A1259
Title: Patient risk profiling with pair-copula constructions
Authors: Ozge Sahin - Delft University of Technology (Netherlands) [presenting]
Abstract: Patient-risk profiling is crucial for optimizing post-operative care and resource allocation, yet traditional binary metrics often fail to capture the full spectrum of patient risk. Pair-copula constructions (PCCs) are applied to model complex dependencies among mixed continuous-discrete clinical variables. Posterior probabilities of outcomes are estimated through discriminant analysis with PCCs and tailored selection and estimation methods. A data-driven framework is introduced to define patient risk groups based on these probabilities. Using a colorectal and small bowel surgery dataset, the method is evaluated against established clinical benchmarks, demonstrating its effectiveness in identifying low-risk patients and highlighting challenges in predicting high-risk cases. The value of PCCs in enhancing predictive accuracy and supporting clinical decisions is shown, such as safe patient discharge.
