Policy makers who implement influenza vaccine programs or interventions to increase influenza vaccine coverage must first understand the epidemiology of influenza in the targeted population. For example, results of studies of severe influenza morbidity and mortality influenced the World Health Organization (WHO) to develop a policy to broaden influenza vaccination programs to cover pregnant women [1]. Similarly, epidemiologic data on the burden of pediatric influenza disease were instrumental in recent decisions by the United States and the United Kingdom to expand influenza vaccine recommendations to cover all children [1]–[3]. Most high-income, temperate countries have identified adults, particularly elderly adults, as being a high-risk group for influenza complications and death, and therefore a priority group for vaccine receipt [1], [2]. In the United States, over 23,000 adults, mostly elderly, are estimated to die from influenza and its complications yearly, though annual mortality estimates vary considerably from year to year [4]. Because little influenza surveillance is conducted in low-resource countries, it is unknown whether there is a similar disproportionate burden of adult influenza in these settings.

In an effort to standardize influenza surveillance and to increase influenza morbidity data collection from low resource settings, the WHO established guidelines to standardize influenza surveillance worldwide and developed tools to use such data to estimate the burden of severe influenza disease [5], [6]. The WHO initiative to standardize hospitalized influenza surveillance has focused on identification of patients with severe acute respiratory infections (SARI) in sentinel hospitals, with collection of epidemiologic data and clinical specimens for influenza-specific laboratory testing limited to these patients [6]. To date, published SARI surveillance studies use evidence of hospitalization for acute febrile, lower respiratory tract illness as the case definition for study entry and influenza testing [7]–[16]. However, influenza virus infection can precipitate a wide range of processes necessitating hospital care, and case definitions that require evidence of acute lower respiratory tract infection will overlook non-pneumonia complications of influenza, including exacerbations of asthma, chronic obstructive lung disease, and congestive heart failure [17].

The goal of this study was to assess the extent to which restricting influenza testing to adults hospitalized with pneumonia may underestimate the total burden of hospitalized influenza disease. We hypothesized that the burden of influenza disease was considerably greater than that captured by influenza pneumonia surveillance.

During the study period, there were 28,176,969 hospitalizations in the three states; 16,044,193 (57% of total) hospitalizations were for respiratory or circulatory outcomes, while 2,066,072 (7% of total) were for critical illness outcomes (Table 1). Almost all of the critical illness hospitalizations (94.1%) had ICD-9 codes for respiratory and circulatory diagnoses, while few respiratory and circulatory diagnoses had ICD-9 codes for critical illness (12.1%). Using multivariate binomial regression models, we estimated that there were 80,834 (95% CI 29,214–174,033) influenza-associated respiratory or circulatory hospitalizations and 26,760 (95% CI 14,541–47,464) influenza-associated critical illness hospitalizations (Table 1). We estimated annual incidence estimates of influenza-associated respiratory and circulatory hospitalizations for 18–49 years (14.8 per 100,000 person years), 50–64 years (24.2 per 100,000 person years, 65–74 years (40.3 per 100,000 person years), 75–84 years (144.9 per 100,000 person years), and ≥85 years (531.3 per 100,000 person years) (Table S2).

We repeated these analyses excluding outcomes that listed a diagnostic code for pneumonia. When a pneumonia diagnosis was excluded, the estimated number of influenza-associated respiratory or circulatory hospitalizations was 24,816 (95% CI 6,342–92,624). The estimated number of influenza-associated critical illness hospitalizations excluding pneumonia diagnoses was 8,213 (95% CI 3,764–20,799). These data suggest that 31% of both the total influenza-associated respiratory and circulatory hospitalizations and the total influenza-associated critical illness hospitalizations do not have pneumonia diagnoses and may not be captured by hospital-based influenza pneumonia surveillance.

Within adult age groups, we estimate that requiring pneumonia diagnoses to define influenza-associated events would exclude from 17.0% (65 to 74 years) to 39.9% (18 to 49 years) of total influenza-associated respiratory and circulatory hospitalizations. For influenza-associated critical illness hospitalizations, requiring a pneumonia diagnoses would exclude from 23.3% (75 to 84 years) to 38.7% (50 to 64 years) of total events.

Our study found that disease modelling approaches will estimate 50% fewer cases of influenza-associated hospitalizations if pneumonia ICD-9-CM codes are required as compared to more sensitive respiratory or circulatory ICD-9-CM codes. This study has implications for influenza disease surveillance in hospitals and for estimates of influenza disease burden. If adult influenza surveillance is limited to patients with pneumonia, our data suggest that a substantial proportion of influenza-associated hospitalizations may be missed. Such underestimates might affect the perceived value or cost-effectiveness of influenza vaccination, particularly in countries with only limited influenza surveillance data.

Non-pneumonia complications of influenza virus infection are more likely to occur in older adults with chronic comorbid conditions [29]. The United States has an older population with a higher prevalence of chronic underlying medical conditions than typical low- or middle-income countries [30]. These demographic differences suggest that influenza-pneumonia case definitions may be more sensitive for all-cause influenza disease in different socio-demographic settings. A recent study from South Africa estimates that influenza associated deaths in the elderly due to cerebrovascular disease, diabetes, or ischemic heart disease were higher than in the United States [31], and younger adults with AIDS in South Africa had elevated incidence of non-pneumonia influenza-attributable mortality [32]. With increasing survival and rising prevalence of non-communicable diseases throughout the world [30], influenza-associated exacerbations of underlying diseases will likely increase in importance.

Our study should be interpreted in the context of its limitations. The ideal study design to address our primary objective would be prospective hospital surveillance with sensitive specimen collection criteria for influenza testing and standardized radiographic assessment for pneumonia. Few such studies have been conducted, however, likely owing to the high costs of such research and the large sample sizes needed for high precision estimates, particularly among the critically ill. Owing to the nature of our modelling approach, caution should be taken with direct extrapolation of our results to the individual level. Because SID databases do not have results for microbiological investigations or particular therapies, we are unable to calculate the number of hospitalized influenza cases, necessitating our disease modeling approach. Similar methods have been used to estimate the burden of influenza disease in many countries [22], [33]–[37], and they are recommended by WHO to estimate influenza burden of disease when administrative hospitalization datasets and robust influenza surveillance data are available [5]. A recent study has validated our approach to estimating influenza-associated hospitalizations against prospective surveillance for laboratory-confirmed influenza in children [38], but there are no such validation studies in adults. The event estimates assumed that influenza types/subtype effects are non-negative and that virus activity is independent by type/subtype. If either of these two assumptions were violated, we may have over-estimated the total number of influenza-associated events. We used administrative codes to identify persons with acute respiratory failure, pneumonia, orsevere sepsis. While we used definitions of these outcomes from previous clinical epidemiology research, only pneumonia and severe sepsis have been validated against medical record review [25], [26], [39], [40].

Influenza vaccines are safe and effective at preventing influenza disease [2]. Vaccination would likely have the highest impact in settings where influenza morbidity is highest. Many countries will want local or regional burden of disease data to inform estimates of vaccine program impact in order to decide among competing public health priorities. Our study suggests that there may be a large proportion of unmeasured severe influenza disease in the influenza literature which must be considered in vaccine program impact calculations.