Office of Research & Development
The field of informatics, also known as biomedical, clinical, or health care informatics, is a relatively new discipline that has been spurred by the rapid growth of health care technology. In its broadest sense, informatics attempts to improve the effectiveness and delivery of health care by using health information technology and many other disciplines, such as cognitive and behavioral sciences. VA informatics research and development initiatives cover several areas of broad focus, including:
All four services of the Office of Research and Development (ORD) support research in the area of informatics. But due to its mission of pursuing health care research related to caring for the "whole patient" (treatment, quality outcomes, cost, and delivery of care), Health Services Research & Development (HSR&D) supports a comprehensive informatics research program, funding intramural research projects and maintaining two resource centers that promote access to clinical and administrative data for VA researchers.
A critical foundation to health information technology research in VA is the Veterans Informatics and Computing Infrastructure (VINCI). VINCI is a research resource center to improve researchers' access to VA data and to facilitate the analysis of those data while ensuring Veterans' privacy and data security. Funded by VA's ORD and Office of Information and Technology, VINCI provides a secure high-performance computing environment and access to comprehensive VA health data, which helps researchers access national data on the entire VA population, and makes it easier to create sophisticated analytics tools.
The VA Information Resource Center (VIReC) is an HSR&D-funded resource center. It is tasked with advancing VA capacity to use data effectively for research and quality improvement and to foster communication between research data users and the VA health care community. VIReC shares this knowledge with VA researchers through its products and services (e.g., cyberseminar series).
Current areas of VA research in health care informatics include: NLP, phenotype studies, adverse event monitoring, clinical decision support systems, connected care technologies, and care management tools.
NLP is a field of study within computer science that uses artificial intelligence and computational linguistics to interpret and understand written language. It is especially useful to medical researchers who wish to use medical data contained within a Veteran's EHR that was written in the form of free text (rather than being entered into a form that contains discrete fields). Two examples of VA studies using NLP techniques are:
Watching for cancer after polyp removal—The objective of this study is to develop NLP techniques that can extract key data from free-text entries in EHRs to identify Veterans at high risk for colorectal cancer (CRC). Researchers propose to study a large group of Veterans who had at least one polyp removed via colonoscopy and that were followed up with a repeat colonoscopy. Using NLP algorithms, they will extract key data that may be associated with CRC from patient records, e.g., bowel preparation and polyp characteristics. Researchers will then develop a predictive model to estimate the risk of recurrent cancer.
Measuring risk for pressure ulcers—The objective of this retrospective study is to develop a NLP program to identify the occurrence of pressure ulcers in Veterans with spinal cord injury (SCI), using both structured and narrative data from the EHR. A secondary aim is to develop predictive models that will categorize the risk of developing pressure ulcers in Veterans with SCI. The currently used risk assessment tool for pressure ulcers, the Braden Scale, suffers from a number of limitations. Pressure ulcers are among the most significant complications in Veterans with SCI, in terms of quality of life and cost of care.
A phenotype is the physical expression of a person's genes through observable traits such as eye color, temperament, or high blood pressure. Phenotype studies search for specific traits across populations of patients by using informatics tools using NLP techniques. For myriad research initiatives, ranging from precision medicine to predictive analytics to population health, extraction of phenotypes from the EHR is a central task. Optimal EHR phenotyping is vital for large-scale research programs underway in VA, including the Million Veteran Program (MVP) and multicenter clinical trials that seek to develop evidence regarding the effectiveness of treatments and strategies to improve the health of Veterans.
Phenotype workshop—Building up thestate of the science conference on NLP, ORD convened a workshop, "The Central Role of Phenotyping in VA Research" in September 2016. At present there are few systems in place to share phenotyping methods across VA. The purpose of the workshop was to break down barriers to scientific collaboration and increase the validity of data used in phenotyping efforts. Discussions at the conference also focused on how NLP might be used to better understand patients' experiences of care; to monitor for changes in function and disability relevant to Veterans' benefits; and to improve surveillance of new symptoms and emerging diseases. As the need for phenotyping grows across VA research programs, it is essential to ensure that scientists collaborate and share best practices across all research services.
Collecting VA system-wide genetic data—VA's Million Veteran Program provides a rich platform for discovering the relationships among genes, environmental exposures, and health. More than 550,000 Veterans have provided DNA specimens, military exposure information, and access to their health records (by authorized researchers) to facilitate studies on topics ranging from the causes of Gulf War illness and posttraumatic stress disorder (PTSD) to functional impairment in schizophrenia and bipolar disorder. HSR&D is particularly interested in supporting studies that promote (1) development and application of big data methods and computational science (including NLP, machine learning, and advanced predictive analytics) to improve data standardization, clinical knowledge, and decision-making involving genomic data and (2) advancement of clinical phenotyping, validation and improvement of accuracy of phenotyping using medical record data, MVP survey data, and genomic data.
Big Data Science Initiative—The VA and Department of Energy created a partnership in early 2017 that allows VA researchers to utilize DOE's high-performance super computers to analyze vast sets of health data in search of new solutions to health problems like suicide prevention and cancer. The partnership will allow VA researchers to analyze data in a secure fashion from multiple agencies and programs, to include the Million Veteran Program, Department of Defense, Centers for Medicare and Medicaid Services, and the Center for Disease Control's National Death Index. Using computer simulation and modeling, researchers will be able to uncover scientific trends and new insights that were unapparent using traditional methods.
Quantifying mystery diseases using computer algorithms—Building on techniques used in previous studies, researchers set out to develop and implement informatics tools to monitor, detect, and prevent health problems in Operation Enduring Freedom and Operation Iraqi Freedom (OEF/OIF) Veterans. Their goals were threefold: (1) Use NLP techniques to extract information from progress notes within Veterans' EHR; (2) develop and validate algorithms to examine phenotypes of Veterans with syndromic diagnoses; and (3) characterize the epidemiology of medically unexplained syndromes in deployed Veterans.
Identifying risk for homelessness in Veterans using data in EHR—In this study, researchers aimed to develop computer algorithms to mine data (both structured and free text) from VA's EHR to identify both homeless Veterans and those at risk for homelessness. Using NLP and a dictionary of key terms, the algorithm will search the health record for known risk factors related to homelessness, and alert care givers so that they can connect Veterans with community resources. Researchers will identify a cohort of Veterans who have been established as homeless; develop a vocabulary of key terms and concepts related to homelessness (with special emphasis on psychosocial phenotyping); develop algorithms to identify Veterans at risk for homelessness; and establish working relationships with community homeless service providers.
Adverse event monitoring (AEM) and reporting is an important part of patient care, research, and clinical trials. Due to the relatively small number of patients typically followed in a clinical trial, adverse events often don't surface until a drug or medical product is in wide commercial use. Success depends in large part on having effective AEM processes and tools available, and communicating those standards to researchers and clinicians.
Examples of VA research to develop informatics-based tools to conduct AEM surveillance include the following projects:
Identifying Veterans at risk for kidney injury following heart procedures—For patients who have had cardiac catheterization, 1 to 31 percent experience acute kidney injury (AKI), which is associated with a 30 percent mortality rate. In a VA study done at Tennessee Valley Health Care System, Nashville, Tennessee, Researchers set out to develop and evaluate informatics tools that could be used to conduct near real-time surveillance for kidney injury in VA patients who undergo cardiac catheterization. Their goals are to (1) develop NLP interactive learning techniques to extract structured data that is relevant to AKI from the EHR; (2) develop a group of statistical prediction models to identify those patients who were most likely to experience AKI after cardiac catheterization; and (3) survey current patterns of care to identify risk for postprocedural AKI.
Ensuring smooth transition between VA and non-VA health care services—This project aims to examine the impact of VA provider notification when older Veterans utilize non-VA inpatient services, followed by posthospital geriatric-care coordination in a controlled trial. It is important to ensure that data from care given outside the VA medical system is captured and included in the VA EHR, to avoid test duplication, medication prescribing errors, and adverse events.
Given the vast body of scientific and medical knowledge available to clinicians, it is becoming increasingly important to develop clinical decision support (CDS) tools to help them deliver the best care to Veterans. These tools are also needed for the growing group of Veterans who are aging and often experience a host of complex diseases that can occur together—like heart disease and diabetes—complicating their care. Two examples of VA research into developing better decision support systems are:
Providing better care for patients with chronic disease—In this study, researchers aimed to develop new informatics techniques to automate quality control using CDS tools that will take into account Veterans who have multiple chronic illnesses. The objectives of the study were to (1) create automated knowledge bases (KBs) that incorporate the latest evidence-based clinical guidelines; (2) interview key VA personnel on recommendations for best-managing patients with multiple chronic diseases; and (3) develop systems to provide disease-specific KBs along with CDS tools to medical staff in Patient-Aligned Care Teams (PACT). The project team developed or updated the KBs for five clinical areas (diabetes, hyperlipidemia, hypertension, chronic kidney disease, and heart failure). They also developed functionality for the CDS tools to process guidelines for all five KBs sequentially for the same patient.
Using computer alerts to ensure timely cancer diagnosis follow-up—Despite technology advances that have helped physicians identify and treat cancer earlier and more effectively, there is a need to improve the process of informing clinicians and patients about cancer diagnoses. Researchers have identified a breakdown in the follow-up process for a cancer diagnosis, despite physician alerts in the EHR. Building on earlier work, researchers propose to create an automatic surveillance program that uses VINCI to generate triggers in the EHR for evidence of potential delays in follow-up of abnormal test results. Their aims are to (1) evaluate the accuracy of a VINCI-based "real-time" automated surveillance system to identify patients at risk of missed or delayed diagnosis of five common cancers, (2) establish how to integrate "real-time" surveillance and communication about at-risk patients into the PACT, and (3) evaluate effects of automated surveillance on timeliness of the diagnostic process.
Using measurement science to standardize VHA data—The primary goal of the Measurement Science QUERI is to integrate measurement science into health care for Veterans. Measurement science (defined as the theory, practice, and application of suitable metrics) is at the core of VA's learning health care system and is a critical component at every stage of the quality improvement and implementation process. Using system-wide data to promote performance measurement, improvement efforts, and electronic tools (i.e., clinical reminders) depends on the uniformity of those metrics, the efficiency with which they can be obtained, and the accuracy with which they are measured. Whenever a health care decision is made, it must be based on valid and reliable data that are linked with all relevant information.
Aging Veterans who have multiple chronic diseases are particularly vulnerable to adverse outcomes when transitioning to a different health care provider or facility, especially when that care is complicated by a co-existing mental health or social issue like homelessness or poverty. Using technology to assist patient transitions and track quality of care is fast becoming the new standard. Two examples of newly funded research into connected care technologies are:
Virtual nurses assist Veterans—This study proposes a technology-assisted care transition intervention founded on the concept of a virtual nurse that interacts with Veterans through different technology channels. The virtual nurse is a computer program designed to simulate a discharge nurse. During the inpatient stay, the virtual nurse will appear on a computer touch screen and will educate Veterans with congestive heart failure or chronic obstructive pulmonary disease about the important components of a care transition. As well as how to send and receive text messages on their mobile phone. Following discharge to home, the virtual nurse will continue to coach Veterans and their family members and improve post-discharge access to care through two-way computer-tailored text messaging made possible by VHA's new HealtheDialog system.
Better diabetes care through technology—Currently over 3 million Veterans have adopted the VA Patient Portal, My HealtheVet; however, actual use of the portal varies considerably, and there is little data on how the portal features can best improve patient outcomes. In this study, researchers will work to define effective portal use, using clinical and administrative data, coding of secure messages, patient surveys, and in-depth interviews with Veterans who have extensive portal experience. Lessons from this work will be used to refine an intervention to support effective use of portal features for chronic disease management. Researchers have chosen to anchor their investigation in a particular disease area: type 2 diabetes, a complex disease affecting many Veterans.
Veterans are in particular need of optimal care coordination, given that many suffer from co-occurring health conditions, mental health problems, and a challenging home environment. Poor care coordination is a principal cause of avoidable illness, death, excess resource use, and dissatisfaction among Veterans and their health-care teams. Three examples of VA research into developing better care management tools (CMT) are:
Improving care coordination for patients with chronic illness—Veterans with complicated health care needs are most often reliant on their primary-care providers (PCP) for care coordination. VA's PACTs have been proposed as uniquely suited to improve care coordination for Veterans in the community setting. In order for that strategy to be effective, researchers say team members must not only have appropriate patient information, but also care-coordination standards to gauge caregiver effectiveness. In this study, researchers set out to identify the information PACTs need to best coordinate care for high-risk patients in three groups: uncontrolled diabetes, congestive heart failure with suboptimal medication control, and those at risk for a delayed cancer diagnosis. They also aimed to develop care coordination standards and measures.
A better way to refer Veterans for specialty care—More than 33 percent of patients in the U.S. are referred to a specialist each year. However, that process is often marred by cancelled appointments, breakdown of communication between specialists and PCPs, and delayed patient care. In VA, 36 percent of referrals from PCPs are cancelled by specialists, and 50 percent lack follow-up within 30 days. Researchers in this study set out to identify barriers to specialty consultations within VA and to design improvements to the referral process. They examined the consultation process at two VA health care facilities by holding in-depth interviews, observing work flows, and assessing consult templates in the EHR. After analyzing the collected data, researchers suggested improvements in the referral process. Their recommendations included communicating specialists' requirements to PCP care teams, limiting the need for information external to the patient record, and expediting communication of urgent patient information.
Helping Veterans better manage their medication—The goal of this project is to improve usability of the Secure Messaging for Medication Reconciliation Tool (SMMRT) for Veterans, pharmacists, and nurses using a three-arm randomized controlled trial to evaluate the effects of My HealtheVet training and SMMRT. The My HealtheVet database is the VA's online personal health record, which provides online access to better manage Veterans' overall health. The SMMRT trial will be a benefit to the VA by decreasing medication discrepancies; decreasing emergency room visits and hospital readmissions following discharge, commonly the result of adverse drug events; and improving measures of patient engagement, patient centeredness, and patient satisfaction.
Document search tool may boost heart care
VA Research Currents, October 2016
VA Research holds conference on natural language processing
VA Research Quarterly Update, October 2015
Genomic medicine: Tackling the next frontiers in VA
VA Research Currents, October 2015
Million Veteran Program hits quarter-million enrollment mark
VA Research Currents, March 2014
Investigator Insights: Using Informatics to Identify Veterans At-Risk for Homelessness
Veterans Health Administration, Dr. Adi Gundlapalli
The Burden of Inbox Notifications in Commercial Electronic Health Records. Murphy DR, Meyer AN, Russo E, Sittig DF, Wei L, Singh H. Health providers receive an overwhelming number of electronic notifications via the EHR, making it difficult to focus on important patient information. Researchers set out to measure the number and type of EHR notifications so that they could calculate physician burden. JAMA Internal Medicine. 2016 Apr;176(4):559-60.
Predictors of Recurrent AKI. Siew ED, Parr SK, Abdel-Kader K, Eden S, Peterson JF, Bansal N, Hung AM, Fly J, Speroff T, Ikizler TA, Matheny M. Using VA health care data, researchers identified clinical risk factors that could predict risk for recurrent kidney injury in Veterans who had been previously hospitalized for acute kidney injury. J Am Soc Nephrol. 2016 Apr;27(4):1190-200.
Validating a strategy for psychosocial phenotyping using a large corpus of clinical text. Gundlapalli AV, Redd A, Carter M, Divita G, Shen S, Palmer M, Samore MH. Researchers used natural language processing algorithms to examine more than 1,500 patient records to extract key psychosocial concepts. Using these techniques will allow researchers to focus their data searches on selected key terms, rather than sort through entire patient databases. J Am Med Inform Assoc. 2013 Dec;20(e2):e355-64.
Study protocol: identifying and delivering point-of-care information to improve care coordination. Hysong SJ, Che X, Weaver SJ, Petersen LA. In the interest of improving care coordination and becoming more patient-centered, VA created the Patient-Aligned Care Team, or PACT. This study partnered with primary-care clinicians at several ambulatory care sites to find out what types of information PACT members need to be most effective in delivering quality care to Veterans. Implement Sci. 2015 Oct 19;10:145.
New Studies Do Not Challenge the American College of Cardiology/American Heart Association Lipid Guidelines. Hofer TP, Sussman JB, Hayward RA. The study authors warn against returning to old guidelines that recommend treating patients with high cholesterol by prescribing medicine to target desired LDL levels. Instead they recommend following recent changes in treatment guidelines that promote the importance of considering cardiovascular risk along with optimal LDL levels. Ann Intern Med. 2016 May 17;164(10):683-4.
Implications for informatics given expanding access to care for Veterans and other populations. Dixon BE, Haggstrom DA, Weiner M. Increasing access to health care both within and without the VA health care system will require a reciprocal growth in informatics research to support care coordination needs. J Am Med Inform Assoc. 2015 Jul;22(4):917-20.
Exploiting the UMLS Metathesaurus for extracting and categorizing concepts representing signs and symptoms to anatomically related organ systems. Tran LT, Divita G, Carter ME, Judd J, Samore MH, Gundlapalli AV. The Unified Medical Language System is a collection of key medical terminology and coding standards that enables interoperability between different computer systems. Using the UMLS, researchers developed a method to extract key data from medical records and map those concepts to anatomical organ systems. The goal is to correlate patient symptoms to organ systems for further epidemiological studies. J Biomed Inform. 2015 Dec;58:19-27.
Heart Failure Medications Detection and Prescription Status Classification in Clinical Narrative Documents. Meystre SM, Kim Y, Heavirland J, Williams J, Bray BE, Garvin J. Researchers set out to document heart failure treatment performance metrics by extracting key data from EHRs. They used two methods of data detection and extraction: a rules-based learning system and a machine learning-based system. Studies in health technology and informatics. 2015;216:609-13.
Transitional Care Outcomes in Older Spanish-English Bilingual Veterans. O'Kula SS, Gottesman E, Jones S, Signor D, Hung WW, Boockvar KS. Care transitions between inpatient and outpatient settings can be fraught with poorer outcomes for older patients. For older, bilingual Veterans, navigating care transitions can potentially be more difficult. Researchers reviewed health outcomes for Spanish-English bilingual Veterans with English-speaking Veterans to see whether there were significant differences in care transitions. J Am Geriatr Soc. 2016 May;64(5):1132-3.
Patient Education for Consumer-Mediated HIE. A Pilot Randomized Controlled Trial of the Department of Veterans Affairs Blue Button. Turvey CL, Klein DM, Witry M, Klutts JS, Hill EL, Alexander B, Nazi KM. As VHA opens up patient access to outside providers, it will be important for Veterans to understand how to use the VA patient portal (My HealtheVet) and to generate a continuity of care document that can be shared with other providers. Researchers found that patient education on using My HealtheVet and looking up health information on the internet was well-received by Veterans and helped to reduce duplicate lab tests. Appl Clin Inform. 2016 Aug 3;7(3):765-76.
Finding Important Terms for Patients in Their Electronic Health Records: A Learning-to-Rank Approach Using Expert Annotations. Chen J, Zheng J, Yu H. In an effort to involve patients in their own health care, providers are now giving Veterans access to their online medical records via patient portals. However, most clinical notes are written in jargon and likely not understandable to patients. Researchers set out to develop a NLP system called Finding impOrtant medical Concepts most Useful to patientS (FOCUS), to identify and rank relevant terms most important to patients. JMIR Med Inform. 2016 Nov 30;4(4):e40.
Registry-Based Prospective, Active Surveillance of Medical-Device Safety. Resnic FS, Majithia A, Marinac-Dabic D, Robbins S, Ssemaganda H, Hewitt K, Ponirakis A, Loyo-Berrios N, Moussa I, Drozda J, Normand SL, Matheny ME. Adverse event reporting for medical devices is a voluntary process. Early identification of performance issues in medical devices would allow for improvement of device design and help to protect patients. In this VA study, investigators used data from a national cardiovascular registry to implement a continuous surveillance program to monitor post-surgical outcomes for implantable vascular-closure devices. N Engl J Med. 2017 Feb 9;376:526-535.
Electronic medication alerts designed with provider in mind reduce prescribing errors
Indiana University, March 25, 2017
Artificial intelligence is learning to predict and prevent suicide
WIRED, March 17, 2017
Online tool helps patients better understand EHR notes
Information Management, March 10, 2017
Wearable biosensors can tell you when to see the doctor
National Institute of Biomedical Imaging and Bioengineering, March 9, 2017
Genomic medicine goes mainstream
Modern Healthcare, Feb. 18, 2017
Eight years of decreased MRSA health care-associated infections associated with Veterans Affairs Prevention Initiative
Elsevier Health Sciences, Jan. 5, 2017