A reproducible, bias-aware informatics pipeline for hospital-associated infection surveillance in a resource-constrained tertiary hospital

Background. Hospital-associated infection (HAI) surveillance in resource-constrained hospitals relies on routinely-collected registries with unstable schemas, incomplete denominators and audit-based under-ascertainment. Objectives. To build a reproducible pipeline harmonising heterogeneous monthly exports, estimate hospital-wide and department-level HAI burden, and quantify audit-detected under-reporting. Methods. Retrospective analysis of 39 monthly surveillance files (February 2023–April 2026) from a Chinese tertiary hospital, reported per RECORD; negative-binomial trend, empirical-Bayes standardized infection ratios, and a prior-based under-ascertainment model. Results. The pipeline harmonised 1240 episodes with zero reconciliation discrepancies. Pooled incidence was 0.82 per 1000 patient-days with no temporal trend; four wards carried over half of episodes; carbapenem-resistant Acinetobacter baumannii dominated the resistant spectrum. Bias-corrected incidence was 0.833 per 1000 patient-days (+1.2%, conditional on the audit prior). Conclusions. Under transparent assumptions, a reproducible workflow converts routine surveillance exports into bias-aware burden estimates — a transferable, openly-released template for resource-constrained hospitals.