Big data and artificial intelligence are reshaping our world. Earlier this month, at Computefest 2018, organized by the Institute for Applied Computational Science at Harvard University, held the symposium, “The Digital Doctor: Health Care in an Age of AI and Big Data.” Speakers were:
- Finale Doshi-Velez, PhD, Assistant Professor of Computer Science, Harvard University
- Matt Might, Director, Hugh Kaul Personalized Medicine Institute, University of Alabama at Birmingham
- John Brownstein, PhD, Chief Innovation Officer and Director, Computational Epidemiology Lab, Boston Children’s Hospital
- Marzyeh Ghassemi, PhD, Visiting Researcher, Google’s Verily; Postdoctoral Fellow, Computer Science and Artificial Intelligence Lab, Massachusetts Institute of Technology
- Jennifer Chayes, Managing Director, Microsoft Research New England and New York City
- Emery Brown, PhD, Professor of Medical Engineering and Computational Neuroscience, Massachusetts Institute of Technology
Here are Vector’s five takeaways from the symposium:
1) The use of big data and artificial intelligence in medicine should start with improving patient experience.
Computational science can improve a patient’s health care journey. From developing an “n-of-1” drug for a rare disease, to managing HIV treatment with machine learning, to creating websites for parents to better understand their children’s symptoms, the core focus needs to be the clinical benefit.
2) Technology is just another member of the team.
Machine learning, artificial intelligence and big data all present both solutions and problems when applied to health care. For example, Chayes noted that cancer immunotherapy – which uses components of a patient’s own immune system to fight off cancer cells – has historically only worked for a small subset of patients. Machine learning technologies can predict immunogenicity, identify existing immunotherapies that will work for some patients and aid in developing new ones for others. But tapping the necessary data poses challenges: It is distributed across a huge range of modalities – genes, proteins, immune responses, clinical symptoms. “Technology is just another member of the team, with its own strengths and flaws,” Chayes said. In medicine, technology and computational science can do some things better than others.
3) Life happens outside of the clinic.
Big data collection can track what patients are really experiencing in their daily lives. Ghassemi spoke about the power of digital phenotyping — moment-by-moment data compiled from personal devices like smartphones and quantified for analysis and application. Data collection can be active, input by patients themselves, or passive, recorded automatically. What patients report to health care providers during clinical visits usually isn’t the full picture, so digital phenotyping can give a more holistic and accurate understanding.
4) Existing technologies and datasets are being harnessed in new and exciting ways.
Brownstein’s talk explored how existing technologies are being used in novel ways to gather health care data, improve patient outcomes and give patients more access and autonomy in making health care decisions. He continually asks, “In what areas can digital data optimize patient care throughout the whole process?” Those areas include public health, where crowdsourcing data already present on widely used websites and applications can yield insights. For example, Yelp reviews for tracking food poisoning and Instagram disclosures about misuse of prescription medications, coupled with machine learning technologies, can be used to identify trends.
5) Data is the greatest drug of the 21st century.
Matt Might’s experience creating a one-of-a-kind drug for his young son exemplifies where data-driven health care is headed. When his son was first diagnosed, his disease was assumed to be one-of-a-kind, until a blog post Might wrote went viral and families across the globe began contacting him. Ultimately, 58 patients joined a protocol for rare diseases at the NIH that generated immense amounts of data, including clear biomarkers for the disease and potential diagnostics. From there, Might and fellow researchers determined the mutation causing the disease, identified an inhibitor that rescues the gene’s function, screened 200,000 molecules that would interact with the inhibitor, used deep learning methods to investigate toxicity, discovered an over-the-counter drug that corresponded to the inhibitor and successfully lobbied for “Right to Try” legislation, bypassing the traditional FDA approval process. Might then asked: could this process be repeated, on an accelerated timeline? Using his precision medicine “algorithm,” his team was able to take five rare diseases from patients to proposed pills in just one year.