RUBIN: A Flexible, Bayesian Network-Based Clinical Decision Support System

Background: Bayesian Networks (BNs) are promising models for clinical decision support. Nevertheless, they have rarely been adopted in clinical practice since only a few practitioners have the expertise to work with them. Carefully designed human-computer interfaces can guide and empower clinicians to use BNs to their full potential. However, most BN-based clinical decision support systems (CDSS) and their interfaces are tailored to a specific use case, hindering reusability. CDSS that support diverse BNs often lack the adaptability to tailor their interfaces to specific decision support scenarios, which is crucial for adoption in clinical practice. Methods: We developed RUBIN, a CDSS that supports diverse BNs while still providing a highly adaptable interface for specific decision support tasks. We achieve this by separating the core CDSS from the necessary adaptations using external customization files. RUBIN is designed with a strong emphasis on usability and reusability, tailored to the specific demands of clinical environments. It enables clinicians to work effectively with BNs without requiring technical expertise. Results: We iteratively evaluated RUBIN for preoperative risk stratification in endometrial cancer therapy and designed customizations to the CDSS for this task in close collaboration with gynecological oncologists. The final evaluation demonstrated the system's high usability for healthcare practitioners. To illustrate that these benefits extend beyond a single domain, we conducted a second case study on cardiovascular disease prediction, highlighting the system’s generalizability. Conclusions: By separating core CDSS functionality from domain-specific adaptations, RUBIN supports BN-based decision support without requiring specific BN knowledge and remains highly adaptable to new medical tasks.