instrument manufacturers, health application developers, and more. LOINC is now ubiquitous in health data systems worldwide and is an essential ingredient of system interoperability.
This workshop will cover the basics of LOINC, including: its origin; how it is developed, distributed, and used; and its relationship to other vocabulary and messaging standards such as SNOMED CT and HL7. We will summarize the highlights of global adoption and use cases.
This workshop will explain the LOINC concept model and key features to help users pick out the differences between LOINC terms with subtle but important distinctions. Many informatics professionals are aware of LOINC’s coverage of laboratory tests, which we will explore, but we will also tour its rich content in other domains such as vital signs, clinical document titles, radiology procedures, patient assessment instruments, and patient reported outcomes measures.
This workshop will introduce the key tools and resources available for implementing LOINC. Our primary focus will be on how to get started with LOINC by mapping your local observation codes to LOINC codes. We will briefly describe some secondary uses of LOINC, such as automated electronic laboratory reporting to public health, use in clinical decision support systems, and clinical quality measurement. And last, we will describe some best practices for using LOINC that will prepare the participants for success in the long run.
M. Edmunds, AcademyHealth; B. Johnson, 2-1-1 San Diego; P. Payne, Washington University in St. Louis; S. Greene, Health Care Systems Research Network
This interactive, collaborative workshop aims to provide practical guidance about how to engage in strategic problem formulation and subsequently manage multi-disciplinary teams in order to accelerate progress in biomedical, clinical, and population health informatics and health systems research. Intended for all career stages and organizational settings, this year’s hands-on session, which builds on the success of our 2018 workshop, will include focused discussion and engaging exercises on strategic thinking, problem formulation, and how collaborative dynamics inform team science activities.
The benefits of collaborative work are well-documented (Stokols et al., 2019; Gregersen, 2018; Payne et al., 2017; Baer et al., 2012; Disis & Slattery, 2010; Hall et al., 2008). In addition to reducing redundancy, collaborative methodologies can increase the speed, efficiency, and impact of research efforts, particularly when attempting to define and solve complex, systems-level problems. However, enabling collaboration requires novel methods for problem formulation as well as attention to critical governance, infrastructure, technology, stakeholder engagement, and other sociocultural influences, and concerted attention to building synergy, in order to assemble and sustain high-functioning teams. These parameters are interdependent and warrant thoughtful attention. Further, optimal approaches to such issues are often outside the core competencies of many biomedical researchers, requiring an intentional approach to understanding and applying relevant theories and methods.
This session will be organized into three parts. First, presenters will engage in an interactive discussion with attendees, presenting a landscape of initiatives that seek to enable team science, and lessons learned therein, drawn from their collective experiences via AcademyHealth, 2-1-1 San Diego, the CTSA program, and the Health Care Systems Research Network. During the second hour, participants will engage in small group activity and leverage strategic thinking methods to design frameworks for collaborative approaches to research problems at their own institutions or within organizations or networks that they participate in. In the final hour, the group will discuss key lessons learned and takeaways to bring back to their institutions, organizations, or networks.
G. Giunti, V. Mylonopoulou, University of Oulu; Y. Solad, Yale New Haven Health
Patient empowerment provides patients with an ability to actively influence their health. Increased adoption of wearable technologies and “smart” devices can simplify the collection of patient-generated data and provide the means for educating patients; however, information alone does not always suffice. Conversational agents offer a new medium to streamline the communication between patients and clinical teams or retrieve the information from the medical record. Conversational agents represent the broad spectrum of user interfacing technologies from chatbots to voice assistant and its design and development require a special mindset incorporating user-centered design techniques and user engagement techniques. In order to be effective, interventions have to be delivered at the appropriate time and be both meaningful and actionable for patients. Targeted actions that inspire and motivate have higher chances modify behaviors in a positive way.
The use of conversational agents as a supplement for digital health interventions or as a stand-alone conversational app has gained traction and will continue to grow in the upcoming year. This workshop offers participants the chance to learn the basics of conversational and voice agents design strategies, and to explore approaches for building effective conversational agents.
I. Sim, University of California, San Francisco; J. Duke, Georgia Tech Research Institute; E. Haas, Health eData, Inc; E. Soto, Georgia Tech Research Institute; S. Carini, University of California, San Francisco
Health care providers have shown interest in the use of mobile health (mHealth) data for expanded monitoring of patient conditions and beneficial activities, such as physical exercise. Generated in large amounts and in vendor-specific formats, mHealth data do not need to be stored in the electronic health record (EHR) but need to be made accessible to health care providers within their workflow, for example via SMART-on-FHIR. In this workshop, we will introduce two standards, Open mHealth (OmH) for mHealth data and HL7’s Fast Healthcare Interoperability Resources (FHIR) for EHR data and describe their bridging via a use-case-driven reference application that enables the incorporation of mHealth data from a diverse set of personal devices into EHRs at the point of care.
R. Gamache, H. Kharrazi, Johns Hopkins University
The U.S. health informatics landscape has seen an increased emphasis on data integration across the previously separate clinical, management, insurance, and public health data silos. The sharing of social determinants of health (SDH) data across public health, population health, and social services is the prime example of how the landscape of public and population health informatics practice is evolving and converging. Data collected by public health departments are key sources of data required to calculate population-level health measures that will eventually be used to determine global budgets for medical care delivery systems and to assess whether they reach their community health targets to achieve substantial financial incentives. With this shift in thinking, there is also an increased acknowledgment of the importance of considering a broad array of “non-medical” factors generally termed as SDH.
Many informatics professionals’ top priority is the advancement of public, population, and community health outcomes. Concurrently, the need for SDH data and the growing opioid abuse crisis require new ways of thinking about public health data and related issues. These factors and others are driving fundamental and impactful advances as well as exposing informatics gaps in many facets of healthcare, public, and population health domains.
In this workshop, we propose to lead and facilitate dynamic sharing of ideas and understanding of the evolving health care ecosystem encompassing community, public, population and patient-centric health care. We anticipate faculty, presenters, and participants alike will experience engaging and insightful discussion of current challenges and innovative solutions as well as new perspectives for advancing future public and population health informatics. Further development of consensus statements to advance progress in areas of informatics, SDH, and bridging the gap between clinical and public health will be developed and communicated broadly 6.
C. McDonald, National Library of Medicine
The LHC form and flowsheet tools are bookends to a FHIR medical record. The right bookend provides the machinery for building forms and storing the data collected into a FHIR EHR. It implements FHIR Questionnaire, Questionnaire Response, and much of the capability of FHIR’s Structured Data Capture (SDC) specifications including pre-population, data extraction, calculated values, multi-column choice lists, adaptive (PROMIS-like) questionnaires and survey instrument scoring etc. We will explain how many of these capabilities are implemented and demonstrate their use.
The left bookend – LHC FHIR Flowsheet App—generates a flowsheet from an FHIR EHR and has features for collapsing date columns and rows of similar variables to for a more condensed display. It has built in unit conversion to enable the merging of variables with commensurate units including mass to molar conversions. It also includes a compact pixel map showing everything in the flowsheet (abnormal, normal) to ease navigation. We will describe the content of our 10K de-identified medical records set, which provides the grist for the FHIR Flowsheet App and will explain the configurable template that controls it. Participants will interact with a public, test version of the LHC-Flowsheet app with sample data.
M. Penning, University of Arkansas for Medical Sciences; V. Huser, National Library of Medicine; C. Craven, Mount Sinai Health System; K. Natarajan, Columbia University; A. Mosa, University of Missouri School of Medicine
Ensuring the quality of clinical data is critical to the mission of building actionable knowledge from healthcare data. However, our understanding of the impact of clinical data quality is still in its infancy. During the AMIA Annual Symposium in 2018 informaticians came together in a CRI working group workshop to discuss the impact of data quality on clinical research and set a research agenda for data quality. The AMIA fall meeting brought distributed communities together to share best practices and knowledge. In this workshop we worked collaboratively to ascertain the importance of data quality as well as aspects of data quality assessments such as the business case for data quality and driving data quality assessment results into practice and publications. These discussions touched on many important factors in addition to those planned for the workshop. This follow-up workshop in 2019 will share the results of the previous meeting and to expand on topics that arose during that initial conversation.
Along with a discussion and vetting of research agenda components synthesized last year at the workshop, our expanded discussion will include data quality and 1) its effect on applications, 2) integration of clinical data, 3) interoperability. An understanding of the impact of data quality on many applications is critical; for example, privacy-protecting and secure deduplication processes are complicated by poor data quality. Also, the quality of data directly influences integration of clinical data with biobank data, genomic data, and open source registries. A deeper understanding of this relationship can be gleaned from sharing successes and failures as a community. Finally, interoperability is a growing concern in relation to data quality. A discussion of CRI gaps and challenges will provide not only a forum for knowledge sharing but a basis for future research and publication.
Sunday, November 17, 2019
8:00 a.m. – 11:30 a.m.
D. Schlegel, SUNY Oswego; M. Brochhausen, University of Arkansas for Medical Sciences; J. Bian, University of Florida
Knowledge representation and semantics represents a thriving and crucial subfield of biomedical informatics. The representation of healthcare data, information, and knowledge in a semantically rich manner which allows for diverse applications is critical to the present and future of healthcare. There have been great successes in areas ranging from drug discovery and -omics to public health and medication safety, but the job is not done. As we continue to build systems in the cognitive computing era, we face issues in ensuring the represented knowledge is up to date, that we have not made representational errors, that we can use and reason over the represented knowledge as needed for specific applications, and so forth. This pre-symposium event provides a forum for these and many other topics and issues dealing with knowledge representation and semantics to be discussed. The event will consist of two sessions: a doctoral consortium in which students will present their in-progress dissertation work; and a session combining highlights in knowledge representation and semantics in which published work from a diverse set of venues will be presented to this wider audience, and late-breaking research in which extremely recent results and software systems under development will be presented for the first time.
G. Grieve, Health Intersections, HL7
This workshop will teach participants how the R statistical analysis language can be used to perform data analysis using data sourced from a FHIR server such as those provided through the Argonaut project.
M. Weiner, Lewis Katz School of Medicine at Temple University; H. Lehmann, Johns Hopkins School of Medicine
In an ideal word, all diagnostic tests would make perfect predictions, all therapies would be completely effective and harmless, and resources would be limitless. However, even with ongoing innovation in machine learning algorithms that underlie the current era of precision medicine, outcomes prediction remains imperfect, therapies are not completely effective and resources remain limited. Therefore, the output of these algorithms must be placed in a realistic clinical context to support truly effective clinical decision-making. Through incorporation of prior evidence-based knowledge of clinical valuation of outcomes and decision points, Bayesian analysis plays a key role in translating results of machine learning algorithms into actionable information. Bayesian analysis is essential for understanding the degree of sensitivity and specificity that a new diagnostic test or predictive algorithm needs to have in the context of the effectiveness and the expense of existing therapies to impact clinical decision-making.
Related to Bayesian analysis, Receiver Operating Characteristic (ROC) analysis is used to illustrate the tradeoff between a test’s sensitivity and specificity, where a better test is often considered to have a greater area under the curve (AUC), equivalent to the “c-statistic.” While most analytic software can calculate the c-statistic, the translation of ROC findings into clinical decision-making rules is not always straightforward given different impact of false negative and false positive results. A calculation of a partial area under the ROC curve informed by expert clinical judgments may provide a better real-world assessment of the value of a predictive algorithm. Furthermore, many machine learning algorithms calculate a single threshold, above which a predicted outcome is likely to occur, and below which, the predicted outcome is not likely to occur. Formal ROC analysis enables a more nuanced view of the predicted outcome by supporting the calculation of stratum-specific likelihood ratios where different ranges of results of predictive algorithms have different strengths of associations with the outcome.
This workshop will help attendees understand and apply Bayesian and decision-analytic fundamentals, as well as derive and interpret ROC curves and stratum specific likelihood ratios. Real-world application of these techniques will be discussed, focusing on use cases derived from predictive analytics literature.
R. Finzel, G. Silverman, University of Minnesota; S. Liu, Rochester; H. Liu, Mayo Clinic; X. Jiang, UTHealth; S. Pakhomov, University of Minnesota
This workshop will provide attendees the information necessary to implement NLP workflows using cloud native technologies by providing practical introductions to UIMA-AS, Docker, Kubernetes, and Argo. It will start with the basics of composing NLP system "ensembles" designed to optimize performance in a particular domain and proceed through an introduction to cloud technologies-- including core concepts and technical terms, and explanation of several alternatives to the Argo/Kubernetes/Docker workflow. Explanations of when, where, and why to use each technology, along with some of the practical challenges of using each in a high-security (PHI) environment will be discussed. Workshop participants will then install Docker (a container protocol and server), Kubernetes (a container orchestration system), minikube (a platform for using Kubernetes locally), and Argo (a Kubernetes workflow manager) on their own computers and run a test NLP workflow on a collection of exemplar clinical notes (from the MTSamples corpus). We will then discuss common architectures for UIMA pipelines and pipelines for technologies that are common in other informatics domains and non-UIMA tools, as time permits.
R. Schreiber, Geisinger Holy Spirit, Geisinger Commonwealth School of Medicine; J. Hollberg, Emory University; P. Fu, City of Hope; N. Safdar, Emory University
There are now 1,868 physicians board-certified in clinical informatics by the American Board of Preventive Medicine and American Board of Pathology. There are 26 Clinical Informatics Fellowship Programs accredited by the Accreditation Council for Graduate Medical Education. There are also many non-board certified/eligible practitioners who need training in state-of-the-art applied clinical informatics. AMIA is uniquely suited to be the academic home for this community, because it provides a combination of personal experience and anecdote with firm grounding in evidence-based biomedical informatics literature, informatics theory, foundational knowledge, and proven best practices. A major part of that support is outreach to Chief Medical Information Officers (CMIOs) and those in similar roles (such as Medical Directors for Information Systems) who are charged with leading informatics change within their organizations, both large and small. More than 300 individuals have attended the CMIO workshop since 2011, some attending more than once, ranging from seasoned CMIOs of large systems to those who are just beginning their applied clinical informatics careers. During the 2017 CMIO workshop, we anonymously surveyed the 80+ participants. 91% responded that they desired to attend the workshop again, and the most requested topics were practical leadership skills and guidance on change management.
The 2019 CMIO Workshop will focus on leadership development, including didactic and small group exercises regarding the skills needed to be successful as an executive and potential career paths for those interested in becoming a CMIO or advancing once in that role. Additionally, we will focus on successful negotiation strategies, change management processes, and interactions with other C-suite and key stakeholders within a single hospital as well as for large health systems. We will then apply the leadership and change management skills to case-based discussion of current contentious yet practical topics for CMIOs including management of problem lists, how to implement open notes, and best practices regarding EHR documentation and scribes. Didactic presentations will be integrated with structured group discussions. Participants in the workshop will engage each other during the group discussions to practice the concepts and teachings from the didactic sessions. The CMIO workshop is a bridge between the more applied Clinical Informatics conference and the breadth of Fall Symposium offerings, with practical offerings participants can integrate into their daily workflow and help their organizations realize all the benefits of health IT.
A. Solomonides, NorthShore University HealthSystem; B. Kaplan, Yale University; E. Pan, Westat; C. Petersen, Mayo Clinic; S. Sheinfeld Gorin, NYPAC
Diverse advances in informatics and its applications in biomedicine, healthcare, and everyday life have made it necessary to review the scope of ethics as it applies to this burgeoning field. There are algorithms that learn from human decisions and practices, absorbing certain biases in those decisions along the way. Our ubiquitous, highly convenient mobile and social media also register every move and every purchase we make, every sentiment we share, and every friend we encounter in person or in a call; they create marketable data sets out data they collected by obscure, if not entirely nefarious means. Unprecedented mobility and increasing economic pressure has left many wondering how to look after elderly relations, just when AI visionaries are promising caring “nurse bots”. As individuals press the case to control their own medical records and as the legal climate begins to favor their position, the call has already gone up for records to be held in a mobile device, once again apparently empowering the individual while also handing over far more power to the network. The Internet, the great liberator it was hoped it would become, has also provided a safe environment for dangerous elements who abuse its freedoms. This collaborative workshop will bring together many strands of conversation in AMIA and elsewhere towards a fresh white paper on the new ethics of informatics.
T. Kuo, L. Ohno-Machado, University of California, San Diego
In this instructional workshop, we propose to briefly introduce blockchain technology and describe its use in biomedical research, healthcare, and genomics applications. We propose to recap the content for our last tutorial, emphasizing didactic contents and answering frequently asked questions about blockchain technology. We will then follow with examples of blockchain applications in biomedical, healthcare and genomic domains. The target audience is composed of informatics researchers, clinicians, IT experts, and leaders in healthcare or other institutions who have little or no experience building systems based on blockchain technology. We aim at benefiting both the audience who has attended our tutorial last year and new attendees. The two instructors have experience implementing the technology and can contrast it to other approaches. In this workshop, we will present three use cases in biomedical data sharing, healthcare privacy-preserving predictive modeling, and genomic data access logging. We plan to deliver didactic contents for 70 minutes, and also contents with interactive demos for 80 minutes. In our demos, we will utilize both Ethereum and MultiChain blockchain platforms, as our tutorial is platform-independent. After attending the workshop, we expect participants to understand blockchain technology, including the benefits, platforms, and variations, as well as an understanding of blockchain use cases, underlying requirements, and state-of-the-art implementations.
J. Chen, N. Seligson, The Ohio State University; J. Warner, Vanderbilt University; R. Miller, ASCO CancerLinO; J. Barkal, M2Gen; D. Patt, Texas Oncology; G. Alterovitz, HL7 Clinical Genomics/Harvard University/National Cancer Institute; K. Kehl, Dana-Farber; A. Wagner, GA4GH, Washington University; A. Abernethy, FDA (invited)
Although finding similar cohorts of patients for purposes of understanding clinical outcomes or shared molecular biomarkers seems intuitive, concretely defining patient’s similarities remains elusive. While one-dimensional identifiers of patient similarity have been successfully implemented in a number of limited clinical circumstances, the development of multi-dimensional patient matching has been less successful due partially to the unique challenges of characterizing multi-dimensional data. This workshop will serve to develop standardized “similarity nomenclature” and definitions. Currently, the term “similar patients” is used by multiple endeavors – from clinical trials matching, cohort matching, patients like mine, to simple clustering of “like” patients. Particularly in the oncology domain, cohort matching initiatives are being used for clinical trial patient identification, by pharmaceutical companies to segregate patients into good and poor responder cohorts, and by medical researchers to identify patients who have shared clinico-genomic features. Yet each of these similarity definitions is subtly different.
A unifying factor is that all patient similarity definitions are defined using a heterogenous combination of (1) shared molecular traits, (2) shared phenotypic traits, or (3) shared outcomes. To complicate matters, these disparate types of data require algorithmically diverse computational methods to compute interpatient similarity. Simple overlap analysis, various clustering algorithms, and deep learning methodologies have all been brought to bear. Yet none of these methodologies has become the standard for cohort matching. To provide clarity in defining patient similarity, this collaborative preconference workshop aims to bring together medical informaticians from different backgrounds to develop a unified nomenclature and methods used for different types of patient similarity. Goals of the workshop are to publish in an associated journal the conference discussions and (1) concretely propose definitions for common types patient similarity and instructive example, (2) identify/describe the data elements needed for the similarity types, (3) enumerate the commonly used algorithms that are needed to process the data elements.