A randomized trial across 13 Ontario hospitals is testing an AI calculator that flags patients at high risk of developing delirium so care teams can prioritize prevention measures. Six sites were already using the platform by August 4, while St. Michael's Hospital planned to begin using it with patients in October as part of the year-long trial.
A randomized clinical trial across 13 Ontario hospitals is testing whether an AI risk calculator can help care teams prevent delirium before it develops. The rollout is an important real-world test of clinical prediction: the tool does not diagnose delirium or prescribe treatment, but identifies higher-risk patients for clinicians to consider for a prevention bundle.
From under-detection to targeted prevention
Delirium is an acute, fluctuating disorder that can cause confusion, disorientation and agitation. Canada's National Observer, in reporting by The Canadian Press on August 4, said routine hospital data identified delirium in just over 6% of admissions in a reviewed dataset, compared with 25% found through manual chart review. That gap matters because missed cases can delay prevention and care.
The calculator uses information already available in hospital records, including age, laboratory results and diagnoses, to estimate a patient's risk. A high-risk result prompts clinicians to consider a prevention bundle that can include support with meals and hydration, daily mobility, sleep-friendly lighting, relaxation and cognitive stimulation. The recommendation remains conditional on clinical judgment.
A 13-hospital deployment test
GEMINI's official project page says the AIM to Prevent Delirium Trial spans 13 Ontario hospitals and evaluates the predictor in live clinical settings. The August 4 report said six sites were already using the platform, while St. Michael's Hospital planned to begin using it with patients in October during the one-year trial. Sites range from large teaching hospitals to smaller community settings.
The trial is designed to test more than predictive performance. Researchers are studying how alerts fit into busy wards, who receives them, and whether hospitals can consistently deliver the prevention bundle. The team has also gathered social and demographic information to examine whether performance differs across patient groups.
What the evidence can and cannot show
Earlier GEMINI research established that routine clinical data can support automated delirium identification, and a June 2026 preprint reported temporally tested models using 3,862 labeled admissions from six Toronto hospitals. Those retrospective results support the technical premise, but they do not prove the live intervention prevents delirium.
That is the central question for the randomized trial. For data and clinical-AI teams, the outcome to watch is not only discrimination or calibration, but whether the full workflow produces safer, more consistent preventive care without creating unmanageable alert burden or uneven performance across hospitals and patient groups.
Key Points #
- 1A randomized trial is evaluating an AI delirium-risk calculator across 13 Ontario hospitals, with six sites already using the platform by August 4.
- 2The calculator supports clinician prioritization of a prevention bundle; it is not an autonomous diagnosis or treatment system.
- 3The trial must establish whether the model and the surrounding hospital workflow actually reduce delirium fairly across varied clinical settings.
Scoring Rationale #
A 13-hospital randomized deployment is a material clinical-AI implementation test with patient-safety implications, though its outcome evidence is not yet available.
Sources #
Primary source and supporting public references used for this report.
Practice with real Health & Insurance data
90 SQL & Python problems · 15 industry datasets
250 free problems · No credit card
See all Health & Insurance problems