KIPRC Conversations: Andrew Farry

Andrew Farrey is a syndromic surveillance epidemiologist at the Kentucky Injury Prevention and Research Center. He primarily works with syndromic surveillance data, monitoring nonfatal drug overdose and self-harm trends in Kentucky residents. In addition to his duties as a syndromic surveillance epidemiologist, Andrew also works with Kentucky poison control center call data and Kentucky Emergency Medical Services data.

This story originally appeared in the OD2A Happenings Newsletter. Check out the full newsletter here.

Q: Where did you grow up and what college did you attend?

I was born in and grew up in Atlanta, GA. I attended Oglethorpe University for my undergrad (and played baseball for 4 years), then went to Georgia State University for my master of public health degree.

Q: What drew you to public health?

I took an infectious disease epidemiology class as an undergrad that exposed me to introductory-level public health concepts, and I continued studying on my own until I applied to my MPH program. I found over time the more I learned about public health, the more rewarding and fulfilling I found it.

Q: What is your current role at KIPRC?

I am our syndromic surveillance/CDC DOSE (Drug Overdose Surveillance & Epidemiology) epidemiologist. My primary responsibilities are monitoring nonfatal drug overdoses using syndromic surveillance data, expanding our statewide drug overdose surveillance, and monitoring Kentucky’s syndromic surveillance data quality, but I also work with our EMS and poison control data.

Q: I know that your work (partially) involves creating an alert system to show when there are overdose spikes in areas. How were you able to create that? How did that idea come about?

One of the OD2A deliverables I’m responsible for is expanding and improving Kentucky’s drug overdose surveillance, so the idea stemmed from meeting that deliverable in a way that would be practical and as helpful/ useful as possible for local stakeholders.

Many of our counties report one overdose or less per day on average in syndromic surveillance data, so county alerts and hospital alerts derived from traditional anomaly detection methods were leading to near daily alerts that weren’t actionable or practical (many counties would trigger an alert if they reached three overdoses in a day, despite that not being particularly unusual).

This led me to SaTScan, which factors in a geospatial component alongside the temporal (time series) component, meaning SaTScan detects clusters of events across time and space/location, as opposed to temporal anomalies alone. This makes it less sensitive to the common pitfalls associated with anomaly detection in low-count time series, and it can also identify overdose spikes presenting across multiple counties or hospitals.

Similar systems have been in place in New York City and Baltimore for years.

Q: I believe you also helped the state in a similar way in eastern Kentucky after the flooding in mid-2022. Can you explain that work?

I’ve assisted KDPH with injury surveillance of two natural disasters; the tornadoes in western Kentucky in December of 2021, and the EKY flooding in July/August of 2022. Syndromic surveillance was the only timely/near real-time public health data source available in both cases, so I developed ESSENCE queries to detect tornado-related injuries and flooding-related injuries, and wrote up reports for distribution to state-level stakeholders that I sent out for several weeks following each disaster.

Q: What’s the most interesting part of your role?

I enjoy syndromic surveillance query development, using syndromic surveillance data to create reports that can help bring the data to a wider audience, and monitoring/analyzing the data quality of Kentucky’s syndromic surveillance feed for data quality issues. Data quality Issues typically present as an incorrect or invalid text string in a specific component of the HL7 messages a particular facility is submitting, but they can also result from use of an invalid or outdated code set, or an error in how a specific facility was onboarded.

Q: Anything else you’d like to add?

Syndromic surveillance can be used to detect patient encounters for just about anything that might be of interest. For example, I have a worker’s compensation-related injury query we aren’t using, and there are existing syndrome definitions for motor vehicle collisions, firearm injuries, falls, and much more. If you have any interest in using syndromic surveillance data for more rapid/timely surveillance of your injury area, I’m more than happy to work up a syndrome definition (if one doesn’t already exist) and generate a report for you.