대한신장학회

논문분류	춘계학술대회 초록집
제목	Reimagining Drug-Related Problems in Hemodialysis Population Using Bayesian Networks
저자	Shan Lii Ching
출판정보	2026; 2026(1):
키워드	Bayesian Network, machine learning, drug related problem, hemodialysis, medication therapy management
초록	Objectives: Drug related problems (DRP) cause significant clinical and economic burden on healthcare systems globally. Conventional regression-based models are fundamentally limited in identifying linear associations, leaving them unable to capture the multi-dimensional, non-linear dependencies that drive DRP occurrence. Bayesian Network’s (BN) transparent probabilistic framework explicitly represents causal assumptions that overcomes this, by capturing both associations and causal structures within a single model. The study describes development of a preliminary BN-based predictive framework of DRP risk and characteristics within the hemodialysis population. Methods: BN development involved structural and parameter learning. Variables were determined through combination of literature, expert-elicited domain knowledge and longitudinal dataset from a previous retrospective study that analysed records of adult hemodialysis patients recruited into a pharmacist-led medication management program. DRP were classified using Pharmaceutical Care Network Europe Association DRP classification, v9.0. A three-membered panel of statisticians and DRP expert finalized the model, that was subsequently computed using Netica© software. Model performance was assessed through error rate, quadratic loss and spherical payoff measurements. Results: Model development involved hybrid method, with BN structure informed by manual “by-hand” elicited domain knowledge. Automated learning from dataset consisting 842 cases was used for BN parameters, where Conditional Probability Table was produced and optimized through Maximum Likelihood Estimation and Expectation-Maximisation algorithm. Nine variables were included that influenced the second layer (DRP event causes) and third layer (outcome of DRP characteristics). The model achieved moderate predictive accuracy level with error rate (32.9%), quadratic loss (0.5079) and spherical payoff (0.6717). The practical application and predictive capacity of the BN model are visualized in Figures 1 and 2, highlighting its utility as a clinical decision-support tool. Conclusion: BN demonstrates strong predictive performance with interpretability, enabling transparent DRP risk reasoning and formalizing DRP identification into a systematic, efficient and data-driven framework that supports informed decision-making and targeted interventions.
원문(PDF)	PDF 원문보기

간행물 검색

대한신장학회 회원관리 규정