Modeling uncertainty of estimated illnesses attributed to non-O157:H7 Shiga toxin-producing escherichia coli and its impact on illness cost

Abstract

Because of numerous reported foodborne illness cases due to non-O157:H7 Shiga toxin-producing Escherichia coli (STEC) bacteria in the United States and elsewhere, interest in requiring better control of these pathogens in the food supply has increased. Successfully putting forth regulations depends upon cost-benefit analyses. Policy decisions often depend upon an evaluation of the uncertainty of the estimates used in such an analysis. This article presents an approach for estimating the uncertainties of estimated expected cost per illness and total annual costs of non-O157 STEC-related illnesses due to uncertainties associated with (i) recent FoodNet data and (ii) methodology proposed by Scallan et al. in 2011. The FoodNet data categorize illnesses regarding hospitalization and death. We obtained the illness-category costs from the foodborne illness cost calculator of the U.S. Department of Agriculture, Economic Research Service. Our approach for estimating attendant uncertainties differs from that of Scallan et al. because we used a classical bootstrap procedure for estimating uncertainty of an estimated parameter value (e.g., mean value), reflecting the design of the FoodNet database, whereas the other approach results in an uncertainty distribution that includes an extraneous contribution due to the underlying variability of the distribution of illnesses among different sites. For data covering 2005 through 2010, we estimate that the average cost per illness was about $450, with a 98% credible interval of $230 to $1,000. This estimate and range are based on estimations of about one death and 100 hospitalizations per 34,000 illnesses. Our estimate of the total annual cost is about $51 million, with a 98% credible interval of $19 million to $122 million. The uncertainty distribution for total annual cost is approximated well by a lognormal distribution, with mean and standard deviations for the log-transformed costs of 10.765 and 0.390, respectively.