This past September, I had the opportunity to participate in an eye-opening event across the river from my hometown of Boston, MA. The event was a discussion of AI in medicine held on the campus of Harvard University. During the discussion, a senior faculty leader at Harvard Medical School commented to the effect that research performed at the medical school for a particular specialty had shown that AI could make a patient diagnosis with 90% greater accuracy than a physician. In addition, this same research suggested that 80% of the time patients perceived communications generated using AI to be more compassionate than communications received from actual physicians.
I found these to be astonishing claims and texted my wife, who had spent nearly 30 years in hospital administration within the same teaching hospital system, asking her whether this could possibly be true. “Probably”, she replied. A few weeks later I hosted a panel which included the Chief Data & Analytics Officer of another leading medical research institution and questioned him about this reported data. His response was that these results were consistent with data he had seen at his own institution. This begs a large question. Can AI transform healthcare to improve and potentially even revolutionize medical outcomes?
In 2019, the National Academy of Medicine published a report entitled Artificial Intelligence in Healthcare: The Hope, The Hype, The Promise, The Peril. The report noted “The emergence of artificial intelligence (AI) as a tool for better health care offers unprecedented opportunities to improve patient and clinical team outcomes, reduce costs, and impact population health.” The report went on to add, “AI is poised to make transformative and disruptive advances in health care, but it is prudent to balance the need for thoughtful, inclusive health care AI that plans for and actively manages and reduces potential unintended consequences, while not yielding to marketing hype and profit motives”.
On March 27 of this year, Mayo Clinic, with the benefit of 5 years of further advancements in AI, published a fresh perspective, AI in Healthcare: The Future of Patient Care and Health Management. The Mayo Clinic article discusses current applications of AI in healthcare. Echoing the Harvard Medical School example, the article notes, “In some situations, AI can do a more accurate job than humans. For example, AI has done a more accurate job than current pathology methods in predicting who will survive malignant mesothelioma.”
Belief in the potential of AI to transform healthcare has been building in recent years. In 2019, Microsoft CEO Satya Nadella was quoted as saying, “AI is perhaps the most transformational technology of our time, and healthcare is perhaps AI’s most pressing application”. In that same year, Google Health stated, “We think that AI is poised to transform medicine, delivering new, assistive technologies that will empower doctors to better serve their patients”. So, how will AI be applied to improve and revolutionize healthcare and medical outcomes?
To answer this question, I sat down last month with Philip Payne, the Janet and Bernard Becker Professor, Chief Data Scientist, and director of the Institute for Informatics, Data Science and Biostatistics at Washington University School of Medicine in St. Louis. Payne holds a doctorate in Biomedical Informatics from Columbia University. His research has covered topics ranging from data quality in medical record improvement to work on how synthetic data mimics real patient data to accurately model the COVID-19 pandemic.
The Institute for Informatics, Data Science, and Biostatistics (I2DB) offers an academic and professional home for research and practice across the School of Medicine at Washington University in St. Louis. The Institute also works collaboratively with the University’s Institute for Public Health and Schools of Engineering, Social Work, and Business, as well as the Cortex Innovation Community, an innovation hub centered in St. Louis. In addition to research activities, the Institute also hosts a series of symposiums, including an AI and Digital Health Summit, which was held in October, and most recently, on April 12, a symposium on The Power of AI in Medicine. The April symposium showcased some of the ways in which AI can enhance diagnostics, personalize treatment, and ultimately improve patient outcomes.
Commenting on the work of the Institute and the application of data and AI to improving healthcare and medical outcomes, Payne noted, “The primary benefit of applying data science and AI methods and technologies in health and healthcare is to augment human capabilities”. Payne continued, “This will allow providers and patients to focus on health promotion and care delivery while minimizing high-friction and low-value activities that often impede those foci.” He notes, “Ultimately, this is about re-emphasizing the humanistic aspects of health and healthcare while simultaneously producing and understanding data at a scale that was previously infeasible”. Payne concludes, “This has the potential to improve the quality, safety, outcomes, and value of care.”
Even with great opportunity and potential, challenges remain. The Mayo Clinic report notes that despite the many exciting possibilities for AI in healthcare, there are risks that must be considered. The article comments, “If not properly trained, AI can lead to bias and discrimination. For example, if AI is trained on electronic health records, it is building only on people that can access healthcare and is perpetuating any human bias captured within the records.” Payne recognizes these challenges, observing, “Three significant challenges impede the optimal use of data science and AI methods and technologies in health and healthcare”. He identifies these as 1) misalignment of business incentives, 2) the absence of cohesive, well-characterized, and readily sharable data, and 3) the current dominance of the AI market by industry.
Payne explains that there continues to be a misalignment of business incentives, with the consequence that “improvements in throughput and the reduction of low-value care are deleterious to the financial well-being of care providers”. The lack of readily sharable data is a further consequence of this misalignment, as Payne notes, “Data gathering has been centered on provider organizations, payers, or both”, rather than on patients and their health journeys. This remains a challenge for healthcare providers and institutions.
Perhaps the greatest impediment to the successful application and adoption of AI within healthcare and medicine, however, is due not to scientific or technical limitations, but to behavioral and human factors. This is a topic that Payne has spoken about and is a common obstacle across domains as well as industries. How do we communicate in understandable human terms? How do we see and understand the potential impact of AI within healthcare and medicine through a human lens? Payne observes, “We are approaching the design and delivery of such capabilities as engineering, rather than as a behavioral challenge.” Payne advocates for a greater appreciation of the human element, and advises, “We must invest in and prioritize efforts to understand and optimize the human-computer interface surrounding data science and AI capabilities for providers and patients.”
Decades of research have shown how important “transparency and explainability” are. Payne notes, “We do not readily utilize such knowledge to inform the design/delivery of technologies at the point-of-care and beyond”. Payne adds, “We need to re-emphasize these theories and methods, and recognize that AI and data-driven interventions are most effective when they augment and complement human capabilities”. He concludes, “Effective communication of these human benefits is crucial in overcoming the challenges of data and AI in healthcare.” This improvement of the human experience will be essential to optimizing healthcare and patient outcomes using AI.
Payne foresees the transformational potential of data and AI, noting, “Given the current state of the art, the two most likely, high-impact opportunities for AI and data-driven interventions are: 1) improving throughput and access by better aligning patients and providers based on need, acuity, and “trajectory”; and 2) creating an actionable, longitudinal view of patient health and healthcare in a manner that is quickly interpretable by both providers and patients”. The primary upside of these approaches will be improved speed, value, quality, safety, and outcomes, as well as reduced time needed to move discovery from the lab to practice. Payne notes that, “The primary downside of these approaches is that they may mask or amplify critical biases as well as inequities or misalignment in business and financial incentives that influence health outcomes.”
Wrapping up our conversation, I couldn’t resist asking Payne about the Harvard Medical School discussion and the reported findings on AI improvement of medical diagnosis and compassionate communications. Payne responded, “As we delve deeper into testing AI in clinical settings, we’re seeing that AI can make astoundingly accurate diagnoses, and we don’t even fully understand how”. Payne continued, “We’ve also been pleasantly surprised by some of AI’s other benefits such as its ability to sound compassionate, which really fills a niche that in many cases we weren’t even actively trying to solve for, at least from a technology perspective”.
In summation, Payne reflects on the future of AI in healthcare and medicine, “It’s exciting to me to see some AI technologies coming to scale and know that it’s really just a shadow of what will emerge over the next decade as we come to understand the proper role and value of AI technology in improving human health and wellbeing — and our essential role as humans in responsibly guiding its use”. It is clear that we are in the very early stages of what can be the most consequential application of AI for greater good – improving human health. It is gratifying to see healthcare and medical leaders taking the initiative as they push the limits of AI to drive innovation and patient care.