Can We Actually Predict Long-Term Patient Satisfaction After Traumatic Brain Injury?

The Traumatic Brain Injury - Patient Reported Outcome (TBI-PRO) model was previously derived to predict long-term patient satisfaction as assessed by the Quality of Life After Brain Injury (QOLIBRI) score. The aim of this study is to externally and prospectively validate the TBI-PRO model to predict long-term patient-reported outcomes and to derive a new model using a larger dataset of older adults with TBI.

Patients admitted to a Level I trauma center with TBI were prospectively followed for 1 y after injury. Outcomes predicted by the TBI-PRO model based on admission findings were compared to actual QOLIBRI scores reported by patients at 3,6, and 12 mo. When deriving a new model, Collaborative European NeuroTrauma Effectiveness Research in TBI and the Transforming Research and Clinical Knowledge in Traumatic Brain Injury databases were used to identify older adults (≥50 y) with TBI from 2014 to 2018. Bayesian additive regression trees were used to identify predictive admission covariates. The coefficient of determination was used to identify the fitness of the model.

For prospective validation, a total of 140 patients were assessed at 3 mo, with follow-up from 69 patients at 6 mo and 13 patients at 12 mo postinjury. The area under receiver operating curve of the TBI-PRO model for predicting favorable outcomes at 3, 6, and 12 mo were 0.65, 0.57, and 0.62, respectively. When attempting to derive a novel predictive model, a total of 1521 patients (80%) was used in the derivation dataset while 384 (20%) were used in the validation dataset. A past medical history of heart conditions, initial hospital length of stay, admission systolic blood pressure, age, number of reactive pupils on admission, and the need for craniectomy were most predictive of long-term QOLIBRI-Overall Scale. The coefficient of determination for the validation model including only the most predictive variables were 0.28, 0.19, and 0.27 at 3, 6, and 12 mo, respectively.

In the present study, the prospective validation of a previously derived TBI-PRO model failed to accurately predict a long-term patient reported outcome measures in TBI. Additionally, the derivation of a novel model in older adults using a larger database showed poor accuracy in predicting long-term health-related quality of life. This study demonstrates limitations to current targeted approaches in TBI care. This study provides a framework for future studies and more targeted datasets looking to assess long-term quality of life based upon early hospital variables and can serve as a starting point for future predictive analysis.