ResistTrack-ID: An Integrated AI-Assisted Antimicrobial Stewardship and Multimodal Spatiotemporal Surveillance Framework for Indonesia

Background: Antimicrobial resistance (AMR) requires action at two levels that are often separated: safe antibiotic decisions for individual patients and timely recognition of changing resistance ecology at population level. ResistTrack-ID is designed as an AI-assisted clinical decision-support, adherence, stewardship and multimodal surveillance architecture for Indonesia. As artificial intelligence becomes more capable of shaping empirical antibiotic selection, surveillance is no longer only a retrospective reporting function. It becomes part of the model-safety infrastructure, because an AI system can repeatedly apply a resistance pattern that is locally outdated with far greater speed and consistency than an individual clinician. Objective: We propose ResistTrack-ID as an integrated framework that combines patient-level clinical decision support with longitudinal, catchment-based triangulation of clinical microbiology, environmental AMR signals, syndromic patterns and antibiotic-use data. Framework: The clinical core harmonizes electronic health record, diagnostic, microbiology, local antibiogram and treatment-guideline data and produces clinician-facing ranked antibiotic options with explicit uncertainty while retaining clinician authority. A patient-facing component supports treatment completion. The surveillance layer combines continuously collected clinical microbiology with monthly trajectory analysis, hospital and community wastewater from predefined sentinel sites, weekly syndromic signals and antibiotic-use metrics. Rather than flatten these streams into one score, a spatiotemporal engine preserves their different evidentiary meanings and temporal resolutions, evaluates baseline deviation, change points, spatial clusters and lagged relationships, and asks whether signals converge, diverge or precede one another. Alerts trigger intensified sampling, stewardship review, model reassessment or targeted genomic investigation rather than being interpreted as proof of transmission. Evaluation: ResistTrack-ID requires separate validation of clinical prediction, prescribing safety, stewardship outcomes, surveillance lead time, false-alert burden, geographic concordance, data completeness, model drift, usability and cost. Conclusion: ResistTrack-ID reframes AMR control as a connected learning loop between bedside decisions and community-level ecological surveillance. The purpose of surveillance in an AI-assisted system is not merely to draw a richer map of resistance, but to keep the information environment underlying antibiotic recommendations current, geographically grounded, uncertainty-aware and open to correction.