Optimizing Influenza Sentinel Surveillance at the State Level
P.M. Polgreen, E. Chen, A.M. Segre, M. Pentella, G. Rushton
International Conference on Emerging Infectious Diseases (March 2008).

Background: Influenza-like illness (ILI) data is collected in the US via an Influenza Sentinel Provider Surveillance Network. Members represent individual healthcare providers, group practices, emergency departments, and student-health clinics. In Iowa, members are recruited by the Iowa Department of Public Health (IDPH). Because participation is voluntary, locations of sentinel providers may not reflect optimal geographic placement. Location- allocation modeling has been used to find optimal locations for banks and retail stores. The purpose of this study was to use this approach to find the optimal location for sentinel providers in Iowa.

Methods: The population in each zip code area (ZCTA) in Iowa was obtained from the 2000 US Census, the geographic location of hospitals in Iowa from the Natural Resources Geographic Information Systems Library, the location of existing sentinel members from IDPH for 2006–2007, and ZCTA to hospital distances were computed from their respective geocodes. Using a maximal coverage model (MCM), we maximized the number of persons in Iowa within 25 miles of N hospitals that were placed within a known population distribution. Starting with N=1, we found the optimal location for one sentinel provider. Holding each previously selected sentinel location, we increased N by 1 unit until we reached 148 (number of existing sites and possible new sites). This gave us the “best” possible locations for adding individual sentinels to a network of known size. Next we calculated the coverage derived from the existing 22 sentinel locations and then added additional sentinel locations using the MCM.

Results: For the first optimal zip code location, the covered population was 17%, for two it was 29%, and additional locations provided more coverage but with diminishing marginal returns. Maximal coverage (99%) was achieved with 46 locations. The existing sentinel locations (22 sites) covered 66% of the population in Iowa. Using the MCM we could have achieved the same coverage with just 17 sites. Using 22 MCM sites would have covered over 81% of the population.

Conclusions: The current system has sentinels in ZCTAs within 25 miles of 66% of the population in Iowa. Given scarce public-health resources in Iowa and other states, using a MCM can help optimally target recruitment efforts for new sentinel locations.