Mad cows and computer models: the U.S. response to BSE.

Area of Science:

• Veterinary epidemiology
• Public health policy
• Statistical modeling

Background:

• The United States tests a smaller proportion of slaughtered cattle for Bovine Spongiform Encephalopathy (BSE) compared to Europe and Japan.
• This necessitates a heavy reliance on statistical models for estimating BSE prevalence and spread within the US.
• Limited data availability poses a significant challenge to the accuracy of these models.

Purpose of the Study:

• To critically examine the statistical models used by the U.S. Department of Agriculture (USDA) for BSE surveillance.
• To assess the reliability of these models in estimating BSE prevalence and predicting its potential spread.
• To evaluate the implications of model-dependent surveillance for public health policy.

Main Methods:

• Analysis of USDA's statistical models for BSE prevalence and spread.
• Evaluation of model sensitivity to parameter variations, particularly under worst-case scenarios.
• Examination of multiple published scenarios incorporating worst-case parameter values.

Main Results:

• The USDA's prevalence model offers only a rough estimate of BSE prevalence due to insufficient data.
• Forecasts of BSE spread are potentially misleading, as they depend on arbitrary constraints on model parameters.
• In three of six tested scenarios, there is a significant (≥25%) probability of rapid BSE spread when multiple worst-case parameters are considered.

Conclusions:

• Reliance on current statistical models for BSE surveillance in the U.S. presents a substantial public health gamble.
• The inherent uncertainties and potential flaws in abstract statistical modeling are not a substitute for robust, precautionary policies.
• Strengthening public health defenses against epidemic threats like BSE requires moving beyond potentially unreliable modeling.