Chinese curses, anthrax, and the risk of bioterrorism☆
Article Outline
You have to love the Chinese. The English translation of their famous ancient curse, “May you live in interesting times,” almost sounds like a blessing. As we all know, however, “interesting” isn't what they really mean.
Enter the study by Fine et al1 published in this issue of Annals. It offers emergency physicians a guide to evaluation and treatment options for patients presenting with fever and flu-like illness on the basis of an increasing probability that the patient has anthrax. The study is an excellent demonstration of the power of sensitivity analysis. The resulting recommendations are reasonable and practical; they also provide the first data-driven (dare I say evidence-based) recommendations on this topic. However, like the Chinese curse, it's not so easy to interpret.
Let's start with the article's strengths. The authors correctly note that when a rapid definitive diagnostic test is lacking, such as with anthrax, the management plan depends on certain disease characteristics and factors that influence the outcomes of various testing and treatment strategies. These include the prevalence of anthrax, its mortality rate, the sensitivity and specificity of diagnostic tests, and the incidence of influenza. If all these factors were established, an effective protocol could be created for varying probabilities of anthrax. The problem is that this information is not known and can only be approximated over a range of values. In addition, some factors, such as the probability that a patient has influenza, vary by the time of year. Therefore, one must test all the values for the various factors and examine the effect on outcomes. Only then can potential management options be accurately evaluated. To accomplish this end, one must use the strategy of sensitivity analysis. This is, in fact, what the authors did … for most factors.
In applying the findings of this investigation, however, one must be wary of several of the authors' assumptions that were not subject to analysis. First are the postulated mortality rates for various treatment strategies. These were taken at face value and not investigated further. The data used to calculate mortality rates for the various evaluation and treatment strategies are derived from a single limited study. The investigators relied on information generated during the October 2001 anthrax event in the United States and the opinions of an expert panel.2 (They did not incorporate data from the victim exposed in Connecticut during November 2001; therefore, my comments will not reflect this case.) How robust are these data? With respect to the October 2001 anthrax attack, the sum total of cases available for analysis is just 10 patients with inhalational anthrax. Of these, 4 died. The 95% confidence intervals (CIs) for this small number of data points are wide enough to drive a truck through. The actual mortality rate could be as low as 17% or as high as 69%. Both these rates are consistent with the data. The same wide CIs apply to the 10% mortality rate (95% CI 2% to 40%) for victims treated early, the 50% mortality rate (95% CI 24% to 76%) for those with positive blood culture results and empiric 3-day treatment, and so on. And what of the expert panel? I think you must remember that, until the anthrax attack of 2001, experts believed the mortality for treated inhalational disease exceeded 90%.3, 4 Although the true rate remains unknown, it is unlikely to be as high as 90%. One must be careful not to assign too much significance to expert panel opinion. It can change as the composition of the panel changes. Unfortunately, no sensitivity analysis was performed on any of these mortality rates. If the real mortality rates are, in fact, different than those assumed by the authors, the conclusions of the investigation could be significantly changed.
Second, and most important, is the assumption that an emergency physician can determine the probability or risk that a patient is infected with anthrax at all. The entire thrust of this article involves treatment decisions based on the use of probability assessment or risk assessment. If a physician cannot determine the risk of anthrax infection in a given patient, the recommendations of this study are unusable, even if all the authors' assumptions are correct. Given the importance of this concept, some discussion of risk assessment is warranted, even if the concept is somewhat complex.
There are many formulas for risk assessment, and the global concept is effectively illustrated by the Figure and by Table 1, Table 2.5 In general, this concept can be represented as
Figure.
Risk-assessment matrix. Adapted from US General Accounting Office Report GAO/NSIAD-98-74.5
Table 1. Definition of severity levels. Reprinted from US General Accounting Office Report GAO/NSIAD-98-74.5
| Severity Level | Characteristics |
|---|---|
| I Catastrophic | Death, system loss, or severe environmental damage |
| II Critical | Severe injury, severe occupational illness, major system or environmental damage |
| III Marginal | Minor injury, minor occupational illness, or minor system or environmental damage |
| IV Negligible | Less than minor injury, occupational illness, or less than minor system or environmental damage |
Table 2. Definition of probability levels. Reprinted from US General Accounting Office Report GAO/NSIAD-98-74.5
| Probability Level | Specific Event |
|---|---|
| A: Frequent | Likely to occur frequently |
| B: Probable | Will occur several times |
| C: Occasional | Likely to occur sometime |
| D: Remote | Unlikely but possible to occur |
| E: Improbable | So unlikely it can be assumed occurrence may not be experienced |
Let's say you are worried about a comet striking the earth. From the observations of Comet Shoemaker-Levy 9's impact with Jupiter, we know this will not be a pleasant experience. The magnitude of the event will be catastrophic. In addition, minimal mitigation efforts are underway, and nothing of substance is planned for the foreseeable future. However, no one is running out to fund billions of dollars in asteroid prevention research. The reason is that the probability of such an event is minuscule. We know this from more than 100 years of fairly meticulous observation of the solar system and other heavenly bodies. Given that the universe is reasonably predictable, past performance is a good predictor of future events (unlike stocks or bonds). Returning to the aforementioned formula, although a comet's impact with the earth will be associated with a very high severity, the probability of this event is small, and therefore, so is the risk.
The risk assessment for earthquakes in California plays out a little differently. Although the magnitude of a large earthquake is significant, so are the mitigation activities used over the years. California has retrofitted everything from bridges to hospitals and is constantly updating its strategies for managing the consequences of earthquakes.7 Therefore, the overall severity of an event is less than its magnitude would suggest. Even if you multiply this by the 95% probability of a 7.5-moment magnitude temblor in the next 30 to 40 years, the risk is almost acceptable.8 This is one of several reasons why I still live in California.
With these examples in mind, let's return to the issue of anthrax. The potential magnitude of an anthrax attack will be significant and has been well-documented by scenario modeling and the US experience.9 Mitigation efforts proceed but are still insufficient to sizably reduce the magnitude of such an event for the near future. Therefore, the risk will be determined largely by the probability of an anthrax exposure. So, is there any way to reliably estimate the probability that a given patient has anthrax? Two recent articles attempt to resolve this issue.
An article by Hupert et al10 tried to create screening criteria to identify patients with inhalational anthrax. The authors examined 3 cohorts of patients: 28 individuals with inhalational anthrax, 2,762 patients with influenza, and 1,932 cases of noninfluenza viral respiratory illness. They compared their presenting symptoms and calculated likelihood ratios for the presence of these symptoms. Unfortunately, most of the likelihood ratios are not high enough to have a significant effect on clinical decisionmaking. Generally, you need to see ratios approaching 10 to make a real difference. Most of their ratios were in the 5 range. In addition, it is hard to go from likelihood ratios to absolute probabilities of a clinical condition. It may tell you that a given patient has a 95% chance of having anthrax versus influenza, but not necessarily that they have a 95% chance of having anthrax. However, the biggest concern with this article is the extremely small number of anthrax cases on which their comparisons depend. Unlike the article by Fine et al,1 no sensitivity analysis was performed on the numbers of anthrax cases with a given symptom. For example, 4 (14%) of 28 patients with inhalational anthrax complained of rhinorrhea. The authors used this point estimate to generate likelihood ratios. However, the 95% CI for this number is between 6% and 31%. If the real number is closer to 31%, the likelihood ratio will not be nearly as impressive (sound familiar?). Without a sensitivity analysis on these numbers, it is very difficult to do much with this information.
The article by Kuehnert et al11 is a little more on target. The authors create a scoring system that would distinguish inhalational anthrax from influenza. Using a combination of clinical and laboratory data, a total of 4 points had a sensitivity of 100% and a specificity of 96%. Sound too good to be true? Well it is. This is a derivation set again based on just 11 cases of inhalational anthrax. No validation set exists. The issue of CIs aside, in the absence of a good validation set, one must apply these criteria with great caution.
Is there historical information that can assist with this problem? Unfortunately, there is not. A growing body of information does exist on the use of explosives by terrorist groups and more accurate risk assessments can be made for this situation. However, there simply has not been enough experience with acts of terrorism involving biologic agents to allow accurate estimates of probability. Within this area of investigation, terrorists remain as predictable as a random number generator.
So, back to the original question. Is there any way to accurately estimate the probability that a patient has anthrax? Not yet. However, there are a few caveats that can help. In the absence of any confirmed cases of anthrax or direct evidence a biologic attack has occurred, you must assume the risk of a patient presenting with anthrax is zero. Otherwise, you will be: (1) ordering chest computed tomographic (CT) scans and triple antibiotics on every febrile elderly patient with an atherosclerotic aorta (and therefore wide mediastinum) and congestive heart failure, and (2) treating everyone with influenza symptoms who does not have influenza A with the quinolone de jour. If an anthrax attack or exposure has occurred in your area, you will have to make clinical judgments based on guidelines such as the articles discussed in this editorial, diagnostic tests such as chest CT scans, and epidemiologic data, if they are available. Certainly, if a cohort of potentially exposed persons is identified, such as postal workers, liberal use of antibiotics is probably a good choice. Unfortunately, this is currently the best we can do.
The article by Fine et al1 takes an important step forward by clarifying the decisions that must be made in evaluating patients with possible inhalational anthrax. However, we must await the guidance derived from further research that will instruct us on how to make them. Until we gain enough experience with biologic terrorist attacks to permit accurate probability assessments or until a rapid, sensitive, and specific diagnostic test for anthrax is developed, we will continue to live in interesting times.
References
- References
- Is it influenza or anthrax? A decision analytic approach to the management of patients with influenza-like illnesses. Ann Emerg Med. 2004;43:318–328
- Bioterrorism-related inhalational anthrax: the first 10 cases reported in the United States. Emerg Infect Dis. 2001;7:933–944
- Domestic Preparedness Training Program Hospital Provider Manual. Edgewood Arsenal, MD: Soldier Biological and Chemical Command, US Army; 1999;
- In: Zajtchuk R, Bellamy RF editor. Textbook of Military Medicine: Medical Aspects of Chemical and Biological Warfare. Washington, DC: Office of the Surgeon General, US Army; 1997;
- . Combating Terrorism: Threat and Risk Assessments Can Help Prioritize and Target Program Investments. (GAO/NSIAD-98-74). Washington, DC: US General Accounting Office; April 1998;
- Kaiser Permanente Hazard and Vulnerability Analysis. Menlo Park, CA: Kaiser Foundation Health Plan, Inc; 2001;
- . Implications of hospital evacuation after the Northridge, California, earthquake. N Engl J Med. 2003;348:1349–1355
- . Planning scenario for a major earthquake on the Newport-Inglewood fault zone. Special Publication 99. Sacramento, CA: California Department of Conservation, Division of Mines and Geology; 1988;
- . Health Aspects of Chemical and Biological Weapons: Report of a WHO Group of Consultants. Geneva, Switzerland: World Health Organization; 1970;
- Accuracy of screening for inhalational anthrax after a bioterrorist attack. Ann Intern Med. 2003;139:337–345
- Clinical features that discriminate inhalational anthrax from other acute respiratory illnesses. Clin Infect Dis. 2003;36:328–336
☆ The author reports this study did not receive any outside funding or support.Reprints not available from the author.
PII: S0196-0644(03)01253-8
doi:10.1016/j.annemergmed.2003.12.004
© 2004 American College of Emergency Physicians. Published by Elsevier Inc. All rights reserved.
Refers to article:
- Is it influenza or anthrax? A decision analytic approach to the treatment of patients with influenza-like illnesses
