Attributable Proportion Extrapolation

Two studies used the available literature to first estimate the proportion of acute gastroenteritis attributable to norovirus then extrapolated that proportion to all-cause acute gastroenteritis incidence. Patel at al. conducted a systematic literature review of the prevalence of norovirus among persons with acute gastroenteritis in the community, outpatient clinics, emergency departments (ED), and hospitals (2). Most of the 31 studies included in the review involved hospitalized children and only 1 of the studies was conducted in the United States, underscoring the limited scope of the available literature. Among hospitalizations or ED visits for acute gastroenteritis in children <5 years of age, a pooled proportion of 12% of cases was attributed to norovirus. This norovirus prevalence was then extrapolated to national estimates of acute gastroenteritis in children <5 years of age from NHDS, the National Ambulatory Medical Care Survey (NAMCS), and the National Hospital Ambulatory Care Survey (NHAMCS) (12). The resulting annual estimates of 235,000 outpatient visits, 91,000 ED visits, and 23,000 hospitalizations associated with norovirus in US children <5 years of age suggested that norovirus was second only to rotavirus (before implementation of the national rotavirus vaccine program) as a cause of severe acute gastroenteritis in children.

Building upon the approach taken by Mead et al. (4), Scallan et al. reported new estimates of the US burden of foodborne disease (6). Although specific data sources had improved over the 12 years separating these 2 reports, the methods for estimating norovirus disease remained largely the same, constrained by the dearth of direct testing data in the United States. On the basis of community studies in the United Kingdom and Australia, and the study in the Netherlands used by Mead et al. (4), Scallan et al. estimated that 11% of acute gastroenteritis cases were caused by norovirus (6). This attributable proportion was then extrapolated to US rates of all-cause acute gastroenteritis from 3 FoodNet population surveys, hospitalizations from 3 databases (NHDS, the Healthcare Cost and Utilization Project Nationwide Inpatient Sample, and NAMCS/NHAMCS), and deaths from the multiple cause-of-death mortality database in the National Vital Statistics System. The resulting norovirus burden estimates across all ages in the United States were slightly lower than those by reported Mead et al. (4) in terms of total illnesses (21 million) but higher with respect to hospitalizations (56,000) and deaths (570). The increased estimate for hospitalizations can be explained in part by the fact that Scallan et al. (6) extrapolated the norovirus-attributable proportion to all-cause acute gastroenteritis across all age groups, whereas Mead et al. (4) applied the norovirus fraction to all-cause acute gastroenteritis only in adults.

Laboratory-confirmed Population-based Surveillance

In recognition of the need to directly assess the incidence of laboratory-confirmed norovirus infections among acute gastroenteritis patients in the United States, 2 surveillance platforms were leveraged to generate this data. In collaboration with FoodNet and a health maintenance organization (HMO) in the state of Georgia, Hall et al. used a passive sampling strategy in a population-based study of acute gastroenteritis incidence among outpatients (7). A random sample of fecal specimens submitted for routine clinical diagnostics (i.e., bacterial culture) were aliquoted for subsequent norovirus testing. Because the samples were derived from a known population catchment based on HMO membership, the resulting norovirus prevalence could then be used to calculate incidence. Health care use rates from 3 FoodNet population surveys were used to scale-up the observed prevalence among patients who submitted fecal specimens to outpatient and community incidence. The resulting adjusted outpatient and community incidence rates for norovirus were 64/10,000 population and 650/10,000 population, respectively (Table 2). If applied to the US population when the samples were collected (2004), these rates correspond to a national estimate of 19 million illnesses and 1.9 million outpatient visits. This total number of norovirus illnesses was within the uncertainty bounds of the estimate of Scallan et al. (6) (90% credibility interval 13–31 million) and provided the first estimate based on direct testing of patients with acute gastroenteritis in the United States. Although this passive sampling approach afforded convenience and required relatively little resources, it is potentially subject to substantial bias for 2 reasons. First, only those fecal samples that had a physician order for bacterial culture were tested. Second, the data may have limited generalizability because the study was conducted in a single, relatively young, privately insured population.

Figure 1

Figure 1. . . . Rates of A) norovirus-associated deaths. B) hospitalizations, C) emergency department (ED) visits, and D) outpatient visits by age group, United States. Data were derived from studies using indirect...

The preferred approach of active surveillance enrollment and laboratory testing of all acute gastroenteritis patients from multiple sites was used in a recent study by Payne et al. from the New Vaccine Surveillance Network (8). This network of 3 pediatric hospitals conducted year-round, population-based, active surveillance for hospitalizations, ED visits, and outpatient clinic visits for acute gastroenteritis among children <5 years of age for whom laboratory confirmation of cases was available. Payne et al. reported annual norovirus hospitalization, ED visit, and outpatient visit rates of 7, 141, and 319/10,000 children <5 years of age, respectively, over a 2-year period (8). Because the outpatient surveillance in the New Vaccine Surveillance Network used sentinel clinics and was not truly population based, the norovirus outpatient visit rate was based on extrapolation of norovirus prevalence to the all-cause acute gastroenteritis outpatient rates from NAMCS/NHAMCS. Extending these norovirus incidence rates to the ≈20 million US children <5 years of age, Payne et al. estimated 14,000 hospitalizations, 281,000 ED visits, and 627,000 outpatient visits for this age group (8). They also reported that the median health care charges for norovirus hospitalizations, ED visits, and outpatient visits were $3,918, $435, and $151, respectively, corresponding to an annual total of $273 million in norovirus-associated treatment costs for US children <5 years of age. Compared with the estimates reported by Patel et al. (2) among children <5 years of age, Payne et al. estimated ≈55% fewer hospitalizations but ≈2 times as many ED and outpatient visits (Figure 1). Moreover, Payne et al. reported that norovirus had become the leading cause of medically attended acute gastroenteritis in children during the post-rotavirus vaccine era.

Indirect Attribution from Regression Modeling

Population-based databases that use International Classification of Disease (ICD) coding are often used to estimate trends of specific syndromes or pathogens. Those databases that are national or nationally representative can be particularly helpful in overcoming the generalizability limitations of studies performed in specific populations that may not be broadly representative. However, ICD coding for specific pathogens is typically used only when there is laboratory confirmation (18). Given the limited availability of direct testing for norovirus among sporadic acute gastroenteritis cases, norovirus-specific coding in these databases is insensitive and unreliable. For example, Payne et al. retrospectively retrieved ICD–9-CM discharge diagnosis codes for 278 medically attended laboratory-confirmed norovirus cases identified by active surveillance and found that none had been assigned the norovirus ICD-9-CM code (008.63) (8).

To overcome this issue and still use these robust sources of data, we conducted a series of modeling studies to indirectly estimate the proportion of cause-unspecified acute gastroenteritis (which represents most acute gastroenteritis–coded events) likely caused by norovirus. In brief, time-series regression models used monthly counts of acute gastroenteritis attributed to specified causes other than norovirus to estimate the number of cause-unspecified acute gastroenteritis cases likely attributable to those causes. Events attributed to these other causes and to background nonseasonal causes were subtracted from the total cause–unspecified acute gastroenteritis, and the remaining unattributed events (i.e., model residuals) were then analyzed to generate norovirus estimates. Models were developed for specific age-groups to the extent this was possible for each specific outcome.

Applying this method to national mortality data from National Vital Statistics System, Hall et al. estimated that norovirus is associated with an average of 797 deaths/year (9). Most (90%) of these norovirus-associated deaths and the highest mortality rate (0.20 deaths/10,000 population) occurred among persons ≥65 years of age (Figure 1, panel A). Using Healthcare Cost and Utilization Project Nationwide Inpatient Sample data, Lopman et al. estimated an average of 71,000 norovirus-associated hospitalizations each year, resulting in $493 million in health care charges (10). Norovirus-associated hospitalization rates exhibit a U-shaped curve (Figure 1, panel B); the highest rates occur among persons <5 years of age (9.4 hospitalizations/10,000 population) and ≥65 years of age (8.1 hospitalizations/10,000 population). To estimate rates of norovirus-associated ambulatory visits, Gastañaduy et al. applied this same method to MarketScan insurance claims databases and reported norovirus associated with 13.5 ED visits and 57.2 outpatient visits/10,000 population across all age groups (11). In contrast to rates of norovirus-associated mortality, rates of ambulatory visits associated with norovirus are highest in children <5 years of age (Figure 1, panels C, D). When Gastañaduy et al. extrapolated these rates to the US population, they estimated a national incidence of 399,000 ED visits and 1.7 million outpatient visits/year, corresponding to $284 million in health care charges.

Figure 2

Figure 2. . . . Number of A) norovirus-associated deaths, B) hospitalizations, C) emergency department (ED) visits, and D) outpatient visits across all age groups, by month and year, United States. Data were...

Although this indirect modeling method has the potential for biases that might overestimate (e.g., assuming all residual seasonality in acute gastroenteritis is caused by norovirus) and underestimate (e.g., assuming none of the background nonseasonal incidence is associated with norovirus) norovirus incidence, it yielded temporal trends highly consistent with what is known about norovirus that help to ensure the validity of this method. These trends included a pronounced winter peak, with 63%–73% of all norovirus-associated events occurring during October–March and increases by <50% during years pandemic strains of norovirus emerged (i.e., 2002–2003 and 2006–2007) (Figure 2). These patterns have been well described through US surveillance of norovirus outbreaks (19–21) but had not been previously described among cases of sporadic norovirus illness. In addition, estimated rates of norovirus-associated outcomes from these models were generally consistent with those generated from population-based testing and attributable proportion extrapolation. For example, the all-ages outpatient rate modeled by Gastañaduy et al. (11) (57 outpatient visits/10,000 population) was within the uncertainty bounds of the estimate of Hall et al. (7) (90% credible interval 36–120 outpatient visits/10,000 population) from direct testing in the Georgia HMO population (Table 2). Likewise, the modeled hospitalization rate in children <5 years of age reported by Lopman et al. (10) (9.4 hospitalizations/10,000 population) was between the estimates obtained through direct testing by Payne et al. (8) (7.2 hospitalizations/10,000 population) and attributable proportion extrapolation by Patel et al. (2) (16 hospitalizations/10,000 population) (Figure 1, panel B).