Foodborne Botulism in the Republic of Georgia: Implications for Preparedness Planning

In this issue of Clinical Infectious Diseases, Varma et al. [1] present data from nearly a quarter century of foodborne botulism cases in the Republic of Georgia [1]. The results may provide Georgia with a cost-saving alternative to the management of botulism cases and may offer guidance for those who are preparing plans for mass numbers of casualties due to botulinum intoxication. The authors reviewed the individual medical records of 1700 patients with bot-ulism during 1980–2002 to determine the initial clinical presentation and the eventual outcome of the disease. They then used classification and regression tree analysis (CART) to determine predictors for survival and death. The incubation period , symptoms at presentation, clinical management, and mortality rate were all similar to what might be expected of food-borne botulism cases in the United States. Trivalent botulinum antitoxin was given to 88% of patients, and 13% required mechanical ventilation. Fifty-four patients (8%) died during the course of treatment for botulism, which is similar to the 7% case fatality rate reported in the United States from 1980 through 1996 [2]. These similarities are reassuring, when considering application of these findings elsewhere. However, as noted by Varma et al. [1], the clinical prediction rules developed from these data should ideally be validated in other populations. CART is a multivariable method that has been used to develop clinical decision rules to separate patients according to risk groups. For example, CART analysis has been used to predict stroke outcomes and the likelihood of acute myocardial infarc-tion and for triage of pediatric trauma victims , but it has not been used commonly in infectious diseases or public health settings [3–5]. CART analysis yields a decision tree that is based on clinically recognizable parameters, which is an advantage over some logistic regression models [5]. Also, the data are not assumed to be normally distributed, and, thus, CART can use data from skewed distributions without transformations. In this study, the large number of clinical parameters associated with death dictated the need for a multivariable analysis, and CART provided a useful solution. Varma et al. [1] determined that characteristics at the time of initial presentation were 100% predictive of survival for patients without shortness of breath, vomiting , or facial weakness. This age-adjusted analysis included only those self-reported signs and physical examination findings recorded at admission to the hospital. Missing from this CART analysis were aspects of clinical management for …