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

Abstract

Background: Theoretical advantages of random-intercept multilevel (ML) logistic regression (LR) modeling over standard LR include separating variability due to patient-level and hospital-level predictors, shrinkage of estimates for lower-volume hospitals toward the overall mean, and fewer hospitals falsely identified as outliers.

Study design: We used Nationwide Inpatient Sample data from 2002 to 2004 to construct LR models of hospital mortality after admission with a principal ICD-9 Clinical Modification injury diagnosis (ICD-9 Clinical Modification codes 800 to 904, 910 to 929, 940 to 957, and 959). After considering various predictors, we used patient-level indicator variables for age groups, gender, maximum Abbreviated Injury Scale (AIS) for the head region (AIS score 3, 4, or 5), maximum AIS for other body regions (AIS score 3, 4, or 5), and mechanisms (eg, fall, gunshot, motor vehicle). Using standard LR and MLLR, we compared predictions based on 2002, 2003, and 2004 data with actual mortality observed in the same hospitals in the 2004, 2005, and 2006 Nationwide Inpatient Samples, respectively.

Results: Patient-level fixed effects were similar for the 2 methods in all years, with mortality associated most strongly with AIS = 5 head injury, other AIS = 5 injury, or higher age groups. ML models identified fewer hospitals as outliers. Differences between actual and predicted mortality were smaller with MLLR models compared with standard LR models.

Conclusions: Multilevel models might have advantages for the measurement and explanation of interhospital differences in trauma patient outcomes.