By Hanae Armitage, Stanford Medicine | May 24, 2019
Invisible sensors, machine learning for disease diagnoses, big data in the clinic and more took the stage as topics at this year’s Big Data in Precision Health Conference.
What can big data do for you?
That question seemed to animate the seventh annual Big Data in Precision Health Conference, which ran May 22-23 at the Li Ka Shing Center for Learning and Knowledge. The event drew health care and data experts from across industry, academia and government to the School of Medicine.
Emma Huang, PhD, director of data sciences external innovation at the Johnson & Johnson California Innovation Center, answered it this way: It depends on who you are. If you’re a patient, big data might be a means to a quicker diagnosis or to an insight into a unique facet of your health. If you’re a doctor, it could help relieve the burden of lengthy or cumbersome tasks: Consider, for example, a machine-learning algorithm that performs administrative duties, like interpreting electronic health records.
With more than 550 attendees, the conference covered new technologies, such as sweat sensors, that capture biological data and algorithms that draw insights from massive amounts of information to better predict, prevent and treat disease precisely — the ultimate goal of precision health.
“This is a really exciting time in data science,” said Lloyd Minor, MD, dean of the School of Medicine, said at the start of the conference. “We’re truly covering the gamut of issues related to precision health and biomedicine. In the next few days, we’ll have speakers who will discuss everything from the social determinants of health to how we can use machine learning and other analytic techniques to improve the drug-discovery process.”
Putting your data where your mouth is
New technologies that collect health data and algorithms that extract biological insights are a big part of harnessing big data, but the ultimate goal is to bring the benefits directly to patients. One of Stanford Medicine’s main efforts in this realm is a pilot project known as Humanwide. The project employs big data to gain a detailed and multifaceted understanding of
individual patients, and uses that knowledge with the aim of improving their health. Led by Stanford clinical professor of medicine Megan Mahoney, MD, Humanwide is one of the first manifestations of big data in precision health at the clinical level. In a panel discussion, Mahoney sat down with Latha Palaniappan, MD, professor of medicine at Stanford; Nancy Shin, PharmD, an ambulatory care clinical pharmacist at Stanford; and Debbie Spaizman, a participant in the Humanwide study.
Spaizman discussed her own health success story, grounded in advances in pharmacogenomics. Using Spaizman’s genetic information, her doctors were able to figure out why Vicodin didn’t work well for her as a pain reliever before she underwent two surgeries. It turned out that her biology could not perform a chemical process that typically turns Vicodin into morphine. Her doctors were then able to provide her with alternative painkillers. “I had a lot of conversations that wouldn’t have ordinarily come up,” Spaizman said. “Getting to spend as much time with my doctor as I did was an added bonus.”
Big data for speedier diagnoses
Matthew Lungren, MD, assistant professor of radiology, is working on research that uses immense sets of medical-imaging data, such as X-rays, to help doctors expedite their diagnoses. Through machine learning, Lungren and a team of scientists have trained an algorithm to recognize X-rays that contain signs of disease in the chest. In his initial studies, Lungren found that the algorithm can diagnose these images just as well as radiologists.
“We’re getting a lot of interest from clinicians,” Lungren said. He gave an example of how he hopes the technology can be used: A primary care doctor sees a patient who she suspects has pneumonia. Per the standard of care, the patient has X-ray images taken of his chest, which must then be read by a radiologist. The problem is that radiologists are in high demand and can often get backed up with work. This, Lungren said, is where he sees potential for his team’s algorithm: to help clinicians read these scans on the spot, without having to wait hours for a radiologist to see them.
Bringing data to the masses
In a conversation with Minor, Euan Ashley, MD, professor of medicine, emphasized the idea of harnessing the power of whole populations to improve health care. It’s an opportunity not lost on Mintu Turakhia, MD, associate professor of medicine, and a team of scientists that collaborated with Apple to conduct a virtual study to determine whether a mobile app that uses data from a heart-rate pulse sensor on the Apple Watch can identify atrial fibrillation. During one of the sessions, Turakhia; Manisha Desai, PhD, professor of medicine; and Marco Perez, MD, associate professor of medicine, discussed the study. “When we started, there hadn’t really been a study done quite like this, and we weren’t sure what the participation was going to be like,” Perez said. “We were blown away by what we saw: Within eight months, we had more than 400,000 people, which is a little unprecedented.”
In a separate session, Kirsten Bibbins Domingo, MD, PhD, a physician at the University of California-San Francisco, pointed to the explosion of genetic data in recent years during a talk on social determinants of health. The data, however, is almost exclusively from people of European descent, which means the information derived from those datasets is biased toward that one population, she said. “The point that I want to make here is not about social determinants of health — it’s about diversifying our representation in our genetic studies,” Bibbins-Domingo said. Understanding when these datasets don’t apply to certain patients is the doctor’s responsibility, and it’s also their responsibility to figure out how we can broaden the diversity of these datasets, Bibbins-Domingo said.
More data, please
It may seem like researchers’ cups runneth over with data, but there are still many untapped sources of health information that scientists are seeking. Sleep, for instance, is still a relatively enigmatic part of human biology. What does it do to rejuvenate us and restore health? “That’s the million-dollar question,” said Emmanuel Mignot, MD, PhD, professor of psychiatry and behavioral sciences during a session on big data in sleep.
In an effort to answer the question, Mignot is collaborating with labs outside Stanford to collect enormous amounts of sleep data from 30,000 participants. “It’s really incredible because we can, for example, find proteins that are known to peak in the blood at very specific circadian times,” Mignot said. By measuring hundreds of proteins that are associated with sleep and circadian rhythm, scientists can start to stitch together a clearer understanding of the molecular biology behind sleep cycles.
Sweat is another potential source for lots of biological information, said Ali Javey, PhD, professor of electrical engineering and computer science at the University of California-Berkeley. “We’re trying to make wearable devices that can analyze sweat on the body noninvasively,” Javey said. There’s a whole library of chemical and physical information that can be distilled from sweat: levels of glucose, ions, heavy metals, drugs and vitamins, as well as skin temperature and even sweat rate. Now, Javey wants to understand how these data points can inform decisions related to health.
While many new technologies that measure human biology focus on wearables, Dina Katabi, PhD, professor of electrical engineering and computer science at the Massachusetts Institute of Technology, is looking to something called “invisibles.” She poses a question: What if you could track your breathing, steps and even pulse without a single piece of technology touching you? “This is exactly what I’ve been doing in my lab at MIT.”
Katabi’s new technology uses the electromagnetic radiation of Wi-Fi to measure these parameters. Every move you make — even every breath you take — creates perturbations in these waves, which Katabi has trained an algorithm to interpret. These algorithms can track where people walk in their homes, if someone has fallen, what their pulse rate is and more.
Unifying these types of technologies, data and analysis methods, Ashley said, is still one of the major hurdles in big data and health. “We’re surrounded by data, and I think we need to improve how we integrate that data into one place so that we can compute across it,” he said. “I think, if we can do that, then we really will be able to make another giant leap forward.”