UCLA, Finland, and Maybe Everywhere: The Future of AI and Analytics in Public Health | Healthcare Innovation

As Senior Contributing Editor David Raths noted in his June 4 report, “With more than 10 million residents, Los Angeles County has a larger population than 41 of the states. A UCLA team has developed a predictive model that pinpoints which populations in which neighborhoods of L.A. County are most at risk from COVID-19 and which should be prioritized for vaccines. Their research, published in the International Journal of Environmental Health, describes the model that maps the county neighborhood-by-neighborhood, based on four indicators known to increase an individual’s vulnerability to COVID-19 infection: preexisting medical conditions, barriers to accessing health care, built-environment characteristics and socioeconomic challenges that create vulnerabilities.”

In his news article, Raths quoted Vickie Mays, Ph.D., UCLA Fielding School of Public Health professor of health policy and management and professor of psychology in the UCLA College of Letters and Sciences, who was quoted in a news release explaining about the research that she and her colleagues have been involved in. “The model we have includes specific resource vulnerabilities that can guide public health officials and local leaders across the nation to harness already available local data to determine which groups in which neighborhoods are most vulnerable and how to prevent new infections, Dr. Mays said in a press release explaining the research.

“When the pandemic hit, we were slowed down by a lack of science and a lack of understanding of the ways in which health disparities in the lives of some of our most vulnerable populations made their risk of COVID-19 infection even greater,” Mays explained. “We thought elderly and people in nursing homes were the most vulnerable, yet we found that lacking a number of social resources contributes to a greater likelihood of getting infected and in some instances leading to death as well.”

What Mays and her colleagues found was that “The research data demonstrate that neighborhoods characterized by significant clustering of racial and ethnic minorities, low-income households, and unmet social needs are still most vulnerable to COVID-19 infection, specifically areas in and around South Los Angeles and the eastern portion of the San Fernando Valley,” as Raths wrote. “Communities along the coast and in the northwestern part of the county, which have more white and higher-income residents, were found to be the least vulnerable.”

As Raths wrote, “Mays, who also directs the National Institutes of Health–funded UCLA BRITE Center for Science, Research and Policy, worked with urban planner Paul Ong, Ph.D., director of the UCLA Center for Neighborhood Knowledge, to develop the indicators model, along with study co-authors Chhandara Pech and Nataly Rios Gutierrez. The maps were created by Abigail Fitzgibbon. Using data from the Fielding School’s UCLA Center for Health Policy Research’s California Health Interview Survey, the U.S. Census Bureau’s American Community Survey, and the California Department of Parks and Recreation, the researchers were able to determine which racial and ethnic groups in Los Angeles County were the most vulnerable, based on their geographical residence.”

And that is research with very exciting potential indeed.

Meanwhile, both the pitfalls and the promise of the leveraging of AI for public health purposes were brought out in an article published in the journal BMC Public Health on January 6. Entitled “’AI’s gonna have an impact on everything in society, so it has to have an impact on public health’: a fundamental qualitative descriptive study of the implications of artificial intelligence for public health,” it was authored by a team of researchers—Jason D. Morgenstern, Laura C. Rosella, Mark J. Daley, Vivek Goel, Holger J. Schnemann and Thomas Piggott.

The researchers noted in their article that, “Beyond the most visible applications of AI among the tech giants of Silicon Valley, it has started to infiltrate healthcare and public health. In healthcare, AI applications have been reported that match or outperform physicians in various domains including radiology, dermatology, and pathology. Additionally, some hospitals have begun to integrate AI into the clinical workflow, as in the case of New York University Langone Health’s predictive analytics unit [9]. While considerable attention has been paid to AI in healthcare, there has been less attention on its impact in public health [10]. Despite this, public health researchers and practitioners have begun applying AI to diverse projects such as scanning the internet for nascent outbreaks, predicting suicide using electronic health records, and identifying risk factors.”

In that regard, they wrote, “As such, there has been growing optimism regarding the potential for AI to improve public health [14]; however, few AI systems have actually been implemented within public health organizations. Moving forward, there are serious concerns regarding AI’s impacts on privacy, interpretability, and potential for bias [15, 16]. Also, there has been criticism that AI as applied in precision public health is merely scaling up the precision medicine approach [17]. The potential to move beyond biomedical applications of AI to models incorporating rich characterizations of the social determinants of health has been identified as a promising and largely unexplored frontier. A clearer understanding of AI’s relevance to public health, which is presently absent from the literature, is needed to navigate the opportunities and risks.”

The researchers developed a complex process that incorporated interview with 15 experts in public health and AI from across North America and Asia, from June 2018 through July 2019. They conducted in-depth semi-structured interviews, iteratively coded the resulting transcripts, and analyzed the results thematically. Their conclusion? “Experts are cautiously optimistic about AI’s impacts on public health practice, particularly for improving disease surveillance. However, they perceived substantial barriers, such as a lack of available expertise, and risks, including inadequate regulation. Therefore, investment and research into AI for public health practice would likely be beneficial. However, increased access to high-quality data, research and education regarding the limitations of AI, and development of rigorous regulation are necessary to realize these benefits.”

Very specifically, the researchers came to the conclusion that “teaching high-level machine learning concepts to public health practitioners could be helpful in catalyzing AI initiatives,” as has already begun via an initiative in Canada that is scheduled to start this summer. The quality and standardization of the data out there will also be issues, they concluded in their study.

What does all of this say? I think that the UCLA case study is emblematic of precisely the kinds of use cases that will be emerging in the coming years, as healthcare leaders start to plumb the vast potential of AI and other forms of predictive analytics to serve the purposes of public health here in the U.S. What’s more, though many of the western European healthcare systems are more advanced in beginning to develop nationwide population health management strategies—one thinks especially of the innovations taking place in the Nordic countries—what seems clear is that the development of AI in healthcare is, overall, far more advanced in the United States.

Does this mean that we might be able to take advantage of the more advanced state of data analytics overall in this country, in order to move forward faster on linking the most forms of analytics to public health work? Yes, that’s exactly what it means.

Of course, we have a far, far, far more complex healthcare system than the Nordic countries do, and far, far larger. Just consider the population differentials: Sweden, by far the largest of the Nordic nations by population has just 10.3 million inhabitants, while Denmark, Finland, Norway, and Iceland have 5.8 million, 5.5 million, 5.3 million, and 368,000 people each. Altogether, the Nordic nations have just over 27 million residents—equivalent to Texas, our second-most-populous state, with 28.9 million residents—yet Texas is just one U.S. state.

I still remember vividly my writing-research trip to Finland in the spring of 2000. I met with some of the leading healthcare quality leaders in the country. Of course, one must keep in mind Finland’s small size in terms of population; with 5.5 million, it hovers in size between Phoenix (5 million metro area) and Atlanta (6 million metro area)—our tenth- and ninth-largest metro areas, respectively. And, as one of the key leaders in the clinical patient safety and quality outcomes movement—an engaging physician who was involved with her colleagues at that time in developing a key set of nationwide outcomes measures, “We really can bring together all the key people into one room here in Helsinki, and work out any issue.”

Yet for all the really excellent things the Finns have been doing around analytics in healthcare, we have certain advantages here in the U.S. that they don’t have.

But first, I should detour and describe a bit of what the Finns have done. In November 2016, I traveled to Barcelona and participated in the HIMSS Europe conference held in that city. While at the conference, I met Jana Sinipuro, the director of Sitra, Finland’s unique healthcare data analytics agency. While we lack a precise equivalent here to Sitra, what Sitra does falls somewhere between what our Agency for Healthcare Research and Quality (AHRQ), our Health Resources & Services Administration (HRSA), and our Center for Medicare & Medicaid Innovation (CMMI) do.

In any case, in November 2016, when I spoke with Jana Sinipuro, Sitra’s Isaacus Digital Health Hub Project was just getting going. One outgrowth of Sitra’s work, as described in a whitepaper entitled “A Finnish Model For The Secure and Effective Use of Data,” has been “the EU-funded MIDAS (Meaningful Integration of Data, Analytics and Services) project, the Finnish partners of which are the VTT Technical Research Centre of Finland Ltd, the National Institute for Health and Welfare and the University of Oulu. The project,” the whitepaper notes, “develops a harmonised solution for using health and well-being data to support leadership and decision-making. The solution includes components for data collation, data analytics and visualisation. With regard to the harmonisation of data sources, the project needed a shared metadata model for data material. The evaluation of different options indicated that the metadata model created in the Isaacus project would be highly suitable for the needs of the MIDAS project.”

Further, the whitepapers states, “In the Isaacus project, the National Institute for Health and Welfare has created not only the model itself but also online services (a material catalogue and material editor) for describing material and sharing metadata. These software solutions are available as open source code, which is a significant factor in choosing a metadata model to be used in the project. The Spanish research and technology organisation Vicomtech, responsible for the harmonisation of data material, has now arranged the above-mentioned services for use in the MIDAS project.”

Obviously, Nordic healthcare systems such as Finland’s are very different in many ways from ours. For one thing, the government runs everything. But we do have public-private partnerships in this country that bear some equivalence to what the Nordic nations do purely governmentally.

In any case, all that having been said, there is vast potential in leveraging data analytics of all kinds, and most excitingly, AI and machine learning, for public health-related purposes here in the U.S. We need to encourage every possible blossoming of such potential here; the future awaits.