AI can drive innovation to achieve health equity and precision population health, experts say
February 1, 2021. Artificial intelligence (AI) tools and applications can drive innovation in population health and reduce health inequities as long as we take advantage of their strengths and cultivate awareness around their limitations.
That was the central message from public health, academic, policy, and industry leaders who participated in a December 8th panel session on AI, Precision Health, and Health Equity: Opportunities and Pitfalls. The panel was the latest event in the New Frontiers in Precision Population Health & Health Equity Seminar Series, sponsored by Stanford Center for Population Health Sciences and the Department of Epidemiology & Population Health, and it was also part of the AI + Health Online Conference, hosted by Stanford Human-Centered Artificial Intelligence (HAI) and Stanford Center for Continuing Medical Education (CME). Stanford Professor of Medicine, Latha Palaniappan, MD, MS, moderated.
Key strategies for leveraging AI applications to advance population health
Chief Health Equity Officer and Deputy Chief Health Officer at IBM Watson, Irene Dankwa-Mullan, MD, MPH, outlined key strategies for leveraging AI applications to produce more innovative interventions in precision population health. She claimed researchers and health practitioners must:
1. Capture more robust, comprehensive data, inclusive of the determinants of community and population health. Dankwa-Mullan explained this data is already being collected in various sectors, but it is generally considered to be too “complex, diverse and extensive” to be handled by our existing analysis, thus requiring AI to decipher it. Our AI and machine learning tools can better identify patterns or trends while uncovering those important associations with health determinants.
2. Think about “the human aspect of AI.” “AI does not function alone, there is always a human in the loop. AI learns from data that we feed it,” Dankwa-Mullan explained. Our “humanity, empathy and compassion” are as important to AI interventions in population health as they are to patient-centered care and population health.
3. Improve our data representation and strive for true “diversity, inclusivity and comprehensiveness.” Dankwa-Mullan argued that most of our research still serves to reinforce norms of homogeneity, particularly surrounding the health of racially-marginalized populations. She encouraged increased focus on differentiation, especially in regards to population-based risks and vulnerabilities, community values and experiences.
4. Embrace the “depth and comprehensiveness of social determinants of health,” and track the ways determinants in early life (e.g. adverse childhood experiences) impact population health.
5. Practice “data empathy,” i.e. integrate empathy and compassion into our study of health disparities, as a way of building community trust and engagement.
Methods for improving AI and ML data models in the service of greater health equity
Director of Health Equity & Product Inclusion at Google, Ivor Horn, MD, MPH, delineated four key points layered upon the conceptual foundation established in Dankwa-Mullan’s remarks. For Horn, data diversity, bias, social determinants of health, and homogeneity are issues that are central to the question of exactly how we approach data in AI and machine learning (ML). She recommended that researchers and health practitioners:
1. “Think about how we look at data to actually close gaps, [including] what the principles are that we need to apply to actually… address and reduce disparities.” Horn pointed out the inherent danger, if we continue to ground our work on biased data, of creating new systems that actually worsen disparities rather than lessen them.
2. “Create mechanisms and models for transparency” so people with AI tools working on the ground can understand their essential limitations–i.e. what they can and cannot show–and perhaps address those limitations with their own additions or modifications.
3. Use a community-based participatory research model when incorporating social determinants of health into our AI models. In Horn’s extensive experience, this approach to data generation will enable us to make significant inroads in understanding the factors contributing to increased individual risk and resilience (or other positive outcomes).
4. “Approach health equity from a global perspective, with humility and a learning mindset.” When developing data models for AI and ML, researchers and health practitioners must take opportunities to learn from other countries so as to effectively move beyond our own assumptions of homogeneity and unconscious biases about people outside of the U.S..
Increasing citizen engagement with AI apps like “Waze for public health”
Ramesh Raskar, PhD, Associate Professor of Media Arts and Sciences at the MIT Media Lab and Founder of PathCheck Foundation for Covid19 Software, understands the challenges of leveraging AI to advance population health and health equity during the current coronavirus pandemic. He recognizes one of the major reasons we have not been able to “tame the current pandemic” is because of a lack of citizen engagement. This lack contributes to gaps in data that public health guidance is based upon, resulting in inefficient policy and leading to low public trust.
To increase citizen engagement and thereby subdue the pandemic, Raskar claimed we need the right AI tools, circulating critical information from people and to people at the time when it is most needed. He recommended that we build a public health system like Waze, a driving/traffic app that provides highly-personalized, high-fidelity information based on participatory data submitted by the user. In addition to AI tools, we need to solve two problems for data collection: adoption and privacy.
Connecting “Waze for public health” with the current pandemic, Raskar argues that it would only take a small percentage (~1%) of users submitting their data to centralized, privacy-protected tools–for instance, in the pandemic, that might include data on exposure, symptoms, testing, tracing, treatments, and so on–to start generating personalized information relevant for other users, including public health decision makers. “To work on the haze of the pandemic, we need ‘Waze for public health,’” he concluded, “and apps can fill the gaps in capture of data, analyzing [data] with AI, and engaging [citizens] in a personalized manner.”
Horn expressed her admiration for Raskar’s “Waze for public health model,” especially in terms of its applicability in a health equity context. She pointed out the importance of ensuring that someone who is walking or riding a bike can get directions as well, providing a useful reminder of the need to develop awareness around the strengths of our research and data models and their significant limitations.
Hope for the future
Moderator Palaniappan closed the event with a message of hope and optimism about the power and potential of AI to advance health equity and precision population health:
I continue to be humbled by all the great minds around the world who continue to work around the clock to address health inequities through AI and who are working to make health equity a reality. I am heartened and confident that, with this brain power and continued collaboration, we will be able to architect innovative solutions for the future. We have learned today how we can take advantage of our strengths and identify our weaknesses to work together for precision public health, while proactively laying the foundation to conquer any health inequities that we may face in the future.
By Katie M. Kanagawa, Ph.D., M.A., Communication & Public Relations Manager for the Center for Population Health Sciences, the Department of Epidemiology & Population Health, and the Department of Biomedical Data Science