Compared to the typical influenza season in HNE which occurs from June through October and peaks in August or September, the first wave of pH1N1 activity began and peaked earlier during a three week period in early July. A similar trend towards an earlier peak in the influenza season was also observed in the southern Australian state of Victoria [6]. Compared to the peak of pH1N1 community illness in HNE, the peak in pH1N1-associated hospitalizations lagged by 1–2 weeks which may be due in part to a delay between illness onset and presentation for medical care and varied transmission patterns influenced by social networks with initial transmission among healthy children and later transmission among more vulnerable populations. While the burden of community illness and pH1N1-associated hospitalizations in HNE increased rapidly resulting in clear peaks, pH1N1-associated deaths occurred more sporadically.

In this analysis, the pH1N1 community ILI attack rate ranged from 4–8%, similar to the estimated attack rate from New Zealand (8%)[7] which experienced the introduction of pH1N1 during the Southern Hemisphere influenza season. During May through August, 2009, outbreaks of pH1N1 infection occurred throughout the United States outside of the typical Northern Hemisphere influenza season, and several estimates of ILI attack rate were made. United States all-cause ILI attack rate estimates ranged from 5% for a four week period in 10 states[8] to 7% for a four week period in New York City [9]. In contrast to these estimates from the United States, our estimate was adjusted using virologic surveillance data suggesting that the ILI attack rate in HNE during the influenza season was higher than in parts of the United States where estimates were made when pH1N1 was circulating outside of the United States influenza season. Attack rates from our analysis, as well as those from most other analyses, underestimate the true pH1N1 attack rate, because asymptomatic infections are not included.

During the twentieth century, three influenza pandemics occurred starting in 1918, 1957, and 1968. Of the twentieth century pandemics, the 1918 pandemic was the most severe with estimated attack rates for the full pandemic period ranging from 20–60% and estimated case fatality ratios ranging from 2–3% [10], [11]. In New South Wales, the 1918 pandemic virus was estimated to result in 300 deaths per 100,000 persons [11]. In comparison, the 1957 and 1968 pandemics were milder with case fatality ratios estimated at less than 0.2% [10]. Our estimated community attack rate of 4–8% and estimate of approximately 1 death per 100,000 persons with a case fatality ratios of 0.009–0.02% suggest that compared to the 20^th century pandemics, the first wave of the current pandemic resulted in clinical infection in a smaller proportion of the population and in fewer severe outcomes. However, our estimates reflect only the first wave of symptomatic pH1N1 activity while estimates from prior pandemics frequently reflect the result of multiple waves of pandemic virus activity. Thus, it remains difficult to predict the overall disease burden that will result from pH1N1.

Our model for estimating pH1N1 disease burden in HNE has several strengths. First, because the Hunter New England community is a well- defined population with a limited number of hospitals where persons with acute respiratory illness are hospitalized, HNEPH was able to implement population-based surveillance for acute respiratory illness hospitalizations and deaths, eliminating the need to extrapolate estimates from a limited sample of the population. HNE hospitals also maintained a consistent catchment area allowing us to easily adjust for hospitalizations of non-HNE residents prior to calculating hospitalization and mortality rates. Second, as surveillance for community illness was internet-based and surveillance for hospitalizations and deaths was based on ICD-10 codes from electronic medical record systems, surveillance data was readily available allowing for a rapid estimate of disease burden and severity.

The limitations of our model reflect the constraints inherent in each of the surveillance systems used for estimating each measure of pH1N1 burden: detection of pH1N1, community illness, hospitalizations, and deaths. First, both sources of laboratory data were based on specimens collected and sent at the discretion of the evaluating physician, and thus were affected by clinician testing practices. In particular, the NSW laboratory data used to estimate the incidence of pH1N1 in the community may have been affected by the shift in the Australian pandemic phase from CONTAIN to PROTECT on June 17, 2009 [12], after which clinicians were advised to focus testing on hospitalized patients and on persons with characteristics conferring a higher risk for severe influenza. However, given the number of specimens submitted for testing at the two NSW reference laboratories and five pathology services compared to the reported number of hospitalizations in NSW [3], it is likely that the majority of specimens were taken from non-hospitalized persons. We also assumed that the incidence of pH1N1 infection was similar among persons with respiratory illness who sought outpatient medical care and among persons with respiratory illness who did not seek medical care. Second, since the majority of participants in the FluTracking surveillance system are employees of the HNE Area Health Service, FluTracking participants may differ from the general HNE population with respect to socioeconomic status and educational background which would affect our estimates of the burden of pH1N1-associated community ILI if these factors were associated with influenza transmission. Third, our estimates of pH1N1-associated hospitalizations and deaths are based on surveillance for hospitalizations and deaths using ICD-10 codes for respiratory illness which are less likely to capture persons with pH1N1 who have cardiovascular presentations or exacerbations of underlying illnesses. An analysis of ICD-10 codes assigned to reported, confirmed case-patients with pH1N1-associated hospitalizations admitted from HNE EDs found that 50% of reported cases were captured by one of the surveillance ICD-10 codes, while the remaining 50% of cases were assigned a broad range of codes [unpublished data], making it difficult for syndromic surveillance to capture these undetected cases while achieving adequate specificity. Surveillance ICD-10 codes also were selected by retrospective review of codes used during prior influenza seasons when coding practices may have been different than during the current pandemic. Lastly, it should be noted that virologic data from ED patients and hospitalized patients was used to calculate our estimates of pH1N1-associated deaths. It is unclear how this might bias our estimates if the proportion of persons with pH1N1 differed among persons who were hospitalized with respiratory illness and those who died with respiratory illness.

This analysis was also unable to explore age-specific differences in disease burden because the number of hospitalizations and deaths was relatively small resulting in small numbers in each age stratum. Seasonal influenza community attack rates and hospitalization and mortality rates have been shown to vary substantially by age [13], [14], [15]. In addition, the age distribution of confirmed pH1N1 cases from many countries suggests that people aged 65 years and older may be at lower risk for pH1N1 infection, and are underrepresented among pH1N1-associated hospitalizations and deaths when compared to seasonal influenza [16], [17], [18]. In this analysis, approximately 20% of hospitalizations and 84% of deaths identified through ICD-10 surveillance occurred in persons aged 65 years and older. If the incidence of pH1N1 infection is lower among persons aged older than 65 years, then we may have over-estimated deaths, and possibly hospitalizations, by applying non-age-specific virologic data to hospitalizations identified through ICD-10 surveillance. However, despite this potential limitation, our estimated case fatality ratio is relatively low compared to prior pandemics and consistent with estimates from other countries [7], [19].