Assessing the Influence of Climate on the Spatial Pattern of West Nile Virus Incidence in the United States

Abstract

Background: West Nile virus (WNV) is the leading cause of mosquito-borne disease in humans in the United States. Since the introduction of the disease in 1999, incidence levels have stabilized in many regions, allowing for analysis of climate conditions that shape the spatial structure of disease incidence.

Objectives: Our goal was to identify the seasonal climate variables that influence the spatial extent and magnitude of WNV incidence in humans.

Methods: We developed a predictive model of contemporary mean annual WNV incidence using U.S. county-level case reports from 2005 to 2019 and seasonally averaged climate variables. We used a random forest model that had an out-of-sample model performance of $R^2=0.61$.

Results: Our model accurately captured the V-shaped area of higher WNV incidence that extends from states on the Canadian border south through the middle of the Great Plains. It also captured a region of moderate WNV incidence in the southern Mississippi Valley. The highest levels of WNV incidence were in regions with dry and cold winters and wet and mild summers. The random forest model classified counties with average winter precipitation levels $<23.3\mathrm{mm}/\text{month}$ as having incidence levels over 11 times greater than those of counties that are wetter. Among the climate predictors, winter precipitation, fall precipitation, and winter temperature were the three most important predictive variables.

Discussion: We consider which aspects of the WNV transmission cycle climate conditions may benefit the most and argued that dry and cold winters are climate conditions optimal for the mosquito species key to amplifying WNV transmission. Our statistical model may be useful in projecting shifts in WNV risk in response to climate change. https://doi.org/10.1289/EHP10986.

Figure 1.

Total number of cases and mean spatial pattern of WNV in the United States during the period 2005–2019, including (A) a time series of WNV cases for the contiguous United States derived from data compiled by the U.S. Centers for Disease Control and Prevention for the sum of neuroinvasive and non-neuroinvasive cases; (B) a county-level map of the number of mean annual WNV cases averaged from 2005–2019; (C) a county-level map of mean annual incidence (cases per 100,000 population per year) derived from the case data shown in panel (B) and annual, county-level U.S. Census population estimates; and (D) the frequency distribution of mean annual incidence across the 3,108 counties in the United States. See Excel Table S1, S2, and S3 for corresponding numerical data. Graphs and maps were created in Python (version 3.7.6), using package “Basemap” (version 1.2.0), and “matplotlib” (version 3.1.3). We edited our figures for layout in Adobe Illustrator 2022. Note: WNV, West Nile virus.

Figure 2.

Incidence and histograms of the county number of WNV cases as a function of precipitation. Four seasonal intervals are shown: (A) winter (DJF), (B) spring (MAM), (C) summer (JJA), and (D) fall (SON). In each histogram, the number of counties was aggregated in $10-\text{mm}/\text{month}$ precipitation bins. Mean annual incidence and mean monthly precipitation variables during the period 2005–2019 were used to create the plots. See Excel Table S3 for corresponding numerical data. Surface air temperature and precipitation data are from the Precipitation elevation Regressions on Independent Slopes Model (PRISM). West Nile virus case data is provided by the U.S. Centers for Disease Control and Prevention. Note: DJF, December, January, February; JJA, June, July, August; MAM, March, April, May; SON, September, October, November; WNV, West Nile virus.

Figure 3.

Incidence and histograms of county number as a function of air temperature. Four seasonal intervals are shown: (A) winter (DJF), (B) spring (MAM), (C) summer (JJA), and (D) fall (SON). In each histogram, the number of counties was aggregated in 2°C temperature bins. Mean annual incidence and mean surface air temperature during the period 2005–2019 were used to create these plots. See Excel Table S3 for corresponding numerical data. Surface air temperature and precipitation data are from the Precipitation elevation Regressions on Independent Slopes Model (PRISM). WNV case data is provided by the U.S. Centers for Disease Control and Prevention. Note: DJF, December, January, February; JJA, June, July, August; MAM, March, April, May; SON, September, October, November; WNV, West Nile virus.

Figure 4.

A spatial correlation matrix between the eight seasonal climate variables across all counties in the contiguous United States. The matrix displays the Pearson correlation coefficients between each pair of predictor variables. Each predictor variable consisted of a vector with 3,108 elements, with each element representing the mean climate condition for a single county. Surface air temperature and precipitation data are from the Precipitation elevation Regressions on Independent Slopes Model (PRISM). Note: DJF, December, January, February; JJA, June, July, August; MAM, March, April, May; SON, September, October, November.

Figure 5.

A map of (A) modeled mean annual WNV incidence predicted by the RF model where each county is the average of the 500 iterations of out-of-sample testing data, (B) the standard deviation across those 500 iterations, and (C) the absolute error from observed WNV incidence levels (2005–2019). All are in units of cases per 100,000 population per year. See Table S3 for corresponding numerical data. Analyses were performed in R (version 3.6.1; R Development Core Team), maps were created in Python (version 3.7.6), using package “Basemap” (version 1.2.0) and “matplotlib” (version 3.1.3). We edited our figures for layout in Adobe Illustrator 2022. Note: RF, random forest; WNV, West Nile virus.

Figure 6.

RF model performance plots showing county-level values for (A) observed WNV incidence vs. the standard deviation among the 500 model iterations and (B) observed WNV incidence vs. the modeled WNV incidence (2005–2019). Each dot represents one of the 3,108 counties in the contiguous United States. A 1-to-1 line was added as a guide to (B). See Table S3 for corresponding numerical data. Note: RF, random forest; WNV, West Nile virus.

Figure 7.

The summary regression tree showing county-level conditional climate splits, created from the mean RF predictions. Precipitation (P) has units of millimeters/month and air temperature (T) has units of °C. Eigenvectors are denoted by E, where 1, 2, or 3 is the corresponding column number in the eigenvector matrix. The seasons are partitioned into winter (DJF), spring (MAM), summer (JJA), and fall (SON). WNV incidence (WNV Inc) is in units of cases per 100,000 population per year. The number of counties that fall into each division is denoted by n. Terminal nodes are in bold font. Note: Maps were created in Python (version 3.7.6), using package “Basemap” (version 1.2.0) and “matplotlib” (version 3.1.3). We edited our figures for layout in Adobe Illustrator 2022. DJF, December, January, February; JJA, June, July, August; MAM, March, April, May; RF, random forest; SON, September, October, November; WNV, West Nile virus.