Since July 2003, all blood donations have been screened for WNV RNA by NAT. If we assume that blood donation and WNV infection are independent events, the proportion of blood donors infected by WNV in 2003 (seasonal incidence of WNV in the blood donor pool) is a function of NAT yield and the average length of time that WNV RNA is detectable after infection. By measuring IgM antibodies in North Dakota blood donors shortly after the epidemic, we estimated the seasonal incidence for that region. After adding NAT screening yield data from the same donor population, we then estimated the length of time that WNV RNA is detectable by NAT.

We combined the length of time that RNA is detectable by NAT with NAT screening yield data by state to estimate state-specific and national WNV seasonal incidence in the blood donor and general population. Finally, by dividing the estimated number of infections in the general population in 2003 by the number of neuroinvasive disease cases reported to the national WNV surveillance system (ArboNET), we estimated the ratio of WNV infections to neuroinvasive disease cases.

US blood donations are screened for WNV RNA by using NAT assays on pools of 6 to 16 donations or on individual samples in high-incidence regions. In 2003, ≈96% of the screening was conducted on pooled samples. Additionally, 2 blood collection organizations (ARC; Blood Systems) retrospectively performed individual donation NAT on cryopreserved plasma from 36,269 donations in 5 states with substantial epidemics to ascertain the proportion of low-level viremia missed by minipool-NAT and to assess, through recipient lookback, the infectivity of units harboring low-level viremia (15,16). Viremic donations detected by individual donation NAT were included in this analysis to compile total NAT yield for 2003, but they were excluded from calculations used to project WNV infection in the general population for the following reasons: 1) inconsistent application of individual donation NAT screening around the country (15,16); 2) variable rate of detection of low-level viremia by individual donation NAT assays (18); and 3) fever and symptoms during the postseroconversion low-level viremia phase (unlike the asymptomatic minipool yield phase), which would result in self-deferral from donation and bias our projections.

We combined 2 large databases consisting of all donations and WNV-confirmed donations obtained by state from July 1 (when most blood centers implemented WNV NAT screening) to October 31, 2003. The first database was derived from 72 of the 74 independent blood centers that constitute ABC, which collect nearly 50% of US donations (14,19). Data elements included total number of donations, donor state of residence, and the minipool and individual donation NAT confirmatory status of all donations collected from July to October 2003. Similar data were obtained from the ARC national donor database, which constitutes 45% of the US supply (16). Donations were classified as confirmed WNV NAT-positive if the index donation was reactive by NAT and 1) positive for IgM or by an alternative NAT procedure or 2) follow-up samples from donors were reactive on a NAT assay or were IgM-positive. The dataset included the subset of confirmed NAT-positive donations that were either originally detected by minipool-NAT, or had been detected by prospective individual donation NAT but were subsequently determined to be detectable by minipool-NAT (were reactive when retested at a 1:16 dilution using minipool-NAT). This extra testing ensured that seasonal incidence estimates were based on data obtained by using a comparably sensitive screening process across all regions of the United States and throughout the epidemic period. The proportion of confirmed positive donations identified by minipool-NAT was determined by month for each state, and ≈95% confidence intervals (CIs) around these minipool-NAT yield estimates were computed (20). The Epi Map component (Environmental Systems Research Institute, Redlands, CA, USA) of EpiInfo version 3.3 (Centers for Disease Control and Prevention [CDC], Atlanta, GA, USA) was used to display results graphically.

To use NAT screening data to estimate state-specific WNV seasonal incidences, we first derived an estimate for the average length of time that RNA is detectable by minipool-NAT after infection occurs (TMP-NAT). TMP-NAT can be approximated if both the minipool-NAT screening yield and seasonal incidence of WNV are known (Appendices 1 and 2; available from http://www.bsrisf.org/eid2006/app1.html and http://www.bsrisf.org/eid2006/app2.html). The seasonal incidence was estimated by measuring the peak WNV IgM prevalence observed in a particular region ≈3 weeks after the end of the region's first epidemic (http://www.bsrisf.org/eid2006/app1.html). Serologic data allowed us to evaluate both minipool-NAT yield and prevalence of IgM for each week from July to September 2003 and to identify peak IgM prevalence. The sum of the weekly minipool-NAT yield estimates divided by the peak IgM prevalence (our estimate of the 2003 seasonal incidence in North Dakota) was used to derive TMP-NAT (http://www.bsrisf.org/eid2006/app2.html). Approximate 95% CIs around peak IgM prevalence and TMP-NAT were calculated by assuming normal distributions with variances approximated by Taylor series (21).

We assumed that WNV infection dynamics are similar in blood donors and in the general population. The monthly WNV incidence in each state for each month from July through October was derived by multiplying the monthly minipool-NAT yield by the number of days in each month and dividing by the average period of time during which RNA is detectable (TMP-NAT) (Appendix 3; available from http://www.bsrisf.org/eid2006/app3.html). Each state-specific seasonal WNV incidence was calculated by summing the 4 monthly WNV incidence estimates. To estimate 2003 WNV infections nationwide, we multiplied each state-specific seasonal WNV incidence by the corresponding July 1, 2003, population estimate from the US Census Bureau (22) and then summed over all states. An ≈95% CI around the estimated 2003 WNV infections nationwide was calculated by assuming a normal distribution with variance approximated by Taylor series (21).

We then calculated the ratio of the estimated number of WNV infections nationwide and the total neuroinvasive disease cases reported to CDC (23). The standard error (SE) of this ratio is dependent on the SE of the total neuroinvasive disease cases (assumed to be Poisson distributed), the SE of TMP-NAT, and the SEs of state-specific minipool-NAT yield estimates (assumed to be binomially distributed) and was approximated by a Taylor series (21). We did not estimate the proportion of infections resulting in West Nile–related febrile illness because it is considerably underreported to ArboNET.

The Investigational New Drug protocols, which included donor consent for WNV NAT screening and follow-up testing, were reviewed and approved by multiple institutional review boards and FDA. Institutional review board approval of this study protocol, including compilation of a national NAT yield database and anonymous IgM and IgG testing (http://www.bsrisf.org/eid2006/app1.html), was obtained from the University of California, San Francisco Committee for Human Research, and from Westat (Rockville, MD, USA).

As shown in Figure 3, minipool-NAT–confirmed positive donations were detected from July 13 to September 6, 2003, in the Bismarck and Minot regions of North Dakota, with minipool-NAT yield peaking at 1.4% (95% CI 0.4–2.3) in late August. IgM-confirmed positive donations were not observed in these same regions during the first 3 weeks of July but were detected toward the end of July. IgM prevalence gradually increased thereafter and reached a plateau around September 7; ≈5% of donations were positive for IgM during most of September. The peak IgM prevalence was observed the last week of September (5.2%, 95% CI 3.0–7.4) and was similar to the IgG prevalence observed 9 months later in June 2004 (5.3%, 95% CI 3.9%–6.7%), when IgM prevalence had declined to 1.2%. Thus, the peak IgM weekly prevalence was assumed to be a good estimate of the seasonal incidence in this region. The average length of time viremia is detectable by minipool-NAT, TMP-NAT, was estimated to be 6.9 days (95% CI 3.0–10.7).

The proportion of the population estimated to have become infected during 2003 in each state was 0%–4.9% (Figure 4A and Table). The highest proportions were observed in Nebraska (4.9%), Colorado (4.3%), North Dakota (4.1%), South Dakota (4.0%), Wyoming (3.5%), and Kansas (2.1%). Nationally, 735,000 persons (95% CI 322,000–1,147,000) were estimated to have been infected in 2003 (Table). Figure 4B shows the distribution of these infections by state. The greatest number of infections were located in Colorado, Texas, Nebraska, Kansas, the Dakotas, and to a lesser extent the states in the Midwest and Northeast, which had only moderate seasonal incidence but have large populations.

We compared the estimated proportion of the population infected with WNV to the proportion of WNV neuroinvasive disease cases reported to CDC for each state. Figure 5 shows that these proportions are highly correlated with one another. A total of 2,866 neuroinvasive WNV cases were reported nationally to CDC's ArboNET system in 2003 (Table). This total was compared to 735,000 persons nationally estimated to have been infected with WNV in 2003. Thus, an estimated 256 WNV incident infections occurred per reported neuroinvasive disease case (95% CI 112–401). An estimate of 353 infections per each reported neuroinvasive disease case (95% CI 190–516) was obtained by analyzing the North Dakota data separately, which had 94 reported neuroinvasive cases among an estimated 33,000 infections (5.2% peak IgM prevalence × the state population of 633,837).

Two potential biases may have affected our estimated ratio of 256 WNV incident infections per reported neuroinvasive disease case. First, neuroinvasive cases may be underreported to ArboNET. A 20% underreporting of neuroinvasive cases to ArboNET alters the ratio to 205 WNV incident infections per reported neuroinvasive disease case (95% CI 90–320). Second, blood donors may underrepresent infections in the general population because prospective donors who are WNV-infected may self-defer or be deferred from donating. If the number of infections in the general population was underestimated by 20%, the ratio of WNV incident infections per reported neuroinvasive disease case would be 320 (95% CI 140–501).