The annual total of patients with symptomatic, laboratory-confirmed cases of NE reported from the state of Baden-Württemberg was provided by the German surveillance system for infectious diseases, which covers all 44 districts of Baden-Württemberg, for 2001–2007. Laboratory diagnosis was based either on detection of viral RNA by reverse transcription–PCR or on detection of immunoglobulin (Ig) M or a marked rise of antihantavirus IgG (15). Details of the German notification system are provided by Faensen et al. (16), and detailed information about the incidence of NE in Germany can be obtained from Piechotowski et al. (15). We excluded from analysis those patients who reported recent travel. We obtained human population density by district from the Statistical Office in Baden-Württemberg (17). Because the population of Baden-Württemberg in the respective districts did not change substantially during 2001–2007, we used census data for 2006 to calculate incidence rates (no. cases/100,000 inhabitants) for each study year.

Factors considered to favor bank vole habitat were obtained from the Forest Research Institute Baden-Württemberg, Department of Biometry and Information Science, from forest inventories conducted during 2001–2002 (18). We assessed percentage of land cover for 5 covariates that have been indicated as preferred habitat for bank voles in Europe (19,20): beech forests, seed plants, bilberries, dwarf shrubs, and blackberries. For oak forests, also a preferred habitat, no data were available. Data for beech forest cover were provided as hectares per district and converted to proportion coverage per district area. For each of the 4 remaining habitat areas by district, data were provided as area variables in 3 categories of coverage: rare (1%–10%), frequent (>10%–50%), and common (>50%). To estimate the availability of the respective habitat in each district, we calculated a weighted sum of the 3 coverage areas by using 0.01, 0.1, and 0.5 as weights for the rare, frequent, and common areas, respectively, and then converted each sum into percentage coverage per district area. For each district, the 5 habitat variables were considered constant during the study period.

We obtained data on annual mast production of beechnuts by district, starting and ending 1 year earlier than the study (2000–2006). Data came from the Ministry for Agriculture Baden-Württemberg and the Public Forest Administration Baden-Württemberg, for which foresters conduct annual counts of beechnuts under beech trees (plot counts) (21). The beechnut mast in the preceding year was used as a potential determinant of the NE incidence because beechnut supply in the fall may influence winter vole survival and, consequently, vole population the following year. Beechnut mast data were stratified into 3 classes: good/excellent crop if 40%–100% of trees produced mast, medium if 10%–39%, and poor if 0%–9%.

Deviations of the monthly temperature for 2000–2007 were referenced against the perennial average for 1961–1990, provided by the German Meteorological Service (22). The station that geographically best represented a respective district was selected from the network of 576 meteorology stations covering Germany. Only the temperature deviations of the winter and early spring months (December–March) were included because mild winters and springs were hypothesized to enhance survival rates of rodents and produce food resources earlier. For each year and district we modeled mean values of temperature deviations for winter (December of the preceding year and January of the same year as NE incidence data) and spring (February and March of the same year).

We modeled associations between potential risk factors and incidence rates of NE during 2001–2007 by multivariate Poisson regression, using the SAS program GENMOD (version 9.1, SAS Institute Inc., Cary, NC, USA). In Poisson regression, we set the logarithm of the expected annual incidence per district and year equal to a linear term of potential determinants: habitat variables, climate factors, human population density, and year of investigation. The number of NE cases per year and district were approximately Poisson distributed. To allow for overdispersion, we introduced a scale parameter in the regression modeling. To account for nonlinearity, the variable “year of investigation” was represented in the regression model by 6 dichotomous variables. All other independent variables were considered continuous and were scaled by units as per 5% increase in district-area coverage by beech forest and seed plants, per increase in human population density of 500 inhabitants per square kilometer, per 1°C change in winter and spring temperature above the long-term average, and per unit step of beechnut mast (good/excellent, medium, poor). The covariates in the Poisson regression model were examined for 2-way interactions, but none could be confirmed.

The criterion for inclusion of a determinant in the final regression model was set at a significance level of p<0.05. For a measure of association between a determinant and NE incidence, the risk ratio (RR) was calculated by using the respective estimated regression parameter of the Poisson regression model and, therefore, adjusted for all other determinants included in the regression model. All estimates of RR were complemented by a 95% confidence interval (CI) and p value. The pseudo-R-squared (R²) was provided as a measure of overall goodness-of-fit of the regression model.

The annual percentage of statewide beechnut mast varied by year and district from a crop failure (0%–9% of optimum mast) in 2005 to good/excellent (40%–100% of optimum mast) mast in 2001 and 2006 (Table 1). Winter temperatures exceeded long-term averages in 2001, 2003, 2004, and 2007. The maximum winter temperature deviation occurred in 2007; median deviation was +3.5°C and maximum was +4.5°C. Winter temperatures were below the long-term averages in 2002, 2005, and 2006 (range of deviation –0.5°C to –1.5°C). Mean spring temperature exceeded the long-term average for all years except 2005 and 2006 (Table 1).

Maximum beech forest cover within a district (17.7%) was found in the region of the Swabian Alb (Table 2; Figure 1, panel C); lowest beech forest cover was found in the eastern and middle regions, as well as in districts containing the major cities of Baden-Württemberg. Maximum seed plant cover (11.3%) was found in the southerly and centrally located districts of Baden-Württemberg; the lowest cover (1.3%) was associated with the northerly districts and in the districts containing the major cities. The median values of the land cover variables of dwarf shrubs, bilberry, and blackberry were <2%; values of ≈10% were restricted to a few districts (Table 2). As the percentages of land cover in bilberry and dwarf shrubs were closely correlated across districts (r ≈ 1), only 1 of these variables (bilberry) was included in the regression modeling. Human population density ranged from 104 inhabitants/km² in rural districts to 2,864 inhabitants/km² in the state capital of Stuttgart (Table 2; Figure 1, panel B). Most districts (80%) contained <1,000 inhabitants/km².

For the final Poisson regression model, only the blackberry and bilberry variables did not pass the inclusion criterion of a positive association with the annual incidence of NE during 2001–2007 and a significance level of p<0.05 (Table 3; Figure 2). The final model explained 75% of the variation of NE incidence (R² = 0.75).

The variables showing annual variation, specifically beechnut mast and spring temperature above the long-term mean, were strong independent predictors of NE. The occurrence of a good/excellent beechnut mast in the previous year increased risk for NE (RR 2.86, 95% CI 1.81–4.50) relative to medium crop (Table 3). An increase of 1°C in spring (February/March) above the long-term average resulted in RR of 4.49 (95% CI 2.86–7.06). The influence of winter (December/January) temperature was less strong, but it was still a significant factor in NE incidence; RR 1.70 (95% CI 1.11–2.61).

The association between NE incidence and year of investigation indicated that incidence increased exponentially during 2002–2006, when all other risk factors were controlled for and 2001 was used as a reference. However, this trend was not apparent in 2007. In 2007, extreme winter and spring temperatures relative to the long-term average, coupled with the best beechnut mast observed during the entire study interval, potentially masked or overwhelmed the temporal trend toward increasing NE incidence in the regression modeling. To check stability of the observed time trend, we repeated the regression analysis with the investigation year 2007 excluded. Again, a highly significant time trend, which indicated a doubling of the NE incidence per year (RR 2.02, 95% CI 1.43–2.83), was observed. Habitat and climate factors as well as human population density had somewhat lower, but still highly significant, estimates of RR (data not shown). When the estimated 2001–2006 regression model was used for prediction, the incidence in 2007 was considerably underestimated. This result underlines the importance of the time-dependent habitat and climate factors for the PUUV reservoir.

As notification of NE cases became a requirement in Germany in 2001, underreporting could have occurred in the first year of investigation (2001). However, exclusion of 2001 in the regression analysis did not change the significance and magnitude of the associations of the included risk factors with the NE incidence (data not shown).

Among the time-invariant determinants, 2 of the land-cover parameters—percentage of beech forest (Figure 1, panel C) and seed plant cover—exhibited a major effect. For each 5% increase in coverage per district, risk for NE approximately doubled (beech forest, RR 1.94, 95% CI 1.69–2.22; seed plants, RR 2.80, 95% CI 2.31–3.40; Table 3). A unit increase in human population density of 500 inhabitants/km² (about the median population density of the 44 districts) increased incidence of NE by ≈12% (RR 1.12, 95% CI 1.01–1.23).