VALID-AI: Validation and Impact of AI for Deterioration

Start Date

5 August 2025

End Date

30 October 2026

Project Participants

Beamtree, University of Sydney

Status: Completed

Tags: AI governance, artificial intelligence (AI), clinical decision support, clinical deterioration, data analytics, early warning systems, hospital, predictive health, quality and safety

Opportunity 

Recognising clinical deterioration early is essential for safe hospital care. Traditional early warning systems respond to changes that have already occurred. AI-driven clinical decision support tools like the Ainsoff Deterioration Index (ADI) have the potential to improve patient outcomes by identifying patients at risk earlier (i.e. predicting deterioration), but robust evaluations are lacking. This project aimed to evaluate the ADI’s performance and clinical impact, supporting evidence-based AI across Australian healthcare. 

Project Objectives  

This project set out to assess ADI’s performance in identifying patients at risk of acute deterioration, its impact on clinical decision-making, and quality and safety outcomes, through both a silent trial and a pilot trial in an Australian hospital. It also explored the effect of implementation on clinical workflows and decision making.

The project implemented ADI in a ward at a NSW hospital and linked complex hospital datasets to examine mortality, unexpected ICU admission and escalation-of-care events. Initial silent trial findings suggest ADI alerts show strong associations with several deterioration outcomes. The analysis also generated valuable practical insights into data quality, governance, alert configuration and implementation within a busy public-hospital environment. 

The VALID-AI project has expanded the evidence available to health services considering AI-enabled deterioration prediction. It demonstrates why clinical AI evaluation must extend beyond algorithm performance to include data quality, workflow fit, clinician experience, governance and operational capacity. The project’s evaluation approach and implementation lessons can inform future hospital-based AI studies and support more evidence-based decisions about adopting and scaling predictive clinical decision-support systems.  

The work has also helped build Australia’s capability, within academic-industry collaborations, to more rigorously evaluate digital health technologies such as healthcare AI in real-world care settings. 

Integrity, Excellence,
Teamwork and Authenticity

Hand holding a smartphone against a colourful, defocused background