A Hybrid Modeling Framework for Generalizable and Interpretable Predictions of ICU Mortality: Leveraging ICD Codes in a Multi-Hospital Study of Mechanically Ventilated Influenza Patients

Abstract

The development of reliable mortality risk stratification models is an active research area in computational healthcare. Mortality risk stratification provides a standard to assist physicians in evaluating a patient's condition or prognosis objectively. Particular interest lies in methods that are transparent to clinical interpretation and that retain predictive power once validated across diverse datasets they were not trained on. We've developed a hybrid model integrating mechanistic, clinical knowledge with mathematical and machine learning models to predict ICU mortality using ICD codes. A tree-structured network connecting independent modules that carry clinical meaning is implemented for interpretability. The trained model is then validated on external datasets from different hospitals, demonstrating successful generalization capabilities.

Noisecut: a python package for noise-tolerant classification of binary data using prior knowledge integration and max-cut solutions

“…Binary endpoints, which have two possible outcomes such as success/failure or present/absent, are commonly used in clinical trials to evaluate the effectiveness and safety of treatments [3]. Binary data also emerge in the context of the International Classification of Diseases (ICD) codes, which represent the presence of distinct medical diagnoses, conditions, and procedures [4]. Moreover, binary outcomes often result from longitudinal data analysis in clinical studies, in which each subject is monitored over a period of time [5,6].…”

“…Fig 4. Classifier accuracy on testing datasets comparison of NoiseCut with various ML models for classifying binary data across the entire spectrum of noise intensities, with a consistent 70% training data size.…”