Duke University has launched a new Apple ResearchKit study, the MS Mosaic Study, aimed at using big data and patient insights to improve medical understanding of multiple sclerosis. Dr. Lee Hartsell, assistant professor of neurology at Duke University Medical Center, spoke about the new study -- and some of the challenges involved in launching it -- at the Digital and Personal Connect preconference event at HIMSS18 last week.

"A patient [with MS] is experiencing any combination of 21 non-specific symptoms, whose severity will fluctuate due to unpredictable bouts of inflammation and due to unmeasured changes in biorhythms, body temperature, activity levels, sun exposure, diet, emotional state, and the side effects from any of 15 different disease modifiers, four different relapse modifiers, and 88 different symptom modifiers," Hartsell said. "These experiences can’t be summarized during a 10-minute visit. We try, and because we’ve tried, MS has become the second most expensive chronic disease to manage in the United States. It’s not significantly data-driven."

MS Mosaic, named because its goal is to create a comprehensive picture of the disease from fragmented patient reports, is an Apple ResearchKit app that launched in September. The app includes a number of tracking tools for people with MS to report their daily symptoms, activities, and experiences. In return, they get various features that help them understand trends in their disease.

"To help solidify engagement we incorporated several feedback tools," Hartsell said. "These include a commonplace symptom and medication journal, weekly reports that summarize symptoms and provide customized links to learn more about your symptoms and how to manage them, provider reports that summarize the data that occurs between clinic visits for quick provider review during the visit itself, and data visualizations that allow participants to find their own correlations within their symptoms."

Meanwhile, the app, which currently has 400 downloads and 350 active users, collects that de-identified data and starts using it to build models that will help researchers better understand the disease.

"All the while, the study is using the same data to solidify our understanding of MS symptoms through a variety of customized machine learning algorithms," Hartsell said. "Examples of our ongoing analyses include ways of identifying the hidden symptom subpopulations through clustering methodology. So a good example of this is actually fatigue. Ninety-seven percent of MS patients experience that on a regular basis. But are they experiencing it because of not getting enough sleep because they were up going to the bathroom throughout the night? Is it because of the sedating effects of their medication? Is it because the disease itself has kept them worn out? We don’t know. And we don’t have the current instruments that help us tease that out. So we don’t have a way of effectively intervening on that because we don’t know enough about this fatigue."

Hartsell said Duke has faced a lot of challenges building the app, some more expected than others. One is that research apps are currently in a weird place when it comes to seeking funding.

"This has been the biggest challenge for us and we imagine it will be a big challenge for other researchers," he said. "Is it a research study or is it a product? Because ultimately this patient-facing research platform can actually evolve into a clinical management tool. So as a result, potential funding sources might reject it. Federal grants will fund the analysis, but not the initial six-figure development cost. Pilot grants are way too small and other sources worry about the return on investment, because it’s originally perceived as a research project."

Another challenge is something of a chicken-and-egg problem. 

"You’ve got this app and it collects a lot of data. But in order to maintain good app engagement you have to provide them with lots of insights. You can do the machine learning for those insights, and that’s great, but machine learning requires massive amounts of data. So how are you going to get that initial catalyst? How are you going to get things moving?" Hartsell said. "I would argue the best approach is to involve the community itself from the very beginning. If they actually are contributing to the entire process, there’s an ownership, it’s much more patient-driven, and you can actually discover a lot of insights you might not otherwise have found."

Duke is continuing to run the study and to improve the app to collect more data -- not only patient-reported data and data incorporated from HealthKit, but eventually MRI and laboratory test data sources from the EHR. Hartsell hopes the finished product will help to disrupt a status quo that's long overdue for innovation.

"As it is right now there is not a single genetic marker, not a single blood test that’s going to help me identify which of the 15 disease modifiers I’m going to use with any one patient," he said. "In essence what really occurs is something akin to this: You have a single exam question with 15 possible answers. You have upwards of a year to answer the question and it may take upwards of four years to figure out you got it wrong — and you'll never find out if you got it right. But hopefully with new data we can inform that decision-making process."