Comparison of Geographically Weighted Generalized Poisson Regression (GWGPR) and Geographically Weighted Negative Binomial Regression (GWNBR) Methods in Determining Factors Affecting Tuberculosis Cases in Indonesia

The findings of this study demonstrate that both the Geographically Weighted Generalized Poisson Regression (GWGPR) and Geographically Weighted Negative Binomial Regression (GWNBR) models are effective in modeling tuberculosis (TB) incidence data characterized by overdispersion and spatial heterogeneity. Although both models yield comparable fit statistics—as indicated by nearly identical Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC) values—GWGPR exhibits a higher sensitivity to regional variability, as evidenced by the formation of four distinct provincial clusters based on significant predictor variables, compared to only two clusters identified by the GWNBR model. This suggests that GWGPR may offer a more nuanced understanding of spatial effects in epidemiological data. Furthermore, several covariates; namely smoking prevalence, average annual humidity, number of rainy days, reported health complaints, and TB case detection and treatment coverage, emerged as consistently significant across all provinces in both modeling approaches. The recurrence of these variables across spatially disaggregated models highlights their fundamental role in influencing TB transmission dynamics at a national scale. Accordingly, the use of spatially adaptive models such as GWGPR can support more targeted and effective disease control strategies by aligning health policy responses with the localized determinants of TB burden.