A Triage Algorithm for Inhalational Anthrax

This issue contains a thoughtful effort to make the most of limited data on a subject of intense public concern: the diagnosis of inhalational anthrax. Soon after September 11, 2001, the United States had its first experience with bioterrorism. The attack was limited but deeply troubling because the perpetrators used the U.S. Postal Service, an accurate and reasonably quick delivery system, to disseminate anthrax spores. It is all too easy to imagine a more determined effort to spread anthrax: emergency departments filled with frightened people, some in the early stages of a deadly disease.

Hupert and colleagues address the problem of efficient, accurate triage of patients with suspected anthrax (1). In this commentary, I focus on the article's principal product: a triage algorithm for people who have symptoms of inhalational anthrax.

An algorithm is a step-by-step instruction for solving a problem (2, 3). One use of algorithms is to solve the problem of sorting patients, by diagnosis or by levels of care, as in the present case. Figure 2 in Hupert's article, also shown here, is an algorithm. The boxes are decision points, which list criteria for an action. The arrows point to the next step in the triage process—another decision point or an action to take.

The ideal diagnostic algorithm for anthrax would use proven decision criteria and give each decision point's sensitivity, specificity, and corresponding probability of anthrax (4). The first step toward this goal would be to study a large number of people with the chief symptom of “I think I may have anthrax” by obtaining a standard set of clinical data and, independently of this information, establishing definite diagnoses. Standard statistical techniques can identify the best predictors of anthrax at each point in the algorithm. A key step is verifying the probabilities by …