Title
Probabilistic patient risk profiling with pair-copula constructions
Abstract
I discuss classifiers using pair-copula constructions for patient risk profiling that incorporate mixed continuous and ordinal perioperative variables. The method fits class-specific joint distributions over the predictors and combines them using Bayes' rule. This gives a flexible generative score that captures nonlinear and asymmetric dependence. Through simulations and an elective bowel surgery case study, I show when such dependence modeling helps in assessing class balance and imbalance, and how posterior scores can be used to define clinically interpretable risk groups under balanced priors. These scores support risk stratification; however, interpreting absolute event probabilities requires an empirical prior or calibrated probabilities.
Bio
Özge Şahin is an assistant professor in the Applied Probability Group at TU Delft, specializing in statistical machine learning and dependence modeling with a strong focus on societal applications in finance and healthcare.
Speaker will present in person.
Practical
Integreat guest talks are open for attendance for everybody.
For those unable to attend in person: https://uio.zoom.us/j/65632027408
Point of contact
- Organiser and moderator: Professor Ingrid Hobæk Haff University of Oslo
- Integreat Administrative Leader Maria Dikova, University of Oslo