Analysis of maternal mortality using geographically weighted Poisson regression with spline truncated estimator

Abstract

This study analyzes maternal mortality in South Sulawesi in 2020 and identifies the determining factors utilizing the Spline Truncated approach in the Geographically Weighted Poisson Regression method (GWPR-ST). This approach builds on nonparametric regression to manage spatial heterogeneity by calculating local parameters at each observation site. Maternal mortality data often follows a Poisson distribution and exhibits overdispersion, which violates the equidispersion assumption of standard Poisson regression. The GWPR-ST method effectively handles overdispersion while accounting for spatial variability. The model estimation uses 1, 2, and 3 knot points, with the Gaussian Kernel as the weighting function. The selection of optimum bandwidth is carried out with Generalized Cross Validation (GCV). The most suitable model is obtained with an order of m=1 and h=3 knot points, resulting in an R-squared value of 80.04. This shows that the GWPR-ST model accounts for 80.04% of the impact of predictor variables on the maternal mortality rate response variable. The influential predictor variables vary across locations, allowing them to be classified into four groups based on the significant predictor variables. These findings provide valuable insights for targeted public health interventions and demonstrate the effectiveness of GWPR-ST in modeling spatially heterogeneous count data with overdispersion.