Modeling the drivers of mumps incidence in China: a spatial multi-method analysis

Introduction Mumps poses a considerable public health burden in China and exhibits significant spatial heterogeneity. Systematically exploring its multi-scale driving factors is crucial for informing targeted public health interventions. However, previous research often relied on global models or single-scale local models, limiting the precise identification of region-specific determinants. Methods Using provincial-level data from China in 2020 (the most recent complete year with publicly available multi-source data at the time of analysis), this study employed spatial autocorrelation analysis to detect clustering patterns of mumps incidence. A multi-model analytical framework was then constructed by integrating multiple linear regression (MLR), spatial lag model (SLM), geographically weighted regression (GWR), and multiscale geographically weighted regression (MGWR). While MLR, SLM, GWR, and MGWR are individually established methods, their integrated application within a single analytical framework - coupled with systematic model comparison and residual spatial diagnostics - represents a methodological advancement that enables explicit evaluation of the added value of capturing spatial heterogeneity and scale-dependent effects. This comprehensive approach enabled systematic elucidation of spatial differentiation patterns and scale-dependent driving mechanisms of mumps incidence, overcoming the limitations of traditional global models. The analysis incorporated multidimensional data encompassing socioeconomic factors, education level, healthcare resources, population structure, and environmental factors. Results Mumps incidence exhibited a west-high/east-low gradient distribution with significant spatial autocorrelation ( Moran’s I = 0.399, p < 0.001), characterized by high-high and low-low clustering. Model comparisons showed that while MGWR achieved the highest explanatory power ( R 2 = 0.769; AIC = 62.409) and eliminated residual spatial autocorrelation (Moran’s I = −0.0476, p = 0.90), its variable-specific bandwidths all exceeded 7,500 km - beyond China’s maximum inter-provincial distance - causing it to degenerate into a global model. In contrast, GWR ( R 2 = 0.738; AIC = 65.918) employed a unified optimal bandwidth of approximately 1,433 km, effectively capturing local spatial heterogeneity. GWR results revealed notable spatial heterogeneity in influencing factors: the negative effect of GDP per capita was strongest in the southwest; years of education showed a pronounced positive effect only in Fujian and Guangdong; the positive association with general practitioner density was most evident in the southeast; PM 2.5 exhibited a strong negative association in the west; and the child dependency ratio’s positive effect was most prominent in the northeast. Conclusion This study quantifies the spatially varying effects of key drivers on mumps incidence across China. Although MGWR offers theoretically advantageous, its practical utility is constrained by limited sample size, making GWR a more robust choice for capturing spatial heterogeneity in small-sample settings. The findings provide a nuanced, scientific basis for developing regionally differentiated prevention and control strategies tailored to local epidemiological profiles.