Artificial Intelligence for Public Health Surveillance

The Challenge

Even as agencies focus more resources on preparation for and management of future outbreaks of previously unknown viruses, they often lack the advanced analytical tools needed to achieve the more integrated, more agile public health surveillance that stakeholders now expect.

For example, mismatched data standards make it difficult for government at the federal, state, and local levels, as well as hospitals and labs across the private sector, to gather and exchange critical information about disease patterns, emerging health incidents, and mortality rates. As the need for flexible disease monitoring and control strategies becomes more urgent, many agencies struggle to access decision-making insights to use in adapting their existing surveillance approaches as situations change.

Harnessing the right data in the right way to identify and contain fast-moving risks to public health is difficult. Even more challenging is using that data to analyze disease trends as soon as they emerge and successfully predicting the outcomes of an array of public health measures agencies need to rapidly implement. In the interrelated pressures it continually created, the COVID-19 pandemic brought this challenge into especially sharp focus.

The Approach

As the pandemic worsened, health organizations across the United States brought an intense commitment to gaining a better understanding of the virus. In a matter of months, decision makers were flooded with data, statistics, models, and forecasts but almost no actionable information. The vast majority of COVID-19 forecasting models projected cases and deaths down to the state level, but they didn’t prioritize and reveal useful insights at the local level—where counties, towns, and even schools needed to take action that reflected unique public health factors.

To address these forecasting limitations, the Booz Allen team proposed a multi-method modeling approach to the community-level spread of COVID-19—the COVID-19 Safe Return Simulator. The goal was to integrate different methods of computational modeling to overcome the limitations of previous approaches and extract the most relevant data from each. Our model would integrate multiple layers of data including population demographics, observed cases of illness and death, and hospital demands at the local county level within different states to make location-specific predictions about COVID-19 cases and deaths. The model relied on a virtual laboratory to:

  • Identify “hot spots” within areas that can turn into potential infection hubs
  • Examine population-specific characteristics, like gender or age, that can result in disproportionate distribution of deaths and disease
  • Prioritize counties based on their perceived disease risks considering multiple decision criteria
  • Evaluate the effectiveness of public mitigation options, like social distancing and widespread testing, aimed at reducing the likelihood of disease transmission within different communities

The Solution

The COVID-19 Safe Return Simulator combined proven epidemiological modeling methodologies with some of today’s most sophisticated forecasting techniques to predict when communities were projected to be low-risk at a county-by-county level. To address unique agency needs, the model can also be enhanced with real-time data such as population movements between counties and states, social media data reflecting updated public sentiment about health measures, hospitalization information, testing and contact tracing capacity, and personal protective equipment availability.

By powerfully clarifying the full picture of the pandemic’s effects, this project marked a significant step forward in the design, testing, and improvement of sophisticated, AI-enabled strategies for public health surveillance overall.

Extending the focus beyond the initial COVID-19 pandemic, federal health agencies can use this kind of advanced analytical approach for a range of public health applications. Examples include:

  • Tracking and assessing outbreaks of COVID-19 variants and of other viruses such as Ebola, Zika, and monkeypox within neighborhoods and throughout the nation as a whole
  • Analyzing opioid abuse in defined locales using overdose data from emergency rooms and modeling the effectiveness of different mitigation strategies
  • Mapping levels of access to healthcare in multiple communities to better understand and address health inequities
  • Mining huge data sets to identify ways to manage Americans’ risks for chronic conditions such as cancer and diabetes and to evaluate the most effective treatments

In each case, virtual laboratories that incorporate AI for accelerated data collection and analysis allow decision makers to use any number of variables to quickly establish many different models. These “digital experiments” move public health surveillance well beyond the realm of cohort studies and other traditional methods to redefine the speed, accuracy, and life-saving potential of public health initiatives, now and in the future. Learn more about how AI is transforming the federal health environment.

Meet the Expert