Customize your JAMA Network experience by selecting one or more topics from the list below.
It is difficult to diagnose influenza infection on clinical grounds alone. Available rapid diagnostic tests have limited sensitivities.
To develop a prediction model that identifies children likely to have influenza infection.
Emergency department of a children's hospital.
All patients with a febrile respiratory illness during the influenza season of winter 2002 were eligible. A prospective sample of 128 children who were suspected of having influenza infection based on predetermined criteria was enrolled. Each patient received a nasal wash for viral culture.
Main Outcome Measure
Clinical features that are most predictive of influenza infection in children.
The mean ± SD age of patients was 6.2 ± 5.2 years; 50% were boys. Viral isolates included the following: influenza A, 45 patients (35%); influenza B, 13 (10%); other viruses, 10 (8%); negative results, 60 (47%). Demographic and clinical findings were not significantly different between the influenza A and influenza B groups. Cough (P = .003), headache (P = .04), and pharyngitis (P = .04) were independently associated with influenza infection. This triad used as a prediction model for influenza infection had a sensitivity of 80% (95% confidence interval [CI], 69%-91%); specificity, 78% (95% CI, 67%-89%); and likelihood ratio for a positive viral culture for influenza, 3.7 (95% CI, 2.3-6.3). The posttest probability of this clinical definition is 77% (95% CI, 63%-91%).
The triad of cough, headache, and pharyngitis is a predictor of influenza infection in children.
Influenza is a common febrile illness with a significant impact on the pediatric population. During annual outbreaks, 15% to 20% of children are infected with influenza.1,2 Estimates of annual outpatient visits attributable to influenza range from 6 to 29 per 100 visits.3,4 School absenteeism, parental work absenteeism, and secondary illness among family members are all significantly higher during influenza season than throughout the rest of the winter.5 Infants and young children are hospitalized for influenza-associated illnesses at rates comparable with those of high-risk adults and elderly patients.3,6 Furthermore, influenza is commonly implicated as a cause of nosocomial infections in pediatric inpatient units.7,8
The nonspecific presentation of influenza infection makes it difficult to distinguish from other febrile or respiratory illnesses. Many clinical features have been associated with influenza illness in children, but no characteristic symptom or symptom complex has been identified. Reported manifestations of influenza in children include abrupt onset of high fever, coryza, cough, sore throat, vomiting, diarrhea, abdominal pain, fatigue, headache, and myalgias. The classic symptoms often associated with influenza in adults are not easily identified in children.
The gold standard for diagnosing influenza infection is viral isolation by culture. However, viral culture results are not available in time to influence patient management. The rapid tests available to diagnose influenza are limited by their diagnostic abilities. A review of the studies performed on commercially available rapid tests reveals great variability in the sensitivities (40%-100%), specificities (63%-100%), positive predictive values (PPVs) (43%-100%), and negative predictive values (56%-100%) of these tests. The higher sensitivities are often difficult to reproduce in clinical practice, and false-negative results are likely to occur.9 As predictive values are affected by disease prevalence, these diagnostic tests are most helpful during periods of high influenza activity. However, it seems impractical and inefficient to test each patient with suspected influenza infection, even during peak season. The diagnostic abilities of these tests can be enhanced if patients with a high likelihood of influenza infection are selected for testing based on a set of clinical criteria. Differentiating influenza infection from infection caused by other respiratory pathogens is important for several reasons. From a public health perspective, understanding the magnitude of illness attributable to influenza is crucial for surveillance data across communities and states. Local and national surveillance is useful for predicting health care resource needs (hospital beds, staffing, supply of diagnostic tests and antivirals needed) and for detecting pandemics.10 Furthermore, children are considered an important factor in this process by introducing infection into the home and transmitting it to adult and elderly patients.1 Timely diagnosis is imperative for the initiation of appropriate antiviral treatment and for the proper isolation of hospitalized patients. Both of these interventions may aid in reducing the overall disease impact. These many factors argue for enhancing the clinical prediction of this disease.
The purpose of this study was to identify a clinical prediction model for influenza infection in children.
Design and definitions
This prospective study was conducted during the 2002 influenza season (January-March). All patients (birth through age 17 years) who were seen in the emergency department of a suburban tertiary pediatric center with a febrile respiratory illness were eligible. A consecutive sample of children suspected of having influenza infection were enrolled. Enrollment criteria included fever and at least one of the following symptoms: coryza, cough, headache, sore throat, or muscle aches. Fever was defined as an emergency department temperature higher than 38°C or history of similar temperature within the previous 24 hours. All enrolled patients received a nasal wash for viral culture. At the time of evaluation, a standardized data collection form was completed by the physician for each enrolled patient. Demographic information, duration of illness, day care attendance, and triage vital signs were recorded for each patient. The presence or absence of 24 clinical features (12 historical features and 12 physical examination findings) was noted on the same form. The historical features were recorded as dichotomous variables as relayed by the caregiver at the time of evaluation. These features included fever, rhinorrhea, cough, vomiting, diarrhea, decreased oral intake, decreased activity, headache, abdominal pain and/or nausea, sore throat, muscle aches, and apnea. Pertinent physical examination findings consisted of coryza (rhinorrhea and/or nasal congestion), cough (observed during examination), wheezing, rales, retractions, nasal flaring, rash (other than chronic dermatitis), conjunctivitis (conjunctival injection and/or discharge), pharyngitis (erythema, swelling and/or exudate of the tonsils and/or pharynx), cervical adenopathy (enlarged, tender anterior cervical nodes without evidence of acute suppurative adenitis), otitis media (abnormal tympanic membrane appearance and decreased mobility), and abdominal tenderness.
This study was approved with consent waiver by the institutional review board of our institution (Alfred I. duPont Hospital for Children).
The beginning of influenza season was defined as the period following the detection of 2 consecutive isolates in one week of either influenza A or influenza B from viral culture in the community. The season ended with the identification of the last of 2 consecutive isolates in one week. Patients were enrolled from January 14, 2002, through March 29, 2002.
Viral culture specimens from nasal washes were inoculated onto human lung tissue (MRC-5) cells, rhesus monkey kidney cells, human embryonic kidney (HEK) cells, human epidermoid laryngeal carcinoma (Hep-2) cells, and human lung carcinoma (A-549) cells. These cells were incubated at 34.5°C and observed daily for 28 days for cytopathic effect. All viruses isolated by culture were confirmed by immunofluorescence staining.
The primary outcome measures were the clinical features that are most predictive of influenza infection in children.
Interobserver agreement was evaluated for each of the clinical examination findings. Two physicians performed separate examinations and recorded their observations independently. Data was analyzed for agreement rates, and by κ statistics.
Sample size was calculated to detect a 30% difference between assumed prevalence of influenza in tested patients (a 50-50 chance, or 50%) and the hypothesis that a prediction model will possess a sensitivity that is approximately equal to rapid test (80%). After continuity correction, a sample size of 45 patients in each of 2 groups representing influenza-positive and influenza-negative patients was estimated.
Demographic and clinical findings of patients with influenza A and influenza B were compared with each other using 2 × 2 contingency tables, independent sample t test, or Mann-Whitney U test, depending on their parametric distribution. A univariate analysis was then performed on all influenza patients as a group compared with those patients who had no viral pathogen isolated. A binary logistic regression analysis using backwards stepwise elimination was performed to identify variables independently associated with influenza infection. Fever greater than or equal to 39°C was entered as a dichotomized variable. To simplify the analysis, the remainder of the vital signs were not included. All tests were 2-tailed and considered significant at P ≤.05. Statistical analyses were performed using SPSS version 11.5 (SPSS Inc, Chicago, Ill).
Demographics and virologic test results
During the study period, influenza tests were ordered on 130 patients who were suspected of having influenza infection based on a predetermined set of criteria. Two patients were excluded because they had no viral culture performed due to inadequate samples. Of the 128 eligible patients, 45 (35%) had viral cultures positive for influenza A and 13 (10%) for influenza B. Ten patients (8%) had other viral pathogens isolated (respiratory syncytial virus, 6; adenovirus, 3; rhinovirus, 1), and 60 specimens (47%) were negative. Study patients had a mean ± SD age of 6.2 ± 5.2 years and an equal sex distribution (50% male).
Clinical characteristics of influenza
There was no significant difference between patients with influenza A and influenza B in demographic or clinical findings (P ≥.05 for all variables). Univariate analysis comparing all influenza patients with the viral-negative group detected a significant difference in their respiratory rates (28 vs 21 breaths/min; P = .03). The remainder of vital signs was not significantly different. However, the influenza patients were more likely to have temperature greater than or equal to 39°C (57% vs 37%; P = .03) (Table 1). History of cough (odds ratio [OR], 2.2; 95% confidence interval [CI], 1.0-5.2), headache (OR, 2.6; 95% CI, 1.2-5.8), abdominal pain and/or nausea (OR, 3.2; 95% CI, 1.2-8.2), and clinical evidence of pharyngitis (OR, 2.0; 95% CI, 0.9-4.3) were more common in the influenza group (Table 2).
The predictive model
Backwards stepwise logistic regression analysis identified cough (adjusted OR, 7.2; 95% CI, 2-27; P = .003), headache (adjusted OR, 4.3; 95% CI, 1-17; P = .04), and pharyngitis (adjusted OR, 3.9; 95% CI, 1-14; P = .04) as independent predictors of influenza infection in children. Hosmer-Lemeshow goodness-of-fit test results were not significant (χ2, 10.2; P = .25), which statistically supports the robustness of this prediction model. The triad of cough and headache by history and clinical finding of pharyngitis used as a prediction model for influenza had a sensitivity of 80% (95% CI, 69%-91%), a specificity of 78% (95% CI, 67%-89%), a PPV of 77% (95% CI, 61%-88%), a negative predictive value of 81% (95% CI, 70%-92%), a likelihood ratio for positive viral culture findings for influenza of 3.7 (95% CI, 2.3-6.3), and a likelihood ratio for negative culture findings of 0.26 (95% CI, 0.14-0.44). The posttest probability of this clinical tool is 77% (95% CI, 63%-91%).
Fifty-six pairs of observations were recorded. The agreement rates of the variables in the model were as follows: cough, 88% (κ = 0.6); headache, 88% (κ = 0.8); and pharyngitis, 76% (κ = 0.5).
Influenza infection is a substantial cause of health care utilization and subsequent morbidity in the pediatric population. The wide range of influenza-associated symptoms in children often makes the clinical diagnosis difficult. The diagnosis is further complicated by other respiratory viruses that cocirculate with influenza and cause similar symptoms. Because influenza is potentially preventable by proper isolation and treatable by antiviral agents, timely diagnosis is important for reduction of disease spread and management of individual cases.
Traditionally, the laboratory diagnosis of influenza is made by isolating the virus in culture. However, this standard is impractical in clinical practice because of the time required to obtain results. Rapid diagnostic tests are available to aid in diagnosis. Their use is limited by their variable availability in health care centers and their relatively low sensitivity compared with viral culture.10 In practice, most physicians rely on clinical symptoms to diagnose influenza infection. However, studies comparing the accuracy of clinical diagnosis with laboratory-confirmed influenza reveal a wide range of PPVs (18%-87%).11-16 Several of these studies in the literature on adult patients include children in their methods; however, we are aware of no study that focuses exclusively on children.
In our study, cough, headache, and pharyngitis were significantly associated with positive viral culture findings for influenza in children. The sensitivity of this model is comparable with the rapid influenza tests (80%; 95% CI, 69%-91%). Additionally, the likelihood ratio is 3.7 (95% CI, 2.3-6.3). Likelihood ratios are a powerful clinical tool that help clinicians determine how much more likely a test result is to occur in someone with the disease than in someone without the disease. The higher the likelihood ratio, the better the test result is for ruling in a diagnosis. This information is used with the pretest probability to estimate the posttest probability (or PPV) of disease for that patient. Based on the gold standard of viral culture, this model had a posttest probability of 77% (95% CI, 63%-91%). This is a significant improvement over our physicians' pretest probability of 45% using an inclusive list of influenza-associated symptoms.
Adult studies have identified the complexes of fever and cough,11-13 or fever, cough, and abrupt onset,14,15 as the best clinical predictors of influenza infection. Although the difference in temperatures between our 2 groups of patients neared statistical significance (39.1°C in the influenza group vs 38.7°C in the negative group; P = .06), the difference is clinically negligible. This difference in the mean temperature could explain the higher respiratory rates in the influenza patients. Myalgias, which are classically associated with influenza infection, were not a clinical predictor in either our model or any of the quoted adult definitions. Notably, the triad of cough, headache, and pharyngitis is simple to elicit clinically, and each feature had a high agreement rate between observers.
A recognized limitation of this study is that it was conducted in a single setting during a single influenza season. The data were collected in a suburban tertiary care hospital, and the results may not be generalizable in all patient populations. Additional studies are needed at different sites during another influenza season to validate this model prospectively.
In conclusion, the clinical triad of cough, headache, and pharyngitis accurately predicts influenza infection in febrile children during a community outbreak. When influenza is circulating in the community, this pediatric prediction model can be used when considering management options for affected children. Furthermore, this model is likely to improve the diagnostic abilities of rapid tests by identifying a subset of patients with a high likelihood of influenza infection to be tested.
What This Study Adds
Influenza is a frequent cause of outpatient visits, hospitalizations, and nosocomial infections in children. It is often difficult to distinguish influenza from other febrile or respiratory illnesses on clinical grounds alone. The classic symptoms of influenza observed in older children and adults are not easily identified in young children. The rapid diagnostic tests available to aid in the diagnosis of influenza are limited by their sensitivities. Previous studies have identified clinical predictors of influenza, but none have focused on infection in children. Our study found that cough, headache, and pharyngitis were significantly associated with influenza infection in children. The sensitivity of this pediatric model is comparable with that of rapid influenza tests, and the 2 may be used in conjunction to improve clinical decision making.
Corresponding author and reprints: Marla J. Friedman, DO, Division
of Emergency Medicine, Miami Children's Hospital, 3100 SW 62nd Ave, Miami,
FL 33155-3009 (e-mail: firstname.lastname@example.org).
Accepted for publication December 11, 2003.
This study was presented in part at the Pediatric Academic Societies
Emergency Medicine Special Interest Group; May 3, 2003; Seattle, Wash; and
at the Society for Academic Emergency Medicine National Meeting; May 29, 2003;
Friedman MJ, Attia MW. Clinical Predictors of Influenza in Children. Arch Pediatr Adolesc Med. 2004;158(4):391–394. doi:10.1001/archpedi.158.4.391
Coronavirus Resource Center