Prediction model for intracranial hypertension demonstrates robust performance during external validation on the CENTER-
- PDF / 828,009 Bytes
- 3 Pages / 595.276 x 790.866 pts Page_size
- 12 Downloads / 194 Views
LETTER
Prediction model for intracranial hypertension demonstrates robust performance during external validation on the CENTER‑TBI dataset Giorgia Carra1 , Fabian Güiza1, Bart Depreitere2, and Geert Meyfroidt1* on behalf of the CENTER-TBI HighResolution ICU (HR ICU) Sub-Study Participants and Investigators © 2020 Springer-Verlag GmbH Germany, part of Springer Nature
Dear Editor, In patients with traumatic brain injury (TBI), the management of elevated intracranial pressure (ICP) is mainly reactive, and the most recent Seattle International Severe Traumatic Brain Injury Consensus Conference (SIBICC) guidelines recommend aggressive treatment when ICP rises above 22 mmHg [1, 2]. To drive TBI care towards proactiveness, Güiza et al. [3] developed a machine learning model for the prediction of extremely elevated ICP. Providing a 30 min forewarning, the model predicted with good discrimination and good calibration, events of ICP above 30 mmHg that lasted more than 10 min. The model, which only requires continuous ICP and mean arterial blood pressure signals, was developed on data prior to 2005. Subsequently, the model was externally validated on a multicenter adult cohort collected between 2009 and 2011, with unchanged performance [4]. Since changes in clinical practice over the past decade could have resulted in progressive degradation of model accuracy, in this study, we evaluated the performance of the model on the recent multicenter Collaborative
*Correspondence: [email protected] 1 Department of Cellular and Molecular Medicine, Clinical Division and Laboratory of Intensive Care Medicine, KU Leuven, Leuven, Belgium Full author information is available at the end of the article
The CENTER-TBI High-Resolution ICU (HR-ICU) Sub-Study Participants and Investigators are listed in the Acknowledgement section.
European NeuroTrauma Effectiveness Research in Traumatic Brain Injury (CENTER-TBI) dataset [5]. The validation dataset included 257 patients with TBI, who were recruited prospectively between 2015 and 2017 as part of the CENTER-TBI high-resolution ICU monitoring cohort [5]. See Online Resource 1 for cohort demographics. Model performance was quantified with the metrics used in the previous publications [3, 4]: Area Under the Receiver Operating Curve (AUC), accuracy, sensitivity, specificity, and calibration analysis, while the clinical utility of the model was assessed through decision curve analysis. In this external validation dataset, which was collected almost 10 years later than the original development cohort, the model was still able to predict future episodes of extremely elevated ICP with good discrimination and calibration (AUC = 0.93, calibration slope 1.22, and calibration–in–the–large = − 0.04), see Fig. 1. At the same cutoff of the original study, the model presented an accuracy of 88%, sensitivity of 83%, and specificity of 91%. These results demonstrate that the prediction model developed by Güiza et al. [3] is extremely robust to inter-center variability and, most importantly, to
Data Loading...
