Though they may seem odd associates, the paths of medicine and geospatial technology have long converged. One of the most popular examples used to explain the field of GIS, in fact, is one in which geospatial analysis helps to solve a medical mystery.
In the mid-1800s, English physician John Snow was certain that contaminated water systems were responsible for the spread of cholera. Unfortunately, his colleagues and public health officials disagreed. An 1854 London outbreak of the disease provided Snow the opportunity to use real data to illustrate his theory. The map he produced combined data about cholera cases and water pumps, revealing the pattern between clusters of the disease and water sources.
Today, the two disciplines continue to interact in novel and exciting ways. The arrival of machine learning has expanded the possibilities–AI is able to process far more data far more rapidly than any human. Geo AI is enriching the practice of medicine by providing researchers and practitioners with increasingly granular health intelligence upon which to make decisions. From predicting outbreaks of infectious disease to predicting the likelihood of an asthma attack, machine learning on geospatial data is reshaping the field of medicine. The transformation promises advances in the practice of both public and private health.
Geospatial Artificial Intelligence (or Geo AI) combines the practices of Geographic Information Systems (GIS) and Artificial Intelligence (AI). If you’re new to the field like I am, your idea of AI may run along the lines of R2-D2. But, in this case, it refers to computer systems that are able to perform tasks that require human intelligence. Think less robots and more how Netflix knows whether or not you’d want to watch the film A.I. The primary tools of AI include data mining and machine learning. GIS can also be a slippery term but can be thought of as a framework for deriving knowledge from geospatial (mappable) data. When Dr. John Snow layered information about cholera cases, local business practices, and water pump locations on his map, he created a geographic information system. Today, both disciplines benefit from the glut of data available for analysis.
What does Geo AI offer the medical field? The places where we live, work, and recreate play important roles in our health. Dr. Snow’s map may have convinced public officials of the link between location and disease, but they were playing catch up. Locals didn’t need an experiment to figure out that the best way to stay healthy during the outbreak was to get out of town.
AI’s ability to parse through and detect patterns in large quantities of data can help confirm (or reject) our assumptions about healthcare. The health app on your smartphone probably knows better than you if you’re getting enough sleep, for example. Exploiting AI to produce “insights into healthcare and medicine” is known as health intelligence. Incorporating GIS practices and data into the development of health intelligence applications allows medical professionals to account for location-specific information such as environmental hazards, access to medical facilities, and demographics in their analysis.
Geo AI is having a significant impact on the fields of environmental health and epidemiology. Urban areas, in particular, are loci for both data and technology. Urban planners and shapers of healthcare policy can use remote observation to identify infrastructure that harms or promotes public health. They can also aggregate data to identify connections between social determinants and health. In many cases, mapping health outcomes in adjoining neighborhoods reveals striking disparities. If you know how to code in Python, you might want to check out this tutorial on mapping disparities in life expectancy. Our Data Analytics team used U.S. tract data to conduct the analysis. Using machine learning to analyze such data could help bring clarity to the causes of persistent inequalities in both urban and rural areas.
Furthermore, cities are increasingly filled with sensors and devices monitoring and collecting data about the environment. Air, soil, and water monitoring systems in addition to a plethora of Nest thermostats and Ring doorbells are continuously keeping track of all sorts of data. These devices, part of the Internet of Things (IoT), are introducing new ways to improve public health. One prominent area of study is the environmental and health factors of air pollution. Researchers are developing models that can identify pollution hotspots and enumerate mortality risk in understudied populations. Next up may be “smart cities” that use machine learning and geospatial technology to achieve environmental and health goals. The city of San Jose is already working with Intel to help achieve its Green Vision Goals.
One area of public health in which Geo AI is producing a notable impact is epidemiology. Artificial Intelligence in Medical Epidemiology (AIME) can exploit various datasets to produce actionable knowledge for medical professionals.
Epidemiologists in Latin America are evaluating risk factors to develop algorithms that can predict dengue, Zika, and other mosquito-borne diseases. Even geo-tagged posts on social media provide rich data for analysis. Canada’s Global Public Health Intelligence Network (GPHIN) monitors over 30,000 media sources worldwide. It then uses an algorithm to predict emerging health crises. In 2004, it was able to predict the SARS outbreak well in advance of the World Health Organization.
A number of technologists are using Tweets to predict and track influenza and other flu-like diseases in real-time. Research is showing that such algorithms can supplement traditional health surveillance apparatuses such as the Center for Disease Control’s Influenza-like Illness Surveillance Network, providing more accurate forecasts of flu outbreaks.
Remote sensing provides some of the most beneficial data for AIME applications. Satellite and drone imagery can reveal impactful environmental features–identifying areas of stagnant water is important to the treatment of guinea worm disease, for example. They can also give researchers and care-givers the tools to better understand traditionally “invisible” communities. These include nomadic groups, remote populations, and those living in conflict zones. Epidemiologist Dr. Victoria M. Gammino used Geo AI to support the Global Polio Eradication Initiative (GPEI). Using satellite imagery to detect potential settlements and navigation patterns, GPEI was able to better execute supplemental immunization activities in the Democratic Republic of Congo. Geospatial data and technology developed by the organization were later adapted to help fight Ebola in Nigeria. Stay tuned for an in-depth conversation between Azavea and Victoria Gammino about machine learning’s impacts on medicine.
At Azavea, we believe strongly in the potential of machine learning and satellite imagery to impact society in positive ways. Not only have we developed a suite of analytic tools, we can also direct you to the best imagery for your needs. Still, the camera’s eye is not actually all-seeing. Those analyzing remote imagery face a number of constraints. This includes a dearth of seasonally accurate and high-resolution imagery, and cloud cover remains problematic for many current models. Nor can satellites distinguish between populated and abandoned structures. It remains crucial to verify assumptions with human intelligence gathered at the sites of interest.
Geo AI is also transforming medicine at the patient level. With most practitioners in developed nations having adopted electronic health records (EHRs) an abundance of data exists about individuals in these populations. Much of it is geo-referenced and can be used to tailor interventions. Geomedicine incorporates a patient’s “place history” in their care. Where someone lives or works can reveal valuable medical information such as the types of environmental hazards they face, nearby healthcare facilities, and access to green space or grocery stores. Clinicians at the University of Pittsburgh Medical Center have developed an algorithm that identifies patients at risk of being readmitted. Individual data can also be aggregated to identify the risk of negative health outcomes at the neighborhood level.
In much the same way it is rebuilding urban infrastructure, the IoT is rebuilding personal “infrastructure.” This includes the mobile biosensors–fitness trackers and smart versions of everything from phones to clothing–that many people now wear. People also share information about their physical and mental health on social media. Individuals now have access to more and better data about how their bodies are interacting with their environments.
Machine learning is being used to create mobile health (mHealth) devices and applications. Currently, several smartphone applications allow those with asthma and allergies to map their symptoms. These maps are next integrated with real-time data about pollen and air quality and can alert users when saved locations are experiencing high-risk risk conditions.
Other applications hope to use machine learning and geospatial data to fight the opioid epidemic in the United States. One mHealth system aims to help those recovering from addiction not only by connecting them to treatment teams but also by predicting when the user is at high risk of relapsing. An algorithm is applied to data such as texts, sleep history, location, email activity, and the type of language the user is employing. If it identifies patterns consistent with the likelihood of a relapse, it alerts the patient’s care team. Another application uses sonar to monitor the user’s breathing patterns and can alert emergency services when it believes the user is overdosing.
In an age of increasing calls for corporate responsibility, social media platforms are using similar algorithms to identify health crises amongst their users. After a spate of public suicides on its service, Facebook developed a so-called “suicide algorithm.” Accounts flagged by the algorithm are then reviewed by human staffers who have the final say in whether to contact authorities.
The vast power of Geo AI brings with it vast opportunities for exploitation and neglect. What are some current ethical concerns?
Data siloing and resource inequity– Much machine learning and geospatial work depend on open datasets. Governments have commonly provided large sources of free data for research. The U.S. government’s publically available data includes housing statistics, a food environment atlas, patient surveys, and childhood mortality rates. The datasets can be accessed at data.gov and healthdata.gov. Not all nations, however, have the same technical capabilities or infrastructure. This disparity is often reinforced through the brain drain–when experts must leave the Global South in order to seek professional opportunities. Such trends are further exacerbated by the increasing value of data as a market commodity.
Privacy– Anonymizing patient data is key to its use in aggregated studies. The IoT, social media algorithms, and the use of remote sensing also raise serious questions of privacy and consent.
Diversity and equity– Neither the technology nor the medical industry are known for their track records on diversity in race, gender, or disability. The connection between bias in engineers and bias in models and algorithms is well-documented. Likewise, disparities in health outcomes between men and women/non-binary persons, between whites and people of color, between cis and trans persons are coming under increasing scrutiny. A similar lack of diversity exists in the datasets used to train machine learning models. Today’s algorithms for healthcare almost certainly do not serve some populations as well as others.
Although they followed Dr. Snow’s advice and closed the pump at the center of London’s 1854 cholera outbreak, English officials later rejected his research fearful of its impact on the public. Maybe if he’d had access to greater data and more powerful technology, he might not have had to wait 155 years to be vindicated. With GeoAI, today’s medical researchers, practitioners, and technologists have the ability to do more than Snow could have ever imagined. The eruptions of disease that plagued Snow’s London can now be predicted in advance, allowing public officials to provide support prior to disaster. In fact, cutting edge GeoAI research augurs a future in which such events could be responded to with individual-level interventions. Whether technology and medical professionals can deliver on that promise in meaningful, ethical, and just ways remains to be seen.