Does This Patient Have Influenza?

A 45-year-old eighth-grade math teacher visits your office in mid-December 2003, complaining of temperature to 38.6°C(101.5°F), dry cough, sore throat, myalgias, and malaise. Her symptoms began approximately 24 hours earlier, but she continued to teach through the end of the school day. A number of children in her classes have been absent due to similar complaints over the past 2 weeks. Her physical examination is notable for readily apparent malaise, temperature of 38.5°C (101°F), mild pharyngeal erythema with no exudates, no adenopathy, and clear lung fields. She has taken acetaminophen and ibuprofen for fever and muscle aches, with modest relief. Her medical history is notable for hypertension and gastroesophageal reflux disease, for which she takes hydrochlorthiazide and lansoprazole, respectively. Aside from 2 normal deliveries more than 10 years previously and an appendectomy during childhood, she has not been hospitalized. This year, as in prior years, she chose not to receive influenza vaccine. She comes to you suspecting that she might have “the flu” and asking whether she needs any stronger medication to help her return to the classroom more quickly.

Ten to twenty percent of US residents contract influenza annually, accounting for an average of 36 000 deaths over the past decade¹ and 133 900 pneumonia and influenza hospitalizations per year from 1979-2001.² Given its propensity for antigenic drifts and shifts, influenza has the capability to cause periodic epidemics and global pandemics. A shortfall in production of vaccine due to problems at one manufacturer’s facilities (http://www.hhs.gov/news/press/2004pres/20041005.html) creates the potential for increased morbidity and mortality in the 2004-2005 influenza season. The impact on society during major outbreaks is substantial in terms of both direct medical costs and indirect costs associated with illness, including missed workdays and reduced productivity.³ In 2003, there were concerns about early season reports of influenza-related severe illnesses and deaths in the United States.⁴ The fixed number of doses of vaccine (approximately 83 million) and the increased demand for its use in 2003 led to a redistribution of vaccine to clinicians caring for individuals with the greatest immediate need.⁴ This situation was compounded by a vaccine that may have had reduced effectiveness due to a suboptimal antigenic match. Already, early in the 2004-2005 season, one of the manufacturers of the trivalent inactivated vaccine will not be providing vaccine to the United States; consequently, the available vaccine for the nation will be only about half that projected for the year.⁵ Under these circumstances, early diagnosis and intervention are even more critical.

Fortunately, there are several specific antiviral agents available to treat infection with influenza viruses: amantadine and rimantadine (for type A viruses only) and the newer agents zanamivir and oseltamivir (for either type A or type B strains). All 4 of these antiviral agents reduce the duration of clinical illness,⁶ but they vary in cost and must be instituted within 48 hours of symptom onset for maximal benefit. Consequently, they should be used only when the probability of infection with influenza and the expected benefit are both high.

Influenza-like illness, defined by the Centers for Disease Control and Prevention (CDC) US Influenza Sentinel Providers Surveillance Network as temperature higher than 37.8°C (100°F) plus either cough or sore throat (www.cdc.gov/flu/weekly/fluactivity.htm) but sometimes defined differently by others, is a syndrome characterized by other nonspecific symptoms that may be seen with a variety of upper respiratory tract infections. The frequency of infections attributable to the various viral agents that cause influenza-like illness varies geographically and from week to week throughout the influenza season. Fortunately, excellent weekly reports are available that help clinicians understand both the incidence of influenza-like illness and the current influenza activity rates applicable to their geographic location. The CDC produces weekly influenza reports that are available online (http://www.cdc.gov/flu/weekly/fluactivity.htm). These reports provide a synopsis of epidemiologic information, including laboratory surveillance data, influenza-like illness frequency as reported by US sentinel providers, and regional variability of outbreaks (Figure). Similar reports are available from individual state health departments, Canada (through Health Canada), the World Health Organization (WHO) International Influenza Program, the WHO Flunet, and the European Influenza Surveillance Scheme (hyperlinks available at http://www.cdc.gov/flu/weekly/intsurv.htm).

Based on CDC sentinel data, patient visits to a primary care practitioner’s office for influenza-like illnesses peak at 3.3% to 7.1% during influenza season.⁴ In the 2003-2004 influenza season, the weekly percentage of patient visits for influenza-like illness exceeded the national baseline of 2.5% for 9 consecutive weeks, with a peak of 7.6% in the week ending December 27, 2003.⁴ Thus, during the peak week of the 2003-2004 outbreak, about 1 of every 13 primary care visits in the United States was for an influenza-like illness.

Laboratory surveillance monitoring in the United States showed that the majority of samples in the 2003-2004 influenza season tested negative for influenza. Although not specifically reported, the implication is that these patients often had other viruses such as rhinoviruses, adenoviruses, and parainfluenza. While many of these are relatively benign and self-limited, others may be quite serious; for example, early infection during an epidemic of the coronavirus causing severe acute respiratory syndrome (SARS) produced influenza-like illness.⁷ Bacterial agents, including Legionella species, Chlamydia pneumoniae, Mycoplasma pneumoniae, and Streptococcus pneumoniae, may also be responsible for influenza-like illnesses.

When faced with a patient with influenza-like illness, a physician must be able to accurately estimate the probability of influenza as opposed to other infections. This probability estimate guides the clinician in further diagnostic testing and treatment. Appropriate and prompt diagnosis and therapy affect not only the individual patient but society as well, in that local outbreaks may be detected and control measures initiated. Influenza is difficult to diagnose due to nonspecific symptoms and the host of other diseases that cause similar symptoms. Our objective in this review was to identify clinical factors that may be valuable in distinguishing which patients with influenza-like illness have a higher probability of truly having influenza.

We searched MEDLINE (January 1966 to September 2004) to identify articles pertaining to the diagnosis of influenza based on individual clinical signs and symptoms. We intentionally limited the search to the period prior to the SARS outbreak to avoid implying that the same operating characteristics could be applied during an outbreak with a highly virulent agent causing similar symptoms. The search strategy used the following Medical Subject Headings: EXP influenza or EXP influenza A virus or EXP influenza A virus human or EXP influenza B virus. These terms were then combined with the Medical Subject Headings and text words EXP sensitivity and specificity or EXP medical history taking or EXP physical examination or EXP reproducibility of results or EXP observer variation or symptoms.mp or clinical signs.mp or sensitivity.mp or specificity.mp. We also searched for academic reviews on influenza (EXP influenza or EXP influenza A virus or EXP influenza B virus, limited to human, English-language academic reviews). From this search we retained only systematic reviews. We reviewed the references and citations to identify other relevant articles. We also reviewed the references in a recent systematic review by Ebell et al.⁸ Unpublished primary data were not sought.

Abstracts of the identified articles were reviewed for relevance. Only articles describing primary studies dealing with the diagnosis of influenza based on clinical signs and symptoms were selected for complete review.

Two of the authors (S.A.C., W.P.M.) independently reviewed the final set of 17 articles for quality.^{9 - 25} Differences in assessment were discussed and resolved by consensus. Studies in the final set were excluded from analysis if they did not meet the following criteria: (1) study design qualifying as prospective cohort, randomized controlled trial, or meta-analysis; (2) inclusion of primary assessment of individual clinical signs and symptoms as predictors of diagnosis; (3) definition of at least 1 of the outcomes as influenza type A or B infection that was proven by either (a) culture; (b) 4-fold increase in diagnostic antibody titer, eg, hemagglutination inhibition, complement fixation, or enzyme immunoassay from acute to convalescent serum; (c) polymerase chain reaction; or (d) immunofluorescent antibody; and (4) study quality graded A or B using the scheme appearing previously in the Rational Clinical Examination series, adapted from Holleman and Simel²⁶ as shown below:

Ten articles met all of the inclusion criteria.^{9 ,12 ,14 ,17 - 20 ,23 - 25}

Because the interpretation of rapid influenza test results is tightly coupled to the interpretation of the examination, we added information to the manuscript about the usefulness of diagnostic testing. This information was obtained through an additional MEDLINE database search (January 1996 to October 2004) for English-language articles pertaining to the rapid diagnostic kits for human influenza. This strategy was devised to focus on articles describing the most current and relevant tests available to clinicians and to find citations in which direct comparisons of the most recent tests might be available. The search strategy used the following Medical Subject Headings: EXP influenza and EXP sensitivity and specificity and EXP reagent kits, diagnostic. Data from manufacturers were also sought to establish the products’ range of sensitivity and specificity. Unpublished primary data were not sought. Abstracts of identified articles were reviewed for relevance.

We used data from the identified articles to calculate the sensitivity, specificity, positive likelihood ratio (LR), and negative LR, as well as a summary LR and the diagnostic odds ratio (DOR) for individual history and physical examination findings. The positive LR is a measure of how strongly a positive result increases the odds of disease; the negative LR is a measure of how well a negative result decreases the odds of disease. An LR greater than 1.0 increases the likelihood of disease; an LR less than 1.0 decreases the likelihood; an LR close to 1.0 does not change the likelihood. FastPro (Academic Press, Boston, Mass) was used for all analyses; P<.05 was used to determine statistical significance.

The DOR is a single indicator of diagnostic test performance, reflecting its accuracy.²⁷ The DOR can also be viewed as presenting the odds (likelihood) of the symptom or finding among individuals with disease (ie, the positive LR) compared with the odds of the symptom or finding among those not having the disease (ie, the negative LR). The DOR should always be assessed in comparison with the paired sensitivity and specificity because the same DOR can be associated with different pairs. The value of the DOR ranges from 0 to infinity, with higher values indicating better test performance. Values less than 1 indicate more negative test results among individuals with disease and reverse test interpretation. The DOR can also be used to develop summary estimates in meta-analyses.

We tested the LRs for heterogeneity between studies using the Mantel-Haenszel Q-statistic.²⁸ We used conservative random-effects models to describe the summary estimates and confidence intervals, making it easier to discern the relative usefulness of symptoms and signs.^{29 - 30}

The search strategy identified 915 articles (bibliography available on request). We found only 10 studies that met all the inclusion criteria.^{9 ,12 ,14 ,17 - 20 ,23 - 25} The majority of the excluded articles were not primary studies. We were unable to obtain primary data for 3 of the 10 studies,^{9 ,18 - 19} and data from 1 study was included in 2 papers; thus, the final data (Table 1) are based on 6 studies and included 7105 patients.^{12 ,14 ,20 ,23 ,25 - 26} We identified a recent systematic review that included several studies for which we were unable to obtain the primary data.⁸ Thus, not all the references in this systematic review met our inclusion criteria. One additional study included in this review, but not identified in our literature search, did meet our inclusion criteria.²⁵

The second search strategy identified 13 articles dealing with rapid diagnostic tests for influenza (bibliography available on request). Only 6 original articles^{31 - 36} describing the comparison of a commercially available rapid diagnostic test for influenza vs viral culture as the criterion standard were selected for complete review. Of these, only 1 article³⁵ presented direct comparison of results among 4 test kits studied; the data from this article were evaluated in detail.

None of the studies assessed the precision of signs or symptoms of influenza. Measurements of objective clinical signs such as temperature are assumed to have high precision.

The studies presented used varying definitions for fever, ranging from 37.8°C to 38.5°C (Table 1). We defined fever as “present” or “absent” based on the individual article definitions. “Feverishness” was reported by the patient and could have been based on either a temperature taken at home or a subjective sense of having an elevated temperature. The sensitivity, specificity, positive LR, negative LR, and DOR for clinical variables evaluated in at least 2 of the 5 studies are reported in Table 2 and Table 3. Summary estimates are also presented. Eleven of the 13 clinical factors had heterogeneous DORs (P<.05). The patient's sense of feverishness and vaccination history provided homogeneous results across studies. Despite the heterogeneity, the studies we reviewed seem representative of the universe of patients with influenza, and most of the differences in estimates created by the statistical heterogeneity were small. The heterogeneity, expressed in the confidence intervals (CIs), never moved a finding from useless (LR approaching 1) to obviously useful (LR so different from 1 that it would make influenza extremely likely or extremely unlikely). Therefore, we present the summary LRs and DORs as an efficient way of conveying the relative diagnostic impact of the symptoms and signs.

No single clinical finding consistently had a positive LR high enough to clinically “rule in” influenza, nor did any single finding have a negative LR low enough to clinically “rule out” influenza (Table 2 and Table 3). However, several patterns do emerge when evaluating the data from the multiple studies. Among studies that enrolled patients without regard to age, no single finding had a summary LR greater than 2. For decreasing the likelihood of influenza, the absence of fever (LR, 0.40; 95% CI, 0.25-0.66), cough (LR, 0.42; 95% CI, 0.31-0.57), or nasal congestion (LR, 0.49; 95% CI, 0.42-0.59) were the only findings with an LR less than 0.5. Feverishness, myalgia, malaise, sore throat, and sneezing each had a positive and negative LR that were indistinguishable from 1.0 and therefore of no diagnostic value for the patients in studies that evaluated the entire age spectrum. Among the studies of patients limited to those aged 60 years or older, the strongest univariate indicators of influenza were fever (LR, 3.8; 95% CI, 2.8-5.0), malaise (LR, 2.6; 95% CI, 2.2-3.1), and chills (LR, 2.6; 95% CI, 2.0-3.2). Among older patients exclusively, the presence of sneezing reduced the likelihood of influenza (LR, 0.47; 95% CI, 0.24-0.92).

Two studies, by Govaert et al¹⁴ and Monto et al,²⁰ assessed the diagnostic usefulness of the symptom combination of fever and cough in persons aged 60 years or older and in the unrestricted age group (Table 3). The LRs when both fever and cough were present were 5.0 and 1.9, respectively. The addition of a third variable, “acute onset” of symptoms, added minimally to the discriminatory accuracy in either study.

The calculation of DORs for the individual variables in each study allows us to compare the diagnostic performance of the different variables and combinations of variables using a single measure (Table 2 and Table 3). The 3 studies with the lowest frequency of influenza tended to have the best overall accuracy as expressed by the DOR. In comparison with the calculated DORs for other symptoms, fever (summary DOR, 4.5; 95% CI, 1.8-11.0) and cough (summary DOR, 2.8; 95% CI, 2.1-3.7) are the most useful single findings for distinguishing patients with influenza from those without the illness among the unrestricted age group. The combination of fever and cough, with or without acute onset, had an intermediate DOR value. The DORs were somewhat higher for all of these parameters, particularly the combined symptoms, among persons aged 60 years or older; malaise (DOR, 4.9; 95% CI, 3.3-7.1) also performed well in this group.

Fever, headache, myalgias, and cough are the classic symptoms that physicians associate with influenza. Unfortunately, these symptoms are frequently seen in patients presenting with other infections during influenza season, making the clinical diagnosis of influenza a challenge to the primary care physician. The data presented indicate that the strongest predictor of influenza was the acute onset of both fever and cough in patients aged 60 years or older.

We included only studies in which a laboratory confirmation of the influenza virus was performed in all patients included in the analysis, thus eliminating verification bias. However, not all studies used the same criterion standard diagnostic test; 1 study used culture data only, without supplementation by titer increase or polymerase chain reaction.¹⁷ This may have been associated with false-negative results and a decreased estimate of prevalence of disease. In addition, several of the studies assessed the type of influenza (A vs B), while others did not. One study found that different clinical presentations were associated with the influenza type,¹² but another study showed no difference.²⁰ In the 2 studies that presented data on both influenza types A and B, the proportion of patients diagnosed with influenza B was small (<10%).^{12 ,20} Data presented here reflect all diagnoses of influenza, regardless of type or subtype; we do not know if the clinical presentation of disease varies depending on type or subtype.

The patient populations in the 6 studies were very different but represented a broad spectrum of patients with influenza-like illnesses. Two of the study populations were derived from randomized controlled trials of treatment or vaccine,^{14 ,20} 3 were prospective cohorts of patients presenting to general practitioners,^{12 ,17 ,23} and 1 was a population-based cohort surveyed for symptoms weekly by telephone.²⁵ The studies were also from several different countries: 2 from the Netherlands,^{14 ,23} 1 from France,¹² 1 from the United States,¹⁷ and 1 from England²⁵ ; in addition, 1 was a multinational study including patients from North America, Europe, and the southern hemisphere. This variability in patient population may have led to less precision in the assessment of symptoms due to cultural and language differences. The different study populations may also have had very different clinical characteristics due to the pool from which they were drawn. The studies from Europe were more likely to include patients from home. It is conceivable that these patients may have been more or less ill, have had more or fewer symptoms, and have had a different prevalence of influenza when compared with the populations from the United States or other countries.

Including the randomized controlled trials may lead to spectrum bias, as patients who enroll in randomized controlled trials assessing either treatment or prevention of influenza may not represent the population of persons presenting to a primary care office. Spectrum bias may be a particular issue in the study by Govaert et al,¹⁴ in that signs and symptoms were assessed in all the persons enrolling in the vaccine trial whether they had complaints of illness or not.¹⁴ This not only leads to spectrum bias but also is consistent with the 6.6% prevalence in this study, which is lower than the prevalences in the other studies (range, 8%-67%).

Other differences in the study populations include the age range within each study. This may be important, as Cox and Subbarao³⁷ have noted that influenza presents differently among various age groups. Although the majority of the patients studied in these reports were adults, several of the studies did include children. Govaert et al¹⁴ and Nicholson et al²⁵ evaluated only individuals aged 60 years or older. Notably, the positive LRs for several of the signs and symptoms evaluated in these studies are higher than those in the other studies. One possible explanation for this is that the clinical findings are more diagnostic of influenza in the elderly population or in a population with a lower prevalence of disease.

While all of the studies recruited patients only during influenza season, some were specifically undertaken during epidemics. Thus, the prevalence of disease varies considerably in the published reports of the clinical findings. It is possible that clinical characteristics of the disease change between seasons based on the strain of influenza. All of the studies were performed before the recent epidemic of SARS.

Monto et al²⁰ suggest that the positive predictive value of clinical signs and symptoms increased with increasing duration from illness onset. The 6 studies presented in this article had variable durations of symptoms. Data were not available from each of the studies to assess whether the other studies supported the results of Monto et al.

The 2003 outbreak of influenza brought the diagnostic dilemmas regarding influenza to the forefront. The reduced availability of vaccine for 2004-2005 creates the potential for increased incidence of disease. When faced with a person with influenza-like illness, clinicians struggle with the decision of whether to test or to empirically treat.

Currently there are several laboratory-based procedures available for diagnosing influenza. Viral culture is the criterion standard for laboratory diagnosis, but it may take several days to see cytopathic effects or for virus to be detected by hemadsorption or hemagglutination. Rapid methods may shorten the time to identification but at some cost in sensitivity. Fluorescent-antibody staining or other immunoassays are used to confirm and to type influenza virus in culture and are frequently used directly on respiratory specimens as part of a respiratory virus battery. Results from direct immunoassays may be available within hours. Molecular methods such as reverse transcriptase polymerase chain reaction and hybridization-based arrays are likely to replace culture as the criterion standard because of their superior sensitivity and rapid turnaround time. However, the availability of technology is limited. For diagnostic dilemmas, research studies, and epidemiologic purposes, influenza infection can also be detected by a 4-fold or greater increase in a variety of diagnostic antibody titers (eg, hemagglutination inhibition, complement fixation, or enzyme immunoassay) between specimens collected at least 10 days apart. Although these laboratory-based methods are highly sensitive and specific, clinicians are increasingly reliant on point-of-care rapid diagnostic tests, which are easier to handle, less costly, and provide test results in less than 30 minutes.

A summary of the rapid diagnostic tests for influenza is provided by the CDC (http://www.cdc.gov/flu/professionals/labdiagnosis.htm). These include Directigen Flu A and Directigen Flu A+B (Becton-Dickinson, Franklin Lakes, NJ), FLU OIA and FLU OIA A/B (Thermo Electron Corp, Waltham, Mass), XPECT Flu A/B (Remel, Lenexa, Kan), NOW Flu A Test and NOW Flu B Test (Binax Inc, Portland, Me), QuickVue Influenza Test and Quick Vue Influenza A + B Test (Quidel Corp, San Diego, Calif), SAS Influenza A Test and SAS Influenza B Test (SA Scientific Ltd, San Antonio, Tex), and ZstatFlu (ZymeTx Inc, Oklahoma City, Okla). The tests require specimens of throat swabs, nasopharyngeal swabs, nasal washes, or nasal aspirates. The sensitivity and specificity of these tests have been reported by manufacturers to be between 40% and 100% and between 52% and 100%, respectively.^{38 - 43} Given the differences between older and younger persons in presenting symptoms of influenza,³⁷ the operating characteristics of these tests could differ among various age groups; however, we found no data confirming this. The QuickVue and ZstatFlu tests have waivers from the Clinical Laboratory Improvement Amendments and can be used in any office setting. The Quick Vue A+B Test is the only amendment-waived test that distinguishes between influenza A and B.

Multiple studies have compared individual test kits vs the reference standard of viral culture (Table 4).^{31 - 36} In 2002, Rodriguez et al³⁵ published a study that directly compared 4 of the most widely used rapid diagnostic test kits in children with influenza-like illness. During the 1999-2000 epidemic, the authors had patients provide specimens for viral culture and direct fluorescent antigen as well as for testing with Directigen Flu A, FLU OIA, QuickVue Influenza Test, and ZstatFlu A/B. Influenza A was found in 49% of the patients; 17% of the cases were detected by viral culture only. Sensitivity and specificity of the 4 tests ranged from 72% to 95% and from 76% to 84%, respectively. For diagnosing influenza, these tests all had similar LRs (P = .69) when the results were positive, with a summary LR of 4.7 (95% CI, 3.6-6.2). The ZstatFlu test has a lower sensitivity than the other tests (P<.01); however, the remaining tests perform similarly (P = .99) and exceedingly well for ruling out influenza when the test result is negative, with a summary LR of 0.06 (95% CI, 0.03-0.12).³⁵

Two recent studies examined the cost-effectiveness of several influenza management strategies in adults, including several strategies in which rapid influenza diagnostic tests were used.^{44 - 45} The estimates used for the sensitivity of the rapid tests ranged from 59% to 81%; the estimates used for specificity ranged from 70% to 99%. The prior probability estimate of influenza was 35% in the analysis by Rothberg et al⁴⁴ and 60% in that by Smith and Roberts.⁴⁵ In both analyses, testing strategies were less effective than empirical treatment because of the low sensitivity of the tests. These analyses were sensitive to the probability of influenza infection; the cost-effectiveness of empirical treatment improved relative to the testing strategy as the probability increased. In fact, in the study by Rothberg et al, empirical treatment with a neuraminidase inhibitor in unvaccinated patients was more cost-effective at any probability of influenza greater than 14%. Testing was only preferred between a probability of 5% and 14% (in unvaccinated patients). The study by Smith and Roberts yielded similar results, favoring rapid testing only at a lower prevalence of influenza. These studies highlight the importance of the physician’s estimate of the likelihood of influenza.

The decision analytic model used by Rothberg et al was sensitive to vaccination status. In a recent systematic review of the literature, the estimated reduction in serologically confirmed cases of influenza A by the live attenuated aerosol vaccines was 48%; the reduction with the use of inactivated parenteral vaccines was 68%.⁴⁶ Vaccine efficacy and effectiveness may be affected by epidemiologic characteristics such as age and institutionalization. At least 1 study showed a vaccine efficacy of 58% in older patients who were not institutionalized.⁴⁷

From these analyses, if one is able to estimate the probability of influenza to be greater than 25% to 30%, rapid diagnostic testing does not add to the overall cost-effectiveness of treatment. Thus, clinicians must develop a pretest probability based on clinical signs and symptoms, vaccination history, and epidemiologic risk factors. During influenza season, the CDC publishes weekly online updates that contain information about the prevalence of visits for influenza-like illness along with data about influenza outbreaks (http://www.cdc.gov/flu/weekly/fluactivity.htm). The same information is generally available for each state through its own surveillance reporting systems. It is important that physicians understand the information available in the reports. The percentage of visits to sentinel providers for influenza-like illness for the week ending December 6, 2003 (week 49) was high (5.1%), and there was regional variation (Figure, B and C). Among laboratory respiratory specimens submitted as part of the CDC surveillance system, 36.8% tested positive for influenza during week 49 (Figure, A). At the beginning of the 2003-2004 influenza season, for the week ending October 4, 2003 (week 40), the percentage of office visits for influenza-like illness was only 0.9% (Figure, B); only 1.4% of laboratory respiratory specimens tested positive for influenza during the same week.

Unfortunately, there is no linkage between the surveillance systems for monitoring influenza-like illness and laboratory results. The CDC surveillance systems are careful to note that the system is designed to report “where, when, and what influenza viruses are circulating,” but the data cannot be used by the clinician to determine the probability that an individual patient with an influenza-like illness actually has influenza. Although the likelihood of influenza may vary along with the frequency of influenza-like illness, no data exist for clinically determining whether the threshold levels of the decision analytic model have been exceeded. Depending on the acuity of illness, vaccination status, and presence of comorbid conditions, some physicians might choose to treat empirically with medication while some might choose testing.

The individual presented in the Clinical Scenario comes to the office during the usual influenza season with “classic” influenza-like symptoms. She has been ill for 24 hours, was not vaccinated, and has been exposed to many children with influenza-like illnesses. The primary care clinician is suspicious of influenza and is faced with the decision of whether to manage her symptomatically, treat her with an antiviral agent, or test her for influenza with a rapid test. Because she is less than 48 hours into the illness, treating her now should allow her to return to work more quickly if she does indeed have influenza. The data most pertinent to this patient were released by the CDC on December 11, 2003 (http://www.cdc.gov/flu/weekly/weeklyarchives2003-2004/weekly49.htm).The CDC data indicated regional outbreaks of influenza in her area of the country, with 5.1% of primary care visits for influenza-like illnesses (Figure, B). In week 49, 36.8% of specimens submitted had laboratory confirmation of influenza (Figure, A).

Once clinicians have used the symptoms in Table 2 and Table 3 to establish that the patient has an influenza-like illness, they should use epidemiologic data to determine if influenza virus is circulating. Clinicians must rely on their clinical judgment in deriving a pretest probability for influenza. If a clinician decides to perform a rapid influenza test, strict adherence to the manufacturer’s protocol is required for accurate interpretation. If the result is positive, the odds of disease increase almost 5-fold (summary LR, 4.7). A negative rapid influenza test result (summary LR, 0.06) lowers the probability of disease and could effectively rule out influenza if the prior probability is low. Astute clinicians will recognize that the decision to use rapid diagnostic testing can vary throughout the influenza season depending on the age of their patient, the setting, and the prevalence of disease in their community.

Influenza presents with a constellation of symptoms including cough, fever, malaise, myalgias, and headache. We reviewed the literature regarding signs and symptoms and their diagnostic accuracy for influenza. Unfortunately, no specific symptom or combination of symptoms is diagnostic of this common infection. Despite the variability in participant nationality, language, culture, and age, as well as in clinical setting and influenza type/subtype in the studies reviewed, the data indicate that, while not perfect, the combination of fever and cough during influenza season suggests a significantly increased likelihood of influenza among elderly individuals.

The usefulness of these signs and symptoms follows from their ability to identify a group with influenza-like illness. However, the prevalence of disease among this population varies from week to week and from year to year throughout the influenza season. Clinicians must pay attention to the surveillance data to understand where, when, and what influenza viruses are circulating. As an example, the peak weeks during the 2002-2003 influenza season for influenza-like illnesses occurred later in the season and at lower rates than in 2003-2004. The role of rapid influenza tests has not been fully established, though it seems likely that clinicians will have many options for testing during future influenza seasons. In a randomized trial of the usefulness of rapid influenza tests in a pediatric emergency department, physicians provided with rapid test results ordered fewer laboratory tests and chest radiographs, prescribed fewer antibiotics but more antiviral agents, kept patients in the emergency department for shorter periods of time, and generated lower patient charges.⁴⁸

Once the clinical criteria are used to establish the presence of an influenza-like illness, there is little information other than epidemiologic data that is useful for guiding diagnostic and therapeutic decision making. During the current era of rapidly evolving infections with pathogens unfamiliar to most physicians, we do not know how well the symptoms, signs, and rapid diagnostic tests would perform if these newer infections were to become epidemic. Clinicians in the United States must pay particular attention to the weekly CDC and state reports regarding regional influenza patterns during influenza seasons. International clinicians should use data from the WHO International Influenza Program, the WHO Flunet, Health Canada, or the European Influenza Surveillance Scheme. The hyperlinks for all these sites are available at the CDC Web site (http://www.cdc.gov/flu/weekly/intsurv.htm). It would be very useful for clinicians if a formal linkage could be established between clinical and laboratory surveillance strategies, such as the collection of influenza virus cultures from a random sample of persons presenting with influenza-like illnesses, to allow more precise estimation of an individual’s likelihood of disease. In the absence of such a system, physicians may consider point-of-care testing among patients in their individual practices to gain an estimate of the prevalence of influenza among their patients presenting with influenza-like illnesses.