Imagine being charged with pulling together data from disease surveillance systems, public and commercial laboratories, hospital surveys, and a health information exchange when none of those systems talk to each other electronically. That is just one of the challenges facing Jessie Tenenbaum, Ph.D., in her role as chief data officer for the North Carolina Department of Health and Human Services (DHHS).
Tenenbaum, who was a founding faculty member of the Division of Translational Biomedical Informatics within the Department of Biostatistics and Bioinformatics at Duke University, was named chief data officer last May, taking over from Aaron McKethan, Ph.D., another Duke faculty member.
During an April 20 AMIA webinar, Tenenbaum described the view from inside the state emergency operations center (EOC) that was set up after a March 10 state of emergency was declared in North Carolina. She showed slides of the building that has become mission control for the pandemic response. “This is usually where they do hurricane response,” she said. “A lot of the processes in place work great for hurricanes when you have a crisis for seven to 10 days and then you are done. This has been extremely different.”
She described some basic questions people in the EOC need answers to: How many cases will we see? When will this peak? Will we run out of ICU beds or ventilators? When should we shut down bars, restaurants and schools and businesses? How much PPE do we need and where can we get it? When can we start things back up again? What should we start back up first?
“Luckily, we have some data sources to help answer those questions,” she said. There is the North Carolina Electronic Disease Surveillance System for any reportable diseases. A system called ReadyOp is used to survey hospitals every day to ask how many beds are free, how many COVID patients they have hospitalized, and how many ventilators are in use. “That has been evolving as different questions become more relevant,” Tenenbaum said.
The state HIE, NC HealthConnex, provides clinical data for a subset of the population, and the North Carolina Disease Event Tracking and Epidemiologic Collection Tool (NC DETECT) is the statewide syndromic surveillance system. “As you might imagine, almost none of these talk to each other very well,” she said.
In a simplified walkthrough, Tenenbaum described the process: A sick person goes to see their doctor; the doctor sends a test request to either the state lab, a hospital lab, or a private lab. If the result is positive, it goes into the surveillance database. Very little information goes with that transaction — the name, the demographics, and the fact they had a positive result and the lab where it was done. Additional information about that person, symptoms or travel or hospitalizations, is all investigated and recorded by local health department staff, she explained. Some of the clinical data goes to the state HIE and the North Carolina hospital association data warehouse. “The tricky part is that we have this wall. There are no convenient linkers between these systems. If you want to link people in these clinical data systems to the surveillance data systems, you have to do record linkage based on names, date of birth and phone number and we all know that is fraught with challenges.”
In addition, with the ReadyOp hospital survey tool, which was designed for hurricanes, you have to manually export the results to Excel. “You can’t do anything in an automated fashion because a person has to go to the system and export to Excel,” Tenenbaum said. “We are working on a pilot with one health system in North Carolina to automate as much of this as possible. We had a letter from Vice President Pence and followed up with some more detail from Alex Azar, that said thou shalt collect these fields. We collect most of them already. Some of them we are missing. But from what we have been told, we are doing a good job, so we are trying to keep the impact on people having to manually enter this minimal. We may have to increase a few questions over time.”
Lab test data is one of the areas that is most manual and painful, she said. “For everything done in the state lab, we have electronic data, so that is not bad. But in terms of knowing how many tests have been done in a given day, there is a guy who sits here calling or e-mailing all the different labs doing these tests and saying ‘please send me how many you did in the last 24 hours.’ We are working on improving this process.”
She thinks of the data they have in three different buckets — cases, tests and capacity, and each is collected in its own way. Someone collects the data, they prepare a manual report, and then they e-mail that to multiple people. “In all these cases, it is done by e-mail and that is clearly not ideal,” Tenenbaum said. “We are trying to move toward automating some of these by bringing data into a central data repository in the cloud, enabling dashboards through Tableau and ESRI, and other tools so we can have pubic-facing dashboards that are drawing from the same data as what we show to internal teams and the governor and the secretary.”
Tenenbaum described a very fast-paced environment in which if a solution to solve a data bottleneck takes more than a few days, the priority might have shifted to something else. Also, in the response effort, people are taking on new roles. “The DHHS org chart is hundreds of pages long and very complicated,” she said. “There are a lot of people playing new and different roles right now. We have our deputy secretary for mental health trying to figure out procurement. It is sometimes hard to know who to loop into things because there is no rulebook for who is included.” In addition to DHHS, the Department of Public Safety the state emergency response team, and the National Guard are involved and new people are being brought in all the time.
There are basically three phases to the emergency, she said. The first was, “We need data, STAT!”
“This was a few weeks ago when we needed to decide if we were shutting down schools and businesses,” she said. “Now we are getting into the steady state where the fire alarm has subsided. We still need new data, but we are trying to automate some of these processes. A lot of these processes are in place for a hurricane that comes and goes. This is going to be here for months, so we need to automate to make them sustainable.” Another phase is modeling the resolution — how are we going to know when to ease up and how to approach that. “We are at beginning of the steady state, and getting some things in place.”
Led by her predecessor as state health CDO, Aaron McKethan, epidemiological researchers around the state have come together to work on models to predict what the case rates are going to be and how that interacts with health system capacity. They come out with reports and go back and do it all again with improved data and refined models.
“Now we are starting to look to the future,” Tenenbaum said. “The current state of everyone staying home is clearly untenable, for sanity, the economy and other reasons. But flipping a switch and saying we are opening it back up would be extremely dangerous. We don’t want to do that. So we have this model of a dimmer switch to say what can we ease back into and when? That is what we are looking at now, and it comes down to part modeling and part politics, unfortunately.”