Real-time sepsis risk stratification on cloud data warehouses: a fuzzy inference system on MIMIC-IV

Intensive care monitoring generates continuous streams of physiological measurements that demand real-time interpretation, yet traditional batch analytics cannot capture the temporal dynamics critical for early clinical intervention. We present an end-to-end, event-based framework that influences managed cloud infrastructure (Google BigQuery) and standard SQL to construct a unified, temporally aware clinical event from the MIMIC-IV v3.1 database. In addition to conventional threshold-based warnings, we implement a Mamdani fuzzy inference system that represents clinical reasoning using three input variables (qSOFA score, hours since suspected infection, and infection confidence) and thirteen linguistic rules. The fuzzy model produces well-calibrated risk scores (Brier = 0.090) on 38,888 ICU events from 17,783 hospital admissions, with 79.7% of rule-based alerts classified as high fuzzy risk and the remaining 20.3% assigned moderate risk based on temporal context. The system shows consistency with existing clinical guidelines while providing graded, interpretable outputs. This architecture proves that real-time, event-based analytics can be supported by managed cloud warehouses without specialized stream processors, and that fuzzy reasoning offers scalable, interpretable risk stratification that binary thresholds cannot achieve.