Date Published: April 16, 2013
Publisher: Blackwell Publishing Ltd
Author(s): Dena L Schanzer, Brian Schwartz, Michael J Mello.
Weekly influenza-like illness (ILI) consultation rates are an integral part of influenza surveillance. However, in most health care settings, only a small proportion of true influenza cases are clinically diagnosed as influenza or ILI. The primary objective of this study was to estimate the number and rate of visits to the emergency department (ED) that are attributable to seasonal and pandemic influenza and to describe the effect of influenza on the ED by age, diagnostic categories, and visit disposition. A secondary objective was to assess the weekly “real-time” time series of ILI ED visits as an indicator of the full burden due to influenza.
The authors performed an ecologic analysis of ED records extracted from the National Ambulatory Care Reporting System (NARCS) database for the province of Ontario, Canada, from September 2003 to March 2010 and stratified by diagnostic characteristics (International Classification of Diseases, 10th Revision [ICD-10]), age, and visit disposition. A regression model was used to estimate the seasonal baseline. The weekly number of influenza-attributable ED visits was calculated as the difference between the weekly number of visits predicted by the statistical model and the estimated baseline.
The estimated rate of ED visits attributable to influenza was elevated during the H1N1/2009 pandemic period at 1,000 per 100,000 (95% confidence interval [CI] = 920 to 1,100) population compared to an average annual rate of 500 per 100,000 (95% CI = 450 to 550) for seasonal influenza. ILI or influenza was clinically diagnosed in one of 2.6 (38%) and one of 14 (7%) of these visits, respectively. While the ILI or clinical influenza diagnosis was the diagnosis most specific to influenza, only 87% and 58% of the clinically diagnosed ILI or influenza visits for pandemic and seasonal influenza, respectively, were likely directly due to an influenza infection. Rates for ILI ED visits were highest for younger age groups, while the likelihood of admission to hospital was highest in older persons. During periods of seasonal influenza activity, there was a significant increase in the number of persons who registered with nonrespiratory complaints, but left without being seen. This effect was more pronounced during the 2009 pandemic. The ratio of influenza-attributed respiratory visits to influenza-attributed ILI visits varied from 2.4:1 for the fall H1N1/2009 wave to 9:1 for the 2003/04 influenza A(H3N2) season and 28:1 for the 2007/08 H1N1 season.
Influenza appears to have had a much larger effect on ED visits than was captured by clinical diagnoses of influenza or ILI. Throughout the study period, ILI ED visits were strongly associated with excess respiratory complaints. However, the relationship between ILI ED visits and the estimated effect of influenza on ED visits was not consistent enough from year to year to predict the effect of influenza on the ED or downstream in-hospital resource requirements.
The estimated rate of ED visits attributable to influenza was elevated during the H1N1/2009 pandemic period at 1,000 per 100,000 (95% CI = 920 to 1,100) population compared to an average annual rate of 500 per 100,000 (95% CI = 450 to 550) population for seasonal influenza. In comparison, rates for ILI ED visits increased from an average annual rate of 55 per 100,000 population to 464 of 100,000 during the H1N1/2009 pandemic. As seen in Figure 1A, total ED visits are generally lower during winter months compared to summer months, and spikes for weeks 52 and 1 are visible. So unless peak influenza activity aligns with weeks 52 and 1 (Christmas/New Year’s period in which ED visits normally increase), the excess due to seasonal influenza typically is not associated with a peak in weekly ED visits. The 2009 fall pandemic wave was a significant exception, with ED visits increasing to 1.3 times the usual peak levels, and influenza accounting for an estimated 30% of weekly ED visits at the peak (Figure 1A). Over the entire pandemic period (May 2009 to March 2010), H1N1/2009 accounted for only 3% of total ED visits.
This study confirms a significant effect of both seasonal and pandemic influenza on ED visits. As expected based on results from previous studies of the morbidity and mortality burden of influenza,6,34 ED visits with clinic diagnoses of influenza or ILI account for only a small portion of the estimated number of influenza-attributed ED visits. In addition, the proportion of excess ED visits attributed to influenza that were clinically diagnosed as influenza or ILI was not consistent from season to season and was considerably higher for the H1N1/2009 pandemic than for seasonal influenza (as illustrated in Figure 1D). The Canadian influenza surveillance system, FluWatch1 reported unusually high ILI consultation rates (physician visits per 1,000 patient visits) during the pandemic period as well. The higher proportion of ILI diagnoses among ED visits attributed to influenza during the pandemic may be related to a combination of factors: a lower threshold for patients seeking medical care during the pandemic compared to a normal influenza season due to fear of severe illness and increased identification of ILI by physicians in patients with nonspecific illness due to raised index of suspicion.12 Excessive use of the ED during the pandemic period was not apparent in this study, although this phenomenon has been identified elsewhere.35 Our analysis suggests that the latter, combined with differences in the virulence of the individual strains and the conformity of clinical symptoms to the ILI definition, likely contributed most to variation in the clinical diagnosis of ILI among persons presenting because of influenza infections. Complications often associated with influenza such as otitis media and asthma had low attributable fractions for influenza, and these weekly time series had distinct patterns not captured by the model, which suggests that viruses other than influenza are also responsible for a substantial portion of the morbidity.
Our approach to estimating the full burden of influenza has provided considerable insight not available using other methods, although this ecologic study design has several limitations. The main uncertainty stems from the use of proxy variables for the level of activity associated with influenza and other respiratory viruses and the ecologic nature of the study design. Earlier studies used weekly virologic data (influenza- or RSV-positive tests or percent positive) as proxies for the weekly level of influenza and RSV activity. With the introduction of ICD-10 coding, we can now replace the proxies based on virologic data with the number of laboratory confirmed influenza or RSV admissions to hospital,29 an option that improves the quality of the measures of weekly influenza and RSV activity. CIs were adjusted for the added uncertainty from any factors not captured by the model by including a scale parameter in the regression model.33 Although the estimated CIs should be suitable for the study population, generalization to other populations would introduce additional uncertainty. While limitations inherent with an ecologic study design are applicable to this study, multiple studies using this approach for the estimation of the influenza burden suggest that uncertainties are reasonably stated, and the use of statistical models has recently been recommended by the WHO as the preferred option for the estimation of the influenza burden. In most cases, an influenza infection would not protect against other emergencies, and a patient could arrive in the ED with both a nonrespiratory complaint and an unrelated acute influenza infection. While we estimated the background prevalence of influenza in the community to account for these events, the absenteeism rates used in this calculation likely underestimated the true clinical attack rate.
In this study, influenza had a larger effect on ED visits than was captured by clinical diagnoses of influenza or influenza-like illness, and the effect of these additional ED visits during periods of peak influenza activity was associated with an increase in the proportion of persons who registered but left without being seen. In addition, rates of ED visits attributed to influenza were highest for infants and children, while the proportion admitted to hospital was highest for persons aged 65 years or older. The effect described in this study suggests that there is potential for improvement with alternative health service delivery management models for influenza. Some models, however, would rely on a robust real-time influenza surveillance system that would provide enough advance notice to set up these specialized clinics. Throughout the study period, influenza-like illness ED visits were strongly associated with excess respiratory complaints, and these data may prove to be a useful component of such an alert system. However, the relationship between influenza-like illness ED visits and the estimated effect of influenza on ED visits was not consistent enough from year to year to predict the size of effect of influenza on the ED or downstream in-hospital resource requirements.