Dual-Knowledge-Driven Interpretable Decision Support System for Stroke Critical-Care Rehabilitation: Design and Multi-Center Study

Abstract

The study aimed to develop a clinical decision support system for stroke critical-care rehabilitation in medical intensive care units (ICUs) to address the challenges of complex patient conditions, diverse rehabilitation needs, and subjectivity in physician prescriptions. This study proposed a dual-knowledge-driven intelligent rehabilitation decision support system (RDSS) for personalized intensive-care rehabilitation of severe-stroke patients. The system uses proposed collaborative reasoning that integrates semantic reasoning and knowledge graph-based reasoning to recommend rehabilitation plan, other relevant treatment options, potential complications and comorbidities, and risk factors to consider based on the patient’s clinical context information. In order to validate whether this system can enhance doctors’ capability in formulating rehabilitation prescriptions, a comparative evaluation study and a retrospective study were conducted respectively. The results of the comparative study showed that the overlap rate (73.3%) between the RDSS and the gold standard was higher than the average overlap rate (59.4%) between most (96.6%) rehabilitation professionals and the gold standard. In the retrospective study, 92 patients from two hospitals were enrolled, the overlap rate between the rehabilitation plans provided by the RDSS and those provided by the gold standard was over 50% for all cases, with nearly one-fifth achieving a 100% match. Moreover, 46.7% of rehabilitation plans had an overlap rate of 80% or higher, and the overall average overlap rate reached 78.10%. These findings suggest that the proposed RDSS demonstrated strong performance in recommending rehabilitation plans and providing risk reminders, indicating its potential feasibility for use in clinical practice.

Keywords

• Clinical decision support
• knowledge graph
• semantic reasoning
• stroke rehabilitation