Explainable AI for public health surveillance: investigating the persistent crisis of intentional injury mortality (suicide and homicide) in the Americas

Abstract

Intentional injury mortality (IIM), comprising homicide and suicide, remains a critical public health crisis in the Americas, which not only has the highest regional homicide rates globally but is also the only region where suicide rates continue to rise. This study employs explainable artificial intelligence (XAI) to examine the structural and temporal drivers of IIM across 25 countries, based on data from the previous two decades. Two complementary models were developed: a snapshot model based on contemporaneous socioeconomic indicators and a persistence-aware model incorporating lagged effects of predictors. Analyses were conducted across both income-level categories and geographic sub-regions to uncover context-specific patterns. While both models performed at acceptable levels in distinguishing immediate and enduring effects, persistence-aware models consistently outperformed snapshot models, thereby reframing IIM as a temporally sustained phenomenon. Feature importance, interpreted through SHapley Additive exPlanations (SHAP), highlighted the varying impacts of unemployment, inflation, corruption, and economic growth across income tiers and sub-regions. The results demonstrate that a combination of short-term shocks and the long-standing effects of governance and social factors drives IIM in the Americas. These findings underscore the need for dual-horizon policy approaches that address both immediate crises and structural root causes.