Related Experiment Video
Updated: Oct 4, 2025
A Preclinical Controlled Cortical Impact Model for Traumatic Hemorrhage Contusion and Neuroinflammation
Published on: June 10, 2020
Predicting Mortality Following Traumatic Brain Injury or Subarachnoid Hemorrhage: An Analysis of the Validity of
Matt Rowe1, Jules Brown, Aidan Marsh
1Department of Anaesthesia and Intensive Care Medicine, Southmead Hospital Intensive Care Unit, Southmead Hospital, North Bristol NHS Trust, Bristol, UK.
Introduction:
Standardized mortality ratios (SMRs), calculated using the Acute Physiology, Age, Chronic Health Evaluation II (APACHE II) and Intensive Care National Audit and Research Centre H-2018 (ICNARC H-2018 ) risk prediction models, are widely used in UK intensive care units (ICUs) to measure and compare the quality of critical care delivery. Both models incorporate an assumption of Glasgow Coma Score (GCS) if an actual GCS without sedation is not recordable in the first 24 hours after ICU admission. This study assesses the validity of the APACHE II and ICNARC H-2018 models to predict mortality in ICU patients with traumatic brain injury (TBI) or aneurysmal subarachnoid hemorrhage (aSAH) in whom GCS is related to outcomes.
Methods:
In a retrospective analysis, the SMR calculated by the APACHE II and ICNARC H-2018 models for all UK ICU admissions in a 1-year period was compared with calculated SMRs in TBI/aSAH patients and at 3 GCS groups. Data for patients admitted to a single tertiary neurocritical care unit were similarly analyzed.
Results:
Both models predicted mortality well for the overall TBI/aSAH population; SMR (95% confidence interval) was 1.00 (0.96-1.04) and 0.99 (0.95-1.03) for the APACHE II and ICNARC H-2018 models, respectively. When analyzed by GCS grouping, both models underpredicted mortality in TBI/aSAH patients with GCS ≤8 (SMR, 1.1 [1.05-1.15]) and "unrecordable" GCS (SMR, 1.88 [1.77-1.99]). Similar findings were identified in the local data analysis.
Discussion:
The APACHE II and ICNARC H-2018 models predicted mortality well for the overall TBI/aSAH ICU population but underpredicted mortality when GCS was ≤8 or "unrecordable." This raises questions about the accuracy of these risk prediction models in TBI/aSAH patients and their use to evaluate treatments and compare outcomes between centers.
