Temporal Sepsis Modeling: a Relational and Explainable-by-Design Framework

TL;DR

This paper introduces a relational, interpretable machine learning framework for early sepsis prediction using EMR data, emphasizing explainability and patient sub-phenotype identification.

Contribution

It presents a novel relational approach with a propositionalisation technique and naive Bayesian classifier for interpretable sepsis prediction from EMRs.

Findings

The framework achieves high interpretability in sepsis prediction.

Experimental results validate the relevance and effectiveness of the approach.

The method provides multiple levels of interpretability, including counterfactual explanations.

Abstract

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.