Predicting risk-adjusted mortality for trauma patients: logistic versus multilevel logistic models

Area of Science:

• Medical Statistics
• Health Services Research
• Trauma Surgery

Background:

• Standard logistic regression (LR) has limitations in analyzing patient and hospital-level factors in trauma outcomes.
• Multilevel (ML) logistic regression (LR) offers theoretical advantages, including separating variability and shrinking estimates for low-volume hospitals.

Purpose of the Study:

• To compare the performance of standard LR and ML logistic regression (MLLR) in predicting hospital mortality for trauma patients.
• To evaluate the ability of each model to explain interhospital differences in trauma outcomes.

Main Methods:

• Utilized Nationwide Inpatient Sample data (2002-2004) to build LR and MLLR models for hospital mortality.
• Included patient-level predictors: age groups, gender, Abbreviated Injury Scale (AIS) scores for head and other regions, and injury mechanisms.
• Compared model predictions using 2002-2004 data against actual mortality observed in 2004-2006.

Main Results:

• Patient-level fixed effects were consistent between standard LR and MLLR.
• Higher mortality was strongly associated with severe head injury (AIS=5), other severe injuries (AIS=5), and older age groups.
• MLLR models identified fewer outlier hospitals and showed smaller differences between predicted and actual mortality compared to standard LR.

Conclusions:

• Multilevel models demonstrate potential advantages for measuring and explaining variations in trauma patient outcomes across hospitals.
• MLLR may provide a more accurate and nuanced approach to analyzing hospital performance in trauma care.