AI models show promising results in ICU mortality predictions

Predicting mortality risk in intensive care unit (ICU) patients is a crucial part of medical treatment, and recent research has found Artificial Intelligence could help clinicians make more efficient and accurate decisions.  

The study, led by academics from Australian Catholic University (ACU) and Charles Darwin University (CDU), tested the accuracy of machine learning (ML) algorithms to diagnose conditions and symptoms affecting the mortality of ICU patients. 

Clinicians use a variety of models to estimate ICU mortality, like the Acute Physiology and Chronic Health Evaluation (APACHE) or the Simplified Acute Physiology Score (SAPS), but these tools are limited in their capacity to capture evolving patient conditions, take time to validate, and need frequent recalibration. 

The effectiveness of ML predictions in ICU settings is well-documented, with algorithms outperform traditional scoring systems like APACHE and SAPS, but the unexplainable results of MLs hinders adoption. This research, however, applies further analysis to explain predictions.  

Two algorithms, extra trees (ET) and gradient boosting (GB), had accuracies of 98.33 per cent and 98.23 per cent respectively, in predicting the mortality of ICU patients and then deciding what conditions affected mortality.  

With its higher accuracy, ET’s results were also fed into explanation models to understand the key factors to mortality decisions, and how well the algorithms aligned with medical knowledge.  

The models found hypertension, tumours, endocrine disease, digestive disease and cardiovascular disease to be key factors in ET’s mortality predictions.  

Lead author and CDU Adjunct Professor Niusha Shafiabady, who is Head of Discipline for IT and the Director of Women in AI for Social Good lab at Australian Catholic University, said these methods could empower healthcare practitioners to better understand, trust and act on the ML model’s decisions. 

“These systems can assist clinicians in identifying high-risk patients who require urgent attention or targeted interventions,” Professor Shafiabady said.  

“Such systems enable continuous monitoring of at-risk patients, supporting proactive care and early intervention to prevent deterioration or adverse events.  

“By embedding interpretable findings into clinical decision-support systems, this study supports the advancement of ML tools that are both accurate and clinically meaningful, ensuring they complement rather than complicate frontline healthcare delivery.” 

Future avenues of research include using the models and algorithms with larger datasets and from varying healthcare settings.  

The study was conducted alongside researchers from Amirkabir University of Technology in Tehran, University of New England, University of Technology Sydney, and Western Sydney University.   

Explainable AI for mortality prediction: a comparative study using the MIMIC-III dataset was published in the journal BMJ Health & Care Informatics.