May 15, 2020—In the race to stem COVID-19, researchers around the world are testing the capacity of artificial intelligence (AI) to assist in tasks such as diagnosis and drug discovery. So far, AI’s biggest success during the pandemic has been in speeding up the process of identifying existing drugs that can be repurposed to help suffering patients, said Deborah DiSanzo, who recently lectured on COVID-19 in the new course she’s leading at Harvard T.H. Chan School of Public Health—Artificial Intelligence in Health.
DiSanzo cited in her lecture an AI knowledge graph developed by researchers at the UK startup BenevolentAI and the Imperial College London, which found that baricitinib, a rheumatoid arthritis drug, had the potential to inhibit the virus that causes COVID-19. It and other drugs identified in similar studies have now gone into clinical trials.
“Two years ago, finding either a new or repurposed drug target would take six to 18 months,” said DiSanzo, a former health care technology executive. “These researchers did this in weeks.”
Diagnosing COVID-19, however, has been less successful for AI so far, she said, with the limited lung imagery currently available from COVID-19 patients making it difficult for neural networks to learn the difference between the effects of the virus and standard pneumonia.
Enhance, not replace
For DiSanzo’s students, these mixed results provided a timely example of one of her course’s main takeaways: AI can enhance health care delivery and research, but it’s not a replacement for the knowledge and skill of providers and scientists.
“I’m really excited about the technology and potential application of AI,” said Nimerta Sandhu, MPH’20, an MD candidate at Drexel University College of Medicine. “This course provided insights on technology solutions that offer added value and others that have room for improvement. One of the biggest challenges is going to be ensuring that, as we incorporate more AI in our work, it doesn’t detract from the empathy essential in the patient-provider relationship.”
“I want students to have a realistic view of what artificial intelligence can bring to public health,” DiSanzo said. “People usually have either a very positive view—that it’s magic and can solve all the world’s problems—or they have a very negative view, that it’s biased and doesn’t give accurate results.” She said that she wants students to leave her course knowing the right questions to ask, because it’s likely to be a part of their jobs, whether they are in practice or policy.
Prior to joining Harvard Chan School, DiSanzo’s roles included CEO of Philips Healthcare, and general manager of IBM Watson Health, the IBM business unit founded to advance artificial intelligence in health. Last year, as a Harvard Advanced Leadership Initiative Fellow, she was encouraged by the program’s faculty chair Meredith Rosenthal, C. Boyden Gray Professor of Health Economics and Policy, to develop a course for MPH students.
DiSanzo hadn’t planned to cover COVID-19 as she worked on her syllabus in January, but as the full extent of the pandemic emerged, she added it to her list of lecture topics—which also included drug discovery, medical imaging, and patient monitoring.
While the spring semester’s move to online learning required the first-time instructor to pivot on the fly, DiSanzo has been delighted with the results so far, she said. Her 24 students—who include physicians, a veterinarian, and a psychologist—have been very engaged, participating actively on discussion boards and in chats with guests including executives from Google and pharmaceutical companies.
DiSanzo hesitates to make predictions about the future of AI in health, noting the field’s history of overly optimistic projections. But things are different today, she said. In recent years, computing power, available data, and neural network capacity have advanced by leaps and bounds. It’s likely that in 10 years—maybe even five—“every health care or public health decision that we make, or care that we give, or diagnosis that we make, will be made with some help from artificial intelligence,” DiSanzo said. And with the COVID-19 pandemic pushing the field forward at even faster rates, she said, the next advancements may be just months away.