This study compared the predictive values of demographic and clinical factors for laboratory-confirmed influenza during the 2007–2008 influenza season in Southern California among different age groups and evaluated the performance of resulting case definitions. Marked variations across age groups were found with robust predictors among 6- to 49-year-olds and few predictors among younger children and older adults. No single case definition was found to have both a sensitivity and specificity of clinical utility, even in an age group exhibiting the strongest predictors; the CDC ILI case definition would have excluded 62% of influenza cases in this study population. With no evidence for a diagnostically useful case definition (used alone) and the potential unreliability of RIDTs [12, 36], new diagnostic strategies are urgently needed.
In our study population, 6- to 49-year-olds had the highest influenza rate (differences not statistically significant), with strong associations found between laboratory-confirmed influenza and acute onset, fever, cough, body aches, and unvaccinated status, consistent with prior studies [16, 17, 21, 23, 25, 37]. Cough had the strongest association with influenza. Influenza positive 6- to 49-year-olds had approximately three times the odds of being unvaccinated (n = 141, 89.8%) compared to negative subjects (n = 270, 73.8%) (p = .001), suggesting a moderate protective effect of the vaccine for this season in this age group. No demographic or clinical predictors of influenza were found among 3- to 5-year-olds, consistent with findings in similar studies . The difficulty of eliciting a symptomatic history when the subject has limited language skills may contribute to this phenomenon . Vaccination appeared strongly protective in this age group. A (non-significant) lower rate of influenza in this group conflicts with the common belief that young children exhibit a higher incidence of influenza, higher viral loads and higher persistence of the virus. However, CDC data from outpatients with ILI in 12 states during the 2010–2011 influenza season suggests that the rate of laboratory-confirmed influenza was lower among children < 5 years old compared to 5- to 17-year-olds . Among 50- to 80-year-olds, only fever was associated with influenza, consistent with prior studies failing to identify reliable clinical predictors among older adults [20, 24]. No protective effect from vaccination was apparent in this group, as might be expected with age-related waning immunity , although denominators were small.
Together, these findings substantiate previous observations that during a community influenza epidemic, acute onset of fever and cough in young to middle-aged adults are the best individual clinical predictors for laboratory-confirmed influenza [17, 21], with young children demonstrating the fewest predictors . Differing influenza detection rates according to specimen type collected warrants further study.
The all-4-criteria clinical case definition (requiring acute onset, fever, cough, and body aches) had a strong specificity among 6- to 49-year-olds (98.6%), but missed 85% of influenza cases. The at-least-2-criteria test (subject must have at least two of the four criteria) captured 94.9% of cases, but only excluded influenza in 20.2% of noncases; 33.8% of test positives had influenza in this group (overall prevalence similar to that of a typical influenza epidemic). The clinical tests defined by exactly two or three criteria likewise performed poorly, though roughly half had higher sensitivity than the CDC ILI case definition (40.1%) in this age group. The CDC ILI definition’s sensitivity varied significantly across age groups, with the poorest performance among 3- to 5-year-olds and the highest among 6- to 49-year-olds. Across the whole population the CDC ILI definition missed >60% of cases and had a PPV <60% in each age group. NPVs ranged from 73-100%, indicating that the absence of most of the symptoms modeled effectively predicted the absence of influenza infection; the at-least-1-criterion test, at-least-2-criteria test, and “cough + any of other 3 criteria” test could all be utilized to exclude influenza (NPV >90%), eliminating the need for a follow-up rapid or laboratory-based test.
All case definitions in Table 4 performed inferiorly to the typically cited sensitivities (50%–70%) or specificities (90%–95%) or both of RIDTs . However, a common RIDT was found to have a sensitivity as low as of 27% (95% CI, 19%–32%) for influenza during 2007–2008 ; recent studies of that same and other RIDTs report sensitivities 38%–69% in detection of pH1N1 [13–15] and 31%–63% against other strains . If RIDTs are less sensitive than previously cited, their optimal diagnostic utility might be among groups with higher influenza prevalence (eg, patients meeting a sensitive clinical case definition like the at-least-2-criteria clinical test during a local epidemic) with confirmatory testing following negative RIDT results .
Vaccination status was excluded in this diagnostic modeling since the effect of vaccination is likely to change annually with variable strain matching. Clinicians should take vaccination status into account when making a clinical diagnosis only once the performance of the vaccine has been established for that region and season.
Influenza type/subtype analysis yielded unexpected results. The meaning of the associations of influenza A with history of smoking and influenza B with increasing age is not clear. Researchers previously reported the two influenza types might infect different age groups at different rates, but may cause a similar clinical syndrome across age groups . Differences in rates across age groups may be related to previous circulation of similar strains. Smoking increases with age among 6- to 49-year-olds; thus confounding is unlikely. Fever was associated with influenza A(H1) infection compared with A(H3). If fever correlates with a higher viral titer, it is not surprising that the initially unsubtypeable specimens were A(H3)-positive. However, if fever reflects more severe or systemic disease, this finding contradicts previous reports suggesting influenza A(H3) was associated with such illness [16, 41]. We are unaware of a physiologic mechanism through which smoking would increase the risk of influenza A(H3), or decrease the risk of A(H1), which our results suggest, but this merits further investigation.
This study’s major strengths include a methodology facilitating comparison to similar studies and inclusion in meta-analyses and an analysis of age group and influenza type/subtype effects. The surveillance population studied provides an ideal setting for this analysis: a large outpatient population at the point of care, wide age range and socioeconomic spectrum, broad and uniform inclusion criteria (resulting in symptomatic diversity), and robust laboratory influenza diagnosis with type and subtype. Type distributions were similar to national (71% influenza A and 29% influenza B)  and local  rates in the same season. The military dependent population studied is similar to the general US population, though dependents may be healthier due to universal access to medical care.
Study limitations include omission of subjects ages <3 years and incomplete data on prior vaccination history in children—potential receipt of two doses in individuals ages ≤8 years was not documented. Other limitations include use of self-reported vaccination data, lack of antiviral or antipyretic use history, lack of information regarding comorbid conditions that increase risk of influenza or adverse outcomes from influenza, and asymmetry between influenza A and B testing methods. Subjects in this study may not be considered high risk for influenza complications, thus may not represent the population for whom antiviral therapy is clearly indicated for treatment of influenza infection. Clinical predictors of influenza and the performance of clinical case definitions may differ in a high risk population. An RT-PCR assay for influenza B was not utilized, thus true influenza B and overall influenza rates may be underestimated. We believe underdiagnosis of influenza B is more likely to create a type II error. As the clinical case definitions modeled here were created utilizing our entire data set, sensitivities and specificities could be overestimates of the true values; these models should be evaluated and validated in other populations and during other time periods. Last, we report on only one influenza season and cannot assess clinical disease variations over time resulting from antigenic drift.