Related Experiment Videos
Predicting risk-adjusted mortality for CABG surgery: logistic versus hierarchical logistic models
Edward L Hannan1, Chuntao Wu, Elizabeth R DeLong
1Department of Health Policy, Management, and Behavior, School of Public Health, University at Albany, State University of New York, Albany, New York 12144-3456, USA. elh03@health.state.ny.us
Medical Care
|June 23, 2005
Summary
Standard logistic regression and hierarchical models performed similarly in predicting hospital mortality rates for coronary artery bypass graft (CABG) surgery. This finding suggests simpler models may suffice for risk-adjusted outcomes in healthcare research.
Area of Science:
- Health Services Research
- Biostatistics
- Medical Informatics
Background:
- Hierarchical statistical models (multilevel or random-effects models) are recommended for nested data analysis in health research.
- These advanced models are computationally intensive and complex to implement.
- Limited literature compares standard logistic regression with hierarchical models for predicting provider performance.
Purpose of the Study:
- To compare the predictive accuracy of standard logistic regression versus hierarchical modeling.
- To assess performance in predicting risk-adjusted hospital mortality rates for coronary artery bypass graft (CABG) surgery in New York State.
Main Methods:
- Utilized New York State CABG Registry data from 1994-1999.
- Related statistical predictions from one year to hospital performance two years later.
- Compared predicted vs. observed mortality rates using root mean square errors, mean absolute difference, and confidence interval coverage.
Main Results:
- Standard logistic regression demonstrated performance comparable to hierarchical models.
- This similarity held true with and without the inclusion of a second-level covariate.
- Statistical testing revealed no significant differences between the models' predictive abilities.
Conclusions:
- Comparing predictive abilities of statistical models is crucial for assessing performance on specific datasets and applications.
- Standard logistic regression may be a viable alternative to more complex hierarchical models for certain healthcare prediction tasks.