Informatics Research Seminar: Data Analytics for Health- Utilizing Large Social Media Data

Speaker: Albert Park, PhD

Presented from UNC-C

Broadcast Link: Seminar

Abstract: 

With the rapid increase in health-related data, we are seeing a revolution in both data sources and data analytics that can supplement and change traditional health care. My research focuses on utilizing our daily interaction on social media platforms to better understand health-related needs and concerns and then encourage positive behavior change at both the individual and population levels in the contexts of health. For this purpose, I process text communication from a highly popular social media platform, such as Reddit as well as a smaller but popular public online health communities like WebMD. In this talk, I share my research on (1) a process towards deriving health implications and identifying health-related changes from social media data and (2) a development towards behavioral change through data analytics.

Biosketch: 

Albert Park, PhD is an Assistant Professor in the Department of Software and Information Systems within the College of Computing and Informatics at the University of North Carolina-Charlotte. His research focuses on understanding and addressing a variety of public and personal health problems by applying computational methods (e.g., Natural Language Processing, Machine Learning) to large datasets. He was a National Institutes of Health-National Library of Medicine Post-Doctoral Fellow at the University of Utah. He holds a bachelor’s and master’s degrees in Computer Science from Virginia Tech, and a Ph.D. in Biomedical and Health Informatics from the University of Washington.

Informatics Research Seminar: Look before you Link-Assessing Health Data Linkage Feasibility and Linkage Results

Speaker: Sudha Raman, PhD

Presented from Duke

Broadcast Link: Seminar

Abstract: 

Linking data sources can help to enrich clinical researchers to more accurately define study populations, enable adjustment for confounding, and improve the capture of health outcomes. While there is much guidance about the technical process of linking data, there is less so about the pre study assessment of feasibility and the post study evaluation of the linkage results.  When creating novel linked datasets, researchers must assess the feasibility of both scientific aspects (data quality and linkage methods) and operational aspects (access, data use and transfer, governance, and cost).  Another key aspect of data linkage evaluation is to articulate how a linkage process was performed and its accuracy so that the potential for bias can be assessed. This presentation will review the results of a group effort to create guidance for the assessment of the feasibility of health data linkage for researchers, as well as recommendations for the evaluation and reporting of health data linkage. Examples of health data linkage from Dr. Raman’s work will be shared.

Biosketch: 

Sudha Raman, PhD, is an Assistant Professor in the Department of Population Health Sciences. She is an epidemiologist who focuses on the use and effects of medications in populations (pharmaco-epidemiology), with careful consideration of a medicine’s benefits as well as harms. She received her PhD in epidemiology from the University of North Carolina at Chapel Hill and completed a fellowship at the Center for Pragmatic Health Systems Research at Duke Clinical Research Institute. Her current research explores the quality and methodological challenges of conducting research using real-world data, such as electronic medical records and administrative claims data, as applied to the evaluation of health care for both children and adults.

Informatics Research Seminar: Using Natural Language Processing to Evaluate Electronic Health Record Documentation of Hypertension Treatment

Speaker: Kim Shoenbill, MD, MS

Presented from UNC-CH

Broadcast Link: Seminar

Abstract: 

Over 45% of the 85.7 million US adults with hypertension have uncontrolled blood pressure resulting in increased risks of cardiovascular disease including stroke, heart failure, and myocardial infarction. Guidelines on hypertension management include lifestyle modification (e.g., diet, exercise) and medication initiation as first line treatment. To understand current hypertension treatment efforts and improve hypertension control, it is important to determine the frequency and inter-relatedness of lifestyle modification and hypertension medication initiation. However, lifestyle modification data is documented in narrative form within the electronic health record, making it “invisible” in evaluation of discrete data or metric measurement of hypertension treatment. Electronic health record data from 14,860 adult hypertension patients at an academic medical center were analyzed using natural language processing and statistical methods to determine documentation of lifestyle modification (i.e., advice and/or assessment) and hypertension medication initiation. Methods and results from this analysis will be discussed.

Biosketch: 

Kimberly Shoenbill, MD, PhD is a physician and informatician at the University of North Carolina – Chapel Hill. She is an Assistant Professor working in the Department of Family Medicine and the Program on Health and Clinical Informatics. She received her MD and her PhD in Clinical Investigation with an emphasis in Informatics from the University of Wisconsin – Madison. She is dually board certified in Family Medicine and Clinical Informatics. Her research focuses on secondary use of electronic health record data using natural language processing and machine learning coupled with statistical analysis. She is committed to using informatics to evaluate, inform, and improve patient care delivery and outcomes.

Informatics Research Seminar: A Direct-to-Patient Alert for Glycated Hemoglobin Screening

Speaker: Brian J. Wells, MD, PhD

Presented from Wake Forest

Broadcast Link: Seminar

Abstract: 

The creation of statistical models for risk prediction has rapidly increased over the past 20 years as electronic health records (EHRs) and other electronic data have become ubiquitous in health care. Unfortunately, these tools have had minimal impact on clinical practice and patient outcomes.  There are many reasons for the ineffectiveness of clinical decision support tools including: “alert-fatigue”, unsuccessful implementation of the tools into existing clinical workflows, clinician workloads, and lack of physician confidence in the tools. Many of these tools might have more impact if they were targeted at other members of the care team, including patients. Dr. Wells has proposed the use of “direct-to-patient alerts” and will give an example of an ongoing research project that will target high risk patients with text messages to suggest hemoglobin A1c (HbA1c) screening. He will describe the creation of a previously published HbA1c prediction tool being used to identify high risk patients as well as the research strategy for the upcoming project.

Biosketch: 

Dr. Wells is an Associate Professor in the Department of Biostatistics and Data Science at the Wake Forest School of Medicine where he also serves as the Associate Program lead for the Biomedical Informatics Program in the Clinical and Translational Science Institute. He is board certified in both Family Medicine and Clinical informatics and has extensive experience in the extraction and analyses of EHR data both locally and for multicenter projects like the CDC funded  SEARCH for Diabetes in Youth. Much of his research has focused on the creation of risk prediction models built from EHR data and the evaluation of outcomes in patients with diabetes. Dr. Wells is passionate about improving the creation and implementation of clinical decision support tools for better medical decision-making.