DESIGN AND IMPLEMENTATION OF A KNOWLEDGE REPRESENTATION-BASED CLINICAL DECISION SUPPORT SYSTEM USING ANALOGICAL REASONING FOR INTELLIGENT HEALTHCARE MANAGEMENT: INSIGHTS FROM MIMIC-IV AND PHYSIONET DATASETS

Abstract

The use of Clinical Decision Support Systems (CDSS) is gaining popularity to improve patient outcomes in ICUs. In this paper, we describe the design, implementation, and evaluation of a hybrid knowledge representation-based clinical decision support system (CDSS) that combines rule-based and analogical reasoning. A total of 10,000 records were analyzed from the patient database of MIMIC-IV and PhysioNet. Out of all, 68% were male, and 32% were female patients. Furthermore, the average age was 62.0 years (SD = 15.1). The agents of CDSS required 150 ontology entities, 75 clinical rules, and 10,000 historical cases through 5 reasoning steps. Using both analogy and rules yielded 85% predictive accuracy, higher than each alone.  The study achieved precision, recall, and F1-score of 0.82, 0.83, and 0.825, respectively; interpretability reached 0.78, and weighted performance was at 0.81. An analysis of the feature importance showed that vital signs (heart rate, blood pressure, and respiratory rate) and lab results (hemoglobin, WBC, and creatinine) contributed over 65% in the model predictions. The case-based reasoning retrieved historical cases that had high similarity (similarity scores 0.91–0.97) and led to adaptive recommendations. All 10000 patients had a decision path that was traceable, making it interpretable. The results indicate that hybrid knowledge representation and analogical reasoning can remarkably enhance the predictive performance, contextual adaptability, and intelligibility of ICU decision support and offer a quantitative framework for real-world deployment