Prediction Models for In-Hospital Delirium Using Routinely Collected Electronic Health Record Data: Systematic Review.
JMIR Med InformSep 16, 2026 (epub)
Hung-Min Huang, Chun-Shun Lu, Geng-Wei Chang et al.
✦ AI-curated · Sources linked
In 30 seconds
This systematic review evaluated prediction models for in-hospital delirium using electronic health record data, analyzing 29 studies. Most studies were retrospective cohorts, with 83% reporting internal discrimination (area under the curve 0.77-0.97). The findings indicate that while EHR data can aid in delirium risk prediction, no model is currently ready for routine clinical use.
Key findings
- 83% of studies reported internal discrimination with area under the curve ranging from 0.77 to 0.97.
- 69% of the studies were retrospective cohorts, primarily in general wards or intensive care units.
- 62% of models utilized machine learning or hybrid approaches, but complex models did not consistently outperform simpler methods.
- Only 3 studies reported decision curve analysis, highlighting limited prospective evaluation.
Why it matters
Understanding the predictive capabilities of EHR-based models for delirium can enhance early identification and prevention strategies in hospitalized patients, potentially improving patient outcomes.