From 3M Health Information Systems
Keeping up with changing technology: The IAB Talks
To promote industry connections and an emphasis on curricular relevance, the University of Utah Department of Biomedical Informatics holds industry-sponsored mini-conferences every April and October, called the IAB (Industry Advisory Board) Talks. The April 2019 session focused on advancing technologies and showcased some of the interesting research projects being conducted.
The first speaker, Janette Vasquez, presented a study on using biometric data to enhance a patient’s engagement with their own health information. The project analyzed how video games could create engagement which would correlate to more playing and potentially better engagement with one’s own health information. At a high-level this idea seems quite practical; being a millennial, my world is full of video games and display or voice based interactive systems, so I can relate to the approach. This study used biometric data compared to surveys to gauge differences in engagement, measuring facial expressions (anger, disgust, fear, happiness, sadness, surprise, etc.), electrodermal activity (sweat gland production), and heart rates. For me as a nurse, assessing someone’s true physical and emotional feelings is highly subjective and at times, extremely complex. Considering I was taught to evaluate someone’s pain using the Wong-Baker FACES Pain Scale rating (as shown below), you can quickly see how difficult and subjective it can be.
This presentation was exciting to me, knowing that the use of biometrics in future assessments may accurately reflect the intrinsic feeling someone is having, allowing me as a clinician to better support their needs.
The second speaker of the event, Ryzen Benson, presented a project centered around using digital phenotyping to better understand population health. The presentation showed specifically how users are aggregating tweets through Natural Language Processing (NLP) algorithms to determine the usage of different brands of electronic cigarettes. Take for example the following tweet: “A high school boy that works with me just showed up with 800 JUUL pods in a louis clutch, u can’t make this stuff up”. The NLP algorithm captures underage possession of the product (“high school boy”), first person experience (“with me”), mention of the brand (“JUUL”), and sentiment towards the tweet itself (“can’t make this stuff up”). Can or should this type of data aggregation be used for population health? Personally, I am encouraged to see how technology is enabling us to view elements of health in emerging markets at the population level.
The presentation reminded me of another at the 2018 American Medical Informatics Association (AMIA) Symposium by keynote speaker Dr. Jessica Mega. Her work with Verily Life Sciences has focused on utilizing hashtags (#) to capture images of food so that diabetic patients can better calculate their insulin dosages based on the carbohydrate count to that of similar meals. Dr. Mega and her team have used social media to enhance their library of available images for meals that would be commonly ordered in restaurant settings, increasing the accuracy of self-dosing for users of their application. This is a reminder that technology can be very helpful when used for the appropriate reasons.
Finally, a presentation I found particularly interesting was given by Eric Just, Senior Vice President and General Manager of Health Catalyst. He talked about digital and electronic phenotyping in the real world, defining a phenotype as a set of observable characteristics of an individual from the interaction of its genotype with the environment. Thus, phenotyping can be used as a process to identify populations with a specific interest in mind. For population health, who are you targeting? For inpatient safety, what do you want to improve? For evidence-based medicine, what are you trying to discover? Phenotyping can be utilized to stratify data into usable information.
In health care today, a majority of electronic phenotyping comes directly from Electronic Health Record (EHR) data found within healthcare systems as opposed to digital phenotyping which covers a wider spectrum of data including things like social media and devices. Implementing digital phenotyping could help us understand what’s happening around us, perhaps contributing to our goals of improving patient safety and preventing harmful events.
Similarities to this work can be seen in some of the work being performed at 3M Health Information Systems. In order to develop machine learning applications and apply phenotyping, data analytics (which includes data standardization and rule-based algorithms) starts with the foundation of clinical terminologies, applying them to both structured and un-structured data. Building a machine learning model starts with stratification of data for a specific use case. Using terminologies, rules, risk score calculations and other algorithms, tools can be built to help detect at-risk individuals, determine populations requiring intervention, learn from ongoing trends and prevent future harm to individuals being cared for.
Technology is changing rapidly and we’re all fighting to keep up. I applaud the University of Utah Department of Biomedical Informatics for bringing academia and industry together, and look forward to participating in future IAB Talk events.
Michael Denton, RN, is a clinical data analyst with the 3M Healthcare Data Dictionary (HDD) team.