Hybrid Rule-Based and Machine Learning Architecture for Reducing Alert Fatigue in Clinical Decision Support Systems

Clinical decision support systems operating in hospital environments generate large numbers of alerts based on continuously monitored clinical data. A major challenge in such systems is alert fatigue, caused by excessive numbers of notifications and unnecessary escalation of alert severity, which may reduce clinicians’ responsiveness to truly critical events. This paper presents a hybrid decision architecture designed for the assessment and prioritization of clinically significant events in inpatient monitoring environments. The proposed approach combines deterministic expert-defined rules with conditional machine learning models within a unified decision pipeline. Expert rules provide transparent parameter-level assessments based on clinically validated thresholds, while machine learning is applied only when rule-based aggregation yields ambiguous outcomes. This design preserves interpretability and safety while enabling the system to capture complex interactions between clinical parameters. The solution was evaluated under conditions approximating real clinical environments using laboratory measurements, vital signs, and patient-specific diagnostic contexts. Experimental results indicate that the hybrid approach improves the balance between sensitivity and reliability, reducing unnecessary alert escalation while maintaining detection of critical events. These findings demonstrate the potential of hybrid decision architectures for improving the quality of alert generation in clinical monitoring systems.