SpatialCausal: a spatially-aware causal inference deep learning model for out-of-hospital cardiac arrest survival prediction

In health geography, understanding how risk factors and health outcomes interact across space is crucial for targeted interventions. Traditional machine learning often fails to address confounding bias, leading to inaccurate effect estimates. Deep learning-based causal inference models help mitigate this by learning balanced covariate representations, enabling more accurate causal effect estimation. However, incorporating spatial and unmeasured confounders into such models remains a major challenge. To address this, we propose SpatialCausal, a spatially-aware deep learning model that integrates spatial, non-spatial, and unmeasured confounders, achieving improved causal effect estimation and outperforming existing methods in predicting Out-of-Hospital Cardiac Arrest survival.