Student Presenter(s): Nikit Patel, Richard LaRocco, Scott Kivitz, Dana Schulz, Sonia Amanat
Faculty Mentor: Stephen Dirusso
Department: Clinical Sciences
School/College: College of Osteopathic Medicine, Long Island

Aim: Generate a prediction model from in-hospital data to assess mortality and discharge disposition for survivors in patients with traumatic brain injuries (TBIs).

Methods: Level II NYC Hospital Trauma Registry data (2016–2019) was used. Patients dead on arrival were excluded. For the rehab discharge disposition prediction model, patients who died prior to discharge were also excluded. Model variables included: demographics, co-morbidities, and physiologic data. Prediction models were generated using logistic regression (LR). Odds ratios (ORs) were calculated for significant variables (p<0.05). Discrimination (Area under the Receiver Operator Curve (AuROC)) and calibration (Hosmer-Lemeshow C-statistic (HL-C)) measured predictive capability.

Results: There were 934 patients with TBIs in the trauma registry. The final mortality prediction model included 843 TBI patients. The final rehab discharge disposition prediction model included 780 TBI patients. Significant predictors for mortality included ED GCS Score, supplemental oxygen, NISS, and insurance. Significant predictors for discharge disposition for survivors included ED GCS Score and NISS. AuROC is 0.955 and HL-C is 0.147 (p>0.05) for the mortality regression model. AuROC is 0.857 and HL-C is 0.541 (p>0.05) for rehab discharge disposition regression model.

Conclusions: We created good prediction models for mortality and rehab discharge disposition for TBI patients. These models can be utilized for trauma service process improvement.