A call for AI programs on surveillance, prevention, and resilience in pandemics from Dr. Ioannis A. Kakadiaris and Dr. Winston Liaw.
We live in unprecedented times. But as Dr. Monika Langeh, rightly asked, we need to choose where we want to be during the COVID-19 crisis. Do we want to be in the fear zone, the learning zone, or the growth zone? As evidenced by the number of AI-related initiatives related to COVID-19, computer scientists, data scientists, and medical researchers have chosen to live in the growth zone. They are focusing on: 1) being a model in their communities – strengthening their relationships, exhibiting empathy and transmitting hope); 2) serving their communities – using their skills in service; and 3) focusing on the future – finding ways to respond to the crisis, practicing creativity.
The timescale of this pandemic remains uncertain. Its fallout will be dire while the disproportionate effects of the crisis on vulnerable populations will deepen health inequalities. What is needed is AI programs that address: (i) Telehealth; (ii) Social Determinants of Health; (iii) Data Infrastructure; and (iv) New Algorithms.
Telehealth: With social distancing, practices are rapidly replacing faceto-face visits with video ones. These visits simultaneously eliminate historical barriers to care, such as lack of transportation and time, and protect those at high risk for COVID-19 complications from unnecessary exposure. As telehealth expands, AI can accelerate its evolution. For example, AI can be used to identify which individuals need to be seen in-person and which can be safely and accurately assessed online. Using chat platforms, AI can answer questions about COVID-19, direct symptomatic patients to the nearest testing sites, or schedule video visits. During these appointments, AI can detect stress and interpret emotions by analyzing the video and alert clinicians when they have missed opportunities to provide comfort or gather important historical clues. Using voice to text, AI can help clinicians with documentation, allowing them to focus on their |bedside presence.
social determinants of health: While this technology has improved access, its adoption remains uneven, with many lacking access to the internet, smartphones, insurance coverage, and clinicians. For the privileged, social isolation has meant working from home, having groceries delivered, and accessing their physicians over the internet. Others are vulnerable, laying bare entrenched divisions across race and class. In Chicago, African Americans compromise 70% of COVID-19 deaths, even though they are only 30% of the population. Similar patterns are emerging in Michigan, Louisiana, and New York. These disparities predate COVID-19, with minorities and the impoverished suffering from high rates of chronic disease, limited access to healthy food, and crowding. Many cannot afford to stop working outside the home and have to choose between exposing themselves to infection and meeting basic needs. Connecting individuals with social needs with appropriate and available community resources is a critical gap, and AI can facilitate the matching process. During pandemics, these referrals can occur alongside contact tracing or the practice of tracking down individuals who have been exposed to infected persons and isolating them before they infect others. Teams of public health professionals are being recruited and trained to do this vital work, and AI can help this army target exposed individuals, provide real-time translation, collect data about symptoms and social networks, and provide social assistance to those in need.
data infrastructure: Infrastructure should focus on data Services, regulatory services, legal, privacy, and ethical issues. The performance of state-of-the-art machine-learning models trained on data from one medical institution drops by large margins when tested on the same task on another institution’s data. This failure to generalize stems from an assumption baked into most machine-learning methods: the training data for the machine-learning model is supposed to be a representative sample of what the model will be applied to in the future. This assumption is rarely valid in practice, especially in a pandemic, where each institution alone has limited data (or view) of the pandemic. The ecosystem of health care is complex, and there are several barriers to the introduction of AI systems: closed EHR systems, lack of data-sharing arrangements, regulatory restrictions, and lack of standardization to a sufficient degree so that similar products work similarly. There is an urgent need to set up a coordinated collection of data, data-sharing agreements, and data governance agreements. Then, research on new algorithms for distributed training can be developed to take into consideration complex data sharing constraints to optimize model performance and preserve patient privacy.
new algorithms: Large portions of clinical medicine consist of decision making in light of longitudinal, heterogeneous data. Demonstrations of individual applications that match or outperform human clinicians in their decision-making accuracy are numerous but are still very narrow in their focus. Most AI tools remain “black box” entities, and it is challenging to identify their reasons for making particular decisions reliably. It is clear that AI’s have enormous potential in a pandemic. Still, significant challenges are yet to be overcome, and much fundamental research needs to be done to realize this potential. Specifically, significant gaps in (i) AIAided Human-Computer Interaction; (ii) Explainability; (iii) Empathy; (iv) Cultural AI; and (v) Distributed Learning. We anticipate that AI stands poised to deliver enormous improvements in improved patient outcomes and to reduce the costs.
If we are to address this pandemic and prevent future ones, additional bridges between AI and health care will need to be built, across communities, health systems, academic medical centers, public health departments, and federal agencies. As Epictetus said, “It is not what happens to you, but how you react to it that matters.” To that end, AI presents a unique opportunity to tackle issues that have confounded public health experts for years. Harnessing its power will allow us to implement appropriate telehealth measures for large healthcare agencies, determine patients most at risk for poor outcomes using social determinants of health, improve the data infrastructure to allow for robust analysis, and develop new decision-assisting tools for clinicians and healthcare providers. We are calling for funding bodies to collaborate and support research funding focused on AI programs on surveillance, prevention, and resilience in pandemics so that we may tackle these challenges head-on.
We have been presented a unique opportunity to grow in our humanity but also our technological prowess and efficiency. It is our duty to our communities and the world at large to seize this opportunity.
Dr. Ioannis Kakadiaris is a Hugh Roy and Lillie Cranz Cullen University Professor of Computer Science, Electrical & Computer Engineering, and Biomedical Engineering at the University of Houston. Dr. Winston Liaw is a family physician, Chairman of the Department of Health Systems and Popul