Improving Hospital Mortality Predictions with Risk Trends
Author Information
Author(s): Jenna Wong, Monica Taljaard, Alan J. Forster, Carl van Walraven
Primary Institution: Ottawa Hospital Research Institute
Hypothesis
Does adding risk-trends to survival models improve in-hospital mortality predictions?
Conclusion
Incorporating trend indicators into the survival model did not significantly improve its predictive performance for hospital mortality.
Supporting Evidence
- Three trend indicators were significantly associated with time to hospital mortality.
- Adding trend indicators resulted in only small improvements in model discrimination and calibration.
- The existing model had good discrimination with a concordance probability of 0.879.
Takeaway
Doctors can use changes in a patient's health over time to help predict if they might die in the hospital, but just adding this information didn't make predictions much better.
Methodology
A cohort study including all adult inpatient hospitalizations from April 2004 to March 2009, using a time-dependent survival model to create trend indicators.
Limitations
The trend indicators did not significantly improve the predictive performance of the existing model, likely due to the already high performance of the original model.
Participant Demographics
Most patients were elderly (median age 61) with few comorbidities (median Elixhauser score of 0).
Statistical Information
P-Value
< .0001
Confidence Interval
95% CI 0.8718-0.8861
Statistical Significance
p<0.05
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website