Spatial Analysis of GOVID-19 Outbreak Using Spatial Regression

Introduction: The COVID-19 epidemic is considered a geographical phenomenon, and its impact on decision-making and daily life is significant. Geographic information systems (GIS) and spatial techniques play crucial roles in analyzing the spread of COVID-19 globally. Studies using spatial analysis have highlighted the importance of social and health variables in infection and mortality rates, despite existing uncertainties about the effects of meteorological variables. Given Iran's climatic diversity, it is valuable to identify the key spatial factors influencing COVID-19. Therefore, this study aims to model and determine the factors affecting the COVID-19 epidemic based on available data.
Materials and Methods: This study investigated the factors influencing the distribution of COVID-19 infection rates using global and local spatial regression methods. Seventy-three cities were selected, with data on COVID-19 infections available from March 10 to June 20, 2019. The factors considered were altitude, population density, average age, the ratio of the population over 55 years to the total population, and meteorological parameters, including humidity, temperature, pressure, and wind speed. Their relationships with the disease were analyzed using spatial statistics methods. Stepwise regression identified population density, air pressure, average age, and wind speed as significant predictors, and the occurrence of the disease was modeled using the Ordinary Least Squares (OLS) technique. Due to the unstable relationship between the independent and dependent variables, the Geographically Weighted Regression (GWR) technique was used. Principal Component Analysis (PCA) and SPSS software were employed to address spatial variability and multicollinearity.
Results and Discussion: The results showed that the OLS model was statistically significant, with variance values explained by the model being non-random. However, the explanatory variables had an inconsistent relationship with the dependent variable in both geographic and data spaces. The residual distribution deviated somewhat from normal, indicating model instability. Thus, the GWR technique was applied for modeling. PCA addressed multicollinearity (due to a cluster pattern in meteorological variables), reducing meteorological factors to one component, explaining nearly 70% of the variance. The model improved by consolidating the average age and the ratio of the population over 55 into one factor. Subsequently, population density, meteorological factors, and age demographics were utilized as predictive variables in the GWR model. A 10% increase in the adjusted R-squared of the GWR model (63%) demonstrated its relative improvement over the OLS model. Moran's spatial autocorrelation test indicated that, while the cluster pattern of residuals was less pronounced in the GWR model than in the OLS model, it remained significant at the 99% confidence level. Hot spot analysis at the 95% confidence level identified the western parts of Kurdistan province and the northern and western parts of Khuzestan province as hot spots (areas of significant underestimation), and the eastern parts of Hamadan province and the northern parts of Bushehr province as cold spots (areas of significant overestimation). Thus, at least one variable affecting disease occurrence was not considered. Other potential factors, such as cultural, health, and genetic variables, were not included due to unavailability or measurement difficulties.
Conclusion: This study highlights the importance of demographic and environmental factors in COVID-19 infection rates and provides a foundation for further research in the study area.