Answers are grounded in this paper and cite their sources.
Background Research in traumatic brain injury (TBI) mediation and outcome have developed many methods of assessing patient state and response. While several outcome calculators for TBI have been suggested, these calculators do not include all collected variables in the management of these patients; with a paucity of studies on how these hundreds of variables contribute statistically to long-term outcome. Thus, this work endeavors to assess the statistical relationships between commonly collected variables at admission, and during the first 24 h of critical care, with overall patient outcome. Methods From a retrospective dataset of 1,266 TBI patients requiring intensive care, unique patient variables (including demographic, injury pattern, pathophysiological, treatment and biochemical/imaging biomarkers) were explored. Intra-variable and variable reduction analysis were completed with importance assessed through SHapley Additive exPlanations (SHAP) method. From all of this we developed predictive models to assess the 6–12-month Glasgow Outcome Scale (GOS) using general linear and XGBoost models. Results Of the 321 variables assessed, a final 36 variables were found to be the important and non-redundant, that included common variables (Glasgow coma scale (GCS), age, heart rate and blood pressure at scene of accident) and individualized measures (S100B level and long pressure reactivity). Survival based on these non-redundant variables resulted in an accuracy of 88.8% 0.723 AUC (0.65—0.795) vs the IMPACT LAB 80.9% 0.608 AUC (0.54—0.677). Notable combining the IMPACT LAB model with the non-redundant variables found an accuracy of 90.3% 0.765 AUC (0.693—0.838) and an improvement of the Nagelkerke R2 value from 0.317 to 0.499. Conclusion These results found that overall general linear and XGBoost models behave similarly with an increase in non-IMPACT LAB variables performing better in XGBoost models. Broad categories of patient states like survival and favorable outcome can be determined by basic patient variables, however more dynamic measures of individualized response improve the predictive capacity for patient outcomes. Thus, to better understand patient state and more effectively monitor patient outcomes, dynamic and individualized patient physiology needs further evaluation.
No takes yet. Share an insight, caveat, or question.
Thelin et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: