ORIGINAL INVESTIGATION Clinical Signs and Symptoms Predicting Influenza Infection Arnold S. Monto, MD; Stefan Gravenstein, MD; Michael Elliott, MD; Michael Colopy, PhD; Jo Schweinle, MD Background: New antiviral drugs are available for els to identify those best predicting laboratory confirma- the treatment of influenza type A and type B infec- tion of influenza. tions. In clinical practice, antiviral use has rarely been guided by antecedent laboratory diagnosis. Defined Results: Of 3744 subjects enrolled with baseline influ- clinical predictors of an influenza infection can help enzalike symptoms, and included in this analysis, 2470 guide timely therapy and avoid unnecessary antibiotic (66%) were confirmed to have influenza. Individuals with use. influenza were more likely to have cough (93% vs 80%), fever (68% vs 40%), cough and fever together (64% vs Objective: To examine which clinical signs and symp- 33%), and/or nasal congestion (91% vs 81%) than those toms are most predictive of influenza infection in pa- without influenza. The best multivariate predictors of in- tients with influenzalike illness using a large data set de- fluenza infections were cough and fever with a positive rived from clinical trials of zanamivir. predictive value of 79% (P,.001). The positive predic- tive value rose with the increase in the temperature at Methods: This analysis is a retrospective, pooled analy- the time of recruitment. sis of baseline signs and symptoms from phase 2 and 3 clinical trial participants. It was conducted in mainly un- Conclusion: When influenza is circulating within the vaccinated (mean age, 35 years) adults and adolescents community, patients with an influenzalike illness who have who had influenzalike illness, defined as having fever or both cough and fever within 48 hours of symptom onset feverishness plus at least 2 of the following influenza- are likely to have influenza and the administration of in- like symptoms: headache, myalgia, cough, or sore throat fluenza antiviral therapy may be appropriate to consider. who underwent laboratory testing for influenza. Clini- cal signs and symptoms were evaluated in statistical mod- Arch Intern Med. 2000;160:3243-3247 NFLUENZA VIRUS infection is a ma- lem.14,15 Zanamivir and oseltamivir, new, From the Department of jor public health problem, occur- recently approved antiviral agents, are in- Epidemiology, University ring, typically, in the Northern hibitors of the influenza virus enzyme of Michigan, Ann Arbor Hemisphere, between the months neuraminidase and are active against in- (Dr Monto); Center for of December and April. Epidem- fluenza type A and type B.3,11,12,16 When ad- Geriatrics and Gerontology, Iics of influenza are characterized by an in- ministered early in the course of infec- East Virginia Medical School, creased morbidity and mortality in the tion, both reduce the time to alleviation Norfolk (Dr Gravenstein); and community and result in increased absen- of clinical symptoms in individuals Infectious Diseases and teeism from school and work.1-6 The clas- infected with influenza.3,11,12 Hepatitis, Glaxo Wellcome Inc, sic influenza syndrome is sudden in on- To be maximally effective against in- Research Triangle Park, NC (Drs Elliott, Colopy, and set and is characterized by fever, headache, fluenza, antiviral therapy must be initi- Schweinle). Dr Monto has cough, sore throat, myalgia, nasal conges- ated as soon as possible after symptom on- 3-8 received honoraria and tion, weakness, and loss of appetite. set, and clearly cannot await traditional research support from Glaxo Vaccines, the most cost-effective pri- laboratory diagnosis.11,12,17,18 This analy- Wellcome Inc and mary prevention for influenza, are effec- sis examined which clinical signs and Hoffmann-LaRoche Inc. tive and readily available but have their symptoms are most predictive of influ- Dr Gravenstein is a consultant limitations.9,10 Two antiviral agents with enza infection in patients with influenza- to Glaxo Wellcome Inc and similar activity, amantadine hydrochlo- like illness using a large data set derived Hoffmann-LaRoche Inc and has ride and rimantadine hydrochloride, have from clinical trials of zanamivir. Knowl- received research support from been available many years for prophy- edge of the most predictive symptoms Glaxo Wellcome Inc. 3,10-13 Drs Elliott, Colopy, and laxis and treatment of influenza. How- of influenza could be used by physicians Schweinle are employee of ever, both of these agents are active only to diagnose influenza more accurately and Glaxo Wellcome Inc and may against influenza type A and not influ- to begin an appropriate course of treat- hold Glaxo Wellcome stock enza type B, and resistance of influenza ment in time to be of most benefit to the or stock options. A viruses to these drugs can be a prob- patient. (REPRINTED) ARCH INTERN MED/ VOL 160, NOV 27, 2000 WWW.ARCHINTERNMED.COM 3243 ©2000 American Medical Association. All rights reserved. Downloaded From: https://jamanetwork.com/ on 09/24/2021 PATIENTS AND METHODS DATA COLLECTION AND ANALYSIS The following baseline signs or symptoms were analyzed as STUDIES AND DESIGN potential predictors of influenza infection: fever, feverishness, cough, headache, sore throat, myalgia, nasal congestion, weak- Eight double-blind, placebo-controlled studies involving ness, and loss of appetite. Although the symptoms were col- 231 study centers in North America, Europe, and the South- lected using a 4-point severity scale, for this analysis they were ern Hemisphere were included in this analysis (Table 1). analyzed as either absent or present. A diagnosis of influenza These studies were phase 2 and 3 clinical trials designed was defined either as a positive culture for influenza virus or to evaluate the use of the antiviral agent zanamivir vs pla- as a 4-fold or greater increase in influenza antibody titer in cebo for the treatment of influenza type A and type B viral convalescent vs acute serum samples as determined by hem- infections and were conducted during the fall and/or win- agglutination inhibition. In some studies, influenza infection ter of 1994 through 1998. could also or alternatively be confirmed by polymerase chain Data regarding patients’ signs and symptoms, col- reaction (NAIA/B3002 [the protocol report number]) or by lected at baseline (ie, prior to any treatment), were immunofluorescence.19,20 pooled across the 8 studies. All available study data were Univariate and multivariate analyses were conducted to included in the analysis to minimize the potential for compare clinical signs and symptoms with the diagnosis of bias. influenza. Demographic factors such as age, sex, high-risk populations (defined as having chronic respiratory disease, PATIENTS cardiovascular disease, or more advanced age [ie, $65 years old]), and hours from symptom onset to collection of the di- To be eligible, study participants were required to have fe- agnostic specimen were also incorporated into the analysis. ver (body temperature $37.8°C or $37.2°C for patients A stepwise logistic regression analysis was performed to de- $65 years old in NAIA/B3002 [the protocol report num- termine which baseline symptoms and patient characteris- ber]) or a symptom of feverishness, plus at least 2 of the tics best predict influenza infection. The stepwise procedure following influenzalike symptoms: headache, myalgia, selected the best explanatory variable into the model first, then cough, or sore throat. Feverishness was defined as the pa- selected the second best explanatory variable, and so forth. tient’s subjective symptom of feeling like they had a fever The stepwise procedure stopped selecting additional vari- or chill. Specific influenzalike symptoms required for pa- ables when they did not reach statistical significance at the tient enrollment into each study are also summarized by a=.05 level, given the other variables already in the model. study in Table 1. The stepwise logistic regression model was tested for good- A requirement for a study site to begin patient ness of fit to determine the most parsimonious model. Odds enrollment was the identification of at least 2 individuals ratios and their 95% confidence intervals were calculated for with culture-confirmed influenza within a 7-day period each variable in the logistic regression model. Measures of prior to enrollment, who resided in the geographic positive predictive value (PPV), negative predictive value, region around the site. Geographic region was defined as sensitivity, and specificity were calculated to also identify the the area within a 50-mile radius of the study location. best predictors of an influenza infection. Positive predictive This requirement was to improve the probability that values were compared for individual symptoms, as well as influenza was actually present in the geographic region combinations of symptoms (SAS Statistical Analysis Soft- of the study. ware, version 6.12; SAS Institute Inc, Cary, NC). RESULTS model, with fever (odds ratio=3.26; P,.001) and cough (odds ratio=2.85; P,.001) as the 2 best explanatory vari- A total of 3744 clinical trial participants with influenza- ables. Headache, myalgia, and high-risk status were poor like symptoms were enrolled during the fall and/or winter explanatory variables (P..05) and were not selected into of 1994 through 1998. Demographic information of study the model. Feverishness was not included since fever (body participants, the percentage who had been vaccinated against temperature $37.8°C) was already in