When should a diagnosis of influenza be considered in adults requiring intensive care unit admission? Results of population-based active surveillance in Toronto

Introduction There is a paucity of data about the clinical characteristics that help identify patients at high risk of influenza infection upon ICU admission. We aimed to identify predictors of influenza infection in patients admitted to ICUs during the 2007/2008 and 2008/2009 influenza seasons and the second wave of the 2009 H1N1 influenza pandemic as well as to identify populations with increased likelihood of seasonal and pandemic 2009 influenza (pH1N1) infection. Methods Six Toronto acute care hospitals participated in active surveillance for laboratory-confirmed influenza requiring ICU admission during periods of influenza activity from 2007 to 2009. Nasopharyngeal swabs were obtained from patients who presented to our hospitals with acute respiratory or cardiac illness or febrile illness without a clear nonrespiratory aetiology. Predictors of influenza were assessed by multivariable logistic regression analysis and the likelihood of influenza in different populations was calculated. Results In 5,482 patients, 126 (2.3%) were found to have influenza. Admission temperature ≥38°C (odds ratio (OR) 4.7 for pH1N1, 2.3 for seasonal influenza) and admission diagnosis of pneumonia or respiratory infection (OR 7.3 for pH1N1, 4.2 for seasonal influenza) were independent predictors for influenza. During the peak weeks of influenza seasons, 17% of afebrile patients and 27% of febrile patients with pneumonia or respiratory infection had influenza. During the second wave of the 2009 pandemic, 26% of afebrile patients and 70% of febrile patients with pneumonia or respiratory infection had influenza. Conclusions The findings of our study may assist clinicians in decision making regarding optimal management of adult patients admitted to ICUs during future influenza seasons. Influenza testing, empiric antiviral therapy and empiric infection control precautions should be considered in those patients who are admitted during influenza season with a diagnosis of pneumonia or respiratory infection and are either febrile or admitted during weeks of peak influenza activity.

Many observational cohort studies, both from the 2009 pandemic and of seasonal influenza pre-pandemic, have found that antiviral therapy for influenza is associated with significantly improved outcomes, particularly when it is initiated within 48 hours of the onset of symptoms [6][7][8]. Optimal management of severe influenza thus depends on the ability to recognize those patients admitted to the ICU who require empiric therapy for influenza pending the results of diagnostic testing. However, data about clinical characteristics that help to identify patients at high risk of influenza infection upon hospital or ICU admission during influenza season are sparse [9,10]. The aim of this study was to identify populations of patients with increased probabilities of influenza infection among subjects admitted to ICUs during the 2007/2008 and 2008/2009 influenza seasons as well as the second wave of the 2009 H1N1 influenza pandemic.

Setting and manoeuvre
The Toronto Invasive Bacterial Diseases Network (TIBDN) is a collaborative network of microbiology laboratories, infection control practitioners and public health departments that performs population-based surveillance for infectious diseases in south-central Ontario [11][12][13]. Six acute care hospitals from the TIBDN participated in active surveillance for laboratory-confirmed influenza requiring ICU admission during the 2007/2008 and 2008/2009 influenza seasons, and three of these hospitals performed active surveillance during the second wave of the pH1N1 influenza pandemic. All admissions to adult medical or medical/surgical ICUs were included.
Prior to the 2007/2008 influenza season, attending physicians agreed that, during influenza seasons, nasopharyngeal (NP) swabs were clinically indicated in patients requiring ICU admission who presented with any acute respiratory or cardiac illness (independent of body temperature) or in patients with any febrile illness without a clear, nonrespiratory aetiology. During each influenza season, study staff screened all admissions daily and suggested orders for NP swabs (if they had not already been ordered) from all patients with any acute cardiac or respiratory illness or any febrile illness without a clear nonrespiratory source. Demographic and medical information was collected from each patient by chart review. Fever upon ICU admission was defined as being present if the first body temperature measured after ICU admission was ≥38.0°C, and the diagnosis was defined as recorded in each chart. Respiratory symptoms were defined as any upper or lower respiratory symptoms such as coryza, cough, wheezing or shortness of breath. NP swabs were tested for the presence of influenza by PCR and viral culture at the Ontario Public Health Laboratory. Some specimens were also tested by direct fluorescent antigen detection (DFA) or enzyme immunoassay (EIA) at individual hospitals.
For the purposes of the study, influenza season was defined as starting when the proportion of positive influenza tests among specimens submitted to the Ontario Public Health Laboratory for viral testing was >5% for two consecutive weeks and as ending when the proportion of positive tests was <5%. The peak season was a priori defined as any week in which the proportion of submitted specimens yielding influenza was >15% [10].

Statistical analyses
Data were double-entered, cleaned and analyzed using SAS version 9.1 software for PC (SAS Institute, Cary, NC, USA). Data were analyzed for all patients who were tested for influenza infection. Differences in medians were analyzed using the Wilcoxon rank-sum test or the Kruskal-Wallis test, and differences in group proportions were assessed using a χ 2 test or Fisher's exact test as appropriate. We performed multivariable logistic regression analyses to evaluate independent predictors of seasonal and pandemic H1N1 influenza. The variables 'age ≥65 years', 'temperature ≥38.0°C upon admission', 'admitting diagnosis respiratory infection', 'admitting diagnosis respiratory failure' and 'week with >15% specimens positive' were considered for inclusion in multivariable models based on clinical judgment and previously published literature [10] in a manner that minimized the Akaike Information Criterion, with final models representing those that best balanced parsimony and fit [14]. The Hosmer-Lemeshow Goodness-of-Fit Test was used to assess model fit. The limited number of outcomes was factored in when building the models to prevent overfitting. Likelihoods were calculated as binomial proportions with 95% confidence intervals. Two-sided P values < 0.05 were considered statistically significant.

Ethics approval
The study was approved by the Research Ethics Boards of all participating hospitals. Written informed consent was obtained as required from all participants or their authorized representatives.

Results of surveillance
During the 2007/2008 and 2008/2009 influenza seasons and the second wave of the 2009 H1N1 pandemic, a total of 5,462 patients were admitted to participating ICUs. Among these patients, 2,416 patients were tested for influenza and 2,360 (97.7%) met the eligibility criteria for testing ( Figure 2). Compliance with testing differed among the three influenza seasons (90.4% of eligible patients tested during the 2009 H1N1 influenza pandemic, 89.4% during the 2008/2009 influenza season and 63.0% during the 2007/2008 influenza season; P < 0.001) and among the six study hospitals (proportion of eligible patients tested ranging from 63.9% to 90.4%; P < 0.001). Specimens from eligible patients were more likely to be submitted if the patient was febrile (80.2% vs. 74.6%; P = 0.035) or reported respiratory symptoms upon admission (78.0% vs. 73.3%; P = 0.006), if the patient was >65 years of age (76.7% vs. 72.4%; P = 0.007), if admission did not occur during peak influenza weeks (77.6% vs. 71.8%; P < 0.001) and if the admission diagnosis was 'pneumonia', 'other respiratory infection', 'asthma exacerbation', 'chronic obstructive pulmonary disease (COPD) exacerbation' or 'respiratory failure' (82.7% vs. 72.9%; P < 0.001).
A total of 126 patients (5.2% of those tested) were identified as being infected with influenza.      Table 2). Analyses according to seasonal influenza A subtypes (influenza A (H1N1) vs. influenza A(H3N2)) could not be performed due to the low numbers of subtyped isolates.

Predictors of seasonal and pandemic (H1N1) 2009 influenza infection in patients admitted to ICUs
In multivariable analysis, body temperature ≥38.0°C and an admission diagnosis of 'pneumonia/other respiratory infection' were independently associated with both seasonal and pH1N1 influenza (Table 3). An admission diagnosis of 'COPD exacerbation/asthma exacerbation/ respiratory failure' independently predicted seasonal influenza but not pH1N1 infection. Of note, the lack of predictive ability of this variable for pH1N1 was independent of patient age. Age <65 years was independently associated with pandemic (H1N1) 2009 influenza but not with seasonal influenza.
Percentage of patients with seasonal and pandemic (H1N1) 2009 influenza in different patient populations Table 4  the percentage of patients with pH1N1 was elevated more than twofold in afebrile patients admitted with 'pneumonia' or 'other respiratory infection' and more than fivefold if these patients were admitted with a fever ≥38.0°C. However, a considerable fraction of patients with influenza were not in high-risk groups, particularly in the case of seasonal influenza. During influenza seasons, patients with 'pneumonia', 'other respiratory infection', asthma exacerbation', 'COPD exacerbation' or 'respiratory failure' admitted during peak weeks comprised only 39% of all patients with influenza. During the pandemic, patients with an admission diagnosis of 'pneumonia' or 'other respiratory infection' comprised 59% of all patients with influenza (Table 4).

Discussion
The findings of our study may assist clinicians in decision making regarding influenza testing and empiric antiviral treatment of adult patients admitted to ICUs during future influenza seasons.  An increasing number of studies suggest that specific antiviral therapy is effective in reducing the morbidity and mortality associated with influenza [18][19][20][21], and current guidelines for both seasonal and pandemic influenza recommend the treatment of patients requiring hospitalization for influenza [6,7,22]. However, it is also clear that antiviral therapy is more effective when started in the early stages of infection [22][23][24][25][26][27][28]. Thus, optimal management of patients dictates the use of empiric therapy, and understanding the pretest probability of infection is necessary for its rational use. The  differences between the positive predictive value for influenza in patients in peak and off-peak influenza weeks during influenza seasons emphasize the need for intensivists to know whether influenza is circulating in their area and where they are in each "season" when deciding on testing and treatment for influenza. As a rule of thumb, influenza seasons last from 10 to 16 weeks, with peak weeks being the central 4 to 6 weeks. The 2007/2008 influenza season in Toronto was an exception. The onset of the season varies by year and by geographic region within countries and is now usually declared locally by public health units in North America. The identification of patients shedding influenza virus is also important for infection prevention, especially in ICUs, where influenza outbreaks may be fatal [29,30]. Hospital influenza outbreaks are not uncommon [31], and expert opinion suggests that additional precautions for patients admitted with influenza are an important element of control of transmission of influenza in ICUs [22,32,33]. Our data support the implementation of additional precautions empirically until influenza is ruled out for all patients admitted with pneumonia or other respiratory infection if they are febrile or are admitted during weeks of peak influenza activity.
RT-PCR is the only diagnostic test with adequate performance characteristics for the diagnosis of influenza in adult patients requiring hospital admission [9,22]. As shown previously by other groups, we have confirmed that RT-PCR is more sensitive than viral culture or DFA for rendering an influenza diagnosis, and these tests are, in turn, more sensitive than commercially available rapid influenza tests (that is, EIA) [34][35][36].
There are a number of limitations to our study. First, sampling from three years in a single geographic area may limit the generalizability of our results. We minimized selection bias by applying broad inclusion criteria for testing: All patients who presented with acute respiratory or cardiac illness or with febrile illness without clear nonrespiratory aetiology were eligible for inclusion. However, only 75% of eligible patients were tested for influenza, suggesting that the proportion of patients actually infected might be higher than our estimate. Data collection by chart review limited the number of risk factors considered, including differences that might have been found between vaccinated and unvaccinated patients. In addition, rather than individual specific respiratory symptoms (for example, cough, shortness of breath), we assessed only respiratory symptoms overall. However, the factors we identified were selected to be easily available for all patients and have the value of simplicity. Finally, it has previously been shown that patients who are admitted to the hospital with influenza infection do not constitute a homogeneous group [37]. The proportion of patients who need specific antiviral therapy to control influenza infection is unknown, in contrast to those who are able to control viral replication but have complications as a result of influenza. There is a need for further study to define the potential of antiviral therapy and the causes of hospitalization due to influenza.

Conclusions
Among those adult patients who are admitted to the ICU during influenza season with a diagnosis of pneumonia or respiratory infection and who are either febrile or admitted during weeks of peak influenza activity, the probability of influenza infection may be high enough to warrant consideration of influenza testing, empiric antiviral therapy and/or empiric infection control precautions. However, although our simple rules may result in improved identification of patients with influenza infection, a significant proportion of patients who present without these characteristics will still be missed and our understanding of which patients will benefit from treatment remains incomplete. These persisting difficulties highlight the need for further research to enable better identification of patients admitted with atypical presentations, to understand when patients with influenza are infectious and to clarify which patients will benefit from antiviral therapy.

Key messages
• Optimal management of severe influenza depends on the ability to recognize those patients admitted to the ICU who require empiric therapy and additional precautions for influenza pending the results of diagnostic testing.
• Influenza testing, empiric antiviral therapy and empiric infection control precautions should be considered in the small proportion of patients admitted during influenza season with a diagnosis of pneumonia or respiratory infection and who are either febrile or admitted during weeks of peak influenza activity.
• Although identification of patients may be improved with the application of this simple rule, a significant proportion of patients with influenza infection will be missed. Further research is needed with regard to strategies for improved identification of influenza patients admitted with atypical presentations. laboratories, as well as the patients, family members and physicians who make TIBDN surveillance possible. Active surveillance for influenza within the TIBDN was an investigator-initiated study funded by an unrestricted contract with Hoffman-La Roche Ltd. SPK is supported by the Swiss National Science Foundation (PBZHP3-125576). The funding organizations had no influence on the design and conduct of the study; the collection, management, analysis and interpretation of the data; or the preparation, review or approval of the manuscript. These data were presented in part at the 2010 Interscience Conference on Antimicrobial Agents and Chemotherapy (ICAAC), Boston, MA, USA (abstract V-427). Part of the 2007/2008 ICU influenza surveillance data have been included in an analysis of influenzaassociated hospitalizations in adults during the 2007/2008 influenza season in Toronto [10].