Explainable machine learning with SHAP decodes the heterogeneous burden of nasopharyngeal carcinoma in high-risk aging Asia

BACKGROUND: Nasopharyngeal carcinoma (NPC) is a global public health concern, with a particularly high burden concentrated in East Asia, Southeast Asia, and high-income Asia-Pacific and disproportionately affecting older adults. However, the diverging long-term epidemiological trajectories across these high-risk subregions remain unquantified. METHODS: Using GBD 2023 data, we analyzed NPC burden (incidence, mortality, prevalence, DALYs) among adults ≥ 55 years in these regions (1990-2023). We employed trend analysis, explainable machine learning (Random Forest with SHAP) and negative binomial regression to identify and quantify key modifiable risk factors, constructed a Composite Risk Index, and projected future trends to 2050. RESULTS: NPC burden showed sharply diverging trends: rising in high-income Asia-Pacific, declining in East Asia, and stable in Southeast Asia. Explainable machine learning identified a core cluster of modifiable risk factors (high intake of red/processed meat and sodium, smoking, high alcohol use) significantly associated with the burden. The geographic pattern of the Composite Risk Index closely matched the highest-burden regions. The disease showed strong male predominance and increased with age. Projections indicate these divergent trajectories will persist. CONCLUSION: The heterogeneous NPC burden in aging Asia is driven by a modifiable risk factor cluster and mediated by regional healthcare disparities. Findings necessitate precision public health strategies: enhanced primary prevention and early diagnosis in Southeast Asia, sustaining therapeutic gains in East Asia, and developing survivorship care in high-income Asia-Pacific.