We briefed the Secretary of Homeland Security, Janet Napolitano, yesterday. She asked a number of very cogent questions about how we could detect threats to health early and bring that information to someone in her position as quickly and clearly as possible. In a very polite way, she was challenging us to demonstrate our capabilities. I think we did well in our response, and it got me thinking that I needed to express some thoughts in this venue to lay out our approach in a longer form.
All incidents relating to health threats begin somewhere on the planet, but may not stay there. Both naturally-occurring zoonotic agents and human-engineered pathogens may start in one place and migrate with human and other kinds of traffic. I separate them into two major categories:
- “Static” data that does not change rapidly or is not augmented with new data very often.
- These data include streets, infrastructure, fixed resources such as hospitals and fire stations.
- These data form a set of foundation layers to allow us to build upon them with the other type of data.
- “Dynamic” data that does change rapidly or grows as we move through time.
- These data can be meteorological, biological, clinical (human and non-human), business data, and social media.
- Some have specific location components (like our EMS data with lat and long elements).
- Some have far less granular location data because of the nature of where they are created. Hospital locations are all known, but the data relating to patients may refer to billing addresses rather than residences. Those can be useful, but less so than EMS data that most often relates to the point of illness or injury.
- Some are specific as to character and location, but are only useful as background or for situational awareness, such as the SMARTT data on hospital resources.
- And some have no geographic specificity such as most online social media.
Assuming that we have the above types of data, the focus turns to the middle layer of aggregation, integration, and analysis to identify threats to health as early as possible and with the greatest specificity as possible. (The Secretary was quite pointed in her need to understand a complex situation quickly—asking us to show more detail about how we detect and characterize incidents.) We have made great strides in the past year to develop methods to incorporate new data sets, develop analytical tools to identify anomalies and characterize them relating to potential health threats.
That brings us to the engagement layer where everything comes together. I see several ways to express information online:
- Static reports—providing a regular update on the status of health threats for professionals. The report shows the activity relating to health threats as alert levels from normal to highest alert, based on what our analytics models determine. Ideally, it’s a single page that features trends, maps, and other indicators. The goal is at-a-glance comprehension of the status for the situation leading up to that day.
- We will tailor these by role (local public health official, emergency department manager, school nurse, physicians in private practice, fire chiefs, emergency management, etc.) by jurisdiction or service area (catchment area, school district, municipality, county, state, market area, etc.) and type of organizational setting (public, private, governmental, NGO, for-profit, not-for-profit).
- Part of the value proposition is to cast the information in terms that relate to the responsibility of the person receiving them. For example, public officials have general responsibility for their jurisdictions to safeguard their fellow citizens. Private firms have responsibility to their companies and shareholders to manage risks to their businesses—including risks of tainted or contaminated products that pose threats to health. We believe that both ends of this continuum are actually mutually reinforcing—if we can get everyone’s data. With comprehensive data we can provide the best available information to guide understanding and inform responses.
- Dynamic reports—providing the above but with extra tools to allow engagement with the alerts and other information the system develops. These are essentially dashboards. The capabilities needed include choosing the data sources, health issues, types of vectors of the agent, type of place where the incident began, disposition of the case, etc. We also want to support changing the time frame of the report from the present or other designated date to look at trends over a week, a month, a quarter, or a year. We want to show alerts over time and place with the best method to speed understanding. These would involve maps, graphs, or line listings. We see these leveraging the output of our system in the form of alerts and characterizing information as to time, place, issue, severity, and the like. Finally, we want these reports to serve as virtual rooms where people can discuss the situation and achieve shared understanding of what is happening.
- Expert level—essentially access to the tools for developing and using our data. While all levels will have roles-based security, the access to this system and its attendant data will be even higher because it includes access to the raw data. We will reserve it to those with legal rights to view such data and with the permission of the data provider. Yet, it will need all the functions listed above and many more to support ongoing investigations and the development of analytics models for alert detection and characterization.
The hardest question the Secretary posed was "Can you tell me what's going to happen next?" More on that later.