Bayesian disease mapping from incomplete hospital-based surveillance: accounting for catchment and missing case origins in dengue surveillance in Makassar, Indonesia

Dengue surveillance data are typically collected at healthcare facilities rather than at patients' residences, so reported case locations reflect healthcare-seeking behaviour as much as the underlying disease process. In practice, cases carry differing levels of location detail: some are geolocated, some are known only to the subdistrict, and some carry no residential information at all. This incompleteness complicates spatial risk estimation. We develop a hierarchical Bayesian point-process framework that maps dengue risk while jointly modelling hospital catchment behaviour and unobserved case origins. Subdistrict-level intensity is modelled on a per-person basis using a Besag--York--Molli\'e (BYM) structure with a population offset; a distance-based gravity (power-law) kernel represents hospital attraction; and cases with incomplete location are handled within a single likelihood that accommodates three levels of spatial resolution and marginalises over unknown origins. Applied to 1{,}842 dengue cases from eight hospitals in Makassar, Indonesia (2021--2025), the framework recovers held-out case origins substantially better than a conventional exponential catchment kernel and than deterministic allocation baselines. The full model identifies Manggala as the subdistrict with the highest estimated relative risk (2.7), with Tamalate, Rappocini and Tallo also meeting the specified high-risk criterion. Rankings by model-estimated relative risk differ from rankings by recorded case count. The framework offers a principled approach to disease mapping from incomplete, facility-based surveillance data.