The Effect of Weather and Population Factors on Dengue Fever Incidence in Saudi Arabia

Background: Dengue fever (DF) is the most important mosquito-transmitted viral disease, causing a large economic and disease burden in many parts of the world. However, most DF research focuses on Latin America and Asia, where burdens are highest, neglecting less impacted regions. Therefore, there is a critical need for studies in other regions where DF is an emerging and important public health problem but less well-characterized, such as the Middle East. The first documented case of DF in Saudi Arabia occurred in 1993. After a decade of sporadic outbreaks, the disease was declared endemic in 2004 and this designation persists. Important knowledge gaps relate to the role of climatic factors as drivers of DF in Saudi Arabia as well as the role of the annual Hajj pilgrimage. This study aims to create a predictive model to fill these gaps and ultimately improve health system preparedness through forecasting and inclusion of data in a system dynamics model of DF transmission.

Methods: This study used time series regression modeling to characterize relationships between incident DF cases, weather data, and pilgrimage data for three different cities in Saudi Arabia for the years 2009 to 2018. We obtained DF incidence data from the public health centers of the respective cities. We collected temperature, precipitation, and humidity re-analysis data from the Global Land Data Assimilation System (GLDAS). We also collected pilgrimage data from the Saudi General Authority for Statistics for the associated years. To create a predictive model we divided our DF data set into a seven year training data set and three year testing data set. We selected the variables to include in our regression models based on the strength of Pearson's correlation coefficients between the number of weekly DF cases and the weather and pilgrimage variables of interest taken at lag periods up to 12 weeks. To assess the accuracy of the model we ran our predictive model on the DF testing data set and compared the predicted number of cases to the true number of cases. Analyses were conducted in R using the fpp2 and forecast packages and the ggplot2 package for graphics.

Results: Over the study period, DF incidence demonstrated seasonal distribution, generally peaking between March and May. In the analyses looking at all three cities combined, the strongest associations between number of DF cases and weather variables were seen with minimum temperature at 12 week lag (r = -0.42, p-value < 2.2^e-16), maximum temperature an no lag (r = 0.18, p-value = 2.69^e-7), minimum humidity at no lag (r = -0.58, p-value < 2.2^e-16), average humidity at 12 week lag (r = 0.3, p-value < 2.2^e-16), and maximum weekly rainfall at 2 week lag (r = -0.17, p-value = 1.49^e-6). Pilgrimage period and overall number of pilgrims were negatively correlated with number of DF cases (r = -0.075, p-value = 0.036) and (r = -0.1, p-value = 0.003) respectively, while proportion of foreign pilgrims was positively correlated with number of DF cases (r = 0.17, p-value = 1.049^e-6). The regression model was able to accurately predict the DF incidence pattern over time but under-predicted the magnitude of disease incidence (number of cases).

Conclusion: This study demonstrates there is a moderately strong relationship between weather variables - specifically temperature, humidity, and to a lesser extent precipitation - and DF incidence in Saudi Arabia. This is also the first study to show the association between DF incidence and population influx during the Hajj pilgrimage. Findings from this effort can be utilized to develop a local DF forecasting model, further building on existing systems dynamics models developed for other geographical settings.