Since the first identification of West Nile virus (WNV) in North America in 1999, the national ArboNET surveillance system has documented that the virus has become endemic throughout most of the USA [1]. Studies conducted in the USA indicate that in persons infected with WNV, about 75% remain asymptomatic, 25% develop West Nile fever (WNF), and <1% develop West Nile neuroinvasive disease (WNND) [Reference Zou2, Reference Mostashari3]. Since routine diagnostic testing is not recommended for WNF patients and many of these patients do not seek medical care, WNF cases are considerably underreported. In contrast, WNND reporting to ArboNet is thought to be nearly complete, particularly in older persons and persons with compatible clinical features [Reference Boehmer4, Reference Weber5]. From 1999 to 2010, the ArboNET surveillance system recorded 12 823 cases of WNND from 47 states.

Several studies have attempted to estimate the true number of WNV infections in the USA. Seroprevalence surveys conducted to determine cumulative infection incidence have found seroprevalences of WNV IgG antibody ranging from 1·9% to 14% [Reference Mostashari3, Reference Mandalakas6, Reference Schweitzer7]; however, these studies were limited by surveying small geographical areas and relatively few subjects. Cumulative infection incidence has also been estimated by multiplying the total number of reported WNND cases by the estimated proportion of total infections per WNND case. Results from a seroprevalence survey in New York after the 1999 epidemic suggested that each reported case of WNND resulted from an average of 140 infections [Reference Mostashari3]. This ratio was later used to estimate a cumulative infection incidence of 1 655 080 up to 2006 in the USA [Reference Lindsey8]. However, this cumulative incidence estimate was subject to considerable uncertainty given that the 140:1 ratio of infection to WNND was based on a seroprevalence estimate (2·6%) whose confidence limits spanned more than threefold (1·2–4·1%) [Reference Mostashari3]. In fact, data derived from blood donor screening suggested that there may be as many as 256–353 infections for each reported WNND case [Reference Busch9]. More recently, data from a serosurvey of blood donors and WNND reporting from an entire state indicated that 213–286 infections are necessary to result in a case of WNND [Reference Carson10]. This last study was of sufficient scope to estimate age- and gender-specific infection-to-WNND ratios; however, the ratio in children could not be ascertained since infection data were derived from blood donors. The risk of WNND following infection was estimated as more than 20 times higher in men aged ⩾65 years than women aged 16–24 years, a finding consistent with national WNND surveillance data [Reference Lindsey8]. Thus, population age and gender structure is an important consideration when computing national WNV infection estimates.

In order to estimate the number of WNV infections in adults in the USA by state to the end of 2010, taking into account the substantial variation in risk in age and gender groups, we multiplied the age- and gender-stratified number of WNND cases with onset up to the end of 2010 (reported through the ArboNET surveillance system) by the recently derived, corresponding estimated infection-to-WNND ratios [Reference Carson10]. Cumulative incidence (percent of population previously infected) was then calculated by dividing these numbers by the corresponding state population (US Census Bureau, 2010). ArboNET was missing age and/or gender data for 79 of the 12 823 WNND cases reported to the end of 2010. For the above calculations, those cases were assigned to the various age and gender strata proportionally based on the cases' state's percentage of WNND cases in each stratum. Variance estimates for the total number of previously infected individuals for each state were also computed by accounting for the stratification by age and gender. The estimated total number of infections (variance estimates) for the whole of USA was computed by summing the state-specific totals (variances), and a Wald 95% confidence interval (CI) was computed using these. Statistical analyses were performed using SAS version 9.2 (SAS Institute Inc., USA), Epi Info (CDC, USA), and R version 2.11.1 (www.r-project.org). R was also used to produce the estimated cumulative incidence map (Fig. 1).

Fig. 1. Estimated cumulative incidence of West Nile virus infection (per cent of population infected) in US adults aged ⩾16 years, 1999–2010.

State-specific cumulative infection incidence estimates indicated that the central plains states had the highest cumulative incidence, with South Dakota (13·4%), Wyoming (8·2%), North Dakota (7·5%), Nebraska (7·4%), and Colorado (6·0%) leading in incidence (Fig. 1, Table 1). An estimated 2 757 029 (95% CI 2 688 327–2 825 730) persons aged ⩾16 years were infected with WNV by the end of 2010 in the USA.

Table 1. Reported cases of West Nile virus (WNV) neuroinvasive disease, estimated number of WNV infections, and estimated cumulative incidence of WNV infection in adults aged ⩾16 years, by state, 1999–2010, USA

* West Nile neuroinvasive disease (WNND) cases reported to CDC through ArboNET; Alaska, Maine, Vermont and Hawaii reported no cases up to 2010.

† Product of WNND cases by age and gender reported in state and corresponding ratios of WNV infections per WNND case reported by Carson et al. [Reference Carson10].

‡ US Census Bureau data for 2010, persons aged ⩾16 years.

Our study estimated that almost 2·8 million adults in the USA had been infected with WNV by 2010. Estimates obtained from serological surveys of the risk of WNV infection in children compared to adults vary [Reference Mostashari3, Reference Mandalakas6]; however, if similar infection rates are assumed in children aged <16 years as in persons aged 16–24 years, the estimated number of infections in all persons in the USA would be closer to 3·2 million. Assuming that about 26% of persons contracting WNV will develop clinical disease by the virus [Reference Zou2], an estimated 780 000 persons in the USA have become ill by the end of 2010. A recent study estimated acute care medical costs of US$46 530 for each case of WNND and US$302 for WNF [Reference Barber, Schleier and Peterson11]. Applying these figures to the 12 823 reported cases of WNND and the estimated 780 000 cases of WNF yields an estimate of the total acute care medical cost of about US$832 million.

The reasons for the geographical pattern in WNV infection incidence are probably complex and are not well understood. The extensive acreage of irrigated farmland in the central plains and some western states may explain the high incidences observed in those locations [Reference De Groot12–Reference Eisen14]. Irrigated farmland provides excellent habitat for Culex tarsalis, an efficient vector for WNV transmission [Reference Turell15, Reference Goddard16]. Other factors associated with WNV incidence in other areas include surface water and wetlands [Reference Ezenwa17–Reference Shaman, Day and Komar19]; avian density, diversity, and abundance [Reference Allan20–Reference Liu, Weng and Gaines22]; and human density, urbanization, and housing [Reference Allan20, 22–Reference Rochlin24].

Because not all patients with WNND are recognized and reported to ArboNET [Reference Boehmer4, Reference Weber5], the actual number of WNND cases was probably higher than the reported number used in our calculations. Thus, the actual number of infections may be higher than that estimated in this study. Although all states use ArboNET's standardized reporting system, completeness of recognition and reporting of cases may vary among states. In addition, uncertainties regarding the age- and gender-specific infection-to-WNND ratios [Reference Carson10] used in our calculations could have resulted in an over- or underestimate of the actual number of infections. Another limitation is that our study spans 12 years, and 2010 census data were used in the extrapolation to derive state-specific and national infection rates; however, our incidence estimates were similar when using census estimates for 2005 (data not presented).

In conclusion, our data indicate that over 3 million people have become infected with WNV in the USA by the end of 2010, almost twice the number estimated based on prior non-stratified estimates up to 2006 [Reference Lindsey8]. Since 2003, the US blood supply has been screened for WNV. Had this not been done, the 2·8 million infected adults would have represented a transfusion safety risk during their acute infection since most are asymptomatic or mildly symptomatic. It is important to note that despite the many people infected so far, over 98% of the US population remains at risk of infection, and hence efforts to prevent infection through vector control and personal protection measures remain important. Our findings indicating high infection incidences in people living in central midwestern states stress the particular importance of prevention measures during the summer WNV transmission season in that region.