Temporal Sepsis Modeling: a Fully Interpretable Relational Way
This work addresses the need for interpretable sepsis prediction models in healthcare, though it is incremental as it builds on existing relational and Bayesian methods.
The authors tackled the problem of predicting sepsis in intensive care patients by developing a fully interpretable relational modeling framework, which achieved competitive performance while providing four types of interpretability.
Sepsis remains one of the most complex and heterogeneous syndromes in intensive care, characterized by diverse physiological trajectories and variable responses to treatment. While deep learning models perform well in the early prediction of sepsis, they often lack interpretability and ignore latent patient sub-phenotypes. In this work, we propose a machine learning framework by opening up a new avenue for addressing this issue: a relational approach. Temporal data from electronic medical records (EMRs) are viewed as multivariate patient logs and represented in a relational data schema. Then, a propositionalisation technique (based on classic aggregation/selection functions from the field of relational data) is applied to construct interpretable features to "flatten" the data. Finally, the flattened data is classified using a selective naive Bayesian classifier. Experimental validation demonstrates the relevance of the suggested approach as well as its extreme interpretability. The interpretation is fourfold: univariate, global, local, and counterfactual.