A Systematic Review of the History and Physical Examination to Diagnose Influenza

Methods

Results

We identified 93 studies in our initial survey Medline search, and 4 additional studies from the bibliographies of these studies. Of this group of 97 studies, 7 met our inclusion criteria.8–14 Study characteristics are shown in Table 1. All were independent cohort studies in the community or outpatient setting, largely primary care and during times of an influenza epidemic. None was explicit about blinding, but because of the usual timing of the HPE and reference laboratory tests, we assume that physicians performing the HPE were not aware of the results of the reference standard test.

The 7 studies reported the sensitivity and specificity for a total of 59 variables. Because there was no clear pattern regarding the accuracy of variables for influenza A versus influenza B, because it would reduce statistical power, and because this would introduce another layer of complexity into the analysis and presentation of data, we chose to combine studies of influenza A or B alone with those of influenza A and B.

Signs and symptoms with a positive likelihood ratio (LR+) greater than 2.0 or a negative likelihood ratio (LR−) less than 0.5 are shown in Table 2. Table 2 also summarizes the findings for all variables reported by more than 1 study, including the area under the ROC curve, when it could be calculated. Signs and symptoms with an LR− greater than 0.5 and an LR+less than 2.0 are generally not useful clinically; those with likelihood ratios falling in this range that were measured only by a single study are not reported and include abdominal pain, antipyretics before consultation, any lower respiratory symptom, any other symptom, asthenia, bronchiolitis, conjunctival injection, dry cough, earache, emesis, face ache, general practitioner consultation, gritty eyes, high risk condition, hoarseness, home visit by physician, lacrimation or conjunctival injection, loss of appetite, lower respiratory tract illness (age >65 years), male gender, moderate or severe fatigue, rhinorrhea, otitis media, pain on respiration, painful cervical adenopathy, received antibiotics, weakness, and wheeze. Monto8 reported on combinations of variables; only 2 combinations (fever, cough, and nasal congestion; fever, cough, and weakness) had a LR+ greater than 2.0.

The elements of the HPE that best ruled in influenza when present were rigors (LR +7.2), the combination of fever and presenting within 3 days of the onset of illness (LR +4.0), and sweating (LR +3.0). The signs and symptoms best able to rule out influenza when absent were having any systemic symptoms (LR −0.36), coughing (LR −0.38), not being able to cope with daily activities (LR −0.39), and being confined to bed (LR −0.50).

The area under the ROC curve could not be calculated for signs and symptoms or clusters of signs and symptoms reported by only a single study. The highest calculable areas under the ROC curve were 0.679 for cough, 0.672 for subjective temperature, 0.654 for nasal congestion, and 0.653 for objectively measured temperature.

Discussion

Physicians have traditionally used information about the presence or absence of an influenza epidemic in the community and certain signs and symptoms such as sudden onset of fever, cough, myalgias, and chills to diagnose influenza. Our systematic review identifies 3 variables that, when present, help rule in influenza (rigors, sweating, and fever and onset of symptoms less than 3 days before) and 4 additional symptoms that rule it out (no systemic symptoms, not coughing, being able to cope with daily activities, and not being confined to bed). Other commonly used symptoms such as sore throat, chills, and myalgias) were less useful.

A bias toward lower estimates of sensitivity and specificity may have been introduced by the fact that most studies only included patients with suspected influenza.8,9,11–13 For example, if fever was part of the inclusion criteria for a study, it will make it impossible for this variable to contribute to discriminating between patients with and without influenza. This bias particularly affects the estimates for fever, headache, myalgias, cough, and sore throat that were part of the inclusion criteria for the large Monto study.8

Our study had several limitations. The size of one study,8 a pooled analysis of the results of several randomized trials, meant that it often dominated the analysis. Any flaws in this study (eg, lack of blinding, an imprecise reference standard, selection bias) would therefore also dominate our analysis. There was considerable heterogeneity between studies, which is why we report the range as well as a summary measure of effect for sensitivity and specificity estimates based on data from more than one study. Finally, several of the variables that had the highest LR+ or lowest LR− came from a single study; again, any flaws in that study’s design would have an important impact on our findings.

The literature review was repeated just before publication (February 2004) and identified only one additional article. This article studied patients over age 65 or with underlying cardiopulmonary disease who were admitted to the hospital with a respiratory diagnosis; approximately 20% had influenza.15 Despite the highly selected nature of the group, these findings were similar to ours. The best predictor of influenza A was the combination of cough, temperature of 38°C or higher, and illness duration of 7 days or less (LR+ 2.9, LR− 0.3).

Previous studies have shown that individual signs and symptoms rarely include or exclude a disease.16,17 A more successful strategy is the use of several key symptoms in a clinical decision rule that stratifies patients into low-, moderate-, and high-risk groups. This information can be used in conjunction with the results of office laboratory tests and perhaps imaging studies to make a more accurate diagnosis while also limiting unnecessary testing and overtreatment. This strategy has been successfully implemented for sore throat, deep vein thrombosis, and pulmonary embolism.16–18 More than anything, this systematic review points out the need for well-designed studies in the primary care setting to develop and validate such a rule.