In an age of unprecedented change, staying current has never been more important. Our team at Chartis is curating news most relevant to the healthcare industry and tracking the topics that are trending on seven key issues: high reliability care, digital and advanced technology, financial sustainability, health disparities, the health ecosystem of the future, partnerships, and the provider enterprise. Each week, we break down what’s happening and why it matters.
The recent announcement that International Business Machines (IBM) may be selling its Watson Health division has caused many to question the future of artificial intelligence (AI) and machine learning (ML) in healthcare. For years, AI and ML have been touted as having the potential to make a great positive impact in healthcare. Emerging applications and some already in use include:
However, despite these use cases, AI is not being employed in more areas of healthcare today, as it faces substantial barriers. Few people, including patients and physicians, truly understand what AI and ML are and how they work, and don’t entirely trust the algorithms, slowing adoption. It is also difficult to effectively incorporate AI into existing clinical workflows. Per a recent NEJM Catalyst article, “A common criticism of machine learning in health care is that it is not easily interpretable.”
AI’s most publicized entrée into clinical practice — the aforementioned IBM Watson Health solution — was found to be error-prone. Watson caught the attention of the world when it won the TV gameshow Jeopardy, but then stumbled when it couldn’t even differentiate between types of cancer, years after it was being sold to healthcare providers with promises of highly complex capabilities that would diagnose cancers and pinpoint more effective treatments. It was reported in 2018 that one of the problems was that Watson’s algorithms were based mostly on hypothetical data, which led it to give bad or harmful clinical advice. The backlash was quick and fierce, halting the use of Watson in many clinical settings, slowing further sales of the technology, and likely contributing to its pending sale. As a recent MedCity News article aptly said, the Watson situation shows “how difficult it can be to apply AI to some of healthcare’s trickiest challenges – and how easy it is to create cynicism among would-be users.”
The lack of understanding of AI and ML, the tendency to distrust usage in clinical practice, and the recent announcement about IBM Watson doesn’t mean this is the end of the road for AI and ML in healthcare. Rather, non-clinical applications can be rolled out first, and they are indeed growing rapidly. Clinical applications should not be built to solve complex problems that humans perform. Rather than replacing a physician and the complex tasks that person performs, ML and algorithms can be used to augment the physician’s knowledge base to lead to a better diagnosis or treatment recommendation. The Wall Street Journal recently quoted Cynthia Burghard, research director at IDC Health Insights, as saying, “The most successful applications of AI in healthcare to date have been when the technology aims to solve discrete and narrow problems.”
Oak Street Health, which operates 80 primary care centers in underserved communities, recently published a piece in NEJM Catalyst about their work in applying ML to create a decision-support tool for the primary care setting. Oak Street sought to create a better system for identifying higher-risk patients based on the recorded medical, behavioral, and social needs of patients. The model was designed to create output that was simple, easy-to-interpret, and actionable. Oak Street Health’s approach was inclusive, collaborative, and involved multi-disciplinary teams (e.g., physicians, nurse practitioners, and social workers). They implemented a staged testing and roll-out period, making adjustments along the way. The tool resulted in progress: “At the beginning of the deployment period, 40 percent of high-risk patients had an action item documented by the care team in the previous 30 days...As of November 2020, this engagement metric had increased to 55 percent of high-risk patients with a recent action item documented.” Still, the authors noted they observed some continued reluctance to use the tool, stating that “more work is needed to reach those providers and social workers who remain uncertain about its utility.”
Though AI and ML have the potential for major improvements to healthcare quality, outcomes, and efficiency, the Watson example and general hesitance around AI/ML demonstrates that simple clinical applications are likely where the technology needs to start. In addition, non-clinical applications, such as tools for back-office administrative tasks, or assisting in research and education, likely have a better chance for earlier adoption and positive impact.
While still in the midst of the pandemic and with most of the current focus on preventing the spread of COVID-19 and vaccination efforts, some groups are already using lessons learned from COVID to plan for the future and prevent another virus from having the same impact.
A new study from UCSF and Careport Health indicates that a critical success factor in future pandemics is ensuring hospitals do not face significant surges. The study compared patient outcomes to surge volumes, and the results showed higher death rates when hospitals were facing excess volumes. Hospitals across the world have seen similar results: increased ICU demand in COVID is linked with increased mortality rates from the disease.
An additional key to limiting pandemic outbreaks is identifying emerging pathogens and determining which pose risk. Abbott is one group building a pandemic defense coalition with the goals of developing capabilities and capacity to identify new viruses and collaborating with leading experts to determine the risk and appropriate reaction for each. Abbott’s current coalition includes infectious disease and surveillance experts from academic health systems and leading research institutes in eight countries around the world, representing key locations “where novel pathogens tend to pop up.” The coalition partners can identify viruses and pathogens to further investigate based on patient patterns they see in their respective geographies and for which Abbott can build prototype testing. Other similar alliances for future pandemic preparedness are also in development, including the Foundation for Innovative New Diagnostics and the University of Washington's Alliance for Pandemic Preparedness.
The case for “flattening the curve” is now clear. A flattened curve utilizes less scarce resources all at once, preventing excess demand and leading to better outcomes, in this case lower mortality rates. Even in the United States, where there is well-developed hospital and medical infrastructure, there has been a shift away from inpatient care in recent years, and numerous small hospitals have closed. The outcomes are worse when suddenly all ICU beds are full and more are needed, and there aren’t appropriate supplies and staff to operate at full capacity.
Future facilities design should anticipate the potential for future surges in patient volume, including easy conversion of medical/surgical rooms to critical care rooms and altering private rooms to semi-private rooms. Additionally, future staffing and training approaches should anticipate potential surges requiring redeployment of health professionals. Global collaboration can also help to flatten the curve. Partnering to quickly identify novel viruses and contain them helps everyone.
There are numerous reasons infectious disease outbreaks are likely to continue beyond COVID, many of which are non-medical — climate change, travel and relocation, war, and urbanization can all lead to disease eruption. The influenza in 1918 spread in part from troop movement. Wars in West Asia have led to outbreaks in cholera, measles, and polio. Where infrastructure is destroyed, and medical professionals are otherwise deployed or unable to work, it leaves an opening for pandemics. Prevention of future pandemics will require geopolitical and scientific advances that allow for “flattening the curve” by focusing on collaboration, infrastructure development, and qualified human resources.
With learnings fresh in mind, health systems have the opportunity to take advantage of current momentum to accelerate the journey to a digitally forward mindset — one that fully and meaningfully intertwines virtual care throughout the broader care delivery system.
The pandemic revealed a significant weakness in the healthcare system: Disparate supply chain, human capital management (HCM), and finance systems and processes are preventing health systems from having the information they need for real-time data-driven decision-making.
A handful of organizations across the country already have figured out what it takes to unlock the transformative value of APPs and are forging a path that others can follow, based on six key elements.
