Methodological challenges to multivariate syndromic surveillance: a case study using Swiss animal health data

Area of Science:

• Veterinary Epidemiology
• Biostatistics
• Animal Health Surveillance

Background:

• Multivariate surveillance systems offer improved disease detection probability over univariate systems.
• Univariate aberration detection algorithms are common in syndromic surveillance (SyS) due to ease of use, despite limitations.
• Stochastic modeling provides a flexible multivariate approach for animal health surveillance, accommodating historical data and complex patterns.

Purpose of the Study:

• To apply a stochastic, two-component model to multivariate animal health data.
• To evaluate the performance of this model against existing methods for disease event detection and prediction.
• To assess the utility of joint modeling for forecasting in animal health surveillance.

Main Methods:

• Application of Held and colleagues' two-component stochastic model to two Swiss animal health datasets.
• Comparison with an improved Farrington algorithm for multivariate time series analysis.
• Analysis of time series data including laboratory test requests and cattle abortions.

Main Results:

• The two-component model demonstrated a low false alarm rate and satisfactory one-step-ahead predictions, suitable for outbreak prediction.
• A two-day lag effect was identified between cattle abortions and test requests for bovine viral diarrhea.
• Joint modeling of multivariate time series showed superior forecasting ability compared to univariate modeling.

Conclusions:

• Stochastic modeling enhances animal syndromic surveillance by allowing for time series specific parameters and covariates.
• Methodological challenges remain in multivariate surveillance, particularly in determining alert thresholds with sparse outbreak data.
• The study highlights the potential of advanced statistical methods for more realistic and effective animal disease surveillance.