Predicting hospital outcomes in concussion and TBI: A mixed-effects analysis utilizing the nationwide readmissions database

Background and Objectives

Traumatic brain injury (TBI) is characterized by a wide range in severity. This variation presents a challenge for predicting outcomes and making management decisions, particularly for patients sustaining less severe injury. We present a novel statistical model for the prediction of hospital outcomes in two propensity-matched cohorts to optimize TBI patient management and counseling.

Methods

Hospitalized patients diagnosed with TBI were selected from the Nationwide Readmissions Database (NRD) from 2010 to 2019 using ICD-9 and ICD-10 codes. Using propensity score matching for baseline characteristics, patients were sorted by GCS score into two cohorts: 1188 patients with mild to moderate TBI (mTBI, GCS > 8) and 1219 patients with severe TBI (sTBI, GCS ≤ 8). Mixed-effects modeling was implemented, and model performance was evaluated using the Area Under the Curve (AUC). Any variance in ROC model prediction between cohorts was compared using DeLong’s test.

Results

After bivariate analysis, the mean length of stay (LOS), hospital cost, and mortality were significantly lower in the mTBI cohort relative to sTBI. GCS scores within the range of 9–15 were predictive of LOS (p < 0.01), with a trend towards significance in the prediction of non-routine discharge (p = 0.06).

Conclusion

Using an advanced mixed-effects model, our study found that GCS is an accurate predictor of hospital outcomes after a TBI diagnosis. These results provide insight that may aid in the development of preventative strategies, management decisions, and patient counseling to ensure a safe return to daily life for patients diagnosed with concussion.