Evaluation of sliding baseline methods for spatial estimation for cluster detection in the biosurveillance system

Area of Science:

• Public Health Surveillance
• Epidemiology
• Health Informatics

Background:

• The Centers for Disease Control and Prevention's (CDC) BioSense system analyzes electronic health data for public health monitoring.
• Detecting spatial and temporal disease clusters is vital for daily public health surveillance.
• This study aimed to identify useful anomalies at manageable alert rates within BioSense data.

Purpose of the Study:

• To evaluate spatial estimation methods for detecting disease clusters in public health surveillance data.
• To compare the effectiveness of different methods based on background cluster rates and signal sensitivity.
• To optimize anomaly detection for public health monitoring.

Main Methods:

• Utilized over three years of daily military outpatient clinic visit data for respiratory and rash syndromes.
• Applied and compared four space-time scan statistics methods using Matlab and C implementations.
• Evaluated methods using facility and residence zip codes for spatial resolution, considering day-of-week patterns.

Main Results:

• Estimation methods incorporating day-of-week patterns demonstrated superior performance in both background cluster rate and signal sensitivity.
• A 28-day baseline proved most effective for robust estimation, balancing daily fluctuations with recent trends.
• Rash syndrome counts exhibited lower background cluster rates than respiratory counts, likely due to seasonality and scale differences.

Conclusions:

• The optimal spatial estimation method depends on the specific data stream characteristics.
• For data with significant day-of-week effects, subregion averages over a 28-day baseline, stratified by weekday/weekend, yielded the best detection.
• Further improvements in anomaly detection can be achieved by adapting estimation methods to different spatial resolutions or disease syndromes.