Evaluating and implementing temporal, spatial, and spatio-temporal methods for outbreak detection in a local...

Area of Science:

• Public Health Surveillance
• Epidemiology
• Biostatistics

Background:

• New York City's Department of Health and Mental Hygiene utilizes a syndromic surveillance system since 2001.
• The system employs daily temporal and spatial scan statistics for disease cluster detection.
• Advancements in cluster detection methods necessitate ongoing evaluation.

Purpose of the Study:

• To evaluate the performance of various temporal and spatial/spatio-temporal cluster detection methods.
• To compare algorithms based on sensitivity, specificity, positive predictive value, coherence, and timeliness.
• To assess implementation factors including programming time, run time, and ease of use.

Main Methods:

• Syndromic surveillance data from New York City was spiked with simulated case injections.
• Six temporal and four spatial/spatio-temporal detection algorithms were tested.
• Performance metrics included detection rates, positive predictive value, and operational efficiency.

Main Results:

• Among temporal methods, Holt-Winters exponential smoother showed the highest detection rate (19%) at 95% specificity.
• Spatial scan statistic performed best among spatial/spatio-temporal methods, detecting 3% of injects.
• All tested methods exhibited low positive predictive values (<7%).

Conclusions:

• Current cluster detection methods, including the spatial scan statistic, show limited effectiveness in identifying simulated disease clusters.
• The spatial scan statistic, implemented in SaTScan, remains the most practical for daily analysis due to ease of programming and implementation.
• Further research is needed to improve the sensitivity and specificity of syndromic surveillance detection algorithms.