Research

Assessing Climate Driven West Nile Virus Risk in Washington State

Gabriella LaBazzo | 2021

Advisor: Christine M. Khosropour

Research Area(s): Environmental & Occupational Health, Infectious Diseases

Full Text


Background: West Nile virus (WNV) is becoming a significant public health concern in Washington State (WA). Climate factors likely impact WNV case numbers in WA, but this has not been widely explored.

Methods: We examined the relationship between weekly averaged metrological conditions and reported WNV cases from 2006 to 2019 for two counties in Eastern WA (Yakima and Benton). We used univariate analyses to identify the time-lag (1-6 weeks) for each climate variable (maximum temperature, minimum temperature, vapor pressure deficit [VPD], specific humidity, and total precipitation) that was associated with the presence of a WNV case with the highest statistical significance. The five climate variables were subsequently included in two different predictive models: a multi-variate logistic regression model and a decision tree model, a type of machine learning method. The results of the models were then qualitatively and quantitively compared based on their accuracy, sensitivity, and specificity.

Results: The five variables most significantly associated with WNV in the univariate analysis were Maximum Temperature Lag Week 2, Minimum Temperature Lag Week 2, Specific Humidity Lag Week 2, Total Precipitation Lag Week 4, and Vapor Pressure Deficit Lag Week 3. In both the logistic regression and decision tree model, minimum temperature was the most important climate variable for determining the probability of a WNV case in Yakima & Benton county, and Vapor Pressure Deficit was identified as the second most important predicative variable. The two models had a similar accuracy, but the decision tree model was more sensitive than the logistic regression model.

Conclusion: The decision tree model and logistic regression model yielded similar results, but the decision tree model may be a better approach to predict WNV as it is more sensitive and may be more intuitive for public health practitioners. The findings of this study can be used to develop early warning systems or seasonal forecast systems that could improve predictions of WNV cases in endemic and nonendemic areas.