Influenza affects up to 40% of healthy children each year and carries with it risks of respiratory tract complications, hospitalization and, rarely, death.1 Infants and young children are at increased risk for complications and hospitalization, as are children of any age who have underlying medical problems, especially those with significant cardiopulmonary disease. Timely diagnosis of influenza permits early initiation of specific antiviral treatment as well as prophylaxis of unimmunized, high-risk contacts. Accurate diagnosis of influenza may also help both clinician and patient avoid unnecessary evaluation and treatment for other causes of illness.
However, because the signs and symptoms of influenza are not specific and often resemble those caused by other respiratory viruses, diagnosing it on the basis of clinical findings can be challenging. In one recent study, the clinical diagnosis of influenza in children was correct only about one third of the time. Diagnostic accuracy was especially low in children younger than 3 years; it was also particularly low early and late in the "flu season," when influenza was less common in the community.2 Because clinical diagnosis is often inaccurate (and because the timing of each year's epidemic varies, making determinations of "early in the season" difficult), clinicians sometimes order a laboratory test to confirm a clinical suspicion of influenza.
Studies that can confirm a diagnosis of influenza include rapid antigen testing, immunofluorescent antigen testing, enzyme immunoassay, reverse transcriptase polymerase chain reaction, viral culture, and serology.1 In the setting of a pediatric office or clinic, rapid antigen tests are most commonly used because they are quick (results in less than 30 minutes) and practical.
In this article, we examine how the accuracy of these rapid tests is affected by the prevalence of influenza in the community at the time of testing. We also present a practical approach to determining whether ordering a rapid antigen test is a reasonable option, based on both disease prevalence and clinical suspicion.
HOW PREDICTIVE VALUES VARY
Several rapid antigen detection kits for influenza are commercially available; each has its own sensitivity and specificity. Test sensitivity refers to the proportion of patients with influenza who will have a positive result on a rapid test, and test specificity refers to the proportion of patients without influenza who will have a negative result on a rapid test.
Rapid antigen detection tests for influenza typically have a sensitivity of 70% or greater and a specificity of 90% or greater.3 These numbers sound fairly reassuring, but clinicians are typically more concerned with a test's predictive values. The positive predictive value is the proportion of patients with a positive result on a rapid test who actually have influenza, and the negative predictive value is the proportion of patients with a negative result on a rapid test who do not have influenza. The predictive values depend on the test's sensitivity and specificity, but also on the frequency of disease in the community. For this reason, the positive and negative predictive values of rapid influenza tests depend on the time during influenza season when the testing is done.
The Advisory Committee on Immunization Practices (ACIP) of the CDC has advised physicians who order rapid influenza tests to consider the current level of influenza activity in their community when interpreting the results of these tests.1,4 However, a busy pediatrician may not have time to calculate the predictive values of positive and negative test results at different frequencies of disease in the community. To help place rapid influenza test results in a useful clinical context, we have calculated the positive and negative predictive values of rapid influenza test results at several points during a hypothetical influenza season.
CALCULATION OF PREDICTIVE VALUES: METHODS
The predictive values of a rapid influenza test depend in part on the test's sensitivity and specificity. For our first calculation of predictive values we used a sensitivity of 70% and specificity of 90% because most rapid tests are approximately that accurate.3 We also performed 2 additional sets of calculations to reflect the broad range of reported sensitivities (45% to 90%) and specificities (60% to 95%) seen with rapid influenza tests.1 The 2 alternative scenarios for which we calculated predictive values were:
• A test with higher sensitivity (90%) but lower specificity (60%).
• A test with lower sensitivity (45%) but higher specificity (95%).
(Cut-off levels for most diagnostic tests represent a compromise between sensitivity and specificity. Lowering a cut-off level, for example, will increase sensitivity at the expense of lowering the specificity, whereas raising the cut-off level will increase specificity at the expense of lowering sensitivity.)
The predictive values of rapid influenza tests also depend on the prevalence of influenza in the population. To estimate the disease prevalence during the course of a typical influenza season, we reviewed national influenza activity data from the CDC for the past 6 winters.5 These data represent the proportion of surveillance specimens obtained each week from patients with respiratory symptoms from which influenza virus type A or B was isolated. Then, for each winter season, we defined the peak week as the week with the highest proportion of specimens from which influenza virus was isolated. For each season, we determined disease activity at 5 different times:
• 12 weeks before peak.
• 6 weeks before peak.
• At peak.
• 6 weeks after peak.
• 12 weeks after peak.
To determine national influenza rates at each of these times during a hypothetical "typical" influenza season, we averaged disease activity at that time over the 6 seasons.
CALCULATION OF PREDICTIVE VALUES: RESULTS
The calculated positive and negative predictive values for the rapid influenza test with a sensitivity of 70% and a specificity of 90% are displayed in Table 1. During periods of less influenza activity, the positive predictive value of the test is modest. Only at the peak of the season does the positive predictive value exceed 70%. The negative predictive value is very high during the periods of lower prevalence, yet drops only to 89% during peak season.
Similar analyses were then performed for the rapid influenza tests with higher and lower sensitivities and specificities (Table 2). The positive predictive values are lower for the rapid test with a higher sensitivity and lower specificity, and are higher for the test with lower sensitivity and higher specificity. Correspondingly, the negative predictive values are higher for the rapid test with a higher sensitivity and lower specificity, and are lower for the test with lower sensitivity and higher specificity.
Taken together, Tables 1 and 2 show the positive and negative predictive values, over the course of the flu season, of rapid influenza diagnostic tests with sensitivities ranging from 45% to 90% and specificities ranging from 60% to 95%.
LIMITATIONS ON RELIABILITY FOR OFF-SEASON TEST RESULTS
Disease activity may change rapidly, especially with a disease like influenza that has marked seasonal variation. Our analysis demonstrates, in a concrete manner, how the predictive values of rapid influenza diagnostic tests can vary over the course of a typical influenza season in which disease activity starts at a low level, increases to a peak, and then declines to a low level again.
Some clinicians tend to trust their clinical judgment when diagnosing influenza at the peak of the flu season in their community but rely more heavily on rapid influenza testing early in the season.6 The Tables show, however, that the positive predictive value of a rapid influenza test can be quite low at off-peak times. False-positive results are very likely early in the season because the test is not perfectly specific and the disease frequency is relatively low.
Of interest, at times of very high disease prevalence, false negatives might become a problem. In one recent study, 50% of children tested for influenza in a pediatric office had a positive result on either viral culture or direct fluorescent antigen testing.7 We therefore repeated our calculations to assess the effect that a prevalence of this magnitude (50%) would have on the predictive values of rapid tests. The positive predictive value of a test with a sensitivity of 70% and a specificity of 90% would be 88% under these circumstances, but its negative predictive value would fall to 75%.
In addition to being affected by disease prevalence, the predictive values of rapid influenza tests are also affected by whether the particular test being used detects both influenza A and influenza B. Most rapid tests detect both types, but some tests detect only one. If a child has influenza B and the rapid test used detects only influenza A, a "negative" test result would technically be a true negative. However, from a clinical perspective, the result would act as a false negative--this child (and his or her close contacts) might lose out on the potential benefits of treatment (or prophylaxis) with a neuraminidase inhibitor. Thus, the functional negative predictive value of any rapid test that detects only one type of influenza is lower than even calculations that take disease prevalence into account would indicate.
Because of constant technological advances, it is to be expected that the sensitivity and specificity of tests such as the rapid antigen test will improve with time. However, while the predictive values at all disease frequencies may increase as test accuracy improves, the seasonal variation in predictive values will persist because of cyclic variations in influenza activity. A test with higher specificity will have the highest positive predictive value during peak disease activity but may still be relatively inaccurate during off-peak times.
EFFECT OF CLINICAL PRESENTATION ON TEST ACCURACY
The likelihood of any individual patient having influenza varies according to season and the frequency of influenza in the community but also according to the patient's clinical presentation. For example, a patient presenting with classic signs and symptoms of influenza early in the flu season may be just as likely to have influenza as a patient who presents at the peak of disease activity with less typical signs and symptoms.
This concept is expressed by the term "prior probability." In this case, the prior probability of influenza is the probability that a patient has influenza before test results are available. In addition to taking into account disease prevalence, prior probability also takes into account how strongly the clinical picture suggests the disease. For example, the clinical suspicion of influenza--and thus the prior probability--would be greater in an unvaccinated child with fever, cough, pharyngitis, headache, and myalgias than in a vaccinated child with rhinorrhea and diarrhea who was seen at the same time of year. Recent exposure to a family member with confirmed influenza would increase the likelihood of influenza even further.
The prior probability of influenza in turn affects the predictive accuracy of positive and negative results on rapid antigen tests (Table 3). Thus, to best determine whether testing is likely to be helpful in a given patient, test sensitivity and specificity, disease frequency, and the level of clinical suspicion must all be taken into consideration.
WHAT IT MEANS FOR YOUR PRACTICE
Table 3 shows how disease frequency and clinical suspicion work together to affect the predictive accuracy of results of a rapid antigen test with typical sensitivity and specificity. This table can help you use rapid influenza testing more wisely in your practice.
For example, suppose you see a child during a period of low influenza frequency and he has some respiratory symptoms but an overall clinical picture that is not very suggestive of influenza. For this patient, the prior probability of influenza would be very low. A positive result on a rapid influenza test in this setting would very likely be a false positive and might be clinically misleading. A negative test result would very likely be accurate and could potentially help you rule out a diagnosis of influenza. However, the test would only serve to confirm what you already strongly suspect. Thus, in this setting, you might decide to rely solely on your clinical judgment to rule out influenza.
At the other extreme, suppose you see a child at the height of annual influenza activity and the clinical findings strongly suggest influenza. A negative result on a rapid influenza test might help you avoid unnecessary antiviral treatment; however, in this setting, a negative result could be a false negative and would be clinically misleading. A positive result would probably be accurate and would support the clinical diagnosis--but here, too, would only serve to confirm what you already strongly suspect. Thus, in this setting as well you might decide to rely on your clinical judgment to make the diagnosis of influenza.
The rapid test is more likely to be helpful in situations where the likelihood (prior probability) of influenza is intermediate. For example, if there is very strong clinical suspicion during the early phase of influenza season or moderate clinical suspicion during peak influenza season, a rapid diagnostic test may help to clarify the correct diagnosis and guide optimal management. Further clinical studies will hopefully clarify even further how in-office use of rapid influenza tests can best complement clinical judgment in the evaluation of children with flu-like illnesses.
Acknowledgment: Ronald B. Turner, MD, made many helpful suggestions.